2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
33 #include <linux/time.h>
34 #include <linux/jbd2.h>
35 #include <linux/highuid.h>
36 #include <linux/pagemap.h>
37 #include <linux/quotaops.h>
38 #include <linux/string.h>
39 #include <linux/slab.h>
40 #include <asm/uaccess.h>
41 #include <linux/fiemap.h>
42 #include "ext4_jbd2.h"
43 #include "ext4_extents.h"
46 #include <trace/events/ext4.h>
49 * used by extent splitting.
51 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
53 #define EXT4_EXT_MARK_UNINIT1 0x2 /* mark first half uninitialized */
54 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
56 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
57 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
59 static __le32
ext4_extent_block_csum(struct inode
*inode
,
60 struct ext4_extent_header
*eh
)
62 struct ext4_inode_info
*ei
= EXT4_I(inode
);
63 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
66 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)eh
,
67 EXT4_EXTENT_TAIL_OFFSET(eh
));
68 return cpu_to_le32(csum
);
71 static int ext4_extent_block_csum_verify(struct inode
*inode
,
72 struct ext4_extent_header
*eh
)
74 struct ext4_extent_tail
*et
;
76 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
77 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
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_RO_COMPAT_FEATURE(inode
->i_sb
,
92 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
95 et
= find_ext4_extent_tail(eh
);
96 et
->et_checksum
= ext4_extent_block_csum(inode
, eh
);
99 static int ext4_split_extent(handle_t
*handle
,
101 struct ext4_ext_path
*path
,
102 struct ext4_map_blocks
*map
,
106 static int ext4_split_extent_at(handle_t
*handle
,
108 struct ext4_ext_path
*path
,
113 static int ext4_find_delayed_extent(struct inode
*inode
,
114 struct extent_status
*newes
);
116 static int ext4_ext_truncate_extend_restart(handle_t
*handle
,
122 if (!ext4_handle_valid(handle
))
124 if (handle
->h_buffer_credits
> needed
)
126 err
= ext4_journal_extend(handle
, needed
);
129 err
= ext4_truncate_restart_trans(handle
, inode
, needed
);
141 static int ext4_ext_get_access(handle_t
*handle
, struct inode
*inode
,
142 struct ext4_ext_path
*path
)
145 /* path points to block */
146 return ext4_journal_get_write_access(handle
, path
->p_bh
);
148 /* path points to leaf/index in inode body */
149 /* we use in-core data, no need to protect them */
159 int __ext4_ext_dirty(const char *where
, unsigned int line
, handle_t
*handle
,
160 struct inode
*inode
, struct ext4_ext_path
*path
)
164 ext4_extent_block_csum_set(inode
, ext_block_hdr(path
->p_bh
));
165 /* path points to block */
166 err
= __ext4_handle_dirty_metadata(where
, line
, handle
,
169 /* path points to leaf/index in inode body */
170 err
= ext4_mark_inode_dirty(handle
, inode
);
175 static ext4_fsblk_t
ext4_ext_find_goal(struct inode
*inode
,
176 struct ext4_ext_path
*path
,
180 int depth
= path
->p_depth
;
181 struct ext4_extent
*ex
;
184 * Try to predict block placement assuming that we are
185 * filling in a file which will eventually be
186 * non-sparse --- i.e., in the case of libbfd writing
187 * an ELF object sections out-of-order but in a way
188 * the eventually results in a contiguous object or
189 * executable file, or some database extending a table
190 * space file. However, this is actually somewhat
191 * non-ideal if we are writing a sparse file such as
192 * qemu or KVM writing a raw image file that is going
193 * to stay fairly sparse, since it will end up
194 * fragmenting the file system's free space. Maybe we
195 * should have some hueristics or some way to allow
196 * userspace to pass a hint to file system,
197 * especially if the latter case turns out to be
200 ex
= path
[depth
].p_ext
;
202 ext4_fsblk_t ext_pblk
= ext4_ext_pblock(ex
);
203 ext4_lblk_t ext_block
= le32_to_cpu(ex
->ee_block
);
205 if (block
> ext_block
)
206 return ext_pblk
+ (block
- ext_block
);
208 return ext_pblk
- (ext_block
- block
);
211 /* it looks like index is empty;
212 * try to find starting block from index itself */
213 if (path
[depth
].p_bh
)
214 return path
[depth
].p_bh
->b_blocknr
;
217 /* OK. use inode's group */
218 return ext4_inode_to_goal_block(inode
);
222 * Allocation for a meta data block
225 ext4_ext_new_meta_block(handle_t
*handle
, struct inode
*inode
,
226 struct ext4_ext_path
*path
,
227 struct ext4_extent
*ex
, int *err
, unsigned int flags
)
229 ext4_fsblk_t goal
, newblock
;
231 goal
= ext4_ext_find_goal(inode
, path
, le32_to_cpu(ex
->ee_block
));
232 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, flags
,
237 static inline int ext4_ext_space_block(struct inode
*inode
, int check
)
241 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
242 / sizeof(struct ext4_extent
);
243 #ifdef AGGRESSIVE_TEST
244 if (!check
&& size
> 6)
250 static inline int ext4_ext_space_block_idx(struct inode
*inode
, int check
)
254 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
255 / sizeof(struct ext4_extent_idx
);
256 #ifdef AGGRESSIVE_TEST
257 if (!check
&& size
> 5)
263 static inline int ext4_ext_space_root(struct inode
*inode
, int check
)
267 size
= sizeof(EXT4_I(inode
)->i_data
);
268 size
-= sizeof(struct ext4_extent_header
);
269 size
/= sizeof(struct ext4_extent
);
270 #ifdef AGGRESSIVE_TEST
271 if (!check
&& size
> 3)
277 static inline int ext4_ext_space_root_idx(struct inode
*inode
, int check
)
281 size
= sizeof(EXT4_I(inode
)->i_data
);
282 size
-= sizeof(struct ext4_extent_header
);
283 size
/= sizeof(struct ext4_extent_idx
);
284 #ifdef AGGRESSIVE_TEST
285 if (!check
&& size
> 4)
292 * Calculate the number of metadata blocks needed
293 * to allocate @blocks
294 * Worse case is one block per extent
296 int ext4_ext_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
298 struct ext4_inode_info
*ei
= EXT4_I(inode
);
301 idxs
= ((inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
302 / sizeof(struct ext4_extent_idx
));
305 * If the new delayed allocation block is contiguous with the
306 * previous da block, it can share index blocks with the
307 * previous block, so we only need to allocate a new index
308 * block every idxs leaf blocks. At ldxs**2 blocks, we need
309 * an additional index block, and at ldxs**3 blocks, yet
310 * another index blocks.
312 if (ei
->i_da_metadata_calc_len
&&
313 ei
->i_da_metadata_calc_last_lblock
+1 == lblock
) {
316 if ((ei
->i_da_metadata_calc_len
% idxs
) == 0)
318 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
)) == 0)
320 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
*idxs
)) == 0) {
322 ei
->i_da_metadata_calc_len
= 0;
324 ei
->i_da_metadata_calc_len
++;
325 ei
->i_da_metadata_calc_last_lblock
++;
330 * In the worst case we need a new set of index blocks at
331 * every level of the inode's extent tree.
333 ei
->i_da_metadata_calc_len
= 1;
334 ei
->i_da_metadata_calc_last_lblock
= lblock
;
335 return ext_depth(inode
) + 1;
339 ext4_ext_max_entries(struct inode
*inode
, int depth
)
343 if (depth
== ext_depth(inode
)) {
345 max
= ext4_ext_space_root(inode
, 1);
347 max
= ext4_ext_space_root_idx(inode
, 1);
350 max
= ext4_ext_space_block(inode
, 1);
352 max
= ext4_ext_space_block_idx(inode
, 1);
358 static int ext4_valid_extent(struct inode
*inode
, struct ext4_extent
*ext
)
360 ext4_fsblk_t block
= ext4_ext_pblock(ext
);
361 int len
= ext4_ext_get_actual_len(ext
);
362 ext4_lblk_t lblock
= le32_to_cpu(ext
->ee_block
);
363 ext4_lblk_t last
= lblock
+ len
- 1;
367 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
370 static int ext4_valid_extent_idx(struct inode
*inode
,
371 struct ext4_extent_idx
*ext_idx
)
373 ext4_fsblk_t block
= ext4_idx_pblock(ext_idx
);
375 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
378 static int ext4_valid_extent_entries(struct inode
*inode
,
379 struct ext4_extent_header
*eh
,
382 unsigned short entries
;
383 if (eh
->eh_entries
== 0)
386 entries
= le16_to_cpu(eh
->eh_entries
);
390 struct ext4_extent
*ext
= EXT_FIRST_EXTENT(eh
);
391 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
392 ext4_fsblk_t pblock
= 0;
393 ext4_lblk_t lblock
= 0;
394 ext4_lblk_t prev
= 0;
397 if (!ext4_valid_extent(inode
, ext
))
400 /* Check for overlapping extents */
401 lblock
= le32_to_cpu(ext
->ee_block
);
402 len
= ext4_ext_get_actual_len(ext
);
403 if ((lblock
<= prev
) && prev
) {
404 pblock
= ext4_ext_pblock(ext
);
405 es
->s_last_error_block
= cpu_to_le64(pblock
);
410 prev
= lblock
+ len
- 1;
413 struct ext4_extent_idx
*ext_idx
= EXT_FIRST_INDEX(eh
);
415 if (!ext4_valid_extent_idx(inode
, ext_idx
))
424 static int __ext4_ext_check(const char *function
, unsigned int line
,
425 struct inode
*inode
, struct ext4_extent_header
*eh
,
426 int depth
, ext4_fsblk_t pblk
)
428 const char *error_msg
;
431 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
432 error_msg
= "invalid magic";
435 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
436 error_msg
= "unexpected eh_depth";
439 if (unlikely(eh
->eh_max
== 0)) {
440 error_msg
= "invalid eh_max";
443 max
= ext4_ext_max_entries(inode
, depth
);
444 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
445 error_msg
= "too large eh_max";
448 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
449 error_msg
= "invalid eh_entries";
452 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
453 error_msg
= "invalid extent entries";
456 /* Verify checksum on non-root extent tree nodes */
457 if (ext_depth(inode
) != depth
&&
458 !ext4_extent_block_csum_verify(inode
, eh
)) {
459 error_msg
= "extent tree corrupted";
465 ext4_error_inode(inode
, function
, line
, 0,
466 "pblk %llu bad header/extent: %s - magic %x, "
467 "entries %u, max %u(%u), depth %u(%u)",
468 (unsigned long long) pblk
, error_msg
,
469 le16_to_cpu(eh
->eh_magic
),
470 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
471 max
, le16_to_cpu(eh
->eh_depth
), depth
);
475 #define ext4_ext_check(inode, eh, depth, pblk) \
476 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
478 int ext4_ext_check_inode(struct inode
*inode
)
480 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
), 0);
483 static struct buffer_head
*
484 __read_extent_tree_block(const char *function
, unsigned int line
,
485 struct inode
*inode
, ext4_fsblk_t pblk
, int depth
,
488 struct buffer_head
*bh
;
491 bh
= sb_getblk(inode
->i_sb
, pblk
);
493 return ERR_PTR(-ENOMEM
);
495 if (!bh_uptodate_or_lock(bh
)) {
496 trace_ext4_ext_load_extent(inode
, pblk
, _RET_IP_
);
497 err
= bh_submit_read(bh
);
501 if (buffer_verified(bh
) && !(flags
& EXT4_EX_FORCE_CACHE
))
503 err
= __ext4_ext_check(function
, line
, inode
,
504 ext_block_hdr(bh
), depth
, pblk
);
507 set_buffer_verified(bh
);
509 * If this is a leaf block, cache all of its entries
511 if (!(flags
& EXT4_EX_NOCACHE
) && depth
== 0) {
512 struct ext4_extent_header
*eh
= ext_block_hdr(bh
);
513 struct ext4_extent
*ex
= EXT_FIRST_EXTENT(eh
);
514 ext4_lblk_t prev
= 0;
517 for (i
= le16_to_cpu(eh
->eh_entries
); i
> 0; i
--, ex
++) {
518 unsigned int status
= EXTENT_STATUS_WRITTEN
;
519 ext4_lblk_t lblk
= le32_to_cpu(ex
->ee_block
);
520 int len
= ext4_ext_get_actual_len(ex
);
522 if (prev
&& (prev
!= lblk
))
523 ext4_es_cache_extent(inode
, prev
,
527 if (ext4_ext_is_uninitialized(ex
))
528 status
= EXTENT_STATUS_UNWRITTEN
;
529 ext4_es_cache_extent(inode
, lblk
, len
,
530 ext4_ext_pblock(ex
), status
);
541 #define read_extent_tree_block(inode, pblk, depth, flags) \
542 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
546 * This function is called to cache a file's extent information in the
549 int ext4_ext_precache(struct inode
*inode
)
551 struct ext4_inode_info
*ei
= EXT4_I(inode
);
552 struct ext4_ext_path
*path
= NULL
;
553 struct buffer_head
*bh
;
554 int i
= 0, depth
, ret
= 0;
556 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
557 return 0; /* not an extent-mapped inode */
559 down_read(&ei
->i_data_sem
);
560 depth
= ext_depth(inode
);
562 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
565 up_read(&ei
->i_data_sem
);
569 /* Don't cache anything if there are no external extent blocks */
572 path
[0].p_hdr
= ext_inode_hdr(inode
);
573 ret
= ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0);
576 path
[0].p_idx
= EXT_FIRST_INDEX(path
[0].p_hdr
);
579 * If this is a leaf block or we've reached the end of
580 * the index block, go up
583 path
[i
].p_idx
> EXT_LAST_INDEX(path
[i
].p_hdr
)) {
584 brelse(path
[i
].p_bh
);
589 bh
= read_extent_tree_block(inode
,
590 ext4_idx_pblock(path
[i
].p_idx
++),
592 EXT4_EX_FORCE_CACHE
);
599 path
[i
].p_hdr
= ext_block_hdr(bh
);
600 path
[i
].p_idx
= EXT_FIRST_INDEX(path
[i
].p_hdr
);
602 ext4_set_inode_state(inode
, EXT4_STATE_EXT_PRECACHED
);
604 up_read(&ei
->i_data_sem
);
605 ext4_ext_drop_refs(path
);
611 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
613 int k
, l
= path
->p_depth
;
616 for (k
= 0; k
<= l
; k
++, path
++) {
618 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
619 ext4_idx_pblock(path
->p_idx
));
620 } else if (path
->p_ext
) {
621 ext_debug(" %d:[%d]%d:%llu ",
622 le32_to_cpu(path
->p_ext
->ee_block
),
623 ext4_ext_is_uninitialized(path
->p_ext
),
624 ext4_ext_get_actual_len(path
->p_ext
),
625 ext4_ext_pblock(path
->p_ext
));
632 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
634 int depth
= ext_depth(inode
);
635 struct ext4_extent_header
*eh
;
636 struct ext4_extent
*ex
;
642 eh
= path
[depth
].p_hdr
;
643 ex
= EXT_FIRST_EXTENT(eh
);
645 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
647 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
648 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
649 ext4_ext_is_uninitialized(ex
),
650 ext4_ext_get_actual_len(ex
), ext4_ext_pblock(ex
));
655 static void ext4_ext_show_move(struct inode
*inode
, struct ext4_ext_path
*path
,
656 ext4_fsblk_t newblock
, int level
)
658 int depth
= ext_depth(inode
);
659 struct ext4_extent
*ex
;
661 if (depth
!= level
) {
662 struct ext4_extent_idx
*idx
;
663 idx
= path
[level
].p_idx
;
664 while (idx
<= EXT_MAX_INDEX(path
[level
].p_hdr
)) {
665 ext_debug("%d: move %d:%llu in new index %llu\n", level
,
666 le32_to_cpu(idx
->ei_block
),
667 ext4_idx_pblock(idx
),
675 ex
= path
[depth
].p_ext
;
676 while (ex
<= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
677 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
678 le32_to_cpu(ex
->ee_block
),
680 ext4_ext_is_uninitialized(ex
),
681 ext4_ext_get_actual_len(ex
),
688 #define ext4_ext_show_path(inode, path)
689 #define ext4_ext_show_leaf(inode, path)
690 #define ext4_ext_show_move(inode, path, newblock, level)
693 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
695 int depth
= path
->p_depth
;
698 for (i
= 0; i
<= depth
; i
++, path
++)
706 * ext4_ext_binsearch_idx:
707 * binary search for the closest index of the given block
708 * the header must be checked before calling this
711 ext4_ext_binsearch_idx(struct inode
*inode
,
712 struct ext4_ext_path
*path
, ext4_lblk_t block
)
714 struct ext4_extent_header
*eh
= path
->p_hdr
;
715 struct ext4_extent_idx
*r
, *l
, *m
;
718 ext_debug("binsearch for %u(idx): ", block
);
720 l
= EXT_FIRST_INDEX(eh
) + 1;
721 r
= EXT_LAST_INDEX(eh
);
724 if (block
< le32_to_cpu(m
->ei_block
))
728 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
729 m
, le32_to_cpu(m
->ei_block
),
730 r
, le32_to_cpu(r
->ei_block
));
734 ext_debug(" -> %u->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
735 ext4_idx_pblock(path
->p_idx
));
737 #ifdef CHECK_BINSEARCH
739 struct ext4_extent_idx
*chix
, *ix
;
742 chix
= ix
= EXT_FIRST_INDEX(eh
);
743 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
745 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
746 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
748 ix
, EXT_FIRST_INDEX(eh
));
749 printk(KERN_DEBUG
"%u <= %u\n",
750 le32_to_cpu(ix
->ei_block
),
751 le32_to_cpu(ix
[-1].ei_block
));
753 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
754 <= le32_to_cpu(ix
[-1].ei_block
));
755 if (block
< le32_to_cpu(ix
->ei_block
))
759 BUG_ON(chix
!= path
->p_idx
);
766 * ext4_ext_binsearch:
767 * binary search for closest extent of the given block
768 * the header must be checked before calling this
771 ext4_ext_binsearch(struct inode
*inode
,
772 struct ext4_ext_path
*path
, ext4_lblk_t block
)
774 struct ext4_extent_header
*eh
= path
->p_hdr
;
775 struct ext4_extent
*r
, *l
, *m
;
777 if (eh
->eh_entries
== 0) {
779 * this leaf is empty:
780 * we get such a leaf in split/add case
785 ext_debug("binsearch for %u: ", block
);
787 l
= EXT_FIRST_EXTENT(eh
) + 1;
788 r
= EXT_LAST_EXTENT(eh
);
792 if (block
< le32_to_cpu(m
->ee_block
))
796 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
797 m
, le32_to_cpu(m
->ee_block
),
798 r
, le32_to_cpu(r
->ee_block
));
802 ext_debug(" -> %d:%llu:[%d]%d ",
803 le32_to_cpu(path
->p_ext
->ee_block
),
804 ext4_ext_pblock(path
->p_ext
),
805 ext4_ext_is_uninitialized(path
->p_ext
),
806 ext4_ext_get_actual_len(path
->p_ext
));
808 #ifdef CHECK_BINSEARCH
810 struct ext4_extent
*chex
, *ex
;
813 chex
= ex
= EXT_FIRST_EXTENT(eh
);
814 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
815 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
816 <= le32_to_cpu(ex
[-1].ee_block
));
817 if (block
< le32_to_cpu(ex
->ee_block
))
821 BUG_ON(chex
!= path
->p_ext
);
827 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
829 struct ext4_extent_header
*eh
;
831 eh
= ext_inode_hdr(inode
);
834 eh
->eh_magic
= EXT4_EXT_MAGIC
;
835 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
836 ext4_mark_inode_dirty(handle
, inode
);
840 struct ext4_ext_path
*
841 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
842 struct ext4_ext_path
*path
, int flags
)
844 struct ext4_extent_header
*eh
;
845 struct buffer_head
*bh
;
846 short int depth
, i
, ppos
= 0, alloc
= 0;
849 eh
= ext_inode_hdr(inode
);
850 depth
= ext_depth(inode
);
852 /* account possible depth increase */
854 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
857 return ERR_PTR(-ENOMEM
);
864 /* walk through the tree */
866 ext_debug("depth %d: num %d, max %d\n",
867 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
869 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
870 path
[ppos
].p_block
= ext4_idx_pblock(path
[ppos
].p_idx
);
871 path
[ppos
].p_depth
= i
;
872 path
[ppos
].p_ext
= NULL
;
874 bh
= read_extent_tree_block(inode
, path
[ppos
].p_block
, --i
,
881 eh
= ext_block_hdr(bh
);
883 if (unlikely(ppos
> depth
)) {
885 EXT4_ERROR_INODE(inode
,
886 "ppos %d > depth %d", ppos
, depth
);
890 path
[ppos
].p_bh
= bh
;
891 path
[ppos
].p_hdr
= eh
;
894 path
[ppos
].p_depth
= i
;
895 path
[ppos
].p_ext
= NULL
;
896 path
[ppos
].p_idx
= NULL
;
899 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
900 /* if not an empty leaf */
901 if (path
[ppos
].p_ext
)
902 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
904 ext4_ext_show_path(inode
, path
);
909 ext4_ext_drop_refs(path
);
916 * ext4_ext_insert_index:
917 * insert new index [@logical;@ptr] into the block at @curp;
918 * check where to insert: before @curp or after @curp
920 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
921 struct ext4_ext_path
*curp
,
922 int logical
, ext4_fsblk_t ptr
)
924 struct ext4_extent_idx
*ix
;
927 err
= ext4_ext_get_access(handle
, inode
, curp
);
931 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
932 EXT4_ERROR_INODE(inode
,
933 "logical %d == ei_block %d!",
934 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
938 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
939 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
940 EXT4_ERROR_INODE(inode
,
941 "eh_entries %d >= eh_max %d!",
942 le16_to_cpu(curp
->p_hdr
->eh_entries
),
943 le16_to_cpu(curp
->p_hdr
->eh_max
));
947 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
949 ext_debug("insert new index %d after: %llu\n", logical
, ptr
);
950 ix
= curp
->p_idx
+ 1;
953 ext_debug("insert new index %d before: %llu\n", logical
, ptr
);
957 len
= EXT_LAST_INDEX(curp
->p_hdr
) - ix
+ 1;
960 ext_debug("insert new index %d: "
961 "move %d indices from 0x%p to 0x%p\n",
962 logical
, len
, ix
, ix
+ 1);
963 memmove(ix
+ 1, ix
, len
* sizeof(struct ext4_extent_idx
));
966 if (unlikely(ix
> EXT_MAX_INDEX(curp
->p_hdr
))) {
967 EXT4_ERROR_INODE(inode
, "ix > EXT_MAX_INDEX!");
971 ix
->ei_block
= cpu_to_le32(logical
);
972 ext4_idx_store_pblock(ix
, ptr
);
973 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
975 if (unlikely(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
976 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
980 err
= ext4_ext_dirty(handle
, inode
, curp
);
981 ext4_std_error(inode
->i_sb
, err
);
988 * inserts new subtree into the path, using free index entry
990 * - allocates all needed blocks (new leaf and all intermediate index blocks)
991 * - makes decision where to split
992 * - moves remaining extents and index entries (right to the split point)
993 * into the newly allocated blocks
994 * - initializes subtree
996 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
998 struct ext4_ext_path
*path
,
999 struct ext4_extent
*newext
, int at
)
1001 struct buffer_head
*bh
= NULL
;
1002 int depth
= ext_depth(inode
);
1003 struct ext4_extent_header
*neh
;
1004 struct ext4_extent_idx
*fidx
;
1005 int i
= at
, k
, m
, a
;
1006 ext4_fsblk_t newblock
, oldblock
;
1008 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
1011 /* make decision: where to split? */
1012 /* FIXME: now decision is simplest: at current extent */
1014 /* if current leaf will be split, then we should use
1015 * border from split point */
1016 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
1017 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
1020 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
1021 border
= path
[depth
].p_ext
[1].ee_block
;
1022 ext_debug("leaf will be split."
1023 " next leaf starts at %d\n",
1024 le32_to_cpu(border
));
1026 border
= newext
->ee_block
;
1027 ext_debug("leaf will be added."
1028 " next leaf starts at %d\n",
1029 le32_to_cpu(border
));
1033 * If error occurs, then we break processing
1034 * and mark filesystem read-only. index won't
1035 * be inserted and tree will be in consistent
1036 * state. Next mount will repair buffers too.
1040 * Get array to track all allocated blocks.
1041 * We need this to handle errors and free blocks
1044 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
1048 /* allocate all needed blocks */
1049 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
1050 for (a
= 0; a
< depth
- at
; a
++) {
1051 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
1052 newext
, &err
, flags
);
1055 ablocks
[a
] = newblock
;
1058 /* initialize new leaf */
1059 newblock
= ablocks
[--a
];
1060 if (unlikely(newblock
== 0)) {
1061 EXT4_ERROR_INODE(inode
, "newblock == 0!");
1065 bh
= sb_getblk(inode
->i_sb
, newblock
);
1066 if (unlikely(!bh
)) {
1072 err
= ext4_journal_get_create_access(handle
, bh
);
1076 neh
= ext_block_hdr(bh
);
1077 neh
->eh_entries
= 0;
1078 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1079 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1082 /* move remainder of path[depth] to the new leaf */
1083 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
1084 path
[depth
].p_hdr
->eh_max
)) {
1085 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
1086 path
[depth
].p_hdr
->eh_entries
,
1087 path
[depth
].p_hdr
->eh_max
);
1091 /* start copy from next extent */
1092 m
= EXT_MAX_EXTENT(path
[depth
].p_hdr
) - path
[depth
].p_ext
++;
1093 ext4_ext_show_move(inode
, path
, newblock
, depth
);
1095 struct ext4_extent
*ex
;
1096 ex
= EXT_FIRST_EXTENT(neh
);
1097 memmove(ex
, path
[depth
].p_ext
, sizeof(struct ext4_extent
) * m
);
1098 le16_add_cpu(&neh
->eh_entries
, m
);
1101 ext4_extent_block_csum_set(inode
, neh
);
1102 set_buffer_uptodate(bh
);
1105 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1111 /* correct old leaf */
1113 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1116 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
1117 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1123 /* create intermediate indexes */
1125 if (unlikely(k
< 0)) {
1126 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
1131 ext_debug("create %d intermediate indices\n", k
);
1132 /* insert new index into current index block */
1133 /* current depth stored in i var */
1136 oldblock
= newblock
;
1137 newblock
= ablocks
[--a
];
1138 bh
= sb_getblk(inode
->i_sb
, newblock
);
1139 if (unlikely(!bh
)) {
1145 err
= ext4_journal_get_create_access(handle
, bh
);
1149 neh
= ext_block_hdr(bh
);
1150 neh
->eh_entries
= cpu_to_le16(1);
1151 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1152 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1153 neh
->eh_depth
= cpu_to_le16(depth
- i
);
1154 fidx
= EXT_FIRST_INDEX(neh
);
1155 fidx
->ei_block
= border
;
1156 ext4_idx_store_pblock(fidx
, oldblock
);
1158 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1159 i
, newblock
, le32_to_cpu(border
), oldblock
);
1161 /* move remainder of path[i] to the new index block */
1162 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
1163 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
1164 EXT4_ERROR_INODE(inode
,
1165 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1166 le32_to_cpu(path
[i
].p_ext
->ee_block
));
1170 /* start copy indexes */
1171 m
= EXT_MAX_INDEX(path
[i
].p_hdr
) - path
[i
].p_idx
++;
1172 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
1173 EXT_MAX_INDEX(path
[i
].p_hdr
));
1174 ext4_ext_show_move(inode
, path
, newblock
, i
);
1176 memmove(++fidx
, path
[i
].p_idx
,
1177 sizeof(struct ext4_extent_idx
) * m
);
1178 le16_add_cpu(&neh
->eh_entries
, m
);
1180 ext4_extent_block_csum_set(inode
, neh
);
1181 set_buffer_uptodate(bh
);
1184 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1190 /* correct old index */
1192 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1195 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1196 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1204 /* insert new index */
1205 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1206 le32_to_cpu(border
), newblock
);
1210 if (buffer_locked(bh
))
1216 /* free all allocated blocks in error case */
1217 for (i
= 0; i
< depth
; i
++) {
1220 ext4_free_blocks(handle
, inode
, NULL
, ablocks
[i
], 1,
1221 EXT4_FREE_BLOCKS_METADATA
);
1230 * ext4_ext_grow_indepth:
1231 * implements tree growing procedure:
1232 * - allocates new block
1233 * - moves top-level data (index block or leaf) into the new block
1234 * - initializes new top-level, creating index that points to the
1235 * just created block
1237 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1239 struct ext4_extent
*newext
)
1241 struct ext4_extent_header
*neh
;
1242 struct buffer_head
*bh
;
1243 ext4_fsblk_t newblock
;
1246 newblock
= ext4_ext_new_meta_block(handle
, inode
, NULL
,
1247 newext
, &err
, flags
);
1251 bh
= sb_getblk(inode
->i_sb
, newblock
);
1256 err
= ext4_journal_get_create_access(handle
, bh
);
1262 /* move top-level index/leaf into new block */
1263 memmove(bh
->b_data
, EXT4_I(inode
)->i_data
,
1264 sizeof(EXT4_I(inode
)->i_data
));
1266 /* set size of new block */
1267 neh
= ext_block_hdr(bh
);
1268 /* old root could have indexes or leaves
1269 * so calculate e_max right way */
1270 if (ext_depth(inode
))
1271 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1273 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1274 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1275 ext4_extent_block_csum_set(inode
, neh
);
1276 set_buffer_uptodate(bh
);
1279 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1283 /* Update top-level index: num,max,pointer */
1284 neh
= ext_inode_hdr(inode
);
1285 neh
->eh_entries
= cpu_to_le16(1);
1286 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh
), newblock
);
1287 if (neh
->eh_depth
== 0) {
1288 /* Root extent block becomes index block */
1289 neh
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1290 EXT_FIRST_INDEX(neh
)->ei_block
=
1291 EXT_FIRST_EXTENT(neh
)->ee_block
;
1293 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1294 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1295 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1296 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1298 le16_add_cpu(&neh
->eh_depth
, 1);
1299 ext4_mark_inode_dirty(handle
, inode
);
1307 * ext4_ext_create_new_leaf:
1308 * finds empty index and adds new leaf.
1309 * if no free index is found, then it requests in-depth growing.
1311 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1312 unsigned int mb_flags
,
1313 unsigned int gb_flags
,
1314 struct ext4_ext_path
*path
,
1315 struct ext4_extent
*newext
)
1317 struct ext4_ext_path
*curp
;
1318 int depth
, i
, err
= 0;
1321 i
= depth
= ext_depth(inode
);
1323 /* walk up to the tree and look for free index entry */
1324 curp
= path
+ depth
;
1325 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1330 /* we use already allocated block for index block,
1331 * so subsequent data blocks should be contiguous */
1332 if (EXT_HAS_FREE_INDEX(curp
)) {
1333 /* if we found index with free entry, then use that
1334 * entry: create all needed subtree and add new leaf */
1335 err
= ext4_ext_split(handle
, inode
, mb_flags
, path
, newext
, i
);
1340 ext4_ext_drop_refs(path
);
1341 path
= ext4_ext_find_extent(inode
,
1342 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1345 err
= PTR_ERR(path
);
1347 /* tree is full, time to grow in depth */
1348 err
= ext4_ext_grow_indepth(handle
, inode
, mb_flags
, newext
);
1353 ext4_ext_drop_refs(path
);
1354 path
= ext4_ext_find_extent(inode
,
1355 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1358 err
= PTR_ERR(path
);
1363 * only first (depth 0 -> 1) produces free space;
1364 * in all other cases we have to split the grown tree
1366 depth
= ext_depth(inode
);
1367 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1368 /* now we need to split */
1378 * search the closest allocated block to the left for *logical
1379 * and returns it at @logical + it's physical address at @phys
1380 * if *logical is the smallest allocated block, the function
1381 * returns 0 at @phys
1382 * return value contains 0 (success) or error code
1384 static int ext4_ext_search_left(struct inode
*inode
,
1385 struct ext4_ext_path
*path
,
1386 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1388 struct ext4_extent_idx
*ix
;
1389 struct ext4_extent
*ex
;
1392 if (unlikely(path
== NULL
)) {
1393 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1396 depth
= path
->p_depth
;
1399 if (depth
== 0 && path
->p_ext
== NULL
)
1402 /* usually extent in the path covers blocks smaller
1403 * then *logical, but it can be that extent is the
1404 * first one in the file */
1406 ex
= path
[depth
].p_ext
;
1407 ee_len
= ext4_ext_get_actual_len(ex
);
1408 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1409 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1410 EXT4_ERROR_INODE(inode
,
1411 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1412 *logical
, le32_to_cpu(ex
->ee_block
));
1415 while (--depth
>= 0) {
1416 ix
= path
[depth
].p_idx
;
1417 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1418 EXT4_ERROR_INODE(inode
,
1419 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1420 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1421 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1422 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1430 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1431 EXT4_ERROR_INODE(inode
,
1432 "logical %d < ee_block %d + ee_len %d!",
1433 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1437 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1438 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1443 * search the closest allocated block to the right for *logical
1444 * and returns it at @logical + it's physical address at @phys
1445 * if *logical is the largest allocated block, the function
1446 * returns 0 at @phys
1447 * return value contains 0 (success) or error code
1449 static int ext4_ext_search_right(struct inode
*inode
,
1450 struct ext4_ext_path
*path
,
1451 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1452 struct ext4_extent
**ret_ex
)
1454 struct buffer_head
*bh
= NULL
;
1455 struct ext4_extent_header
*eh
;
1456 struct ext4_extent_idx
*ix
;
1457 struct ext4_extent
*ex
;
1459 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1462 if (unlikely(path
== NULL
)) {
1463 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1466 depth
= path
->p_depth
;
1469 if (depth
== 0 && path
->p_ext
== NULL
)
1472 /* usually extent in the path covers blocks smaller
1473 * then *logical, but it can be that extent is the
1474 * first one in the file */
1476 ex
= path
[depth
].p_ext
;
1477 ee_len
= ext4_ext_get_actual_len(ex
);
1478 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1479 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1480 EXT4_ERROR_INODE(inode
,
1481 "first_extent(path[%d].p_hdr) != ex",
1485 while (--depth
>= 0) {
1486 ix
= path
[depth
].p_idx
;
1487 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1488 EXT4_ERROR_INODE(inode
,
1489 "ix != EXT_FIRST_INDEX *logical %d!",
1497 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1498 EXT4_ERROR_INODE(inode
,
1499 "logical %d < ee_block %d + ee_len %d!",
1500 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1504 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1505 /* next allocated block in this leaf */
1510 /* go up and search for index to the right */
1511 while (--depth
>= 0) {
1512 ix
= path
[depth
].p_idx
;
1513 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1517 /* we've gone up to the root and found no index to the right */
1521 /* we've found index to the right, let's
1522 * follow it and find the closest allocated
1523 * block to the right */
1525 block
= ext4_idx_pblock(ix
);
1526 while (++depth
< path
->p_depth
) {
1527 /* subtract from p_depth to get proper eh_depth */
1528 bh
= read_extent_tree_block(inode
, block
,
1529 path
->p_depth
- depth
, 0);
1532 eh
= ext_block_hdr(bh
);
1533 ix
= EXT_FIRST_INDEX(eh
);
1534 block
= ext4_idx_pblock(ix
);
1538 bh
= read_extent_tree_block(inode
, block
, path
->p_depth
- depth
, 0);
1541 eh
= ext_block_hdr(bh
);
1542 ex
= EXT_FIRST_EXTENT(eh
);
1544 *logical
= le32_to_cpu(ex
->ee_block
);
1545 *phys
= ext4_ext_pblock(ex
);
1553 * ext4_ext_next_allocated_block:
1554 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1555 * NOTE: it considers block number from index entry as
1556 * allocated block. Thus, index entries have to be consistent
1560 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1564 BUG_ON(path
== NULL
);
1565 depth
= path
->p_depth
;
1567 if (depth
== 0 && path
->p_ext
== NULL
)
1568 return EXT_MAX_BLOCKS
;
1570 while (depth
>= 0) {
1571 if (depth
== path
->p_depth
) {
1573 if (path
[depth
].p_ext
&&
1574 path
[depth
].p_ext
!=
1575 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1576 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1579 if (path
[depth
].p_idx
!=
1580 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1581 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1586 return EXT_MAX_BLOCKS
;
1590 * ext4_ext_next_leaf_block:
1591 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1593 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1597 BUG_ON(path
== NULL
);
1598 depth
= path
->p_depth
;
1600 /* zero-tree has no leaf blocks at all */
1602 return EXT_MAX_BLOCKS
;
1604 /* go to index block */
1607 while (depth
>= 0) {
1608 if (path
[depth
].p_idx
!=
1609 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1610 return (ext4_lblk_t
)
1611 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1615 return EXT_MAX_BLOCKS
;
1619 * ext4_ext_correct_indexes:
1620 * if leaf gets modified and modified extent is first in the leaf,
1621 * then we have to correct all indexes above.
1622 * TODO: do we need to correct tree in all cases?
1624 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1625 struct ext4_ext_path
*path
)
1627 struct ext4_extent_header
*eh
;
1628 int depth
= ext_depth(inode
);
1629 struct ext4_extent
*ex
;
1633 eh
= path
[depth
].p_hdr
;
1634 ex
= path
[depth
].p_ext
;
1636 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1637 EXT4_ERROR_INODE(inode
,
1638 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1643 /* there is no tree at all */
1647 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1648 /* we correct tree if first leaf got modified only */
1653 * TODO: we need correction if border is smaller than current one
1656 border
= path
[depth
].p_ext
->ee_block
;
1657 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1660 path
[k
].p_idx
->ei_block
= border
;
1661 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1666 /* change all left-side indexes */
1667 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1669 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1672 path
[k
].p_idx
->ei_block
= border
;
1673 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1682 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1683 struct ext4_extent
*ex2
)
1685 unsigned short ext1_ee_len
, ext2_ee_len
;
1688 * Make sure that both extents are initialized. We don't merge
1689 * uninitialized extents so that we can be sure that end_io code has
1690 * the extent that was written properly split out and conversion to
1691 * initialized is trivial.
1693 if (ext4_ext_is_uninitialized(ex1
) != ext4_ext_is_uninitialized(ex2
))
1696 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1697 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1699 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1700 le32_to_cpu(ex2
->ee_block
))
1704 * To allow future support for preallocated extents to be added
1705 * as an RO_COMPAT feature, refuse to merge to extents if
1706 * this can result in the top bit of ee_len being set.
1708 if (ext1_ee_len
+ ext2_ee_len
> EXT_INIT_MAX_LEN
)
1710 if (ext4_ext_is_uninitialized(ex1
) &&
1711 (ext4_test_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
) ||
1712 atomic_read(&EXT4_I(inode
)->i_unwritten
) ||
1713 (ext1_ee_len
+ ext2_ee_len
> EXT_UNINIT_MAX_LEN
)))
1715 #ifdef AGGRESSIVE_TEST
1716 if (ext1_ee_len
>= 4)
1720 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1726 * This function tries to merge the "ex" extent to the next extent in the tree.
1727 * It always tries to merge towards right. If you want to merge towards
1728 * left, pass "ex - 1" as argument instead of "ex".
1729 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1730 * 1 if they got merged.
1732 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1733 struct ext4_ext_path
*path
,
1734 struct ext4_extent
*ex
)
1736 struct ext4_extent_header
*eh
;
1737 unsigned int depth
, len
;
1738 int merge_done
= 0, uninit
;
1740 depth
= ext_depth(inode
);
1741 BUG_ON(path
[depth
].p_hdr
== NULL
);
1742 eh
= path
[depth
].p_hdr
;
1744 while (ex
< EXT_LAST_EXTENT(eh
)) {
1745 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1747 /* merge with next extent! */
1748 uninit
= ext4_ext_is_uninitialized(ex
);
1749 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1750 + ext4_ext_get_actual_len(ex
+ 1));
1752 ext4_ext_mark_uninitialized(ex
);
1754 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1755 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1756 * sizeof(struct ext4_extent
);
1757 memmove(ex
+ 1, ex
+ 2, len
);
1759 le16_add_cpu(&eh
->eh_entries
, -1);
1761 WARN_ON(eh
->eh_entries
== 0);
1762 if (!eh
->eh_entries
)
1763 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1770 * This function does a very simple check to see if we can collapse
1771 * an extent tree with a single extent tree leaf block into the inode.
1773 static void ext4_ext_try_to_merge_up(handle_t
*handle
,
1774 struct inode
*inode
,
1775 struct ext4_ext_path
*path
)
1778 unsigned max_root
= ext4_ext_space_root(inode
, 0);
1781 if ((path
[0].p_depth
!= 1) ||
1782 (le16_to_cpu(path
[0].p_hdr
->eh_entries
) != 1) ||
1783 (le16_to_cpu(path
[1].p_hdr
->eh_entries
) > max_root
))
1787 * We need to modify the block allocation bitmap and the block
1788 * group descriptor to release the extent tree block. If we
1789 * can't get the journal credits, give up.
1791 if (ext4_journal_extend(handle
, 2))
1795 * Copy the extent data up to the inode
1797 blk
= ext4_idx_pblock(path
[0].p_idx
);
1798 s
= le16_to_cpu(path
[1].p_hdr
->eh_entries
) *
1799 sizeof(struct ext4_extent_idx
);
1800 s
+= sizeof(struct ext4_extent_header
);
1802 memcpy(path
[0].p_hdr
, path
[1].p_hdr
, s
);
1803 path
[0].p_depth
= 0;
1804 path
[0].p_ext
= EXT_FIRST_EXTENT(path
[0].p_hdr
) +
1805 (path
[1].p_ext
- EXT_FIRST_EXTENT(path
[1].p_hdr
));
1806 path
[0].p_hdr
->eh_max
= cpu_to_le16(max_root
);
1808 brelse(path
[1].p_bh
);
1809 ext4_free_blocks(handle
, inode
, NULL
, blk
, 1,
1810 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
|
1811 EXT4_FREE_BLOCKS_RESERVE
);
1815 * This function tries to merge the @ex extent to neighbours in the tree.
1816 * return 1 if merge left else 0.
1818 static void ext4_ext_try_to_merge(handle_t
*handle
,
1819 struct inode
*inode
,
1820 struct ext4_ext_path
*path
,
1821 struct ext4_extent
*ex
) {
1822 struct ext4_extent_header
*eh
;
1826 depth
= ext_depth(inode
);
1827 BUG_ON(path
[depth
].p_hdr
== NULL
);
1828 eh
= path
[depth
].p_hdr
;
1830 if (ex
> EXT_FIRST_EXTENT(eh
))
1831 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1834 (void) ext4_ext_try_to_merge_right(inode
, path
, ex
);
1836 ext4_ext_try_to_merge_up(handle
, inode
, path
);
1840 * check if a portion of the "newext" extent overlaps with an
1843 * If there is an overlap discovered, it updates the length of the newext
1844 * such that there will be no overlap, and then returns 1.
1845 * If there is no overlap found, it returns 0.
1847 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1848 struct inode
*inode
,
1849 struct ext4_extent
*newext
,
1850 struct ext4_ext_path
*path
)
1853 unsigned int depth
, len1
;
1854 unsigned int ret
= 0;
1856 b1
= le32_to_cpu(newext
->ee_block
);
1857 len1
= ext4_ext_get_actual_len(newext
);
1858 depth
= ext_depth(inode
);
1859 if (!path
[depth
].p_ext
)
1861 b2
= EXT4_LBLK_CMASK(sbi
, le32_to_cpu(path
[depth
].p_ext
->ee_block
));
1864 * get the next allocated block if the extent in the path
1865 * is before the requested block(s)
1868 b2
= ext4_ext_next_allocated_block(path
);
1869 if (b2
== EXT_MAX_BLOCKS
)
1871 b2
= EXT4_LBLK_CMASK(sbi
, b2
);
1874 /* check for wrap through zero on extent logical start block*/
1875 if (b1
+ len1
< b1
) {
1876 len1
= EXT_MAX_BLOCKS
- b1
;
1877 newext
->ee_len
= cpu_to_le16(len1
);
1881 /* check for overlap */
1882 if (b1
+ len1
> b2
) {
1883 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1891 * ext4_ext_insert_extent:
1892 * tries to merge requsted extent into the existing extent or
1893 * inserts requested extent as new one into the tree,
1894 * creating new leaf in the no-space case.
1896 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1897 struct ext4_ext_path
*path
,
1898 struct ext4_extent
*newext
, int gb_flags
)
1900 struct ext4_extent_header
*eh
;
1901 struct ext4_extent
*ex
, *fex
;
1902 struct ext4_extent
*nearex
; /* nearest extent */
1903 struct ext4_ext_path
*npath
= NULL
;
1904 int depth
, len
, err
;
1906 int mb_flags
= 0, uninit
;
1908 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1909 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1912 depth
= ext_depth(inode
);
1913 ex
= path
[depth
].p_ext
;
1914 eh
= path
[depth
].p_hdr
;
1915 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1916 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1920 /* try to insert block into found extent and return */
1921 if (ex
&& !(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
1924 * Try to see whether we should rather test the extent on
1925 * right from ex, or from the left of ex. This is because
1926 * ext4_ext_find_extent() can return either extent on the
1927 * left, or on the right from the searched position. This
1928 * will make merging more effective.
1930 if (ex
< EXT_LAST_EXTENT(eh
) &&
1931 (le32_to_cpu(ex
->ee_block
) +
1932 ext4_ext_get_actual_len(ex
) <
1933 le32_to_cpu(newext
->ee_block
))) {
1936 } else if ((ex
> EXT_FIRST_EXTENT(eh
)) &&
1937 (le32_to_cpu(newext
->ee_block
) +
1938 ext4_ext_get_actual_len(newext
) <
1939 le32_to_cpu(ex
->ee_block
)))
1942 /* Try to append newex to the ex */
1943 if (ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1944 ext_debug("append [%d]%d block to %u:[%d]%d"
1946 ext4_ext_is_uninitialized(newext
),
1947 ext4_ext_get_actual_len(newext
),
1948 le32_to_cpu(ex
->ee_block
),
1949 ext4_ext_is_uninitialized(ex
),
1950 ext4_ext_get_actual_len(ex
),
1951 ext4_ext_pblock(ex
));
1952 err
= ext4_ext_get_access(handle
, inode
,
1956 uninit
= ext4_ext_is_uninitialized(ex
);
1957 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1958 + ext4_ext_get_actual_len(newext
));
1960 ext4_ext_mark_uninitialized(ex
);
1961 eh
= path
[depth
].p_hdr
;
1967 /* Try to prepend newex to the ex */
1968 if (ext4_can_extents_be_merged(inode
, newext
, ex
)) {
1969 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1971 le32_to_cpu(newext
->ee_block
),
1972 ext4_ext_is_uninitialized(newext
),
1973 ext4_ext_get_actual_len(newext
),
1974 le32_to_cpu(ex
->ee_block
),
1975 ext4_ext_is_uninitialized(ex
),
1976 ext4_ext_get_actual_len(ex
),
1977 ext4_ext_pblock(ex
));
1978 err
= ext4_ext_get_access(handle
, inode
,
1983 uninit
= ext4_ext_is_uninitialized(ex
);
1984 ex
->ee_block
= newext
->ee_block
;
1985 ext4_ext_store_pblock(ex
, ext4_ext_pblock(newext
));
1986 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1987 + ext4_ext_get_actual_len(newext
));
1989 ext4_ext_mark_uninitialized(ex
);
1990 eh
= path
[depth
].p_hdr
;
1996 depth
= ext_depth(inode
);
1997 eh
= path
[depth
].p_hdr
;
1998 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
2001 /* probably next leaf has space for us? */
2002 fex
= EXT_LAST_EXTENT(eh
);
2003 next
= EXT_MAX_BLOCKS
;
2004 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
2005 next
= ext4_ext_next_leaf_block(path
);
2006 if (next
!= EXT_MAX_BLOCKS
) {
2007 ext_debug("next leaf block - %u\n", next
);
2008 BUG_ON(npath
!= NULL
);
2009 npath
= ext4_ext_find_extent(inode
, next
, NULL
, 0);
2011 return PTR_ERR(npath
);
2012 BUG_ON(npath
->p_depth
!= path
->p_depth
);
2013 eh
= npath
[depth
].p_hdr
;
2014 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
2015 ext_debug("next leaf isn't full(%d)\n",
2016 le16_to_cpu(eh
->eh_entries
));
2020 ext_debug("next leaf has no free space(%d,%d)\n",
2021 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
2025 * There is no free space in the found leaf.
2026 * We're gonna add a new leaf in the tree.
2028 if (gb_flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
2029 mb_flags
= EXT4_MB_USE_RESERVED
;
2030 err
= ext4_ext_create_new_leaf(handle
, inode
, mb_flags
, gb_flags
,
2034 depth
= ext_depth(inode
);
2035 eh
= path
[depth
].p_hdr
;
2038 nearex
= path
[depth
].p_ext
;
2040 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2045 /* there is no extent in this leaf, create first one */
2046 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2047 le32_to_cpu(newext
->ee_block
),
2048 ext4_ext_pblock(newext
),
2049 ext4_ext_is_uninitialized(newext
),
2050 ext4_ext_get_actual_len(newext
));
2051 nearex
= EXT_FIRST_EXTENT(eh
);
2053 if (le32_to_cpu(newext
->ee_block
)
2054 > le32_to_cpu(nearex
->ee_block
)) {
2056 ext_debug("insert %u:%llu:[%d]%d before: "
2058 le32_to_cpu(newext
->ee_block
),
2059 ext4_ext_pblock(newext
),
2060 ext4_ext_is_uninitialized(newext
),
2061 ext4_ext_get_actual_len(newext
),
2066 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
2067 ext_debug("insert %u:%llu:[%d]%d after: "
2069 le32_to_cpu(newext
->ee_block
),
2070 ext4_ext_pblock(newext
),
2071 ext4_ext_is_uninitialized(newext
),
2072 ext4_ext_get_actual_len(newext
),
2075 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
2077 ext_debug("insert %u:%llu:[%d]%d: "
2078 "move %d extents from 0x%p to 0x%p\n",
2079 le32_to_cpu(newext
->ee_block
),
2080 ext4_ext_pblock(newext
),
2081 ext4_ext_is_uninitialized(newext
),
2082 ext4_ext_get_actual_len(newext
),
2083 len
, nearex
, nearex
+ 1);
2084 memmove(nearex
+ 1, nearex
,
2085 len
* sizeof(struct ext4_extent
));
2089 le16_add_cpu(&eh
->eh_entries
, 1);
2090 path
[depth
].p_ext
= nearex
;
2091 nearex
->ee_block
= newext
->ee_block
;
2092 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
2093 nearex
->ee_len
= newext
->ee_len
;
2096 /* try to merge extents */
2097 if (!(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
))
2098 ext4_ext_try_to_merge(handle
, inode
, path
, nearex
);
2101 /* time to correct all indexes above */
2102 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2106 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
2110 ext4_ext_drop_refs(npath
);
2116 static int ext4_fill_fiemap_extents(struct inode
*inode
,
2117 ext4_lblk_t block
, ext4_lblk_t num
,
2118 struct fiemap_extent_info
*fieinfo
)
2120 struct ext4_ext_path
*path
= NULL
;
2121 struct ext4_extent
*ex
;
2122 struct extent_status es
;
2123 ext4_lblk_t next
, next_del
, start
= 0, end
= 0;
2124 ext4_lblk_t last
= block
+ num
;
2125 int exists
, depth
= 0, err
= 0;
2126 unsigned int flags
= 0;
2127 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
2129 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
2131 /* find extent for this block */
2132 down_read(&EXT4_I(inode
)->i_data_sem
);
2134 if (path
&& ext_depth(inode
) != depth
) {
2135 /* depth was changed. we have to realloc path */
2140 path
= ext4_ext_find_extent(inode
, block
, path
, 0);
2142 up_read(&EXT4_I(inode
)->i_data_sem
);
2143 err
= PTR_ERR(path
);
2148 depth
= ext_depth(inode
);
2149 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2150 up_read(&EXT4_I(inode
)->i_data_sem
);
2151 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2155 ex
= path
[depth
].p_ext
;
2156 next
= ext4_ext_next_allocated_block(path
);
2157 ext4_ext_drop_refs(path
);
2162 /* there is no extent yet, so try to allocate
2163 * all requested space */
2166 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
2167 /* need to allocate space before found extent */
2169 end
= le32_to_cpu(ex
->ee_block
);
2170 if (block
+ num
< end
)
2172 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2173 + ext4_ext_get_actual_len(ex
)) {
2174 /* need to allocate space after found extent */
2179 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
2181 * some part of requested space is covered
2185 end
= le32_to_cpu(ex
->ee_block
)
2186 + ext4_ext_get_actual_len(ex
);
2187 if (block
+ num
< end
)
2193 BUG_ON(end
<= start
);
2197 es
.es_len
= end
- start
;
2200 es
.es_lblk
= le32_to_cpu(ex
->ee_block
);
2201 es
.es_len
= ext4_ext_get_actual_len(ex
);
2202 es
.es_pblk
= ext4_ext_pblock(ex
);
2203 if (ext4_ext_is_uninitialized(ex
))
2204 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
2208 * Find delayed extent and update es accordingly. We call
2209 * it even in !exists case to find out whether es is the
2210 * last existing extent or not.
2212 next_del
= ext4_find_delayed_extent(inode
, &es
);
2213 if (!exists
&& next_del
) {
2215 flags
|= (FIEMAP_EXTENT_DELALLOC
|
2216 FIEMAP_EXTENT_UNKNOWN
);
2218 up_read(&EXT4_I(inode
)->i_data_sem
);
2220 if (unlikely(es
.es_len
== 0)) {
2221 EXT4_ERROR_INODE(inode
, "es.es_len == 0");
2227 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2228 * we need to check next == EXT_MAX_BLOCKS because it is
2229 * possible that an extent is with unwritten and delayed
2230 * status due to when an extent is delayed allocated and
2231 * is allocated by fallocate status tree will track both of
2234 * So we could return a unwritten and delayed extent, and
2235 * its block is equal to 'next'.
2237 if (next
== next_del
&& next
== EXT_MAX_BLOCKS
) {
2238 flags
|= FIEMAP_EXTENT_LAST
;
2239 if (unlikely(next_del
!= EXT_MAX_BLOCKS
||
2240 next
!= EXT_MAX_BLOCKS
)) {
2241 EXT4_ERROR_INODE(inode
,
2242 "next extent == %u, next "
2243 "delalloc extent = %u",
2251 err
= fiemap_fill_next_extent(fieinfo
,
2252 (__u64
)es
.es_lblk
<< blksize_bits
,
2253 (__u64
)es
.es_pblk
<< blksize_bits
,
2254 (__u64
)es
.es_len
<< blksize_bits
,
2264 block
= es
.es_lblk
+ es
.es_len
;
2268 ext4_ext_drop_refs(path
);
2276 * ext4_ext_put_gap_in_cache:
2277 * calculate boundaries of the gap that the requested block fits into
2278 * and cache this gap
2281 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
2284 int depth
= ext_depth(inode
);
2285 unsigned long len
= 0;
2286 ext4_lblk_t lblock
= 0;
2287 struct ext4_extent
*ex
;
2289 ex
= path
[depth
].p_ext
;
2292 * there is no extent yet, so gap is [0;-] and we
2295 ext_debug("cache gap(whole file):");
2296 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
2298 len
= le32_to_cpu(ex
->ee_block
) - block
;
2299 ext_debug("cache gap(before): %u [%u:%u]",
2301 le32_to_cpu(ex
->ee_block
),
2302 ext4_ext_get_actual_len(ex
));
2303 if (!ext4_find_delalloc_range(inode
, lblock
, lblock
+ len
- 1))
2304 ext4_es_insert_extent(inode
, lblock
, len
, ~0,
2305 EXTENT_STATUS_HOLE
);
2306 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2307 + ext4_ext_get_actual_len(ex
)) {
2309 lblock
= le32_to_cpu(ex
->ee_block
)
2310 + ext4_ext_get_actual_len(ex
);
2312 next
= ext4_ext_next_allocated_block(path
);
2313 ext_debug("cache gap(after): [%u:%u] %u",
2314 le32_to_cpu(ex
->ee_block
),
2315 ext4_ext_get_actual_len(ex
),
2317 BUG_ON(next
== lblock
);
2318 len
= next
- lblock
;
2319 if (!ext4_find_delalloc_range(inode
, lblock
, lblock
+ len
- 1))
2320 ext4_es_insert_extent(inode
, lblock
, len
, ~0,
2321 EXTENT_STATUS_HOLE
);
2326 ext_debug(" -> %u:%lu\n", lblock
, len
);
2331 * removes index from the index block.
2333 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2334 struct ext4_ext_path
*path
, int depth
)
2339 /* free index block */
2341 path
= path
+ depth
;
2342 leaf
= ext4_idx_pblock(path
->p_idx
);
2343 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2344 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2347 err
= ext4_ext_get_access(handle
, inode
, path
);
2351 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2352 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2353 len
*= sizeof(struct ext4_extent_idx
);
2354 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2357 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2358 err
= ext4_ext_dirty(handle
, inode
, path
);
2361 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2362 trace_ext4_ext_rm_idx(inode
, leaf
);
2364 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2365 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2367 while (--depth
>= 0) {
2368 if (path
->p_idx
!= EXT_FIRST_INDEX(path
->p_hdr
))
2371 err
= ext4_ext_get_access(handle
, inode
, path
);
2374 path
->p_idx
->ei_block
= (path
+1)->p_idx
->ei_block
;
2375 err
= ext4_ext_dirty(handle
, inode
, path
);
2383 * ext4_ext_calc_credits_for_single_extent:
2384 * This routine returns max. credits that needed to insert an extent
2385 * to the extent tree.
2386 * When pass the actual path, the caller should calculate credits
2389 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2390 struct ext4_ext_path
*path
)
2393 int depth
= ext_depth(inode
);
2396 /* probably there is space in leaf? */
2397 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2398 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2401 * There are some space in the leaf tree, no
2402 * need to account for leaf block credit
2404 * bitmaps and block group descriptor blocks
2405 * and other metadata blocks still need to be
2408 /* 1 bitmap, 1 block group descriptor */
2409 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2414 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2418 * How many index/leaf blocks need to change/allocate to add @extents extents?
2420 * If we add a single extent, then in the worse case, each tree level
2421 * index/leaf need to be changed in case of the tree split.
2423 * If more extents are inserted, they could cause the whole tree split more
2424 * than once, but this is really rare.
2426 int ext4_ext_index_trans_blocks(struct inode
*inode
, int extents
)
2431 /* If we are converting the inline data, only one is needed here. */
2432 if (ext4_has_inline_data(inode
))
2435 depth
= ext_depth(inode
);
2445 static inline int get_default_free_blocks_flags(struct inode
*inode
)
2447 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2448 return EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
;
2449 else if (ext4_should_journal_data(inode
))
2450 return EXT4_FREE_BLOCKS_FORGET
;
2454 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2455 struct ext4_extent
*ex
,
2456 long long *partial_cluster
,
2457 ext4_lblk_t from
, ext4_lblk_t to
)
2459 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2460 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2462 int flags
= get_default_free_blocks_flags(inode
);
2465 * For bigalloc file systems, we never free a partial cluster
2466 * at the beginning of the extent. Instead, we make a note
2467 * that we tried freeing the cluster, and check to see if we
2468 * need to free it on a subsequent call to ext4_remove_blocks,
2469 * or at the end of the ext4_truncate() operation.
2471 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2473 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2475 * If we have a partial cluster, and it's different from the
2476 * cluster of the last block, we need to explicitly free the
2477 * partial cluster here.
2479 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2480 if ((*partial_cluster
> 0) &&
2481 (EXT4_B2C(sbi
, pblk
) != *partial_cluster
)) {
2482 ext4_free_blocks(handle
, inode
, NULL
,
2483 EXT4_C2B(sbi
, *partial_cluster
),
2484 sbi
->s_cluster_ratio
, flags
);
2485 *partial_cluster
= 0;
2488 #ifdef EXTENTS_STATS
2490 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2491 spin_lock(&sbi
->s_ext_stats_lock
);
2492 sbi
->s_ext_blocks
+= ee_len
;
2493 sbi
->s_ext_extents
++;
2494 if (ee_len
< sbi
->s_ext_min
)
2495 sbi
->s_ext_min
= ee_len
;
2496 if (ee_len
> sbi
->s_ext_max
)
2497 sbi
->s_ext_max
= ee_len
;
2498 if (ext_depth(inode
) > sbi
->s_depth_max
)
2499 sbi
->s_depth_max
= ext_depth(inode
);
2500 spin_unlock(&sbi
->s_ext_stats_lock
);
2503 if (from
>= le32_to_cpu(ex
->ee_block
)
2504 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2507 unsigned int unaligned
;
2509 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2510 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2512 * Usually we want to free partial cluster at the end of the
2513 * extent, except for the situation when the cluster is still
2514 * used by any other extent (partial_cluster is negative).
2516 if (*partial_cluster
< 0 &&
2517 -(*partial_cluster
) == EXT4_B2C(sbi
, pblk
+ num
- 1))
2518 flags
|= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER
;
2520 ext_debug("free last %u blocks starting %llu partial %lld\n",
2521 num
, pblk
, *partial_cluster
);
2522 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2524 * If the block range to be freed didn't start at the
2525 * beginning of a cluster, and we removed the entire
2526 * extent and the cluster is not used by any other extent,
2527 * save the partial cluster here, since we might need to
2528 * delete if we determine that the truncate operation has
2529 * removed all of the blocks in the cluster.
2531 * On the other hand, if we did not manage to free the whole
2532 * extent, we have to mark the cluster as used (store negative
2533 * cluster number in partial_cluster).
2535 unaligned
= EXT4_PBLK_COFF(sbi
, pblk
);
2536 if (unaligned
&& (ee_len
== num
) &&
2537 (*partial_cluster
!= -((long long)EXT4_B2C(sbi
, pblk
))))
2538 *partial_cluster
= EXT4_B2C(sbi
, pblk
);
2540 *partial_cluster
= -((long long)EXT4_B2C(sbi
, pblk
));
2541 else if (*partial_cluster
> 0)
2542 *partial_cluster
= 0;
2544 ext4_error(sbi
->s_sb
, "strange request: removal(2) "
2545 "%u-%u from %u:%u\n",
2546 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2552 * ext4_ext_rm_leaf() Removes the extents associated with the
2553 * blocks appearing between "start" and "end", and splits the extents
2554 * if "start" and "end" appear in the same extent
2556 * @handle: The journal handle
2557 * @inode: The files inode
2558 * @path: The path to the leaf
2559 * @partial_cluster: The cluster which we'll have to free if all extents
2560 * has been released from it. It gets negative in case
2561 * that the cluster is still used.
2562 * @start: The first block to remove
2563 * @end: The last block to remove
2566 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2567 struct ext4_ext_path
*path
,
2568 long long *partial_cluster
,
2569 ext4_lblk_t start
, ext4_lblk_t end
)
2571 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2572 int err
= 0, correct_index
= 0;
2573 int depth
= ext_depth(inode
), credits
;
2574 struct ext4_extent_header
*eh
;
2577 ext4_lblk_t ex_ee_block
;
2578 unsigned short ex_ee_len
;
2579 unsigned uninitialized
= 0;
2580 struct ext4_extent
*ex
;
2583 /* the header must be checked already in ext4_ext_remove_space() */
2584 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2585 if (!path
[depth
].p_hdr
)
2586 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2587 eh
= path
[depth
].p_hdr
;
2588 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2589 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2592 /* find where to start removing */
2593 ex
= path
[depth
].p_ext
;
2595 ex
= EXT_LAST_EXTENT(eh
);
2597 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2598 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2601 * If we're starting with an extent other than the last one in the
2602 * node, we need to see if it shares a cluster with the extent to
2603 * the right (towards the end of the file). If its leftmost cluster
2604 * is this extent's rightmost cluster and it is not cluster aligned,
2605 * we'll mark it as a partial that is not to be deallocated.
2608 if (ex
!= EXT_LAST_EXTENT(eh
)) {
2609 ext4_fsblk_t current_pblk
, right_pblk
;
2610 long long current_cluster
, right_cluster
;
2612 current_pblk
= ext4_ext_pblock(ex
) + ex_ee_len
- 1;
2613 current_cluster
= (long long)EXT4_B2C(sbi
, current_pblk
);
2614 right_pblk
= ext4_ext_pblock(ex
+ 1);
2615 right_cluster
= (long long)EXT4_B2C(sbi
, right_pblk
);
2616 if (current_cluster
== right_cluster
&&
2617 EXT4_PBLK_COFF(sbi
, right_pblk
))
2618 *partial_cluster
= -right_cluster
;
2621 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2623 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2624 ex_ee_block
+ ex_ee_len
> start
) {
2626 if (ext4_ext_is_uninitialized(ex
))
2631 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2632 uninitialized
, ex_ee_len
);
2633 path
[depth
].p_ext
= ex
;
2635 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2636 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2637 ex_ee_block
+ex_ee_len
- 1 : end
;
2639 ext_debug(" border %u:%u\n", a
, b
);
2641 /* If this extent is beyond the end of the hole, skip it */
2642 if (end
< ex_ee_block
) {
2644 * We're going to skip this extent and move to another,
2645 * so if this extent is not cluster aligned we have
2646 * to mark the current cluster as used to avoid
2647 * accidentally freeing it later on
2649 pblk
= ext4_ext_pblock(ex
);
2650 if (EXT4_PBLK_COFF(sbi
, pblk
))
2652 -((long long)EXT4_B2C(sbi
, pblk
));
2654 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2655 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2657 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2658 EXT4_ERROR_INODE(inode
,
2659 "can not handle truncate %u:%u "
2661 start
, end
, ex_ee_block
,
2662 ex_ee_block
+ ex_ee_len
- 1);
2665 } else if (a
!= ex_ee_block
) {
2666 /* remove tail of the extent */
2667 num
= a
- ex_ee_block
;
2669 /* remove whole extent: excellent! */
2673 * 3 for leaf, sb, and inode plus 2 (bmap and group
2674 * descriptor) for each block group; assume two block
2675 * groups plus ex_ee_len/blocks_per_block_group for
2678 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2679 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2681 credits
+= (ext_depth(inode
)) + 1;
2683 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2685 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2689 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2693 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2699 /* this extent is removed; mark slot entirely unused */
2700 ext4_ext_store_pblock(ex
, 0);
2702 ex
->ee_len
= cpu_to_le16(num
);
2704 * Do not mark uninitialized if all the blocks in the
2705 * extent have been removed.
2707 if (uninitialized
&& num
)
2708 ext4_ext_mark_uninitialized(ex
);
2710 * If the extent was completely released,
2711 * we need to remove it from the leaf
2714 if (end
!= EXT_MAX_BLOCKS
- 1) {
2716 * For hole punching, we need to scoot all the
2717 * extents up when an extent is removed so that
2718 * we dont have blank extents in the middle
2720 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2721 sizeof(struct ext4_extent
));
2723 /* Now get rid of the one at the end */
2724 memset(EXT_LAST_EXTENT(eh
), 0,
2725 sizeof(struct ext4_extent
));
2727 le16_add_cpu(&eh
->eh_entries
, -1);
2728 } else if (*partial_cluster
> 0)
2729 *partial_cluster
= 0;
2731 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2735 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2736 ext4_ext_pblock(ex
));
2738 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2739 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2742 if (correct_index
&& eh
->eh_entries
)
2743 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2746 * If there's a partial cluster and at least one extent remains in
2747 * the leaf, free the partial cluster if it isn't shared with the
2748 * current extent. If there's a partial cluster and no extents
2749 * remain in the leaf, it can't be freed here. It can only be
2750 * freed when it's possible to determine if it's not shared with
2751 * any other extent - when the next leaf is processed or when space
2752 * removal is complete.
2754 if (*partial_cluster
> 0 && eh
->eh_entries
&&
2755 (EXT4_B2C(sbi
, ext4_ext_pblock(ex
) + ex_ee_len
- 1) !=
2756 *partial_cluster
)) {
2757 int flags
= get_default_free_blocks_flags(inode
);
2759 ext4_free_blocks(handle
, inode
, NULL
,
2760 EXT4_C2B(sbi
, *partial_cluster
),
2761 sbi
->s_cluster_ratio
, flags
);
2762 *partial_cluster
= 0;
2765 /* if this leaf is free, then we should
2766 * remove it from index block above */
2767 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2768 err
= ext4_ext_rm_idx(handle
, inode
, path
, depth
);
2775 * ext4_ext_more_to_rm:
2776 * returns 1 if current index has to be freed (even partial)
2779 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2781 BUG_ON(path
->p_idx
== NULL
);
2783 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2787 * if truncate on deeper level happened, it wasn't partial,
2788 * so we have to consider current index for truncation
2790 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2795 int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2798 struct super_block
*sb
= inode
->i_sb
;
2799 int depth
= ext_depth(inode
);
2800 struct ext4_ext_path
*path
= NULL
;
2801 long long partial_cluster
= 0;
2805 ext_debug("truncate since %u to %u\n", start
, end
);
2807 /* probably first extent we're gonna free will be last in block */
2808 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, depth
+ 1);
2810 return PTR_ERR(handle
);
2813 trace_ext4_ext_remove_space(inode
, start
, end
, depth
);
2816 * Check if we are removing extents inside the extent tree. If that
2817 * is the case, we are going to punch a hole inside the extent tree
2818 * so we have to check whether we need to split the extent covering
2819 * the last block to remove so we can easily remove the part of it
2820 * in ext4_ext_rm_leaf().
2822 if (end
< EXT_MAX_BLOCKS
- 1) {
2823 struct ext4_extent
*ex
;
2824 ext4_lblk_t ee_block
;
2826 /* find extent for this block */
2827 path
= ext4_ext_find_extent(inode
, end
, NULL
, EXT4_EX_NOCACHE
);
2829 ext4_journal_stop(handle
);
2830 return PTR_ERR(path
);
2832 depth
= ext_depth(inode
);
2833 /* Leaf not may not exist only if inode has no blocks at all */
2834 ex
= path
[depth
].p_ext
;
2837 EXT4_ERROR_INODE(inode
,
2838 "path[%d].p_hdr == NULL",
2845 ee_block
= le32_to_cpu(ex
->ee_block
);
2848 * See if the last block is inside the extent, if so split
2849 * the extent at 'end' block so we can easily remove the
2850 * tail of the first part of the split extent in
2851 * ext4_ext_rm_leaf().
2853 if (end
>= ee_block
&&
2854 end
< ee_block
+ ext4_ext_get_actual_len(ex
) - 1) {
2857 if (ext4_ext_is_uninitialized(ex
))
2858 split_flag
= EXT4_EXT_MARK_UNINIT1
|
2859 EXT4_EXT_MARK_UNINIT2
;
2862 * Split the extent in two so that 'end' is the last
2863 * block in the first new extent. Also we should not
2864 * fail removing space due to ENOSPC so try to use
2865 * reserved block if that happens.
2867 err
= ext4_split_extent_at(handle
, inode
, path
,
2868 end
+ 1, split_flag
,
2870 EXT4_GET_BLOCKS_PRE_IO
|
2871 EXT4_GET_BLOCKS_METADATA_NOFAIL
);
2878 * We start scanning from right side, freeing all the blocks
2879 * after i_size and walking into the tree depth-wise.
2881 depth
= ext_depth(inode
);
2886 le16_to_cpu(path
[k
].p_hdr
->eh_entries
)+1;
2888 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
2891 ext4_journal_stop(handle
);
2894 path
[0].p_depth
= depth
;
2895 path
[0].p_hdr
= ext_inode_hdr(inode
);
2898 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0)) {
2905 while (i
>= 0 && err
== 0) {
2907 /* this is leaf block */
2908 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2909 &partial_cluster
, start
,
2911 /* root level has p_bh == NULL, brelse() eats this */
2912 brelse(path
[i
].p_bh
);
2913 path
[i
].p_bh
= NULL
;
2918 /* this is index block */
2919 if (!path
[i
].p_hdr
) {
2920 ext_debug("initialize header\n");
2921 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2924 if (!path
[i
].p_idx
) {
2925 /* this level hasn't been touched yet */
2926 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2927 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2928 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2930 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2932 /* we were already here, see at next index */
2936 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2937 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2939 if (ext4_ext_more_to_rm(path
+ i
)) {
2940 struct buffer_head
*bh
;
2941 /* go to the next level */
2942 ext_debug("move to level %d (block %llu)\n",
2943 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
2944 memset(path
+ i
+ 1, 0, sizeof(*path
));
2945 bh
= read_extent_tree_block(inode
,
2946 ext4_idx_pblock(path
[i
].p_idx
), depth
- i
- 1,
2949 /* should we reset i_size? */
2953 /* Yield here to deal with large extent trees.
2954 * Should be a no-op if we did IO above. */
2956 if (WARN_ON(i
+ 1 > depth
)) {
2960 path
[i
+ 1].p_bh
= bh
;
2962 /* save actual number of indexes since this
2963 * number is changed at the next iteration */
2964 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2967 /* we finished processing this index, go up */
2968 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2969 /* index is empty, remove it;
2970 * handle must be already prepared by the
2971 * truncatei_leaf() */
2972 err
= ext4_ext_rm_idx(handle
, inode
, path
, i
);
2974 /* root level has p_bh == NULL, brelse() eats this */
2975 brelse(path
[i
].p_bh
);
2976 path
[i
].p_bh
= NULL
;
2978 ext_debug("return to level %d\n", i
);
2982 trace_ext4_ext_remove_space_done(inode
, start
, end
, depth
,
2983 partial_cluster
, path
->p_hdr
->eh_entries
);
2985 /* If we still have something in the partial cluster and we have removed
2986 * even the first extent, then we should free the blocks in the partial
2987 * cluster as well. */
2988 if (partial_cluster
> 0 && path
->p_hdr
->eh_entries
== 0) {
2989 int flags
= get_default_free_blocks_flags(inode
);
2991 ext4_free_blocks(handle
, inode
, NULL
,
2992 EXT4_C2B(EXT4_SB(sb
), partial_cluster
),
2993 EXT4_SB(sb
)->s_cluster_ratio
, flags
);
2994 partial_cluster
= 0;
2997 /* TODO: flexible tree reduction should be here */
2998 if (path
->p_hdr
->eh_entries
== 0) {
3000 * truncate to zero freed all the tree,
3001 * so we need to correct eh_depth
3003 err
= ext4_ext_get_access(handle
, inode
, path
);
3005 ext_inode_hdr(inode
)->eh_depth
= 0;
3006 ext_inode_hdr(inode
)->eh_max
=
3007 cpu_to_le16(ext4_ext_space_root(inode
, 0));
3008 err
= ext4_ext_dirty(handle
, inode
, path
);
3012 ext4_ext_drop_refs(path
);
3014 if (err
== -EAGAIN
) {
3018 ext4_journal_stop(handle
);
3024 * called at mount time
3026 void ext4_ext_init(struct super_block
*sb
)
3029 * possible initialization would be here
3032 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
3033 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3034 printk(KERN_INFO
"EXT4-fs: file extents enabled"
3035 #ifdef AGGRESSIVE_TEST
3036 ", aggressive tests"
3038 #ifdef CHECK_BINSEARCH
3041 #ifdef EXTENTS_STATS
3046 #ifdef EXTENTS_STATS
3047 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
3048 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
3049 EXT4_SB(sb
)->s_ext_max
= 0;
3055 * called at umount time
3057 void ext4_ext_release(struct super_block
*sb
)
3059 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
3062 #ifdef EXTENTS_STATS
3063 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
3064 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3065 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3066 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
3067 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
3068 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3069 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
3074 static int ext4_zeroout_es(struct inode
*inode
, struct ext4_extent
*ex
)
3076 ext4_lblk_t ee_block
;
3077 ext4_fsblk_t ee_pblock
;
3078 unsigned int ee_len
;
3080 ee_block
= le32_to_cpu(ex
->ee_block
);
3081 ee_len
= ext4_ext_get_actual_len(ex
);
3082 ee_pblock
= ext4_ext_pblock(ex
);
3087 return ext4_es_insert_extent(inode
, ee_block
, ee_len
, ee_pblock
,
3088 EXTENT_STATUS_WRITTEN
);
3091 /* FIXME!! we need to try to merge to left or right after zero-out */
3092 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
3094 ext4_fsblk_t ee_pblock
;
3095 unsigned int ee_len
;
3098 ee_len
= ext4_ext_get_actual_len(ex
);
3099 ee_pblock
= ext4_ext_pblock(ex
);
3101 ret
= sb_issue_zeroout(inode
->i_sb
, ee_pblock
, ee_len
, GFP_NOFS
);
3109 * ext4_split_extent_at() splits an extent at given block.
3111 * @handle: the journal handle
3112 * @inode: the file inode
3113 * @path: the path to the extent
3114 * @split: the logical block where the extent is splitted.
3115 * @split_flags: indicates if the extent could be zeroout if split fails, and
3116 * the states(init or uninit) of new extents.
3117 * @flags: flags used to insert new extent to extent tree.
3120 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3121 * of which are deterimined by split_flag.
3123 * There are two cases:
3124 * a> the extent are splitted into two extent.
3125 * b> split is not needed, and just mark the extent.
3127 * return 0 on success.
3129 static int ext4_split_extent_at(handle_t
*handle
,
3130 struct inode
*inode
,
3131 struct ext4_ext_path
*path
,
3136 ext4_fsblk_t newblock
;
3137 ext4_lblk_t ee_block
;
3138 struct ext4_extent
*ex
, newex
, orig_ex
, zero_ex
;
3139 struct ext4_extent
*ex2
= NULL
;
3140 unsigned int ee_len
, depth
;
3143 BUG_ON((split_flag
& (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
)) ==
3144 (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
));
3146 ext_debug("ext4_split_extents_at: inode %lu, logical"
3147 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
3149 ext4_ext_show_leaf(inode
, path
);
3151 depth
= ext_depth(inode
);
3152 ex
= path
[depth
].p_ext
;
3153 ee_block
= le32_to_cpu(ex
->ee_block
);
3154 ee_len
= ext4_ext_get_actual_len(ex
);
3155 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
3157 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
3158 BUG_ON(!ext4_ext_is_uninitialized(ex
) &&
3159 split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3160 EXT4_EXT_MARK_UNINIT1
|
3161 EXT4_EXT_MARK_UNINIT2
));
3163 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3167 if (split
== ee_block
) {
3169 * case b: block @split is the block that the extent begins with
3170 * then we just change the state of the extent, and splitting
3173 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3174 ext4_ext_mark_uninitialized(ex
);
3176 ext4_ext_mark_initialized(ex
);
3178 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
3179 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3181 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3186 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
3187 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
3188 if (split_flag
& EXT4_EXT_MARK_UNINIT1
)
3189 ext4_ext_mark_uninitialized(ex
);
3192 * path may lead to new leaf, not to original leaf any more
3193 * after ext4_ext_insert_extent() returns,
3195 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3197 goto fix_extent_len
;
3200 ex2
->ee_block
= cpu_to_le32(split
);
3201 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
3202 ext4_ext_store_pblock(ex2
, newblock
);
3203 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3204 ext4_ext_mark_uninitialized(ex2
);
3206 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
3207 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3208 if (split_flag
& (EXT4_EXT_DATA_VALID1
|EXT4_EXT_DATA_VALID2
)) {
3209 if (split_flag
& EXT4_EXT_DATA_VALID1
) {
3210 err
= ext4_ext_zeroout(inode
, ex2
);
3211 zero_ex
.ee_block
= ex2
->ee_block
;
3212 zero_ex
.ee_len
= cpu_to_le16(
3213 ext4_ext_get_actual_len(ex2
));
3214 ext4_ext_store_pblock(&zero_ex
,
3215 ext4_ext_pblock(ex2
));
3217 err
= ext4_ext_zeroout(inode
, ex
);
3218 zero_ex
.ee_block
= ex
->ee_block
;
3219 zero_ex
.ee_len
= cpu_to_le16(
3220 ext4_ext_get_actual_len(ex
));
3221 ext4_ext_store_pblock(&zero_ex
,
3222 ext4_ext_pblock(ex
));
3225 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3226 zero_ex
.ee_block
= orig_ex
.ee_block
;
3227 zero_ex
.ee_len
= cpu_to_le16(
3228 ext4_ext_get_actual_len(&orig_ex
));
3229 ext4_ext_store_pblock(&zero_ex
,
3230 ext4_ext_pblock(&orig_ex
));
3234 goto fix_extent_len
;
3235 /* update the extent length and mark as initialized */
3236 ex
->ee_len
= cpu_to_le16(ee_len
);
3237 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3238 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3240 goto fix_extent_len
;
3242 /* update extent status tree */
3243 err
= ext4_zeroout_es(inode
, &zero_ex
);
3247 goto fix_extent_len
;
3250 ext4_ext_show_leaf(inode
, path
);
3254 ex
->ee_len
= orig_ex
.ee_len
;
3255 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3260 * ext4_split_extents() splits an extent and mark extent which is covered
3261 * by @map as split_flags indicates
3263 * It may result in splitting the extent into multiple extents (up to three)
3264 * There are three possibilities:
3265 * a> There is no split required
3266 * b> Splits in two extents: Split is happening at either end of the extent
3267 * c> Splits in three extents: Somone is splitting in middle of the extent
3270 static int ext4_split_extent(handle_t
*handle
,
3271 struct inode
*inode
,
3272 struct ext4_ext_path
*path
,
3273 struct ext4_map_blocks
*map
,
3277 ext4_lblk_t ee_block
;
3278 struct ext4_extent
*ex
;
3279 unsigned int ee_len
, depth
;
3282 int split_flag1
, flags1
;
3283 int allocated
= map
->m_len
;
3285 depth
= ext_depth(inode
);
3286 ex
= path
[depth
].p_ext
;
3287 ee_block
= le32_to_cpu(ex
->ee_block
);
3288 ee_len
= ext4_ext_get_actual_len(ex
);
3289 uninitialized
= ext4_ext_is_uninitialized(ex
);
3291 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
3292 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
;
3293 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
3295 split_flag1
|= EXT4_EXT_MARK_UNINIT1
|
3296 EXT4_EXT_MARK_UNINIT2
;
3297 if (split_flag
& EXT4_EXT_DATA_VALID2
)
3298 split_flag1
|= EXT4_EXT_DATA_VALID1
;
3299 err
= ext4_split_extent_at(handle
, inode
, path
,
3300 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
3304 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3307 * Update path is required because previous ext4_split_extent_at() may
3308 * result in split of original leaf or extent zeroout.
3310 ext4_ext_drop_refs(path
);
3311 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
, 0);
3313 return PTR_ERR(path
);
3314 depth
= ext_depth(inode
);
3315 ex
= path
[depth
].p_ext
;
3317 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
3318 (unsigned long) map
->m_lblk
);
3321 uninitialized
= ext4_ext_is_uninitialized(ex
);
3324 if (map
->m_lblk
>= ee_block
) {
3325 split_flag1
= split_flag
& EXT4_EXT_DATA_VALID2
;
3326 if (uninitialized
) {
3327 split_flag1
|= EXT4_EXT_MARK_UNINIT1
;
3328 split_flag1
|= split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3329 EXT4_EXT_MARK_UNINIT2
);
3331 err
= ext4_split_extent_at(handle
, inode
, path
,
3332 map
->m_lblk
, split_flag1
, flags
);
3337 ext4_ext_show_leaf(inode
, path
);
3339 return err
? err
: allocated
;
3343 * This function is called by ext4_ext_map_blocks() if someone tries to write
3344 * to an uninitialized extent. It may result in splitting the uninitialized
3345 * extent into multiple extents (up to three - one initialized and two
3347 * There are three possibilities:
3348 * a> There is no split required: Entire extent should be initialized
3349 * b> Splits in two extents: Write is happening at either end of the extent
3350 * c> Splits in three extents: Somone is writing in middle of the extent
3353 * - The extent pointed to by 'path' is uninitialized.
3354 * - The extent pointed to by 'path' contains a superset
3355 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3357 * Post-conditions on success:
3358 * - the returned value is the number of blocks beyond map->l_lblk
3359 * that are allocated and initialized.
3360 * It is guaranteed to be >= map->m_len.
3362 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3363 struct inode
*inode
,
3364 struct ext4_map_blocks
*map
,
3365 struct ext4_ext_path
*path
,
3368 struct ext4_sb_info
*sbi
;
3369 struct ext4_extent_header
*eh
;
3370 struct ext4_map_blocks split_map
;
3371 struct ext4_extent zero_ex
;
3372 struct ext4_extent
*ex
, *abut_ex
;
3373 ext4_lblk_t ee_block
, eof_block
;
3374 unsigned int ee_len
, depth
, map_len
= map
->m_len
;
3375 int allocated
= 0, max_zeroout
= 0;
3379 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3380 "block %llu, max_blocks %u\n", inode
->i_ino
,
3381 (unsigned long long)map
->m_lblk
, map_len
);
3383 sbi
= EXT4_SB(inode
->i_sb
);
3384 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3385 inode
->i_sb
->s_blocksize_bits
;
3386 if (eof_block
< map
->m_lblk
+ map_len
)
3387 eof_block
= map
->m_lblk
+ map_len
;
3389 depth
= ext_depth(inode
);
3390 eh
= path
[depth
].p_hdr
;
3391 ex
= path
[depth
].p_ext
;
3392 ee_block
= le32_to_cpu(ex
->ee_block
);
3393 ee_len
= ext4_ext_get_actual_len(ex
);
3396 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3398 /* Pre-conditions */
3399 BUG_ON(!ext4_ext_is_uninitialized(ex
));
3400 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3403 * Attempt to transfer newly initialized blocks from the currently
3404 * uninitialized extent to its neighbor. This is much cheaper
3405 * than an insertion followed by a merge as those involve costly
3406 * memmove() calls. Transferring to the left is the common case in
3407 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3408 * followed by append writes.
3410 * Limitations of the current logic:
3411 * - L1: we do not deal with writes covering the whole extent.
3412 * This would require removing the extent if the transfer
3414 * - L2: we only attempt to merge with an extent stored in the
3415 * same extent tree node.
3417 if ((map
->m_lblk
== ee_block
) &&
3418 /* See if we can merge left */
3419 (map_len
< ee_len
) && /*L1*/
3420 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L2*/
3421 ext4_lblk_t prev_lblk
;
3422 ext4_fsblk_t prev_pblk
, ee_pblk
;
3423 unsigned int prev_len
;
3426 prev_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3427 prev_len
= ext4_ext_get_actual_len(abut_ex
);
3428 prev_pblk
= ext4_ext_pblock(abut_ex
);
3429 ee_pblk
= ext4_ext_pblock(ex
);
3432 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3433 * upon those conditions:
3434 * - C1: abut_ex is initialized,
3435 * - C2: abut_ex is logically abutting ex,
3436 * - C3: abut_ex is physically abutting ex,
3437 * - C4: abut_ex can receive the additional blocks without
3438 * overflowing the (initialized) length limit.
3440 if ((!ext4_ext_is_uninitialized(abut_ex
)) && /*C1*/
3441 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3442 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3443 (prev_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3444 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3448 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3451 /* Shift the start of ex by 'map_len' blocks */
3452 ex
->ee_block
= cpu_to_le32(ee_block
+ map_len
);
3453 ext4_ext_store_pblock(ex
, ee_pblk
+ map_len
);
3454 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3455 ext4_ext_mark_uninitialized(ex
); /* Restore the flag */
3457 /* Extend abut_ex by 'map_len' blocks */
3458 abut_ex
->ee_len
= cpu_to_le16(prev_len
+ map_len
);
3460 /* Result: number of initialized blocks past m_lblk */
3461 allocated
= map_len
;
3463 } else if (((map
->m_lblk
+ map_len
) == (ee_block
+ ee_len
)) &&
3464 (map_len
< ee_len
) && /*L1*/
3465 ex
< EXT_LAST_EXTENT(eh
)) { /*L2*/
3466 /* See if we can merge right */
3467 ext4_lblk_t next_lblk
;
3468 ext4_fsblk_t next_pblk
, ee_pblk
;
3469 unsigned int next_len
;
3472 next_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3473 next_len
= ext4_ext_get_actual_len(abut_ex
);
3474 next_pblk
= ext4_ext_pblock(abut_ex
);
3475 ee_pblk
= ext4_ext_pblock(ex
);
3478 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3479 * upon those conditions:
3480 * - C1: abut_ex is initialized,
3481 * - C2: abut_ex is logically abutting ex,
3482 * - C3: abut_ex is physically abutting ex,
3483 * - C4: abut_ex can receive the additional blocks without
3484 * overflowing the (initialized) length limit.
3486 if ((!ext4_ext_is_uninitialized(abut_ex
)) && /*C1*/
3487 ((map
->m_lblk
+ map_len
) == next_lblk
) && /*C2*/
3488 ((ee_pblk
+ ee_len
) == next_pblk
) && /*C3*/
3489 (next_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3490 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3494 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3497 /* Shift the start of abut_ex by 'map_len' blocks */
3498 abut_ex
->ee_block
= cpu_to_le32(next_lblk
- map_len
);
3499 ext4_ext_store_pblock(abut_ex
, next_pblk
- map_len
);
3500 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3501 ext4_ext_mark_uninitialized(ex
); /* Restore the flag */
3503 /* Extend abut_ex by 'map_len' blocks */
3504 abut_ex
->ee_len
= cpu_to_le16(next_len
+ map_len
);
3506 /* Result: number of initialized blocks past m_lblk */
3507 allocated
= map_len
;
3511 /* Mark the block containing both extents as dirty */
3512 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3514 /* Update path to point to the right extent */
3515 path
[depth
].p_ext
= abut_ex
;
3518 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3520 WARN_ON(map
->m_lblk
< ee_block
);
3522 * It is safe to convert extent to initialized via explicit
3523 * zeroout only if extent is fully inside i_size or new_size.
3525 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3527 if (EXT4_EXT_MAY_ZEROOUT
& split_flag
)
3528 max_zeroout
= sbi
->s_extent_max_zeroout_kb
>>
3529 (inode
->i_sb
->s_blocksize_bits
- 10);
3531 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3532 if (max_zeroout
&& (ee_len
<= max_zeroout
)) {
3533 err
= ext4_ext_zeroout(inode
, ex
);
3536 zero_ex
.ee_block
= ex
->ee_block
;
3537 zero_ex
.ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
));
3538 ext4_ext_store_pblock(&zero_ex
, ext4_ext_pblock(ex
));
3540 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3543 ext4_ext_mark_initialized(ex
);
3544 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3545 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3551 * 1. split the extent into three extents.
3552 * 2. split the extent into two extents, zeroout the first half.
3553 * 3. split the extent into two extents, zeroout the second half.
3554 * 4. split the extent into two extents with out zeroout.
3556 split_map
.m_lblk
= map
->m_lblk
;
3557 split_map
.m_len
= map
->m_len
;
3559 if (max_zeroout
&& (allocated
> map
->m_len
)) {
3560 if (allocated
<= max_zeroout
) {
3563 cpu_to_le32(map
->m_lblk
);
3564 zero_ex
.ee_len
= cpu_to_le16(allocated
);
3565 ext4_ext_store_pblock(&zero_ex
,
3566 ext4_ext_pblock(ex
) + map
->m_lblk
- ee_block
);
3567 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3570 split_map
.m_lblk
= map
->m_lblk
;
3571 split_map
.m_len
= allocated
;
3572 } else if (map
->m_lblk
- ee_block
+ map
->m_len
< max_zeroout
) {
3574 if (map
->m_lblk
!= ee_block
) {
3575 zero_ex
.ee_block
= ex
->ee_block
;
3576 zero_ex
.ee_len
= cpu_to_le16(map
->m_lblk
-
3578 ext4_ext_store_pblock(&zero_ex
,
3579 ext4_ext_pblock(ex
));
3580 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3585 split_map
.m_lblk
= ee_block
;
3586 split_map
.m_len
= map
->m_lblk
- ee_block
+ map
->m_len
;
3587 allocated
= map
->m_len
;
3591 allocated
= ext4_split_extent(handle
, inode
, path
,
3592 &split_map
, split_flag
, flags
);
3597 /* If we have gotten a failure, don't zero out status tree */
3599 err
= ext4_zeroout_es(inode
, &zero_ex
);
3600 return err
? err
: allocated
;
3604 * This function is called by ext4_ext_map_blocks() from
3605 * ext4_get_blocks_dio_write() when DIO to write
3606 * to an uninitialized extent.
3608 * Writing to an uninitialized extent may result in splitting the uninitialized
3609 * extent into multiple initialized/uninitialized extents (up to three)
3610 * There are three possibilities:
3611 * a> There is no split required: Entire extent should be uninitialized
3612 * b> Splits in two extents: Write is happening at either end of the extent
3613 * c> Splits in three extents: Somone is writing in middle of the extent
3615 * This works the same way in the case of initialized -> unwritten conversion.
3617 * One of more index blocks maybe needed if the extent tree grow after
3618 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3619 * complete, we need to split the uninitialized extent before DIO submit
3620 * the IO. The uninitialized extent called at this time will be split
3621 * into three uninitialized extent(at most). After IO complete, the part
3622 * being filled will be convert to initialized by the end_io callback function
3623 * via ext4_convert_unwritten_extents().
3625 * Returns the size of uninitialized extent to be written on success.
3627 static int ext4_split_convert_extents(handle_t
*handle
,
3628 struct inode
*inode
,
3629 struct ext4_map_blocks
*map
,
3630 struct ext4_ext_path
*path
,
3633 ext4_lblk_t eof_block
;
3634 ext4_lblk_t ee_block
;
3635 struct ext4_extent
*ex
;
3636 unsigned int ee_len
;
3637 int split_flag
= 0, depth
;
3639 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3640 __func__
, inode
->i_ino
,
3641 (unsigned long long)map
->m_lblk
, map
->m_len
);
3643 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3644 inode
->i_sb
->s_blocksize_bits
;
3645 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3646 eof_block
= map
->m_lblk
+ map
->m_len
;
3648 * It is safe to convert extent to initialized via explicit
3649 * zeroout only if extent is fully insde i_size or new_size.
3651 depth
= ext_depth(inode
);
3652 ex
= path
[depth
].p_ext
;
3653 ee_block
= le32_to_cpu(ex
->ee_block
);
3654 ee_len
= ext4_ext_get_actual_len(ex
);
3656 /* Convert to unwritten */
3657 if (flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
) {
3658 split_flag
|= EXT4_EXT_DATA_VALID1
;
3659 /* Convert to initialized */
3660 } else if (flags
& EXT4_GET_BLOCKS_CONVERT
) {
3661 split_flag
|= ee_block
+ ee_len
<= eof_block
?
3662 EXT4_EXT_MAY_ZEROOUT
: 0;
3663 split_flag
|= (EXT4_EXT_MARK_UNINIT2
| EXT4_EXT_DATA_VALID2
);
3665 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3666 return ext4_split_extent(handle
, inode
, path
, map
, split_flag
, flags
);
3669 static int ext4_convert_initialized_extents(handle_t
*handle
,
3670 struct inode
*inode
,
3671 struct ext4_map_blocks
*map
,
3672 struct ext4_ext_path
*path
)
3674 struct ext4_extent
*ex
;
3675 ext4_lblk_t ee_block
;
3676 unsigned int ee_len
;
3680 depth
= ext_depth(inode
);
3681 ex
= path
[depth
].p_ext
;
3682 ee_block
= le32_to_cpu(ex
->ee_block
);
3683 ee_len
= ext4_ext_get_actual_len(ex
);
3685 ext_debug("%s: inode %lu, logical"
3686 "block %llu, max_blocks %u\n", __func__
, inode
->i_ino
,
3687 (unsigned long long)ee_block
, ee_len
);
3689 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3690 err
= ext4_split_convert_extents(handle
, inode
, map
, path
,
3691 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
);
3694 ext4_ext_drop_refs(path
);
3695 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
, 0);
3697 err
= PTR_ERR(path
);
3700 depth
= ext_depth(inode
);
3701 ex
= path
[depth
].p_ext
;
3703 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
3704 (unsigned long) map
->m_lblk
);
3710 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3713 /* first mark the extent as uninitialized */
3714 ext4_ext_mark_uninitialized(ex
);
3716 /* note: ext4_ext_correct_indexes() isn't needed here because
3717 * borders are not changed
3719 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3721 /* Mark modified extent as dirty */
3722 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3724 ext4_ext_show_leaf(inode
, path
);
3729 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3730 struct inode
*inode
,
3731 struct ext4_map_blocks
*map
,
3732 struct ext4_ext_path
*path
)
3734 struct ext4_extent
*ex
;
3735 ext4_lblk_t ee_block
;
3736 unsigned int ee_len
;
3740 depth
= ext_depth(inode
);
3741 ex
= path
[depth
].p_ext
;
3742 ee_block
= le32_to_cpu(ex
->ee_block
);
3743 ee_len
= ext4_ext_get_actual_len(ex
);
3745 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3746 "block %llu, max_blocks %u\n", inode
->i_ino
,
3747 (unsigned long long)ee_block
, ee_len
);
3749 /* If extent is larger than requested it is a clear sign that we still
3750 * have some extent state machine issues left. So extent_split is still
3752 * TODO: Once all related issues will be fixed this situation should be
3755 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3757 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3758 " len %u; IO logical block %llu, len %u\n",
3759 inode
->i_ino
, (unsigned long long)ee_block
, ee_len
,
3760 (unsigned long long)map
->m_lblk
, map
->m_len
);
3762 err
= ext4_split_convert_extents(handle
, inode
, map
, path
,
3763 EXT4_GET_BLOCKS_CONVERT
);
3766 ext4_ext_drop_refs(path
);
3767 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
, 0);
3769 err
= PTR_ERR(path
);
3772 depth
= ext_depth(inode
);
3773 ex
= path
[depth
].p_ext
;
3776 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3779 /* first mark the extent as initialized */
3780 ext4_ext_mark_initialized(ex
);
3782 /* note: ext4_ext_correct_indexes() isn't needed here because
3783 * borders are not changed
3785 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3787 /* Mark modified extent as dirty */
3788 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3790 ext4_ext_show_leaf(inode
, path
);
3794 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3795 sector_t block
, int count
)
3798 for (i
= 0; i
< count
; i
++)
3799 unmap_underlying_metadata(bdev
, block
+ i
);
3803 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3805 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3807 struct ext4_ext_path
*path
,
3811 struct ext4_extent_header
*eh
;
3812 struct ext4_extent
*last_ex
;
3814 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3817 depth
= ext_depth(inode
);
3818 eh
= path
[depth
].p_hdr
;
3821 * We're going to remove EOFBLOCKS_FL entirely in future so we
3822 * do not care for this case anymore. Simply remove the flag
3823 * if there are no extents.
3825 if (unlikely(!eh
->eh_entries
))
3827 last_ex
= EXT_LAST_EXTENT(eh
);
3829 * We should clear the EOFBLOCKS_FL flag if we are writing the
3830 * last block in the last extent in the file. We test this by
3831 * first checking to see if the caller to
3832 * ext4_ext_get_blocks() was interested in the last block (or
3833 * a block beyond the last block) in the current extent. If
3834 * this turns out to be false, we can bail out from this
3835 * function immediately.
3837 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3838 ext4_ext_get_actual_len(last_ex
))
3841 * If the caller does appear to be planning to write at or
3842 * beyond the end of the current extent, we then test to see
3843 * if the current extent is the last extent in the file, by
3844 * checking to make sure it was reached via the rightmost node
3845 * at each level of the tree.
3847 for (i
= depth
-1; i
>= 0; i
--)
3848 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3851 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3852 return ext4_mark_inode_dirty(handle
, inode
);
3856 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3858 * Return 1 if there is a delalloc block in the range, otherwise 0.
3860 int ext4_find_delalloc_range(struct inode
*inode
,
3861 ext4_lblk_t lblk_start
,
3862 ext4_lblk_t lblk_end
)
3864 struct extent_status es
;
3866 ext4_es_find_delayed_extent_range(inode
, lblk_start
, lblk_end
, &es
);
3868 return 0; /* there is no delay extent in this tree */
3869 else if (es
.es_lblk
<= lblk_start
&&
3870 lblk_start
< es
.es_lblk
+ es
.es_len
)
3872 else if (lblk_start
<= es
.es_lblk
&& es
.es_lblk
<= lblk_end
)
3878 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
)
3880 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3881 ext4_lblk_t lblk_start
, lblk_end
;
3882 lblk_start
= EXT4_LBLK_CMASK(sbi
, lblk
);
3883 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3885 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
);
3889 * Determines how many complete clusters (out of those specified by the 'map')
3890 * are under delalloc and were reserved quota for.
3891 * This function is called when we are writing out the blocks that were
3892 * originally written with their allocation delayed, but then the space was
3893 * allocated using fallocate() before the delayed allocation could be resolved.
3894 * The cases to look for are:
3895 * ('=' indicated delayed allocated blocks
3896 * '-' indicates non-delayed allocated blocks)
3897 * (a) partial clusters towards beginning and/or end outside of allocated range
3898 * are not delalloc'ed.
3900 * |----c---=|====c====|====c====|===-c----|
3901 * |++++++ allocated ++++++|
3902 * ==> 4 complete clusters in above example
3904 * (b) partial cluster (outside of allocated range) towards either end is
3905 * marked for delayed allocation. In this case, we will exclude that
3908 * |----====c========|========c========|
3909 * |++++++ allocated ++++++|
3910 * ==> 1 complete clusters in above example
3913 * |================c================|
3914 * |++++++ allocated ++++++|
3915 * ==> 0 complete clusters in above example
3917 * The ext4_da_update_reserve_space will be called only if we
3918 * determine here that there were some "entire" clusters that span
3919 * this 'allocated' range.
3920 * In the non-bigalloc case, this function will just end up returning num_blks
3921 * without ever calling ext4_find_delalloc_range.
3924 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3925 unsigned int num_blks
)
3927 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3928 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3929 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3930 unsigned int allocated_clusters
= 0;
3932 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3933 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3935 /* max possible clusters for this allocation */
3936 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3938 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3940 /* Check towards left side */
3941 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
);
3943 lblk_from
= EXT4_LBLK_CMASK(sbi
, lblk_start
);
3944 lblk_to
= lblk_from
+ c_offset
- 1;
3946 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3947 allocated_clusters
--;
3950 /* Now check towards right. */
3951 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
+ num_blks
);
3952 if (allocated_clusters
&& c_offset
) {
3953 lblk_from
= lblk_start
+ num_blks
;
3954 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3956 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3957 allocated_clusters
--;
3960 return allocated_clusters
;
3964 ext4_ext_convert_initialized_extent(handle_t
*handle
, struct inode
*inode
,
3965 struct ext4_map_blocks
*map
,
3966 struct ext4_ext_path
*path
, int flags
,
3967 unsigned int allocated
, ext4_fsblk_t newblock
)
3973 * Make sure that the extent is no bigger than we support with
3974 * uninitialized extent
3976 if (map
->m_len
> EXT_UNINIT_MAX_LEN
)
3977 map
->m_len
= EXT_UNINIT_MAX_LEN
/ 2;
3979 ret
= ext4_convert_initialized_extents(handle
, inode
, map
,
3982 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3983 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
3987 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3988 if (allocated
> map
->m_len
)
3989 allocated
= map
->m_len
;
3990 map
->m_len
= allocated
;
3992 return err
? err
: allocated
;
3996 ext4_ext_handle_uninitialized_extents(handle_t
*handle
, struct inode
*inode
,
3997 struct ext4_map_blocks
*map
,
3998 struct ext4_ext_path
*path
, int flags
,
3999 unsigned int allocated
, ext4_fsblk_t newblock
)
4003 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
4005 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
4006 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4007 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
4009 ext4_ext_show_leaf(inode
, path
);
4012 * When writing into uninitialized space, we should not fail to
4013 * allocate metadata blocks for the new extent block if needed.
4015 flags
|= EXT4_GET_BLOCKS_METADATA_NOFAIL
;
4017 trace_ext4_ext_handle_uninitialized_extents(inode
, map
, flags
,
4018 allocated
, newblock
);
4020 /* get_block() before submit the IO, split the extent */
4021 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
4022 ret
= ext4_split_convert_extents(handle
, inode
, map
,
4023 path
, flags
| EXT4_GET_BLOCKS_CONVERT
);
4027 * Flag the inode(non aio case) or end_io struct (aio case)
4028 * that this IO needs to conversion to written when IO is
4032 ext4_set_io_unwritten_flag(inode
, io
);
4034 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
4035 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4036 if (ext4_should_dioread_nolock(inode
))
4037 map
->m_flags
|= EXT4_MAP_UNINIT
;
4040 /* IO end_io complete, convert the filled extent to written */
4041 if ((flags
& EXT4_GET_BLOCKS_CONVERT
)) {
4042 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
, map
,
4045 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4046 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4050 map
->m_flags
|= EXT4_MAP_MAPPED
;
4051 map
->m_pblk
= newblock
;
4052 if (allocated
> map
->m_len
)
4053 allocated
= map
->m_len
;
4054 map
->m_len
= allocated
;
4057 /* buffered IO case */
4059 * repeat fallocate creation request
4060 * we already have an unwritten extent
4062 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) {
4063 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4067 /* buffered READ or buffered write_begin() lookup */
4068 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4070 * We have blocks reserved already. We
4071 * return allocated blocks so that delalloc
4072 * won't do block reservation for us. But
4073 * the buffer head will be unmapped so that
4074 * a read from the block returns 0s.
4076 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4080 /* buffered write, writepage time, convert*/
4081 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, path
, flags
);
4083 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4090 map
->m_flags
|= EXT4_MAP_NEW
;
4092 * if we allocated more blocks than requested
4093 * we need to make sure we unmap the extra block
4094 * allocated. The actual needed block will get
4095 * unmapped later when we find the buffer_head marked
4098 if (allocated
> map
->m_len
) {
4099 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
4100 newblock
+ map
->m_len
,
4101 allocated
- map
->m_len
);
4102 allocated
= map
->m_len
;
4104 map
->m_len
= allocated
;
4107 * If we have done fallocate with the offset that is already
4108 * delayed allocated, we would have block reservation
4109 * and quota reservation done in the delayed write path.
4110 * But fallocate would have already updated quota and block
4111 * count for this offset. So cancel these reservation
4113 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4114 unsigned int reserved_clusters
;
4115 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4116 map
->m_lblk
, map
->m_len
);
4117 if (reserved_clusters
)
4118 ext4_da_update_reserve_space(inode
,
4124 map
->m_flags
|= EXT4_MAP_MAPPED
;
4125 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
4126 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
4132 if (allocated
> map
->m_len
)
4133 allocated
= map
->m_len
;
4134 ext4_ext_show_leaf(inode
, path
);
4135 map
->m_pblk
= newblock
;
4136 map
->m_len
= allocated
;
4138 return err
? err
: allocated
;
4142 * get_implied_cluster_alloc - check to see if the requested
4143 * allocation (in the map structure) overlaps with a cluster already
4144 * allocated in an extent.
4145 * @sb The filesystem superblock structure
4146 * @map The requested lblk->pblk mapping
4147 * @ex The extent structure which might contain an implied
4148 * cluster allocation
4150 * This function is called by ext4_ext_map_blocks() after we failed to
4151 * find blocks that were already in the inode's extent tree. Hence,
4152 * we know that the beginning of the requested region cannot overlap
4153 * the extent from the inode's extent tree. There are three cases we
4154 * want to catch. The first is this case:
4156 * |--- cluster # N--|
4157 * |--- extent ---| |---- requested region ---|
4160 * The second case that we need to test for is this one:
4162 * |--------- cluster # N ----------------|
4163 * |--- requested region --| |------- extent ----|
4164 * |=======================|
4166 * The third case is when the requested region lies between two extents
4167 * within the same cluster:
4168 * |------------- cluster # N-------------|
4169 * |----- ex -----| |---- ex_right ----|
4170 * |------ requested region ------|
4171 * |================|
4173 * In each of the above cases, we need to set the map->m_pblk and
4174 * map->m_len so it corresponds to the return the extent labelled as
4175 * "|====|" from cluster #N, since it is already in use for data in
4176 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4177 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4178 * as a new "allocated" block region. Otherwise, we will return 0 and
4179 * ext4_ext_map_blocks() will then allocate one or more new clusters
4180 * by calling ext4_mb_new_blocks().
4182 static int get_implied_cluster_alloc(struct super_block
*sb
,
4183 struct ext4_map_blocks
*map
,
4184 struct ext4_extent
*ex
,
4185 struct ext4_ext_path
*path
)
4187 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
4188 ext4_lblk_t c_offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4189 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
4190 ext4_lblk_t rr_cluster_start
;
4191 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4192 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4193 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
4195 /* The extent passed in that we are trying to match */
4196 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
4197 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
4199 /* The requested region passed into ext4_map_blocks() */
4200 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
4202 if ((rr_cluster_start
== ex_cluster_end
) ||
4203 (rr_cluster_start
== ex_cluster_start
)) {
4204 if (rr_cluster_start
== ex_cluster_end
)
4205 ee_start
+= ee_len
- 1;
4206 map
->m_pblk
= EXT4_PBLK_CMASK(sbi
, ee_start
) + c_offset
;
4207 map
->m_len
= min(map
->m_len
,
4208 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
4210 * Check for and handle this case:
4212 * |--------- cluster # N-------------|
4213 * |------- extent ----|
4214 * |--- requested region ---|
4218 if (map
->m_lblk
< ee_block
)
4219 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
4222 * Check for the case where there is already another allocated
4223 * block to the right of 'ex' but before the end of the cluster.
4225 * |------------- cluster # N-------------|
4226 * |----- ex -----| |---- ex_right ----|
4227 * |------ requested region ------|
4228 * |================|
4230 if (map
->m_lblk
> ee_block
) {
4231 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
4232 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
4235 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
4239 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
4245 * Block allocation/map/preallocation routine for extents based files
4248 * Need to be called with
4249 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4250 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4252 * return > 0, number of of blocks already mapped/allocated
4253 * if create == 0 and these are pre-allocated blocks
4254 * buffer head is unmapped
4255 * otherwise blocks are mapped
4257 * return = 0, if plain look up failed (blocks have not been allocated)
4258 * buffer head is unmapped
4260 * return < 0, error case.
4262 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
4263 struct ext4_map_blocks
*map
, int flags
)
4265 struct ext4_ext_path
*path
= NULL
;
4266 struct ext4_extent newex
, *ex
, *ex2
;
4267 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4268 ext4_fsblk_t newblock
= 0;
4269 int free_on_err
= 0, err
= 0, depth
, ret
;
4270 unsigned int allocated
= 0, offset
= 0;
4271 unsigned int allocated_clusters
= 0;
4272 struct ext4_allocation_request ar
;
4273 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
4274 ext4_lblk_t cluster_offset
;
4275 int set_unwritten
= 0;
4277 ext_debug("blocks %u/%u requested for inode %lu\n",
4278 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
4279 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
4281 /* find extent for this block */
4282 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, NULL
, 0);
4284 err
= PTR_ERR(path
);
4289 depth
= ext_depth(inode
);
4292 * consistent leaf must not be empty;
4293 * this situation is possible, though, _during_ tree modification;
4294 * this is why assert can't be put in ext4_ext_find_extent()
4296 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
4297 EXT4_ERROR_INODE(inode
, "bad extent address "
4298 "lblock: %lu, depth: %d pblock %lld",
4299 (unsigned long) map
->m_lblk
, depth
,
4300 path
[depth
].p_block
);
4305 ex
= path
[depth
].p_ext
;
4307 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4308 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4309 unsigned short ee_len
;
4313 * Uninitialized extents are treated as holes, except that
4314 * we split out initialized portions during a write.
4316 ee_len
= ext4_ext_get_actual_len(ex
);
4318 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
4320 /* if found extent covers block, simply return it */
4321 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
4322 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
4323 /* number of remaining blocks in the extent */
4324 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
4325 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
4326 ee_block
, ee_len
, newblock
);
4329 * If the extent is initialized check whether the
4330 * caller wants to convert it to unwritten.
4332 if ((!ext4_ext_is_uninitialized(ex
)) &&
4333 (flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
)) {
4334 allocated
= ext4_ext_convert_initialized_extent(
4335 handle
, inode
, map
, path
, flags
,
4336 allocated
, newblock
);
4338 } else if (!ext4_ext_is_uninitialized(ex
))
4341 ret
= ext4_ext_handle_uninitialized_extents(
4342 handle
, inode
, map
, path
, flags
,
4343 allocated
, newblock
);
4352 if ((sbi
->s_cluster_ratio
> 1) &&
4353 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
4354 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4357 * requested block isn't allocated yet;
4358 * we couldn't try to create block if create flag is zero
4360 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4362 * put just found gap into cache to speed up
4363 * subsequent requests
4365 if ((flags
& EXT4_GET_BLOCKS_NO_PUT_HOLE
) == 0)
4366 ext4_ext_put_gap_in_cache(inode
, path
, map
->m_lblk
);
4371 * Okay, we need to do block allocation.
4373 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
4374 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
4375 cluster_offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4378 * If we are doing bigalloc, check to see if the extent returned
4379 * by ext4_ext_find_extent() implies a cluster we can use.
4381 if (cluster_offset
&& ex
&&
4382 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
4383 ar
.len
= allocated
= map
->m_len
;
4384 newblock
= map
->m_pblk
;
4385 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4386 goto got_allocated_blocks
;
4389 /* find neighbour allocated blocks */
4390 ar
.lleft
= map
->m_lblk
;
4391 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
4394 ar
.lright
= map
->m_lblk
;
4396 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
4400 /* Check if the extent after searching to the right implies a
4401 * cluster we can use. */
4402 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
4403 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
4404 ar
.len
= allocated
= map
->m_len
;
4405 newblock
= map
->m_pblk
;
4406 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4407 goto got_allocated_blocks
;
4411 * See if request is beyond maximum number of blocks we can have in
4412 * a single extent. For an initialized extent this limit is
4413 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4414 * EXT_UNINIT_MAX_LEN.
4416 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
4417 !(flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4418 map
->m_len
= EXT_INIT_MAX_LEN
;
4419 else if (map
->m_len
> EXT_UNINIT_MAX_LEN
&&
4420 (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4421 map
->m_len
= EXT_UNINIT_MAX_LEN
;
4423 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4424 newex
.ee_len
= cpu_to_le16(map
->m_len
);
4425 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
4427 allocated
= ext4_ext_get_actual_len(&newex
);
4429 allocated
= map
->m_len
;
4431 /* allocate new block */
4433 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
4434 ar
.logical
= map
->m_lblk
;
4436 * We calculate the offset from the beginning of the cluster
4437 * for the logical block number, since when we allocate a
4438 * physical cluster, the physical block should start at the
4439 * same offset from the beginning of the cluster. This is
4440 * needed so that future calls to get_implied_cluster_alloc()
4443 offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4444 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4446 ar
.logical
-= offset
;
4447 if (S_ISREG(inode
->i_mode
))
4448 ar
.flags
= EXT4_MB_HINT_DATA
;
4450 /* disable in-core preallocation for non-regular files */
4452 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4453 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4454 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4457 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4458 ar
.goal
, newblock
, allocated
);
4460 allocated_clusters
= ar
.len
;
4461 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4462 if (ar
.len
> allocated
)
4465 got_allocated_blocks
:
4466 /* try to insert new extent into found leaf and return */
4467 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4468 newex
.ee_len
= cpu_to_le16(ar
.len
);
4469 /* Mark uninitialized */
4470 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
){
4471 ext4_ext_mark_uninitialized(&newex
);
4472 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4474 * io_end structure was created for every IO write to an
4475 * uninitialized extent. To avoid unnecessary conversion,
4476 * here we flag the IO that really needs the conversion.
4477 * For non asycn direct IO case, flag the inode state
4478 * that we need to perform conversion when IO is done.
4480 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
))
4482 if (ext4_should_dioread_nolock(inode
))
4483 map
->m_flags
|= EXT4_MAP_UNINIT
;
4487 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4488 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4491 err
= ext4_ext_insert_extent(handle
, inode
, path
,
4494 if (!err
&& set_unwritten
) {
4496 ext4_set_io_unwritten_flag(inode
, io
);
4498 ext4_set_inode_state(inode
,
4499 EXT4_STATE_DIO_UNWRITTEN
);
4502 if (err
&& free_on_err
) {
4503 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4504 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4505 /* free data blocks we just allocated */
4506 /* not a good idea to call discard here directly,
4507 * but otherwise we'd need to call it every free() */
4508 ext4_discard_preallocations(inode
);
4509 ext4_free_blocks(handle
, inode
, NULL
, newblock
,
4510 EXT4_C2B(sbi
, allocated_clusters
), fb_flags
);
4514 /* previous routine could use block we allocated */
4515 newblock
= ext4_ext_pblock(&newex
);
4516 allocated
= ext4_ext_get_actual_len(&newex
);
4517 if (allocated
> map
->m_len
)
4518 allocated
= map
->m_len
;
4519 map
->m_flags
|= EXT4_MAP_NEW
;
4522 * Update reserved blocks/metadata blocks after successful
4523 * block allocation which had been deferred till now.
4525 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4526 unsigned int reserved_clusters
;
4528 * Check how many clusters we had reserved this allocated range
4530 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4531 map
->m_lblk
, allocated
);
4532 if (map
->m_flags
& EXT4_MAP_FROM_CLUSTER
) {
4533 if (reserved_clusters
) {
4535 * We have clusters reserved for this range.
4536 * But since we are not doing actual allocation
4537 * and are simply using blocks from previously
4538 * allocated cluster, we should release the
4539 * reservation and not claim quota.
4541 ext4_da_update_reserve_space(inode
,
4542 reserved_clusters
, 0);
4545 BUG_ON(allocated_clusters
< reserved_clusters
);
4546 if (reserved_clusters
< allocated_clusters
) {
4547 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4548 int reservation
= allocated_clusters
-
4551 * It seems we claimed few clusters outside of
4552 * the range of this allocation. We should give
4553 * it back to the reservation pool. This can
4554 * happen in the following case:
4556 * * Suppose s_cluster_ratio is 4 (i.e., each
4557 * cluster has 4 blocks. Thus, the clusters
4558 * are [0-3],[4-7],[8-11]...
4559 * * First comes delayed allocation write for
4560 * logical blocks 10 & 11. Since there were no
4561 * previous delayed allocated blocks in the
4562 * range [8-11], we would reserve 1 cluster
4564 * * Next comes write for logical blocks 3 to 8.
4565 * In this case, we will reserve 2 clusters
4566 * (for [0-3] and [4-7]; and not for [8-11] as
4567 * that range has a delayed allocated blocks.
4568 * Thus total reserved clusters now becomes 3.
4569 * * Now, during the delayed allocation writeout
4570 * time, we will first write blocks [3-8] and
4571 * allocate 3 clusters for writing these
4572 * blocks. Also, we would claim all these
4573 * three clusters above.
4574 * * Now when we come here to writeout the
4575 * blocks [10-11], we would expect to claim
4576 * the reservation of 1 cluster we had made
4577 * (and we would claim it since there are no
4578 * more delayed allocated blocks in the range
4579 * [8-11]. But our reserved cluster count had
4580 * already gone to 0.
4582 * Thus, at the step 4 above when we determine
4583 * that there are still some unwritten delayed
4584 * allocated blocks outside of our current
4585 * block range, we should increment the
4586 * reserved clusters count so that when the
4587 * remaining blocks finally gets written, we
4590 dquot_reserve_block(inode
,
4591 EXT4_C2B(sbi
, reservation
));
4592 spin_lock(&ei
->i_block_reservation_lock
);
4593 ei
->i_reserved_data_blocks
+= reservation
;
4594 spin_unlock(&ei
->i_block_reservation_lock
);
4597 * We will claim quota for all newly allocated blocks.
4598 * We're updating the reserved space *after* the
4599 * correction above so we do not accidentally free
4600 * all the metadata reservation because we might
4601 * actually need it later on.
4603 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4609 * Cache the extent and update transaction to commit on fdatasync only
4610 * when it is _not_ an uninitialized extent.
4612 if ((flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) == 0)
4613 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4615 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4617 if (allocated
> map
->m_len
)
4618 allocated
= map
->m_len
;
4619 ext4_ext_show_leaf(inode
, path
);
4620 map
->m_flags
|= EXT4_MAP_MAPPED
;
4621 map
->m_pblk
= newblock
;
4622 map
->m_len
= allocated
;
4625 ext4_ext_drop_refs(path
);
4629 trace_ext4_ext_map_blocks_exit(inode
, flags
, map
,
4630 err
? err
: allocated
);
4631 ext4_es_lru_add(inode
);
4632 return err
? err
: allocated
;
4635 void ext4_ext_truncate(handle_t
*handle
, struct inode
*inode
)
4637 struct super_block
*sb
= inode
->i_sb
;
4638 ext4_lblk_t last_block
;
4642 * TODO: optimization is possible here.
4643 * Probably we need not scan at all,
4644 * because page truncation is enough.
4647 /* we have to know where to truncate from in crash case */
4648 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4649 ext4_mark_inode_dirty(handle
, inode
);
4651 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4652 >> EXT4_BLOCK_SIZE_BITS(sb
);
4654 err
= ext4_es_remove_extent(inode
, last_block
,
4655 EXT_MAX_BLOCKS
- last_block
);
4656 if (err
== -ENOMEM
) {
4658 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
4662 ext4_std_error(inode
->i_sb
, err
);
4665 err
= ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4666 ext4_std_error(inode
->i_sb
, err
);
4669 static int ext4_alloc_file_blocks(struct file
*file
, ext4_lblk_t offset
,
4670 ext4_lblk_t len
, int flags
, int mode
)
4672 struct inode
*inode
= file_inode(file
);
4677 struct ext4_map_blocks map
;
4678 unsigned int credits
;
4680 map
.m_lblk
= offset
;
4682 * Don't normalize the request if it can fit in one extent so
4683 * that it doesn't get unnecessarily split into multiple
4686 if (len
<= EXT_UNINIT_MAX_LEN
)
4687 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4690 * credits to insert 1 extent into extent tree
4692 credits
= ext4_chunk_trans_blocks(inode
, len
);
4695 while (ret
>= 0 && ret
< len
) {
4696 map
.m_lblk
= map
.m_lblk
+ ret
;
4697 map
.m_len
= len
= len
- ret
;
4698 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
4700 if (IS_ERR(handle
)) {
4701 ret
= PTR_ERR(handle
);
4704 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4706 ext4_debug("inode #%lu: block %u: len %u: "
4707 "ext4_ext_map_blocks returned %d",
4708 inode
->i_ino
, map
.m_lblk
,
4710 ext4_mark_inode_dirty(handle
, inode
);
4711 ret2
= ext4_journal_stop(handle
);
4714 ret2
= ext4_journal_stop(handle
);
4718 if (ret
== -ENOSPC
&&
4719 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4724 return ret
> 0 ? ret2
: ret
;
4727 static long ext4_zero_range(struct file
*file
, loff_t offset
,
4728 loff_t len
, int mode
)
4730 struct inode
*inode
= file_inode(file
);
4731 handle_t
*handle
= NULL
;
4732 unsigned int max_blocks
;
4733 loff_t new_size
= 0;
4739 struct address_space
*mapping
= inode
->i_mapping
;
4740 unsigned int blkbits
= inode
->i_blkbits
;
4742 trace_ext4_zero_range(inode
, offset
, len
, mode
);
4744 if (!S_ISREG(inode
->i_mode
))
4748 * Write out all dirty pages to avoid race conditions
4749 * Then release them.
4751 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4752 ret
= filemap_write_and_wait_range(mapping
, offset
,
4759 * Round up offset. This is not fallocate, we neet to zero out
4760 * blocks, so convert interior block aligned part of the range to
4761 * unwritten and possibly manually zero out unaligned parts of the
4764 start
= round_up(offset
, 1 << blkbits
);
4765 end
= round_down((offset
+ len
), 1 << blkbits
);
4767 if (start
< offset
|| end
> offset
+ len
)
4769 partial
= (offset
+ len
) & ((1 << blkbits
) - 1);
4771 lblk
= start
>> blkbits
;
4772 max_blocks
= (end
>> blkbits
);
4773 if (max_blocks
< lblk
)
4778 flags
= EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
|
4779 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
;
4780 if (mode
& FALLOC_FL_KEEP_SIZE
)
4781 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4783 mutex_lock(&inode
->i_mutex
);
4786 * Indirect files do not support unwritten extnets
4788 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4793 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4794 offset
+ len
> i_size_read(inode
)) {
4795 new_size
= offset
+ len
;
4796 ret
= inode_newsize_ok(inode
, new_size
);
4800 * If we have a partial block after EOF we have to allocate
4807 if (max_blocks
> 0) {
4809 /* Now release the pages and zero block aligned part of pages*/
4810 truncate_pagecache_range(inode
, start
, end
- 1);
4812 /* Wait all existing dio workers, newcomers will block on i_mutex */
4813 ext4_inode_block_unlocked_dio(inode
);
4814 inode_dio_wait(inode
);
4817 * Remove entire range from the extent status tree.
4819 ret
= ext4_es_remove_extent(inode
, lblk
, max_blocks
);
4823 ret
= ext4_alloc_file_blocks(file
, lblk
, max_blocks
, flags
,
4829 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 4);
4830 if (IS_ERR(handle
)) {
4831 ret
= PTR_ERR(handle
);
4832 ext4_std_error(inode
->i_sb
, ret
);
4836 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4839 if (new_size
> i_size_read(inode
))
4840 i_size_write(inode
, new_size
);
4841 if (new_size
> EXT4_I(inode
)->i_disksize
)
4842 ext4_update_i_disksize(inode
, new_size
);
4845 * Mark that we allocate beyond EOF so the subsequent truncate
4846 * can proceed even if the new size is the same as i_size.
4848 if ((offset
+ len
) > i_size_read(inode
))
4849 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4852 ext4_mark_inode_dirty(handle
, inode
);
4854 /* Zero out partial block at the edges of the range */
4855 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
, len
);
4857 if (file
->f_flags
& O_SYNC
)
4858 ext4_handle_sync(handle
);
4860 ext4_journal_stop(handle
);
4862 ext4_inode_resume_unlocked_dio(inode
);
4864 mutex_unlock(&inode
->i_mutex
);
4869 * preallocate space for a file. This implements ext4's fallocate file
4870 * operation, which gets called from sys_fallocate system call.
4871 * For block-mapped files, posix_fallocate should fall back to the method
4872 * of writing zeroes to the required new blocks (the same behavior which is
4873 * expected for file systems which do not support fallocate() system call).
4875 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4877 struct inode
*inode
= file_inode(file
);
4879 loff_t new_size
= 0;
4880 unsigned int max_blocks
;
4885 unsigned int blkbits
= inode
->i_blkbits
;
4887 /* Return error if mode is not supported */
4888 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
|
4889 FALLOC_FL_COLLAPSE_RANGE
| FALLOC_FL_ZERO_RANGE
))
4892 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4893 return ext4_punch_hole(inode
, offset
, len
);
4895 ret
= ext4_convert_inline_data(inode
);
4900 * currently supporting (pre)allocate mode for extent-based
4903 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4906 if (mode
& FALLOC_FL_COLLAPSE_RANGE
)
4907 return ext4_collapse_range(inode
, offset
, len
);
4909 if (mode
& FALLOC_FL_ZERO_RANGE
)
4910 return ext4_zero_range(file
, offset
, len
, mode
);
4912 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4913 lblk
= offset
>> blkbits
;
4915 * We can't just convert len to max_blocks because
4916 * If blocksize = 4096 offset = 3072 and len = 2048
4918 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
4921 flags
= EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
;
4922 if (mode
& FALLOC_FL_KEEP_SIZE
)
4923 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4925 mutex_lock(&inode
->i_mutex
);
4927 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4928 offset
+ len
> i_size_read(inode
)) {
4929 new_size
= offset
+ len
;
4930 ret
= inode_newsize_ok(inode
, new_size
);
4935 ret
= ext4_alloc_file_blocks(file
, lblk
, max_blocks
, flags
, mode
);
4939 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
4943 tv
= inode
->i_ctime
= ext4_current_time(inode
);
4946 if (new_size
> i_size_read(inode
)) {
4947 i_size_write(inode
, new_size
);
4948 inode
->i_mtime
= tv
;
4950 if (new_size
> EXT4_I(inode
)->i_disksize
)
4951 ext4_update_i_disksize(inode
, new_size
);
4954 * Mark that we allocate beyond EOF so the subsequent truncate
4955 * can proceed even if the new size is the same as i_size.
4957 if ((offset
+ len
) > i_size_read(inode
))
4958 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4960 ext4_mark_inode_dirty(handle
, inode
);
4961 if (file
->f_flags
& O_SYNC
)
4962 ext4_handle_sync(handle
);
4964 ext4_journal_stop(handle
);
4966 mutex_unlock(&inode
->i_mutex
);
4967 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
4972 * This function convert a range of blocks to written extents
4973 * The caller of this function will pass the start offset and the size.
4974 * all unwritten extents within this range will be converted to
4977 * This function is called from the direct IO end io call back
4978 * function, to convert the fallocated extents after IO is completed.
4979 * Returns 0 on success.
4981 int ext4_convert_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
4982 loff_t offset
, ssize_t len
)
4984 unsigned int max_blocks
;
4987 struct ext4_map_blocks map
;
4988 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4990 map
.m_lblk
= offset
>> blkbits
;
4992 * We can't just convert len to max_blocks because
4993 * If blocksize = 4096 offset = 3072 and len = 2048
4995 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
4998 * This is somewhat ugly but the idea is clear: When transaction is
4999 * reserved, everything goes into it. Otherwise we rather start several
5000 * smaller transactions for conversion of each extent separately.
5003 handle
= ext4_journal_start_reserved(handle
,
5004 EXT4_HT_EXT_CONVERT
);
5006 return PTR_ERR(handle
);
5010 * credits to insert 1 extent into extent tree
5012 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
5014 while (ret
>= 0 && ret
< max_blocks
) {
5016 map
.m_len
= (max_blocks
-= ret
);
5018 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
5020 if (IS_ERR(handle
)) {
5021 ret
= PTR_ERR(handle
);
5025 ret
= ext4_map_blocks(handle
, inode
, &map
,
5026 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
5028 ext4_warning(inode
->i_sb
,
5029 "inode #%lu: block %u: len %u: "
5030 "ext4_ext_map_blocks returned %d",
5031 inode
->i_ino
, map
.m_lblk
,
5033 ext4_mark_inode_dirty(handle
, inode
);
5035 ret2
= ext4_journal_stop(handle
);
5036 if (ret
<= 0 || ret2
)
5040 ret2
= ext4_journal_stop(handle
);
5041 return ret
> 0 ? ret2
: ret
;
5045 * If newes is not existing extent (newes->ec_pblk equals zero) find
5046 * delayed extent at start of newes and update newes accordingly and
5047 * return start of the next delayed extent.
5049 * If newes is existing extent (newes->ec_pblk is not equal zero)
5050 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5051 * extent found. Leave newes unmodified.
5053 static int ext4_find_delayed_extent(struct inode
*inode
,
5054 struct extent_status
*newes
)
5056 struct extent_status es
;
5057 ext4_lblk_t block
, next_del
;
5059 if (newes
->es_pblk
== 0) {
5060 ext4_es_find_delayed_extent_range(inode
, newes
->es_lblk
,
5061 newes
->es_lblk
+ newes
->es_len
- 1, &es
);
5064 * No extent in extent-tree contains block @newes->es_pblk,
5065 * then the block may stay in 1)a hole or 2)delayed-extent.
5071 if (es
.es_lblk
> newes
->es_lblk
) {
5073 newes
->es_len
= min(es
.es_lblk
- newes
->es_lblk
,
5078 newes
->es_len
= es
.es_lblk
+ es
.es_len
- newes
->es_lblk
;
5081 block
= newes
->es_lblk
+ newes
->es_len
;
5082 ext4_es_find_delayed_extent_range(inode
, block
, EXT_MAX_BLOCKS
, &es
);
5084 next_del
= EXT_MAX_BLOCKS
;
5086 next_del
= es
.es_lblk
;
5090 /* fiemap flags we can handle specified here */
5091 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5093 static int ext4_xattr_fiemap(struct inode
*inode
,
5094 struct fiemap_extent_info
*fieinfo
)
5098 __u32 flags
= FIEMAP_EXTENT_LAST
;
5099 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
5103 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
5104 struct ext4_iloc iloc
;
5105 int offset
; /* offset of xattr in inode */
5107 error
= ext4_get_inode_loc(inode
, &iloc
);
5110 physical
= (__u64
)iloc
.bh
->b_blocknr
<< blockbits
;
5111 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
5112 EXT4_I(inode
)->i_extra_isize
;
5114 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
5115 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
5117 } else { /* external block */
5118 physical
= (__u64
)EXT4_I(inode
)->i_file_acl
<< blockbits
;
5119 length
= inode
->i_sb
->s_blocksize
;
5123 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
5125 return (error
< 0 ? error
: 0);
5128 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
5129 __u64 start
, __u64 len
)
5131 ext4_lblk_t start_blk
;
5134 if (ext4_has_inline_data(inode
)) {
5137 error
= ext4_inline_data_fiemap(inode
, fieinfo
, &has_inline
);
5143 if (fieinfo
->fi_flags
& FIEMAP_FLAG_CACHE
) {
5144 error
= ext4_ext_precache(inode
);
5149 /* fallback to generic here if not in extents fmt */
5150 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5151 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
5154 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
5157 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
5158 error
= ext4_xattr_fiemap(inode
, fieinfo
);
5160 ext4_lblk_t len_blks
;
5163 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
5164 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
5165 if (last_blk
>= EXT_MAX_BLOCKS
)
5166 last_blk
= EXT_MAX_BLOCKS
-1;
5167 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
5170 * Walk the extent tree gathering extent information
5171 * and pushing extents back to the user.
5173 error
= ext4_fill_fiemap_extents(inode
, start_blk
,
5176 ext4_es_lru_add(inode
);
5182 * Function to access the path buffer for marking it dirty.
5183 * It also checks if there are sufficient credits left in the journal handle
5187 ext4_access_path(handle_t
*handle
, struct inode
*inode
,
5188 struct ext4_ext_path
*path
)
5192 if (!ext4_handle_valid(handle
))
5196 * Check if need to extend journal credits
5197 * 3 for leaf, sb, and inode plus 2 (bmap and group
5198 * descriptor) for each block group; assume two block
5201 if (handle
->h_buffer_credits
< 7) {
5202 credits
= ext4_writepage_trans_blocks(inode
);
5203 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
5204 /* EAGAIN is success */
5205 if (err
&& err
!= -EAGAIN
)
5209 err
= ext4_ext_get_access(handle
, inode
, path
);
5214 * ext4_ext_shift_path_extents:
5215 * Shift the extents of a path structure lying between path[depth].p_ext
5216 * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift
5217 * from starting block for each extent.
5220 ext4_ext_shift_path_extents(struct ext4_ext_path
*path
, ext4_lblk_t shift
,
5221 struct inode
*inode
, handle_t
*handle
,
5225 struct ext4_extent
*ex_start
, *ex_last
;
5227 depth
= path
->p_depth
;
5229 while (depth
>= 0) {
5230 if (depth
== path
->p_depth
) {
5231 ex_start
= path
[depth
].p_ext
;
5235 ex_last
= EXT_LAST_EXTENT(path
[depth
].p_hdr
);
5239 err
= ext4_access_path(handle
, inode
, path
+ depth
);
5243 if (ex_start
== EXT_FIRST_EXTENT(path
[depth
].p_hdr
))
5246 *start
= le32_to_cpu(ex_last
->ee_block
) +
5247 ext4_ext_get_actual_len(ex_last
);
5249 while (ex_start
<= ex_last
) {
5250 le32_add_cpu(&ex_start
->ee_block
, -shift
);
5251 /* Try to merge to the left. */
5253 EXT_FIRST_EXTENT(path
[depth
].p_hdr
)) &&
5254 ext4_ext_try_to_merge_right(inode
,
5255 path
, ex_start
- 1))
5260 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
5264 if (--depth
< 0 || !update
)
5268 /* Update index too */
5269 err
= ext4_access_path(handle
, inode
, path
+ depth
);
5273 le32_add_cpu(&path
[depth
].p_idx
->ei_block
, -shift
);
5274 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
5278 /* we are done if current index is not a starting index */
5279 if (path
[depth
].p_idx
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))
5290 * ext4_ext_shift_extents:
5291 * All the extents which lies in the range from start to the last allocated
5292 * block for the file are shifted downwards by shift blocks.
5293 * On success, 0 is returned, error otherwise.
5296 ext4_ext_shift_extents(struct inode
*inode
, handle_t
*handle
,
5297 ext4_lblk_t start
, ext4_lblk_t shift
)
5299 struct ext4_ext_path
*path
;
5301 struct ext4_extent
*extent
;
5302 ext4_lblk_t stop_block
, current_block
;
5303 ext4_lblk_t ex_start
, ex_end
;
5305 /* Let path point to the last extent */
5306 path
= ext4_ext_find_extent(inode
, EXT_MAX_BLOCKS
- 1, NULL
, 0);
5308 return PTR_ERR(path
);
5310 depth
= path
->p_depth
;
5311 extent
= path
[depth
].p_ext
;
5313 ext4_ext_drop_refs(path
);
5318 stop_block
= le32_to_cpu(extent
->ee_block
) +
5319 ext4_ext_get_actual_len(extent
);
5320 ext4_ext_drop_refs(path
);
5323 /* Nothing to shift, if hole is at the end of file */
5324 if (start
>= stop_block
)
5328 * Don't start shifting extents until we make sure the hole is big
5329 * enough to accomodate the shift.
5331 path
= ext4_ext_find_extent(inode
, start
- 1, NULL
, 0);
5333 return PTR_ERR(path
);
5334 depth
= path
->p_depth
;
5335 extent
= path
[depth
].p_ext
;
5337 ex_start
= le32_to_cpu(extent
->ee_block
);
5338 ex_end
= le32_to_cpu(extent
->ee_block
) +
5339 ext4_ext_get_actual_len(extent
);
5344 ext4_ext_drop_refs(path
);
5347 if ((start
== ex_start
&& shift
> ex_start
) ||
5348 (shift
> start
- ex_end
))
5351 /* Its safe to start updating extents */
5352 while (start
< stop_block
) {
5353 path
= ext4_ext_find_extent(inode
, start
, NULL
, 0);
5355 return PTR_ERR(path
);
5356 depth
= path
->p_depth
;
5357 extent
= path
[depth
].p_ext
;
5359 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
5360 (unsigned long) start
);
5364 current_block
= le32_to_cpu(extent
->ee_block
);
5365 if (start
> current_block
) {
5366 /* Hole, move to the next extent */
5367 ret
= mext_next_extent(inode
, path
, &extent
);
5369 ext4_ext_drop_refs(path
);
5376 ret
= ext4_ext_shift_path_extents(path
, shift
, inode
,
5378 ext4_ext_drop_refs(path
);
5388 * ext4_collapse_range:
5389 * This implements the fallocate's collapse range functionality for ext4
5390 * Returns: 0 and non-zero on error.
5392 int ext4_collapse_range(struct inode
*inode
, loff_t offset
, loff_t len
)
5394 struct super_block
*sb
= inode
->i_sb
;
5395 ext4_lblk_t punch_start
, punch_stop
;
5397 unsigned int credits
;
5401 /* Collapse range works only on fs block size aligned offsets. */
5402 if (offset
& (EXT4_BLOCK_SIZE(sb
) - 1) ||
5403 len
& (EXT4_BLOCK_SIZE(sb
) - 1))
5406 if (!S_ISREG(inode
->i_mode
))
5409 trace_ext4_collapse_range(inode
, offset
, len
);
5411 punch_start
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
5412 punch_stop
= (offset
+ len
) >> EXT4_BLOCK_SIZE_BITS(sb
);
5414 /* Call ext4_force_commit to flush all data in case of data=journal. */
5415 if (ext4_should_journal_data(inode
)) {
5416 ret
= ext4_force_commit(inode
->i_sb
);
5421 /* Write out all dirty pages */
5422 ret
= filemap_write_and_wait_range(inode
->i_mapping
, offset
, LLONG_MAX
);
5426 /* Take mutex lock */
5427 mutex_lock(&inode
->i_mutex
);
5430 * There is no need to overlap collapse range with EOF, in which case
5431 * it is effectively a truncate operation
5433 if (offset
+ len
>= i_size_read(inode
)) {
5438 /* Currently just for extent based files */
5439 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5444 truncate_pagecache(inode
, offset
);
5446 /* Wait for existing dio to complete */
5447 ext4_inode_block_unlocked_dio(inode
);
5448 inode_dio_wait(inode
);
5450 credits
= ext4_writepage_trans_blocks(inode
);
5451 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
5452 if (IS_ERR(handle
)) {
5453 ret
= PTR_ERR(handle
);
5457 down_write(&EXT4_I(inode
)->i_data_sem
);
5458 ext4_discard_preallocations(inode
);
5460 ret
= ext4_es_remove_extent(inode
, punch_start
,
5461 EXT_MAX_BLOCKS
- punch_start
);
5463 up_write(&EXT4_I(inode
)->i_data_sem
);
5467 ret
= ext4_ext_remove_space(inode
, punch_start
, punch_stop
- 1);
5469 up_write(&EXT4_I(inode
)->i_data_sem
);
5472 ext4_discard_preallocations(inode
);
5474 ret
= ext4_ext_shift_extents(inode
, handle
, punch_stop
,
5475 punch_stop
- punch_start
);
5477 up_write(&EXT4_I(inode
)->i_data_sem
);
5481 new_size
= i_size_read(inode
) - len
;
5482 i_size_write(inode
, new_size
);
5483 EXT4_I(inode
)->i_disksize
= new_size
;
5485 up_write(&EXT4_I(inode
)->i_data_sem
);
5487 ext4_handle_sync(handle
);
5488 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
5489 ext4_mark_inode_dirty(handle
, inode
);
5492 ext4_journal_stop(handle
);
5494 ext4_inode_resume_unlocked_dio(inode
);
5496 mutex_unlock(&inode
->i_mutex
);