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 <linux/backing-dev.h>
43 #include "ext4_jbd2.h"
44 #include "ext4_extents.h"
47 #include <trace/events/ext4.h>
50 * used by extent splitting.
52 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
54 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
55 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
57 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
58 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
60 static __le32
ext4_extent_block_csum(struct inode
*inode
,
61 struct ext4_extent_header
*eh
)
63 struct ext4_inode_info
*ei
= EXT4_I(inode
);
64 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
67 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)eh
,
68 EXT4_EXTENT_TAIL_OFFSET(eh
));
69 return cpu_to_le32(csum
);
72 static int ext4_extent_block_csum_verify(struct inode
*inode
,
73 struct ext4_extent_header
*eh
)
75 struct ext4_extent_tail
*et
;
77 if (!ext4_has_metadata_csum(inode
->i_sb
))
80 et
= find_ext4_extent_tail(eh
);
81 if (et
->et_checksum
!= ext4_extent_block_csum(inode
, eh
))
86 static void ext4_extent_block_csum_set(struct inode
*inode
,
87 struct ext4_extent_header
*eh
)
89 struct ext4_extent_tail
*et
;
91 if (!ext4_has_metadata_csum(inode
->i_sb
))
94 et
= find_ext4_extent_tail(eh
);
95 et
->et_checksum
= ext4_extent_block_csum(inode
, eh
);
98 static int ext4_split_extent(handle_t
*handle
,
100 struct ext4_ext_path
**ppath
,
101 struct ext4_map_blocks
*map
,
105 static int ext4_split_extent_at(handle_t
*handle
,
107 struct ext4_ext_path
**ppath
,
112 static int ext4_find_delayed_extent(struct inode
*inode
,
113 struct extent_status
*newes
);
115 static int ext4_ext_truncate_extend_restart(handle_t
*handle
,
121 if (!ext4_handle_valid(handle
))
123 if (handle
->h_buffer_credits
> needed
)
125 err
= ext4_journal_extend(handle
, needed
);
128 err
= ext4_truncate_restart_trans(handle
, inode
, needed
);
140 static int ext4_ext_get_access(handle_t
*handle
, struct inode
*inode
,
141 struct ext4_ext_path
*path
)
144 /* path points to block */
145 BUFFER_TRACE(path
->p_bh
, "get_write_access");
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 WARN_ON(!rwsem_is_locked(&EXT4_I(inode
)->i_data_sem
));
166 ext4_extent_block_csum_set(inode
, ext_block_hdr(path
->p_bh
));
167 /* path points to block */
168 err
= __ext4_handle_dirty_metadata(where
, line
, handle
,
171 /* path points to leaf/index in inode body */
172 err
= ext4_mark_inode_dirty(handle
, inode
);
177 static ext4_fsblk_t
ext4_ext_find_goal(struct inode
*inode
,
178 struct ext4_ext_path
*path
,
182 int depth
= path
->p_depth
;
183 struct ext4_extent
*ex
;
186 * Try to predict block placement assuming that we are
187 * filling in a file which will eventually be
188 * non-sparse --- i.e., in the case of libbfd writing
189 * an ELF object sections out-of-order but in a way
190 * the eventually results in a contiguous object or
191 * executable file, or some database extending a table
192 * space file. However, this is actually somewhat
193 * non-ideal if we are writing a sparse file such as
194 * qemu or KVM writing a raw image file that is going
195 * to stay fairly sparse, since it will end up
196 * fragmenting the file system's free space. Maybe we
197 * should have some hueristics or some way to allow
198 * userspace to pass a hint to file system,
199 * especially if the latter case turns out to be
202 ex
= path
[depth
].p_ext
;
204 ext4_fsblk_t ext_pblk
= ext4_ext_pblock(ex
);
205 ext4_lblk_t ext_block
= le32_to_cpu(ex
->ee_block
);
207 if (block
> ext_block
)
208 return ext_pblk
+ (block
- ext_block
);
210 return ext_pblk
- (ext_block
- block
);
213 /* it looks like index is empty;
214 * try to find starting block from index itself */
215 if (path
[depth
].p_bh
)
216 return path
[depth
].p_bh
->b_blocknr
;
219 /* OK. use inode's group */
220 return ext4_inode_to_goal_block(inode
);
224 * Allocation for a meta data block
227 ext4_ext_new_meta_block(handle_t
*handle
, struct inode
*inode
,
228 struct ext4_ext_path
*path
,
229 struct ext4_extent
*ex
, int *err
, unsigned int flags
)
231 ext4_fsblk_t goal
, newblock
;
233 goal
= ext4_ext_find_goal(inode
, path
, le32_to_cpu(ex
->ee_block
));
234 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, flags
,
239 static inline int ext4_ext_space_block(struct inode
*inode
, int check
)
243 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
244 / sizeof(struct ext4_extent
);
245 #ifdef AGGRESSIVE_TEST
246 if (!check
&& size
> 6)
252 static inline int ext4_ext_space_block_idx(struct inode
*inode
, int check
)
256 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
257 / sizeof(struct ext4_extent_idx
);
258 #ifdef AGGRESSIVE_TEST
259 if (!check
&& size
> 5)
265 static inline int ext4_ext_space_root(struct inode
*inode
, int check
)
269 size
= sizeof(EXT4_I(inode
)->i_data
);
270 size
-= sizeof(struct ext4_extent_header
);
271 size
/= sizeof(struct ext4_extent
);
272 #ifdef AGGRESSIVE_TEST
273 if (!check
&& size
> 3)
279 static inline int ext4_ext_space_root_idx(struct inode
*inode
, int check
)
283 size
= sizeof(EXT4_I(inode
)->i_data
);
284 size
-= sizeof(struct ext4_extent_header
);
285 size
/= sizeof(struct ext4_extent_idx
);
286 #ifdef AGGRESSIVE_TEST
287 if (!check
&& size
> 4)
294 ext4_force_split_extent_at(handle_t
*handle
, struct inode
*inode
,
295 struct ext4_ext_path
**ppath
, ext4_lblk_t lblk
,
298 struct ext4_ext_path
*path
= *ppath
;
299 int unwritten
= ext4_ext_is_unwritten(path
[path
->p_depth
].p_ext
);
301 return ext4_split_extent_at(handle
, inode
, ppath
, lblk
, unwritten
?
302 EXT4_EXT_MARK_UNWRIT1
|EXT4_EXT_MARK_UNWRIT2
: 0,
303 EXT4_EX_NOCACHE
| EXT4_GET_BLOCKS_PRE_IO
|
304 (nofail
? EXT4_GET_BLOCKS_METADATA_NOFAIL
:0));
308 * Calculate the number of metadata blocks needed
309 * to allocate @blocks
310 * Worse case is one block per extent
312 int ext4_ext_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
314 struct ext4_inode_info
*ei
= EXT4_I(inode
);
317 idxs
= ((inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
318 / sizeof(struct ext4_extent_idx
));
321 * If the new delayed allocation block is contiguous with the
322 * previous da block, it can share index blocks with the
323 * previous block, so we only need to allocate a new index
324 * block every idxs leaf blocks. At ldxs**2 blocks, we need
325 * an additional index block, and at ldxs**3 blocks, yet
326 * another index blocks.
328 if (ei
->i_da_metadata_calc_len
&&
329 ei
->i_da_metadata_calc_last_lblock
+1 == lblock
) {
332 if ((ei
->i_da_metadata_calc_len
% idxs
) == 0)
334 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
)) == 0)
336 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
*idxs
)) == 0) {
338 ei
->i_da_metadata_calc_len
= 0;
340 ei
->i_da_metadata_calc_len
++;
341 ei
->i_da_metadata_calc_last_lblock
++;
346 * In the worst case we need a new set of index blocks at
347 * every level of the inode's extent tree.
349 ei
->i_da_metadata_calc_len
= 1;
350 ei
->i_da_metadata_calc_last_lblock
= lblock
;
351 return ext_depth(inode
) + 1;
355 ext4_ext_max_entries(struct inode
*inode
, int depth
)
359 if (depth
== ext_depth(inode
)) {
361 max
= ext4_ext_space_root(inode
, 1);
363 max
= ext4_ext_space_root_idx(inode
, 1);
366 max
= ext4_ext_space_block(inode
, 1);
368 max
= ext4_ext_space_block_idx(inode
, 1);
374 static int ext4_valid_extent(struct inode
*inode
, struct ext4_extent
*ext
)
376 ext4_fsblk_t block
= ext4_ext_pblock(ext
);
377 int len
= ext4_ext_get_actual_len(ext
);
378 ext4_lblk_t lblock
= le32_to_cpu(ext
->ee_block
);
383 * - overflow/wrap-around
385 if (lblock
+ len
<= lblock
)
387 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
390 static int ext4_valid_extent_idx(struct inode
*inode
,
391 struct ext4_extent_idx
*ext_idx
)
393 ext4_fsblk_t block
= ext4_idx_pblock(ext_idx
);
395 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
398 static int ext4_valid_extent_entries(struct inode
*inode
,
399 struct ext4_extent_header
*eh
,
402 unsigned short entries
;
403 if (eh
->eh_entries
== 0)
406 entries
= le16_to_cpu(eh
->eh_entries
);
410 struct ext4_extent
*ext
= EXT_FIRST_EXTENT(eh
);
411 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
412 ext4_fsblk_t pblock
= 0;
413 ext4_lblk_t lblock
= 0;
414 ext4_lblk_t prev
= 0;
417 if (!ext4_valid_extent(inode
, ext
))
420 /* Check for overlapping extents */
421 lblock
= le32_to_cpu(ext
->ee_block
);
422 len
= ext4_ext_get_actual_len(ext
);
423 if ((lblock
<= prev
) && prev
) {
424 pblock
= ext4_ext_pblock(ext
);
425 es
->s_last_error_block
= cpu_to_le64(pblock
);
430 prev
= lblock
+ len
- 1;
433 struct ext4_extent_idx
*ext_idx
= EXT_FIRST_INDEX(eh
);
435 if (!ext4_valid_extent_idx(inode
, ext_idx
))
444 static int __ext4_ext_check(const char *function
, unsigned int line
,
445 struct inode
*inode
, struct ext4_extent_header
*eh
,
446 int depth
, ext4_fsblk_t pblk
)
448 const char *error_msg
;
449 int max
= 0, err
= -EFSCORRUPTED
;
451 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
452 error_msg
= "invalid magic";
455 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
456 error_msg
= "unexpected eh_depth";
459 if (unlikely(eh
->eh_max
== 0)) {
460 error_msg
= "invalid eh_max";
463 max
= ext4_ext_max_entries(inode
, depth
);
464 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
465 error_msg
= "too large eh_max";
468 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
469 error_msg
= "invalid eh_entries";
472 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
473 error_msg
= "invalid extent entries";
476 if (unlikely(depth
> 32)) {
477 error_msg
= "too large eh_depth";
480 /* Verify checksum on non-root extent tree nodes */
481 if (ext_depth(inode
) != depth
&&
482 !ext4_extent_block_csum_verify(inode
, eh
)) {
483 error_msg
= "extent tree corrupted";
490 ext4_error_inode(inode
, function
, line
, 0,
491 "pblk %llu bad header/extent: %s - magic %x, "
492 "entries %u, max %u(%u), depth %u(%u)",
493 (unsigned long long) pblk
, error_msg
,
494 le16_to_cpu(eh
->eh_magic
),
495 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
496 max
, le16_to_cpu(eh
->eh_depth
), depth
);
500 #define ext4_ext_check(inode, eh, depth, pblk) \
501 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
503 int ext4_ext_check_inode(struct inode
*inode
)
505 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
), 0);
508 static struct buffer_head
*
509 __read_extent_tree_block(const char *function
, unsigned int line
,
510 struct inode
*inode
, ext4_fsblk_t pblk
, int depth
,
513 struct buffer_head
*bh
;
516 bh
= sb_getblk_gfp(inode
->i_sb
, pblk
, __GFP_MOVABLE
| GFP_NOFS
);
518 return ERR_PTR(-ENOMEM
);
520 if (!bh_uptodate_or_lock(bh
)) {
521 trace_ext4_ext_load_extent(inode
, pblk
, _RET_IP_
);
522 err
= bh_submit_read(bh
);
526 if (buffer_verified(bh
) && !(flags
& EXT4_EX_FORCE_CACHE
))
528 err
= __ext4_ext_check(function
, line
, inode
,
529 ext_block_hdr(bh
), depth
, pblk
);
532 set_buffer_verified(bh
);
534 * If this is a leaf block, cache all of its entries
536 if (!(flags
& EXT4_EX_NOCACHE
) && depth
== 0) {
537 struct ext4_extent_header
*eh
= ext_block_hdr(bh
);
538 struct ext4_extent
*ex
= EXT_FIRST_EXTENT(eh
);
539 ext4_lblk_t prev
= 0;
542 for (i
= le16_to_cpu(eh
->eh_entries
); i
> 0; i
--, ex
++) {
543 unsigned int status
= EXTENT_STATUS_WRITTEN
;
544 ext4_lblk_t lblk
= le32_to_cpu(ex
->ee_block
);
545 int len
= ext4_ext_get_actual_len(ex
);
547 if (prev
&& (prev
!= lblk
))
548 ext4_es_cache_extent(inode
, prev
,
552 if (ext4_ext_is_unwritten(ex
))
553 status
= EXTENT_STATUS_UNWRITTEN
;
554 ext4_es_cache_extent(inode
, lblk
, len
,
555 ext4_ext_pblock(ex
), status
);
566 #define read_extent_tree_block(inode, pblk, depth, flags) \
567 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
571 * This function is called to cache a file's extent information in the
574 int ext4_ext_precache(struct inode
*inode
)
576 struct ext4_inode_info
*ei
= EXT4_I(inode
);
577 struct ext4_ext_path
*path
= NULL
;
578 struct buffer_head
*bh
;
579 int i
= 0, depth
, ret
= 0;
581 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
582 return 0; /* not an extent-mapped inode */
584 down_read(&ei
->i_data_sem
);
585 depth
= ext_depth(inode
);
587 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
590 up_read(&ei
->i_data_sem
);
594 /* Don't cache anything if there are no external extent blocks */
597 path
[0].p_hdr
= ext_inode_hdr(inode
);
598 ret
= ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0);
601 path
[0].p_idx
= EXT_FIRST_INDEX(path
[0].p_hdr
);
604 * If this is a leaf block or we've reached the end of
605 * the index block, go up
608 path
[i
].p_idx
> EXT_LAST_INDEX(path
[i
].p_hdr
)) {
609 brelse(path
[i
].p_bh
);
614 bh
= read_extent_tree_block(inode
,
615 ext4_idx_pblock(path
[i
].p_idx
++),
617 EXT4_EX_FORCE_CACHE
);
624 path
[i
].p_hdr
= ext_block_hdr(bh
);
625 path
[i
].p_idx
= EXT_FIRST_INDEX(path
[i
].p_hdr
);
627 ext4_set_inode_state(inode
, EXT4_STATE_EXT_PRECACHED
);
629 up_read(&ei
->i_data_sem
);
630 ext4_ext_drop_refs(path
);
636 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
638 int k
, l
= path
->p_depth
;
641 for (k
= 0; k
<= l
; k
++, path
++) {
643 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
644 ext4_idx_pblock(path
->p_idx
));
645 } else if (path
->p_ext
) {
646 ext_debug(" %d:[%d]%d:%llu ",
647 le32_to_cpu(path
->p_ext
->ee_block
),
648 ext4_ext_is_unwritten(path
->p_ext
),
649 ext4_ext_get_actual_len(path
->p_ext
),
650 ext4_ext_pblock(path
->p_ext
));
657 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
659 int depth
= ext_depth(inode
);
660 struct ext4_extent_header
*eh
;
661 struct ext4_extent
*ex
;
667 eh
= path
[depth
].p_hdr
;
668 ex
= EXT_FIRST_EXTENT(eh
);
670 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
672 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
673 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
674 ext4_ext_is_unwritten(ex
),
675 ext4_ext_get_actual_len(ex
), ext4_ext_pblock(ex
));
680 static void ext4_ext_show_move(struct inode
*inode
, struct ext4_ext_path
*path
,
681 ext4_fsblk_t newblock
, int level
)
683 int depth
= ext_depth(inode
);
684 struct ext4_extent
*ex
;
686 if (depth
!= level
) {
687 struct ext4_extent_idx
*idx
;
688 idx
= path
[level
].p_idx
;
689 while (idx
<= EXT_MAX_INDEX(path
[level
].p_hdr
)) {
690 ext_debug("%d: move %d:%llu in new index %llu\n", level
,
691 le32_to_cpu(idx
->ei_block
),
692 ext4_idx_pblock(idx
),
700 ex
= path
[depth
].p_ext
;
701 while (ex
<= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
702 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
703 le32_to_cpu(ex
->ee_block
),
705 ext4_ext_is_unwritten(ex
),
706 ext4_ext_get_actual_len(ex
),
713 #define ext4_ext_show_path(inode, path)
714 #define ext4_ext_show_leaf(inode, path)
715 #define ext4_ext_show_move(inode, path, newblock, level)
718 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
724 depth
= path
->p_depth
;
725 for (i
= 0; i
<= depth
; i
++, path
++)
733 * ext4_ext_binsearch_idx:
734 * binary search for the closest index of the given block
735 * the header must be checked before calling this
738 ext4_ext_binsearch_idx(struct inode
*inode
,
739 struct ext4_ext_path
*path
, ext4_lblk_t block
)
741 struct ext4_extent_header
*eh
= path
->p_hdr
;
742 struct ext4_extent_idx
*r
, *l
, *m
;
745 ext_debug("binsearch for %u(idx): ", block
);
747 l
= EXT_FIRST_INDEX(eh
) + 1;
748 r
= EXT_LAST_INDEX(eh
);
751 if (block
< le32_to_cpu(m
->ei_block
))
755 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
756 m
, le32_to_cpu(m
->ei_block
),
757 r
, le32_to_cpu(r
->ei_block
));
761 ext_debug(" -> %u->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
762 ext4_idx_pblock(path
->p_idx
));
764 #ifdef CHECK_BINSEARCH
766 struct ext4_extent_idx
*chix
, *ix
;
769 chix
= ix
= EXT_FIRST_INDEX(eh
);
770 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
772 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
773 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
775 ix
, EXT_FIRST_INDEX(eh
));
776 printk(KERN_DEBUG
"%u <= %u\n",
777 le32_to_cpu(ix
->ei_block
),
778 le32_to_cpu(ix
[-1].ei_block
));
780 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
781 <= le32_to_cpu(ix
[-1].ei_block
));
782 if (block
< le32_to_cpu(ix
->ei_block
))
786 BUG_ON(chix
!= path
->p_idx
);
793 * ext4_ext_binsearch:
794 * binary search for closest extent of the given block
795 * the header must be checked before calling this
798 ext4_ext_binsearch(struct inode
*inode
,
799 struct ext4_ext_path
*path
, ext4_lblk_t block
)
801 struct ext4_extent_header
*eh
= path
->p_hdr
;
802 struct ext4_extent
*r
, *l
, *m
;
804 if (eh
->eh_entries
== 0) {
806 * this leaf is empty:
807 * we get such a leaf in split/add case
812 ext_debug("binsearch for %u: ", block
);
814 l
= EXT_FIRST_EXTENT(eh
) + 1;
815 r
= EXT_LAST_EXTENT(eh
);
819 if (block
< le32_to_cpu(m
->ee_block
))
823 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
824 m
, le32_to_cpu(m
->ee_block
),
825 r
, le32_to_cpu(r
->ee_block
));
829 ext_debug(" -> %d:%llu:[%d]%d ",
830 le32_to_cpu(path
->p_ext
->ee_block
),
831 ext4_ext_pblock(path
->p_ext
),
832 ext4_ext_is_unwritten(path
->p_ext
),
833 ext4_ext_get_actual_len(path
->p_ext
));
835 #ifdef CHECK_BINSEARCH
837 struct ext4_extent
*chex
, *ex
;
840 chex
= ex
= EXT_FIRST_EXTENT(eh
);
841 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
842 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
843 <= le32_to_cpu(ex
[-1].ee_block
));
844 if (block
< le32_to_cpu(ex
->ee_block
))
848 BUG_ON(chex
!= path
->p_ext
);
854 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
856 struct ext4_extent_header
*eh
;
858 eh
= ext_inode_hdr(inode
);
861 eh
->eh_magic
= EXT4_EXT_MAGIC
;
862 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
863 ext4_mark_inode_dirty(handle
, inode
);
867 struct ext4_ext_path
*
868 ext4_find_extent(struct inode
*inode
, ext4_lblk_t block
,
869 struct ext4_ext_path
**orig_path
, int flags
)
871 struct ext4_extent_header
*eh
;
872 struct buffer_head
*bh
;
873 struct ext4_ext_path
*path
= orig_path
? *orig_path
: NULL
;
874 short int depth
, i
, ppos
= 0;
877 eh
= ext_inode_hdr(inode
);
878 depth
= ext_depth(inode
);
879 if (depth
< 0 || depth
> EXT4_MAX_EXTENT_DEPTH
) {
880 EXT4_ERROR_INODE(inode
, "inode has invalid extent depth: %d",
887 ext4_ext_drop_refs(path
);
888 if (depth
> path
[0].p_maxdepth
) {
890 *orig_path
= path
= NULL
;
894 /* account possible depth increase */
895 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
898 return ERR_PTR(-ENOMEM
);
899 path
[0].p_maxdepth
= depth
+ 1;
905 /* walk through the tree */
907 ext_debug("depth %d: num %d, max %d\n",
908 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
910 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
911 path
[ppos
].p_block
= ext4_idx_pblock(path
[ppos
].p_idx
);
912 path
[ppos
].p_depth
= i
;
913 path
[ppos
].p_ext
= NULL
;
915 bh
= read_extent_tree_block(inode
, path
[ppos
].p_block
, --i
,
922 eh
= ext_block_hdr(bh
);
924 if (unlikely(ppos
> depth
)) {
926 EXT4_ERROR_INODE(inode
,
927 "ppos %d > depth %d", ppos
, depth
);
931 path
[ppos
].p_bh
= bh
;
932 path
[ppos
].p_hdr
= eh
;
935 path
[ppos
].p_depth
= i
;
936 path
[ppos
].p_ext
= NULL
;
937 path
[ppos
].p_idx
= NULL
;
940 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
941 /* if not an empty leaf */
942 if (path
[ppos
].p_ext
)
943 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
945 ext4_ext_show_path(inode
, path
);
950 ext4_ext_drop_refs(path
);
958 * ext4_ext_insert_index:
959 * insert new index [@logical;@ptr] into the block at @curp;
960 * check where to insert: before @curp or after @curp
962 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
963 struct ext4_ext_path
*curp
,
964 int logical
, ext4_fsblk_t ptr
)
966 struct ext4_extent_idx
*ix
;
969 err
= ext4_ext_get_access(handle
, inode
, curp
);
973 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
974 EXT4_ERROR_INODE(inode
,
975 "logical %d == ei_block %d!",
976 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
977 return -EFSCORRUPTED
;
980 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
981 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
982 EXT4_ERROR_INODE(inode
,
983 "eh_entries %d >= eh_max %d!",
984 le16_to_cpu(curp
->p_hdr
->eh_entries
),
985 le16_to_cpu(curp
->p_hdr
->eh_max
));
986 return -EFSCORRUPTED
;
989 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
991 ext_debug("insert new index %d after: %llu\n", logical
, ptr
);
992 ix
= curp
->p_idx
+ 1;
995 ext_debug("insert new index %d before: %llu\n", logical
, ptr
);
999 len
= EXT_LAST_INDEX(curp
->p_hdr
) - ix
+ 1;
1002 ext_debug("insert new index %d: "
1003 "move %d indices from 0x%p to 0x%p\n",
1004 logical
, len
, ix
, ix
+ 1);
1005 memmove(ix
+ 1, ix
, len
* sizeof(struct ext4_extent_idx
));
1008 if (unlikely(ix
> EXT_MAX_INDEX(curp
->p_hdr
))) {
1009 EXT4_ERROR_INODE(inode
, "ix > EXT_MAX_INDEX!");
1010 return -EFSCORRUPTED
;
1013 ix
->ei_block
= cpu_to_le32(logical
);
1014 ext4_idx_store_pblock(ix
, ptr
);
1015 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
1017 if (unlikely(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
1018 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
1019 return -EFSCORRUPTED
;
1022 err
= ext4_ext_dirty(handle
, inode
, curp
);
1023 ext4_std_error(inode
->i_sb
, err
);
1030 * inserts new subtree into the path, using free index entry
1032 * - allocates all needed blocks (new leaf and all intermediate index blocks)
1033 * - makes decision where to split
1034 * - moves remaining extents and index entries (right to the split point)
1035 * into the newly allocated blocks
1036 * - initializes subtree
1038 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
1040 struct ext4_ext_path
*path
,
1041 struct ext4_extent
*newext
, int at
)
1043 struct buffer_head
*bh
= NULL
;
1044 int depth
= ext_depth(inode
);
1045 struct ext4_extent_header
*neh
;
1046 struct ext4_extent_idx
*fidx
;
1047 int i
= at
, k
, m
, a
;
1048 ext4_fsblk_t newblock
, oldblock
;
1050 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
1053 /* make decision: where to split? */
1054 /* FIXME: now decision is simplest: at current extent */
1056 /* if current leaf will be split, then we should use
1057 * border from split point */
1058 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
1059 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
1060 return -EFSCORRUPTED
;
1062 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
1063 border
= path
[depth
].p_ext
[1].ee_block
;
1064 ext_debug("leaf will be split."
1065 " next leaf starts at %d\n",
1066 le32_to_cpu(border
));
1068 border
= newext
->ee_block
;
1069 ext_debug("leaf will be added."
1070 " next leaf starts at %d\n",
1071 le32_to_cpu(border
));
1075 * If error occurs, then we break processing
1076 * and mark filesystem read-only. index won't
1077 * be inserted and tree will be in consistent
1078 * state. Next mount will repair buffers too.
1082 * Get array to track all allocated blocks.
1083 * We need this to handle errors and free blocks
1086 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
1090 /* allocate all needed blocks */
1091 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
1092 for (a
= 0; a
< depth
- at
; a
++) {
1093 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
1094 newext
, &err
, flags
);
1097 ablocks
[a
] = newblock
;
1100 /* initialize new leaf */
1101 newblock
= ablocks
[--a
];
1102 if (unlikely(newblock
== 0)) {
1103 EXT4_ERROR_INODE(inode
, "newblock == 0!");
1104 err
= -EFSCORRUPTED
;
1107 bh
= sb_getblk_gfp(inode
->i_sb
, newblock
, __GFP_MOVABLE
| GFP_NOFS
);
1108 if (unlikely(!bh
)) {
1114 err
= ext4_journal_get_create_access(handle
, bh
);
1118 neh
= ext_block_hdr(bh
);
1119 neh
->eh_entries
= 0;
1120 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1121 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1124 /* move remainder of path[depth] to the new leaf */
1125 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
1126 path
[depth
].p_hdr
->eh_max
)) {
1127 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
1128 path
[depth
].p_hdr
->eh_entries
,
1129 path
[depth
].p_hdr
->eh_max
);
1130 err
= -EFSCORRUPTED
;
1133 /* start copy from next extent */
1134 m
= EXT_MAX_EXTENT(path
[depth
].p_hdr
) - path
[depth
].p_ext
++;
1135 ext4_ext_show_move(inode
, path
, newblock
, depth
);
1137 struct ext4_extent
*ex
;
1138 ex
= EXT_FIRST_EXTENT(neh
);
1139 memmove(ex
, path
[depth
].p_ext
, sizeof(struct ext4_extent
) * m
);
1140 le16_add_cpu(&neh
->eh_entries
, m
);
1143 ext4_extent_block_csum_set(inode
, neh
);
1144 set_buffer_uptodate(bh
);
1147 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1153 /* correct old leaf */
1155 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1158 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
1159 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1165 /* create intermediate indexes */
1167 if (unlikely(k
< 0)) {
1168 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
1169 err
= -EFSCORRUPTED
;
1173 ext_debug("create %d intermediate indices\n", k
);
1174 /* insert new index into current index block */
1175 /* current depth stored in i var */
1178 oldblock
= newblock
;
1179 newblock
= ablocks
[--a
];
1180 bh
= sb_getblk(inode
->i_sb
, newblock
);
1181 if (unlikely(!bh
)) {
1187 err
= ext4_journal_get_create_access(handle
, bh
);
1191 neh
= ext_block_hdr(bh
);
1192 neh
->eh_entries
= cpu_to_le16(1);
1193 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1194 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1195 neh
->eh_depth
= cpu_to_le16(depth
- i
);
1196 fidx
= EXT_FIRST_INDEX(neh
);
1197 fidx
->ei_block
= border
;
1198 ext4_idx_store_pblock(fidx
, oldblock
);
1200 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1201 i
, newblock
, le32_to_cpu(border
), oldblock
);
1203 /* move remainder of path[i] to the new index block */
1204 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
1205 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
1206 EXT4_ERROR_INODE(inode
,
1207 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1208 le32_to_cpu(path
[i
].p_ext
->ee_block
));
1209 err
= -EFSCORRUPTED
;
1212 /* start copy indexes */
1213 m
= EXT_MAX_INDEX(path
[i
].p_hdr
) - path
[i
].p_idx
++;
1214 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
1215 EXT_MAX_INDEX(path
[i
].p_hdr
));
1216 ext4_ext_show_move(inode
, path
, newblock
, i
);
1218 memmove(++fidx
, path
[i
].p_idx
,
1219 sizeof(struct ext4_extent_idx
) * m
);
1220 le16_add_cpu(&neh
->eh_entries
, m
);
1222 ext4_extent_block_csum_set(inode
, neh
);
1223 set_buffer_uptodate(bh
);
1226 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1232 /* correct old index */
1234 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1237 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1238 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1246 /* insert new index */
1247 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1248 le32_to_cpu(border
), newblock
);
1252 if (buffer_locked(bh
))
1258 /* free all allocated blocks in error case */
1259 for (i
= 0; i
< depth
; i
++) {
1262 ext4_free_blocks(handle
, inode
, NULL
, ablocks
[i
], 1,
1263 EXT4_FREE_BLOCKS_METADATA
);
1272 * ext4_ext_grow_indepth:
1273 * implements tree growing procedure:
1274 * - allocates new block
1275 * - moves top-level data (index block or leaf) into the new block
1276 * - initializes new top-level, creating index that points to the
1277 * just created block
1279 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1282 struct ext4_extent_header
*neh
;
1283 struct buffer_head
*bh
;
1284 ext4_fsblk_t newblock
, goal
= 0;
1285 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
1288 /* Try to prepend new index to old one */
1289 if (ext_depth(inode
))
1290 goal
= ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode
)));
1291 if (goal
> le32_to_cpu(es
->s_first_data_block
)) {
1292 flags
|= EXT4_MB_HINT_TRY_GOAL
;
1295 goal
= ext4_inode_to_goal_block(inode
);
1296 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, flags
,
1301 bh
= sb_getblk_gfp(inode
->i_sb
, newblock
, __GFP_MOVABLE
| GFP_NOFS
);
1306 err
= ext4_journal_get_create_access(handle
, bh
);
1312 /* move top-level index/leaf into new block */
1313 memmove(bh
->b_data
, EXT4_I(inode
)->i_data
,
1314 sizeof(EXT4_I(inode
)->i_data
));
1316 /* set size of new block */
1317 neh
= ext_block_hdr(bh
);
1318 /* old root could have indexes or leaves
1319 * so calculate e_max right way */
1320 if (ext_depth(inode
))
1321 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1323 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1324 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1325 ext4_extent_block_csum_set(inode
, neh
);
1326 set_buffer_uptodate(bh
);
1329 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1333 /* Update top-level index: num,max,pointer */
1334 neh
= ext_inode_hdr(inode
);
1335 neh
->eh_entries
= cpu_to_le16(1);
1336 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh
), newblock
);
1337 if (neh
->eh_depth
== 0) {
1338 /* Root extent block becomes index block */
1339 neh
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1340 EXT_FIRST_INDEX(neh
)->ei_block
=
1341 EXT_FIRST_EXTENT(neh
)->ee_block
;
1343 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1344 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1345 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1346 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1348 le16_add_cpu(&neh
->eh_depth
, 1);
1349 ext4_mark_inode_dirty(handle
, inode
);
1357 * ext4_ext_create_new_leaf:
1358 * finds empty index and adds new leaf.
1359 * if no free index is found, then it requests in-depth growing.
1361 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1362 unsigned int mb_flags
,
1363 unsigned int gb_flags
,
1364 struct ext4_ext_path
**ppath
,
1365 struct ext4_extent
*newext
)
1367 struct ext4_ext_path
*path
= *ppath
;
1368 struct ext4_ext_path
*curp
;
1369 int depth
, i
, err
= 0;
1372 i
= depth
= ext_depth(inode
);
1374 /* walk up to the tree and look for free index entry */
1375 curp
= path
+ depth
;
1376 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1381 /* we use already allocated block for index block,
1382 * so subsequent data blocks should be contiguous */
1383 if (EXT_HAS_FREE_INDEX(curp
)) {
1384 /* if we found index with free entry, then use that
1385 * entry: create all needed subtree and add new leaf */
1386 err
= ext4_ext_split(handle
, inode
, mb_flags
, path
, newext
, i
);
1391 path
= ext4_find_extent(inode
,
1392 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1395 err
= PTR_ERR(path
);
1397 /* tree is full, time to grow in depth */
1398 err
= ext4_ext_grow_indepth(handle
, inode
, mb_flags
);
1403 path
= ext4_find_extent(inode
,
1404 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1407 err
= PTR_ERR(path
);
1412 * only first (depth 0 -> 1) produces free space;
1413 * in all other cases we have to split the grown tree
1415 depth
= ext_depth(inode
);
1416 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1417 /* now we need to split */
1427 * search the closest allocated block to the left for *logical
1428 * and returns it at @logical + it's physical address at @phys
1429 * if *logical is the smallest allocated block, the function
1430 * returns 0 at @phys
1431 * return value contains 0 (success) or error code
1433 static int ext4_ext_search_left(struct inode
*inode
,
1434 struct ext4_ext_path
*path
,
1435 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1437 struct ext4_extent_idx
*ix
;
1438 struct ext4_extent
*ex
;
1441 if (unlikely(path
== NULL
)) {
1442 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1443 return -EFSCORRUPTED
;
1445 depth
= path
->p_depth
;
1448 if (depth
== 0 && path
->p_ext
== NULL
)
1451 /* usually extent in the path covers blocks smaller
1452 * then *logical, but it can be that extent is the
1453 * first one in the file */
1455 ex
= path
[depth
].p_ext
;
1456 ee_len
= ext4_ext_get_actual_len(ex
);
1457 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1458 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1459 EXT4_ERROR_INODE(inode
,
1460 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1461 *logical
, le32_to_cpu(ex
->ee_block
));
1462 return -EFSCORRUPTED
;
1464 while (--depth
>= 0) {
1465 ix
= path
[depth
].p_idx
;
1466 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1467 EXT4_ERROR_INODE(inode
,
1468 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1469 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1470 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1471 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1473 return -EFSCORRUPTED
;
1479 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1480 EXT4_ERROR_INODE(inode
,
1481 "logical %d < ee_block %d + ee_len %d!",
1482 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1483 return -EFSCORRUPTED
;
1486 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1487 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1492 * search the closest allocated block to the right for *logical
1493 * and returns it at @logical + it's physical address at @phys
1494 * if *logical is the largest allocated block, the function
1495 * returns 0 at @phys
1496 * return value contains 0 (success) or error code
1498 static int ext4_ext_search_right(struct inode
*inode
,
1499 struct ext4_ext_path
*path
,
1500 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1501 struct ext4_extent
**ret_ex
)
1503 struct buffer_head
*bh
= NULL
;
1504 struct ext4_extent_header
*eh
;
1505 struct ext4_extent_idx
*ix
;
1506 struct ext4_extent
*ex
;
1508 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1511 if (unlikely(path
== NULL
)) {
1512 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1513 return -EFSCORRUPTED
;
1515 depth
= path
->p_depth
;
1518 if (depth
== 0 && path
->p_ext
== NULL
)
1521 /* usually extent in the path covers blocks smaller
1522 * then *logical, but it can be that extent is the
1523 * first one in the file */
1525 ex
= path
[depth
].p_ext
;
1526 ee_len
= ext4_ext_get_actual_len(ex
);
1527 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1528 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1529 EXT4_ERROR_INODE(inode
,
1530 "first_extent(path[%d].p_hdr) != ex",
1532 return -EFSCORRUPTED
;
1534 while (--depth
>= 0) {
1535 ix
= path
[depth
].p_idx
;
1536 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1537 EXT4_ERROR_INODE(inode
,
1538 "ix != EXT_FIRST_INDEX *logical %d!",
1540 return -EFSCORRUPTED
;
1546 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1547 EXT4_ERROR_INODE(inode
,
1548 "logical %d < ee_block %d + ee_len %d!",
1549 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1550 return -EFSCORRUPTED
;
1553 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1554 /* next allocated block in this leaf */
1559 /* go up and search for index to the right */
1560 while (--depth
>= 0) {
1561 ix
= path
[depth
].p_idx
;
1562 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1566 /* we've gone up to the root and found no index to the right */
1570 /* we've found index to the right, let's
1571 * follow it and find the closest allocated
1572 * block to the right */
1574 block
= ext4_idx_pblock(ix
);
1575 while (++depth
< path
->p_depth
) {
1576 /* subtract from p_depth to get proper eh_depth */
1577 bh
= read_extent_tree_block(inode
, block
,
1578 path
->p_depth
- depth
, 0);
1581 eh
= ext_block_hdr(bh
);
1582 ix
= EXT_FIRST_INDEX(eh
);
1583 block
= ext4_idx_pblock(ix
);
1587 bh
= read_extent_tree_block(inode
, block
, path
->p_depth
- depth
, 0);
1590 eh
= ext_block_hdr(bh
);
1591 ex
= EXT_FIRST_EXTENT(eh
);
1593 *logical
= le32_to_cpu(ex
->ee_block
);
1594 *phys
= ext4_ext_pblock(ex
);
1602 * ext4_ext_next_allocated_block:
1603 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1604 * NOTE: it considers block number from index entry as
1605 * allocated block. Thus, index entries have to be consistent
1609 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1613 BUG_ON(path
== NULL
);
1614 depth
= path
->p_depth
;
1616 if (depth
== 0 && path
->p_ext
== NULL
)
1617 return EXT_MAX_BLOCKS
;
1619 while (depth
>= 0) {
1620 if (depth
== path
->p_depth
) {
1622 if (path
[depth
].p_ext
&&
1623 path
[depth
].p_ext
!=
1624 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1625 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1628 if (path
[depth
].p_idx
!=
1629 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1630 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1635 return EXT_MAX_BLOCKS
;
1639 * ext4_ext_next_leaf_block:
1640 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1642 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1646 BUG_ON(path
== NULL
);
1647 depth
= path
->p_depth
;
1649 /* zero-tree has no leaf blocks at all */
1651 return EXT_MAX_BLOCKS
;
1653 /* go to index block */
1656 while (depth
>= 0) {
1657 if (path
[depth
].p_idx
!=
1658 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1659 return (ext4_lblk_t
)
1660 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1664 return EXT_MAX_BLOCKS
;
1668 * ext4_ext_correct_indexes:
1669 * if leaf gets modified and modified extent is first in the leaf,
1670 * then we have to correct all indexes above.
1671 * TODO: do we need to correct tree in all cases?
1673 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1674 struct ext4_ext_path
*path
)
1676 struct ext4_extent_header
*eh
;
1677 int depth
= ext_depth(inode
);
1678 struct ext4_extent
*ex
;
1682 eh
= path
[depth
].p_hdr
;
1683 ex
= path
[depth
].p_ext
;
1685 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1686 EXT4_ERROR_INODE(inode
,
1687 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1688 return -EFSCORRUPTED
;
1692 /* there is no tree at all */
1696 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1697 /* we correct tree if first leaf got modified only */
1702 * TODO: we need correction if border is smaller than current one
1705 border
= path
[depth
].p_ext
->ee_block
;
1706 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1709 path
[k
].p_idx
->ei_block
= border
;
1710 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1715 /* change all left-side indexes */
1716 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1718 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1721 path
[k
].p_idx
->ei_block
= border
;
1722 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1731 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1732 struct ext4_extent
*ex2
)
1734 unsigned short ext1_ee_len
, ext2_ee_len
;
1736 if (ext4_ext_is_unwritten(ex1
) != ext4_ext_is_unwritten(ex2
))
1739 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1740 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1742 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1743 le32_to_cpu(ex2
->ee_block
))
1747 * To allow future support for preallocated extents to be added
1748 * as an RO_COMPAT feature, refuse to merge to extents if
1749 * this can result in the top bit of ee_len being set.
1751 if (ext1_ee_len
+ ext2_ee_len
> EXT_INIT_MAX_LEN
)
1753 if (ext4_ext_is_unwritten(ex1
) &&
1754 (ext4_test_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
) ||
1755 atomic_read(&EXT4_I(inode
)->i_unwritten
) ||
1756 (ext1_ee_len
+ ext2_ee_len
> EXT_UNWRITTEN_MAX_LEN
)))
1758 #ifdef AGGRESSIVE_TEST
1759 if (ext1_ee_len
>= 4)
1763 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1769 * This function tries to merge the "ex" extent to the next extent in the tree.
1770 * It always tries to merge towards right. If you want to merge towards
1771 * left, pass "ex - 1" as argument instead of "ex".
1772 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1773 * 1 if they got merged.
1775 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1776 struct ext4_ext_path
*path
,
1777 struct ext4_extent
*ex
)
1779 struct ext4_extent_header
*eh
;
1780 unsigned int depth
, len
;
1781 int merge_done
= 0, unwritten
;
1783 depth
= ext_depth(inode
);
1784 BUG_ON(path
[depth
].p_hdr
== NULL
);
1785 eh
= path
[depth
].p_hdr
;
1787 while (ex
< EXT_LAST_EXTENT(eh
)) {
1788 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1790 /* merge with next extent! */
1791 unwritten
= ext4_ext_is_unwritten(ex
);
1792 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1793 + ext4_ext_get_actual_len(ex
+ 1));
1795 ext4_ext_mark_unwritten(ex
);
1797 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1798 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1799 * sizeof(struct ext4_extent
);
1800 memmove(ex
+ 1, ex
+ 2, len
);
1802 le16_add_cpu(&eh
->eh_entries
, -1);
1804 WARN_ON(eh
->eh_entries
== 0);
1805 if (!eh
->eh_entries
)
1806 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1813 * This function does a very simple check to see if we can collapse
1814 * an extent tree with a single extent tree leaf block into the inode.
1816 static void ext4_ext_try_to_merge_up(handle_t
*handle
,
1817 struct inode
*inode
,
1818 struct ext4_ext_path
*path
)
1821 unsigned max_root
= ext4_ext_space_root(inode
, 0);
1824 if ((path
[0].p_depth
!= 1) ||
1825 (le16_to_cpu(path
[0].p_hdr
->eh_entries
) != 1) ||
1826 (le16_to_cpu(path
[1].p_hdr
->eh_entries
) > max_root
))
1830 * We need to modify the block allocation bitmap and the block
1831 * group descriptor to release the extent tree block. If we
1832 * can't get the journal credits, give up.
1834 if (ext4_journal_extend(handle
, 2))
1838 * Copy the extent data up to the inode
1840 blk
= ext4_idx_pblock(path
[0].p_idx
);
1841 s
= le16_to_cpu(path
[1].p_hdr
->eh_entries
) *
1842 sizeof(struct ext4_extent_idx
);
1843 s
+= sizeof(struct ext4_extent_header
);
1845 path
[1].p_maxdepth
= path
[0].p_maxdepth
;
1846 memcpy(path
[0].p_hdr
, path
[1].p_hdr
, s
);
1847 path
[0].p_depth
= 0;
1848 path
[0].p_ext
= EXT_FIRST_EXTENT(path
[0].p_hdr
) +
1849 (path
[1].p_ext
- EXT_FIRST_EXTENT(path
[1].p_hdr
));
1850 path
[0].p_hdr
->eh_max
= cpu_to_le16(max_root
);
1852 brelse(path
[1].p_bh
);
1853 ext4_free_blocks(handle
, inode
, NULL
, blk
, 1,
1854 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
1858 * This function tries to merge the @ex extent to neighbours in the tree.
1859 * return 1 if merge left else 0.
1861 static void ext4_ext_try_to_merge(handle_t
*handle
,
1862 struct inode
*inode
,
1863 struct ext4_ext_path
*path
,
1864 struct ext4_extent
*ex
) {
1865 struct ext4_extent_header
*eh
;
1869 depth
= ext_depth(inode
);
1870 BUG_ON(path
[depth
].p_hdr
== NULL
);
1871 eh
= path
[depth
].p_hdr
;
1873 if (ex
> EXT_FIRST_EXTENT(eh
))
1874 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1877 (void) ext4_ext_try_to_merge_right(inode
, path
, ex
);
1879 ext4_ext_try_to_merge_up(handle
, inode
, path
);
1883 * check if a portion of the "newext" extent overlaps with an
1886 * If there is an overlap discovered, it updates the length of the newext
1887 * such that there will be no overlap, and then returns 1.
1888 * If there is no overlap found, it returns 0.
1890 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1891 struct inode
*inode
,
1892 struct ext4_extent
*newext
,
1893 struct ext4_ext_path
*path
)
1896 unsigned int depth
, len1
;
1897 unsigned int ret
= 0;
1899 b1
= le32_to_cpu(newext
->ee_block
);
1900 len1
= ext4_ext_get_actual_len(newext
);
1901 depth
= ext_depth(inode
);
1902 if (!path
[depth
].p_ext
)
1904 b2
= EXT4_LBLK_CMASK(sbi
, le32_to_cpu(path
[depth
].p_ext
->ee_block
));
1907 * get the next allocated block if the extent in the path
1908 * is before the requested block(s)
1911 b2
= ext4_ext_next_allocated_block(path
);
1912 if (b2
== EXT_MAX_BLOCKS
)
1914 b2
= EXT4_LBLK_CMASK(sbi
, b2
);
1917 /* check for wrap through zero on extent logical start block*/
1918 if (b1
+ len1
< b1
) {
1919 len1
= EXT_MAX_BLOCKS
- b1
;
1920 newext
->ee_len
= cpu_to_le16(len1
);
1924 /* check for overlap */
1925 if (b1
+ len1
> b2
) {
1926 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1934 * ext4_ext_insert_extent:
1935 * tries to merge requsted extent into the existing extent or
1936 * inserts requested extent as new one into the tree,
1937 * creating new leaf in the no-space case.
1939 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1940 struct ext4_ext_path
**ppath
,
1941 struct ext4_extent
*newext
, int gb_flags
)
1943 struct ext4_ext_path
*path
= *ppath
;
1944 struct ext4_extent_header
*eh
;
1945 struct ext4_extent
*ex
, *fex
;
1946 struct ext4_extent
*nearex
; /* nearest extent */
1947 struct ext4_ext_path
*npath
= NULL
;
1948 int depth
, len
, err
;
1950 int mb_flags
= 0, unwritten
;
1952 if (gb_flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
1953 mb_flags
|= EXT4_MB_DELALLOC_RESERVED
;
1954 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1955 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1956 return -EFSCORRUPTED
;
1958 depth
= ext_depth(inode
);
1959 ex
= path
[depth
].p_ext
;
1960 eh
= path
[depth
].p_hdr
;
1961 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1962 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1963 return -EFSCORRUPTED
;
1966 /* try to insert block into found extent and return */
1967 if (ex
&& !(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
1970 * Try to see whether we should rather test the extent on
1971 * right from ex, or from the left of ex. This is because
1972 * ext4_find_extent() can return either extent on the
1973 * left, or on the right from the searched position. This
1974 * will make merging more effective.
1976 if (ex
< EXT_LAST_EXTENT(eh
) &&
1977 (le32_to_cpu(ex
->ee_block
) +
1978 ext4_ext_get_actual_len(ex
) <
1979 le32_to_cpu(newext
->ee_block
))) {
1982 } else if ((ex
> EXT_FIRST_EXTENT(eh
)) &&
1983 (le32_to_cpu(newext
->ee_block
) +
1984 ext4_ext_get_actual_len(newext
) <
1985 le32_to_cpu(ex
->ee_block
)))
1988 /* Try to append newex to the ex */
1989 if (ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1990 ext_debug("append [%d]%d block to %u:[%d]%d"
1992 ext4_ext_is_unwritten(newext
),
1993 ext4_ext_get_actual_len(newext
),
1994 le32_to_cpu(ex
->ee_block
),
1995 ext4_ext_is_unwritten(ex
),
1996 ext4_ext_get_actual_len(ex
),
1997 ext4_ext_pblock(ex
));
1998 err
= ext4_ext_get_access(handle
, inode
,
2002 unwritten
= ext4_ext_is_unwritten(ex
);
2003 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
2004 + ext4_ext_get_actual_len(newext
));
2006 ext4_ext_mark_unwritten(ex
);
2007 eh
= path
[depth
].p_hdr
;
2013 /* Try to prepend newex to the ex */
2014 if (ext4_can_extents_be_merged(inode
, newext
, ex
)) {
2015 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
2017 le32_to_cpu(newext
->ee_block
),
2018 ext4_ext_is_unwritten(newext
),
2019 ext4_ext_get_actual_len(newext
),
2020 le32_to_cpu(ex
->ee_block
),
2021 ext4_ext_is_unwritten(ex
),
2022 ext4_ext_get_actual_len(ex
),
2023 ext4_ext_pblock(ex
));
2024 err
= ext4_ext_get_access(handle
, inode
,
2029 unwritten
= ext4_ext_is_unwritten(ex
);
2030 ex
->ee_block
= newext
->ee_block
;
2031 ext4_ext_store_pblock(ex
, ext4_ext_pblock(newext
));
2032 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
2033 + ext4_ext_get_actual_len(newext
));
2035 ext4_ext_mark_unwritten(ex
);
2036 eh
= path
[depth
].p_hdr
;
2042 depth
= ext_depth(inode
);
2043 eh
= path
[depth
].p_hdr
;
2044 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
2047 /* probably next leaf has space for us? */
2048 fex
= EXT_LAST_EXTENT(eh
);
2049 next
= EXT_MAX_BLOCKS
;
2050 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
2051 next
= ext4_ext_next_leaf_block(path
);
2052 if (next
!= EXT_MAX_BLOCKS
) {
2053 ext_debug("next leaf block - %u\n", next
);
2054 BUG_ON(npath
!= NULL
);
2055 npath
= ext4_find_extent(inode
, next
, NULL
, 0);
2057 return PTR_ERR(npath
);
2058 BUG_ON(npath
->p_depth
!= path
->p_depth
);
2059 eh
= npath
[depth
].p_hdr
;
2060 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
2061 ext_debug("next leaf isn't full(%d)\n",
2062 le16_to_cpu(eh
->eh_entries
));
2066 ext_debug("next leaf has no free space(%d,%d)\n",
2067 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
2071 * There is no free space in the found leaf.
2072 * We're gonna add a new leaf in the tree.
2074 if (gb_flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
2075 mb_flags
|= EXT4_MB_USE_RESERVED
;
2076 err
= ext4_ext_create_new_leaf(handle
, inode
, mb_flags
, gb_flags
,
2080 depth
= ext_depth(inode
);
2081 eh
= path
[depth
].p_hdr
;
2084 nearex
= path
[depth
].p_ext
;
2086 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2091 /* there is no extent in this leaf, create first one */
2092 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2093 le32_to_cpu(newext
->ee_block
),
2094 ext4_ext_pblock(newext
),
2095 ext4_ext_is_unwritten(newext
),
2096 ext4_ext_get_actual_len(newext
));
2097 nearex
= EXT_FIRST_EXTENT(eh
);
2099 if (le32_to_cpu(newext
->ee_block
)
2100 > le32_to_cpu(nearex
->ee_block
)) {
2102 ext_debug("insert %u:%llu:[%d]%d before: "
2104 le32_to_cpu(newext
->ee_block
),
2105 ext4_ext_pblock(newext
),
2106 ext4_ext_is_unwritten(newext
),
2107 ext4_ext_get_actual_len(newext
),
2112 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
2113 ext_debug("insert %u:%llu:[%d]%d after: "
2115 le32_to_cpu(newext
->ee_block
),
2116 ext4_ext_pblock(newext
),
2117 ext4_ext_is_unwritten(newext
),
2118 ext4_ext_get_actual_len(newext
),
2121 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
2123 ext_debug("insert %u:%llu:[%d]%d: "
2124 "move %d extents from 0x%p to 0x%p\n",
2125 le32_to_cpu(newext
->ee_block
),
2126 ext4_ext_pblock(newext
),
2127 ext4_ext_is_unwritten(newext
),
2128 ext4_ext_get_actual_len(newext
),
2129 len
, nearex
, nearex
+ 1);
2130 memmove(nearex
+ 1, nearex
,
2131 len
* sizeof(struct ext4_extent
));
2135 le16_add_cpu(&eh
->eh_entries
, 1);
2136 path
[depth
].p_ext
= nearex
;
2137 nearex
->ee_block
= newext
->ee_block
;
2138 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
2139 nearex
->ee_len
= newext
->ee_len
;
2142 /* try to merge extents */
2143 if (!(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
))
2144 ext4_ext_try_to_merge(handle
, inode
, path
, nearex
);
2147 /* time to correct all indexes above */
2148 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2152 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
2155 ext4_ext_drop_refs(npath
);
2160 static int ext4_fill_fiemap_extents(struct inode
*inode
,
2161 ext4_lblk_t block
, ext4_lblk_t num
,
2162 struct fiemap_extent_info
*fieinfo
)
2164 struct ext4_ext_path
*path
= NULL
;
2165 struct ext4_extent
*ex
;
2166 struct extent_status es
;
2167 ext4_lblk_t next
, next_del
, start
= 0, end
= 0;
2168 ext4_lblk_t last
= block
+ num
;
2169 int exists
, depth
= 0, err
= 0;
2170 unsigned int flags
= 0;
2171 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
2173 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
2175 /* find extent for this block */
2176 down_read(&EXT4_I(inode
)->i_data_sem
);
2178 path
= ext4_find_extent(inode
, block
, &path
, 0);
2180 up_read(&EXT4_I(inode
)->i_data_sem
);
2181 err
= PTR_ERR(path
);
2186 depth
= ext_depth(inode
);
2187 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2188 up_read(&EXT4_I(inode
)->i_data_sem
);
2189 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2190 err
= -EFSCORRUPTED
;
2193 ex
= path
[depth
].p_ext
;
2194 next
= ext4_ext_next_allocated_block(path
);
2199 /* there is no extent yet, so try to allocate
2200 * all requested space */
2203 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
2204 /* need to allocate space before found extent */
2206 end
= le32_to_cpu(ex
->ee_block
);
2207 if (block
+ num
< end
)
2209 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2210 + ext4_ext_get_actual_len(ex
)) {
2211 /* need to allocate space after found extent */
2216 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
2218 * some part of requested space is covered
2222 end
= le32_to_cpu(ex
->ee_block
)
2223 + ext4_ext_get_actual_len(ex
);
2224 if (block
+ num
< end
)
2230 BUG_ON(end
<= start
);
2234 es
.es_len
= end
- start
;
2237 es
.es_lblk
= le32_to_cpu(ex
->ee_block
);
2238 es
.es_len
= ext4_ext_get_actual_len(ex
);
2239 es
.es_pblk
= ext4_ext_pblock(ex
);
2240 if (ext4_ext_is_unwritten(ex
))
2241 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
2245 * Find delayed extent and update es accordingly. We call
2246 * it even in !exists case to find out whether es is the
2247 * last existing extent or not.
2249 next_del
= ext4_find_delayed_extent(inode
, &es
);
2250 if (!exists
&& next_del
) {
2252 flags
|= (FIEMAP_EXTENT_DELALLOC
|
2253 FIEMAP_EXTENT_UNKNOWN
);
2255 up_read(&EXT4_I(inode
)->i_data_sem
);
2257 if (unlikely(es
.es_len
== 0)) {
2258 EXT4_ERROR_INODE(inode
, "es.es_len == 0");
2259 err
= -EFSCORRUPTED
;
2264 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2265 * we need to check next == EXT_MAX_BLOCKS because it is
2266 * possible that an extent is with unwritten and delayed
2267 * status due to when an extent is delayed allocated and
2268 * is allocated by fallocate status tree will track both of
2271 * So we could return a unwritten and delayed extent, and
2272 * its block is equal to 'next'.
2274 if (next
== next_del
&& next
== EXT_MAX_BLOCKS
) {
2275 flags
|= FIEMAP_EXTENT_LAST
;
2276 if (unlikely(next_del
!= EXT_MAX_BLOCKS
||
2277 next
!= EXT_MAX_BLOCKS
)) {
2278 EXT4_ERROR_INODE(inode
,
2279 "next extent == %u, next "
2280 "delalloc extent = %u",
2282 err
= -EFSCORRUPTED
;
2288 err
= fiemap_fill_next_extent(fieinfo
,
2289 (__u64
)es
.es_lblk
<< blksize_bits
,
2290 (__u64
)es
.es_pblk
<< blksize_bits
,
2291 (__u64
)es
.es_len
<< blksize_bits
,
2301 block
= es
.es_lblk
+ es
.es_len
;
2304 ext4_ext_drop_refs(path
);
2310 * ext4_ext_put_gap_in_cache:
2311 * calculate boundaries of the gap that the requested block fits into
2312 * and cache this gap
2315 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
2318 int depth
= ext_depth(inode
);
2321 struct ext4_extent
*ex
;
2322 struct extent_status es
;
2324 ex
= path
[depth
].p_ext
;
2326 /* there is no extent yet, so gap is [0;-] */
2328 len
= EXT_MAX_BLOCKS
;
2329 ext_debug("cache gap(whole file):");
2330 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
2332 len
= le32_to_cpu(ex
->ee_block
) - block
;
2333 ext_debug("cache gap(before): %u [%u:%u]",
2335 le32_to_cpu(ex
->ee_block
),
2336 ext4_ext_get_actual_len(ex
));
2337 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2338 + ext4_ext_get_actual_len(ex
)) {
2340 lblock
= le32_to_cpu(ex
->ee_block
)
2341 + ext4_ext_get_actual_len(ex
);
2343 next
= ext4_ext_next_allocated_block(path
);
2344 ext_debug("cache gap(after): [%u:%u] %u",
2345 le32_to_cpu(ex
->ee_block
),
2346 ext4_ext_get_actual_len(ex
),
2348 BUG_ON(next
== lblock
);
2349 len
= next
- lblock
;
2354 ext4_es_find_delayed_extent_range(inode
, lblock
, lblock
+ len
- 1, &es
);
2356 /* There's delayed extent containing lblock? */
2357 if (es
.es_lblk
<= lblock
)
2359 len
= min(es
.es_lblk
- lblock
, len
);
2361 ext_debug(" -> %u:%u\n", lblock
, len
);
2362 ext4_es_insert_extent(inode
, lblock
, len
, ~0, EXTENT_STATUS_HOLE
);
2367 * removes index from the index block.
2369 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2370 struct ext4_ext_path
*path
, int depth
)
2375 /* free index block */
2377 path
= path
+ depth
;
2378 leaf
= ext4_idx_pblock(path
->p_idx
);
2379 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2380 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2381 return -EFSCORRUPTED
;
2383 err
= ext4_ext_get_access(handle
, inode
, path
);
2387 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2388 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2389 len
*= sizeof(struct ext4_extent_idx
);
2390 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2393 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2394 err
= ext4_ext_dirty(handle
, inode
, path
);
2397 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2398 trace_ext4_ext_rm_idx(inode
, leaf
);
2400 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2401 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2403 while (--depth
>= 0) {
2404 if (path
->p_idx
!= EXT_FIRST_INDEX(path
->p_hdr
))
2407 err
= ext4_ext_get_access(handle
, inode
, path
);
2410 path
->p_idx
->ei_block
= (path
+1)->p_idx
->ei_block
;
2411 err
= ext4_ext_dirty(handle
, inode
, path
);
2419 * ext4_ext_calc_credits_for_single_extent:
2420 * This routine returns max. credits that needed to insert an extent
2421 * to the extent tree.
2422 * When pass the actual path, the caller should calculate credits
2425 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2426 struct ext4_ext_path
*path
)
2429 int depth
= ext_depth(inode
);
2432 /* probably there is space in leaf? */
2433 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2434 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2437 * There are some space in the leaf tree, no
2438 * need to account for leaf block credit
2440 * bitmaps and block group descriptor blocks
2441 * and other metadata blocks still need to be
2444 /* 1 bitmap, 1 block group descriptor */
2445 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2450 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2454 * How many index/leaf blocks need to change/allocate to add @extents extents?
2456 * If we add a single extent, then in the worse case, each tree level
2457 * index/leaf need to be changed in case of the tree split.
2459 * If more extents are inserted, they could cause the whole tree split more
2460 * than once, but this is really rare.
2462 int ext4_ext_index_trans_blocks(struct inode
*inode
, int extents
)
2467 /* If we are converting the inline data, only one is needed here. */
2468 if (ext4_has_inline_data(inode
))
2471 depth
= ext_depth(inode
);
2481 static inline int get_default_free_blocks_flags(struct inode
*inode
)
2483 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2484 return EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
;
2485 else if (ext4_should_journal_data(inode
))
2486 return EXT4_FREE_BLOCKS_FORGET
;
2490 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2491 struct ext4_extent
*ex
,
2492 long long *partial_cluster
,
2493 ext4_lblk_t from
, ext4_lblk_t to
)
2495 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2496 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2498 int flags
= get_default_free_blocks_flags(inode
);
2501 * For bigalloc file systems, we never free a partial cluster
2502 * at the beginning of the extent. Instead, we make a note
2503 * that we tried freeing the cluster, and check to see if we
2504 * need to free it on a subsequent call to ext4_remove_blocks,
2505 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2507 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2509 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2511 * If we have a partial cluster, and it's different from the
2512 * cluster of the last block, we need to explicitly free the
2513 * partial cluster here.
2515 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2516 if (*partial_cluster
> 0 &&
2517 *partial_cluster
!= (long long) EXT4_B2C(sbi
, pblk
)) {
2518 ext4_free_blocks(handle
, inode
, NULL
,
2519 EXT4_C2B(sbi
, *partial_cluster
),
2520 sbi
->s_cluster_ratio
, flags
);
2521 *partial_cluster
= 0;
2524 #ifdef EXTENTS_STATS
2526 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2527 spin_lock(&sbi
->s_ext_stats_lock
);
2528 sbi
->s_ext_blocks
+= ee_len
;
2529 sbi
->s_ext_extents
++;
2530 if (ee_len
< sbi
->s_ext_min
)
2531 sbi
->s_ext_min
= ee_len
;
2532 if (ee_len
> sbi
->s_ext_max
)
2533 sbi
->s_ext_max
= ee_len
;
2534 if (ext_depth(inode
) > sbi
->s_depth_max
)
2535 sbi
->s_depth_max
= ext_depth(inode
);
2536 spin_unlock(&sbi
->s_ext_stats_lock
);
2539 if (from
>= le32_to_cpu(ex
->ee_block
)
2540 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2543 long long first_cluster
;
2545 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2546 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2548 * Usually we want to free partial cluster at the end of the
2549 * extent, except for the situation when the cluster is still
2550 * used by any other extent (partial_cluster is negative).
2552 if (*partial_cluster
< 0 &&
2553 *partial_cluster
== -(long long) EXT4_B2C(sbi
, pblk
+num
-1))
2554 flags
|= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER
;
2556 ext_debug("free last %u blocks starting %llu partial %lld\n",
2557 num
, pblk
, *partial_cluster
);
2558 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2560 * If the block range to be freed didn't start at the
2561 * beginning of a cluster, and we removed the entire
2562 * extent and the cluster is not used by any other extent,
2563 * save the partial cluster here, since we might need to
2564 * delete if we determine that the truncate or punch hole
2565 * operation has removed all of the blocks in the cluster.
2566 * If that cluster is used by another extent, preserve its
2567 * negative value so it isn't freed later on.
2569 * If the whole extent wasn't freed, we've reached the
2570 * start of the truncated/punched region and have finished
2571 * removing blocks. If there's a partial cluster here it's
2572 * shared with the remainder of the extent and is no longer
2573 * a candidate for removal.
2575 if (EXT4_PBLK_COFF(sbi
, pblk
) && ee_len
== num
) {
2576 first_cluster
= (long long) EXT4_B2C(sbi
, pblk
);
2577 if (first_cluster
!= -*partial_cluster
)
2578 *partial_cluster
= first_cluster
;
2580 *partial_cluster
= 0;
2583 ext4_error(sbi
->s_sb
, "strange request: removal(2) "
2584 "%u-%u from %u:%u\n",
2585 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2591 * ext4_ext_rm_leaf() Removes the extents associated with the
2592 * blocks appearing between "start" and "end". Both "start"
2593 * and "end" must appear in the same extent or EIO is returned.
2595 * @handle: The journal handle
2596 * @inode: The files inode
2597 * @path: The path to the leaf
2598 * @partial_cluster: The cluster which we'll have to free if all extents
2599 * has been released from it. However, if this value is
2600 * negative, it's a cluster just to the right of the
2601 * punched region and it must not be freed.
2602 * @start: The first block to remove
2603 * @end: The last block to remove
2606 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2607 struct ext4_ext_path
*path
,
2608 long long *partial_cluster
,
2609 ext4_lblk_t start
, ext4_lblk_t end
)
2611 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2612 int err
= 0, correct_index
= 0;
2613 int depth
= ext_depth(inode
), credits
;
2614 struct ext4_extent_header
*eh
;
2617 ext4_lblk_t ex_ee_block
;
2618 unsigned short ex_ee_len
;
2619 unsigned unwritten
= 0;
2620 struct ext4_extent
*ex
;
2623 /* the header must be checked already in ext4_ext_remove_space() */
2624 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2625 if (!path
[depth
].p_hdr
)
2626 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2627 eh
= path
[depth
].p_hdr
;
2628 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2629 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2630 return -EFSCORRUPTED
;
2632 /* find where to start removing */
2633 ex
= path
[depth
].p_ext
;
2635 ex
= EXT_LAST_EXTENT(eh
);
2637 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2638 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2640 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2642 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2643 ex_ee_block
+ ex_ee_len
> start
) {
2645 if (ext4_ext_is_unwritten(ex
))
2650 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2651 unwritten
, ex_ee_len
);
2652 path
[depth
].p_ext
= ex
;
2654 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2655 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2656 ex_ee_block
+ex_ee_len
- 1 : end
;
2658 ext_debug(" border %u:%u\n", a
, b
);
2660 /* If this extent is beyond the end of the hole, skip it */
2661 if (end
< ex_ee_block
) {
2663 * We're going to skip this extent and move to another,
2664 * so note that its first cluster is in use to avoid
2665 * freeing it when removing blocks. Eventually, the
2666 * right edge of the truncated/punched region will
2667 * be just to the left.
2669 if (sbi
->s_cluster_ratio
> 1) {
2670 pblk
= ext4_ext_pblock(ex
);
2672 -(long long) EXT4_B2C(sbi
, pblk
);
2675 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2676 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2678 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2679 EXT4_ERROR_INODE(inode
,
2680 "can not handle truncate %u:%u "
2682 start
, end
, ex_ee_block
,
2683 ex_ee_block
+ ex_ee_len
- 1);
2684 err
= -EFSCORRUPTED
;
2686 } else if (a
!= ex_ee_block
) {
2687 /* remove tail of the extent */
2688 num
= a
- ex_ee_block
;
2690 /* remove whole extent: excellent! */
2694 * 3 for leaf, sb, and inode plus 2 (bmap and group
2695 * descriptor) for each block group; assume two block
2696 * groups plus ex_ee_len/blocks_per_block_group for
2699 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2700 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2702 credits
+= (ext_depth(inode
)) + 1;
2704 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2706 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2710 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2714 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2720 /* this extent is removed; mark slot entirely unused */
2721 ext4_ext_store_pblock(ex
, 0);
2723 ex
->ee_len
= cpu_to_le16(num
);
2725 * Do not mark unwritten if all the blocks in the
2726 * extent have been removed.
2728 if (unwritten
&& num
)
2729 ext4_ext_mark_unwritten(ex
);
2731 * If the extent was completely released,
2732 * we need to remove it from the leaf
2735 if (end
!= EXT_MAX_BLOCKS
- 1) {
2737 * For hole punching, we need to scoot all the
2738 * extents up when an extent is removed so that
2739 * we dont have blank extents in the middle
2741 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2742 sizeof(struct ext4_extent
));
2744 /* Now get rid of the one at the end */
2745 memset(EXT_LAST_EXTENT(eh
), 0,
2746 sizeof(struct ext4_extent
));
2748 le16_add_cpu(&eh
->eh_entries
, -1);
2751 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2755 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2756 ext4_ext_pblock(ex
));
2758 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2759 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2762 if (correct_index
&& eh
->eh_entries
)
2763 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2766 * If there's a partial cluster and at least one extent remains in
2767 * the leaf, free the partial cluster if it isn't shared with the
2768 * current extent. If it is shared with the current extent
2769 * we zero partial_cluster because we've reached the start of the
2770 * truncated/punched region and we're done removing blocks.
2772 if (*partial_cluster
> 0 && ex
>= EXT_FIRST_EXTENT(eh
)) {
2773 pblk
= ext4_ext_pblock(ex
) + ex_ee_len
- 1;
2774 if (*partial_cluster
!= (long long) EXT4_B2C(sbi
, pblk
)) {
2775 ext4_free_blocks(handle
, inode
, NULL
,
2776 EXT4_C2B(sbi
, *partial_cluster
),
2777 sbi
->s_cluster_ratio
,
2778 get_default_free_blocks_flags(inode
));
2780 *partial_cluster
= 0;
2783 /* if this leaf is free, then we should
2784 * remove it from index block above */
2785 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2786 err
= ext4_ext_rm_idx(handle
, inode
, path
, depth
);
2793 * ext4_ext_more_to_rm:
2794 * returns 1 if current index has to be freed (even partial)
2797 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2799 BUG_ON(path
->p_idx
== NULL
);
2801 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2805 * if truncate on deeper level happened, it wasn't partial,
2806 * so we have to consider current index for truncation
2808 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2813 int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2816 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2817 int depth
= ext_depth(inode
);
2818 struct ext4_ext_path
*path
= NULL
;
2819 long long partial_cluster
= 0;
2823 ext_debug("truncate since %u to %u\n", start
, end
);
2825 /* probably first extent we're gonna free will be last in block */
2826 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, depth
+ 1);
2828 return PTR_ERR(handle
);
2831 trace_ext4_ext_remove_space(inode
, start
, end
, depth
);
2834 * Check if we are removing extents inside the extent tree. If that
2835 * is the case, we are going to punch a hole inside the extent tree
2836 * so we have to check whether we need to split the extent covering
2837 * the last block to remove so we can easily remove the part of it
2838 * in ext4_ext_rm_leaf().
2840 if (end
< EXT_MAX_BLOCKS
- 1) {
2841 struct ext4_extent
*ex
;
2842 ext4_lblk_t ee_block
, ex_end
, lblk
;
2845 /* find extent for or closest extent to this block */
2846 path
= ext4_find_extent(inode
, end
, NULL
, EXT4_EX_NOCACHE
);
2848 ext4_journal_stop(handle
);
2849 return PTR_ERR(path
);
2851 depth
= ext_depth(inode
);
2852 /* Leaf not may not exist only if inode has no blocks at all */
2853 ex
= path
[depth
].p_ext
;
2856 EXT4_ERROR_INODE(inode
,
2857 "path[%d].p_hdr == NULL",
2859 err
= -EFSCORRUPTED
;
2864 ee_block
= le32_to_cpu(ex
->ee_block
);
2865 ex_end
= ee_block
+ ext4_ext_get_actual_len(ex
) - 1;
2868 * See if the last block is inside the extent, if so split
2869 * the extent at 'end' block so we can easily remove the
2870 * tail of the first part of the split extent in
2871 * ext4_ext_rm_leaf().
2873 if (end
>= ee_block
&& end
< ex_end
) {
2876 * If we're going to split the extent, note that
2877 * the cluster containing the block after 'end' is
2878 * in use to avoid freeing it when removing blocks.
2880 if (sbi
->s_cluster_ratio
> 1) {
2881 pblk
= ext4_ext_pblock(ex
) + end
- ee_block
+ 2;
2883 -(long long) EXT4_B2C(sbi
, pblk
);
2887 * Split the extent in two so that 'end' is the last
2888 * block in the first new extent. Also we should not
2889 * fail removing space due to ENOSPC so try to use
2890 * reserved block if that happens.
2892 err
= ext4_force_split_extent_at(handle
, inode
, &path
,
2897 } else if (sbi
->s_cluster_ratio
> 1 && end
>= ex_end
) {
2899 * If there's an extent to the right its first cluster
2900 * contains the immediate right boundary of the
2901 * truncated/punched region. Set partial_cluster to
2902 * its negative value so it won't be freed if shared
2903 * with the current extent. The end < ee_block case
2904 * is handled in ext4_ext_rm_leaf().
2907 err
= ext4_ext_search_right(inode
, path
, &lblk
, &pblk
,
2913 -(long long) EXT4_B2C(sbi
, pblk
);
2917 * We start scanning from right side, freeing all the blocks
2918 * after i_size and walking into the tree depth-wise.
2920 depth
= ext_depth(inode
);
2925 le16_to_cpu(path
[k
].p_hdr
->eh_entries
)+1;
2927 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
2930 ext4_journal_stop(handle
);
2933 path
[0].p_maxdepth
= path
[0].p_depth
= depth
;
2934 path
[0].p_hdr
= ext_inode_hdr(inode
);
2937 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0)) {
2938 err
= -EFSCORRUPTED
;
2944 while (i
>= 0 && err
== 0) {
2946 /* this is leaf block */
2947 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2948 &partial_cluster
, start
,
2950 /* root level has p_bh == NULL, brelse() eats this */
2951 brelse(path
[i
].p_bh
);
2952 path
[i
].p_bh
= NULL
;
2957 /* this is index block */
2958 if (!path
[i
].p_hdr
) {
2959 ext_debug("initialize header\n");
2960 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2963 if (!path
[i
].p_idx
) {
2964 /* this level hasn't been touched yet */
2965 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2966 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2967 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2969 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2971 /* we were already here, see at next index */
2975 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2976 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2978 if (ext4_ext_more_to_rm(path
+ i
)) {
2979 struct buffer_head
*bh
;
2980 /* go to the next level */
2981 ext_debug("move to level %d (block %llu)\n",
2982 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
2983 memset(path
+ i
+ 1, 0, sizeof(*path
));
2984 bh
= read_extent_tree_block(inode
,
2985 ext4_idx_pblock(path
[i
].p_idx
), depth
- i
- 1,
2988 /* should we reset i_size? */
2992 /* Yield here to deal with large extent trees.
2993 * Should be a no-op if we did IO above. */
2995 if (WARN_ON(i
+ 1 > depth
)) {
2996 err
= -EFSCORRUPTED
;
2999 path
[i
+ 1].p_bh
= bh
;
3001 /* save actual number of indexes since this
3002 * number is changed at the next iteration */
3003 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
3006 /* we finished processing this index, go up */
3007 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
3008 /* index is empty, remove it;
3009 * handle must be already prepared by the
3010 * truncatei_leaf() */
3011 err
= ext4_ext_rm_idx(handle
, inode
, path
, i
);
3013 /* root level has p_bh == NULL, brelse() eats this */
3014 brelse(path
[i
].p_bh
);
3015 path
[i
].p_bh
= NULL
;
3017 ext_debug("return to level %d\n", i
);
3021 trace_ext4_ext_remove_space_done(inode
, start
, end
, depth
,
3022 partial_cluster
, path
->p_hdr
->eh_entries
);
3025 * If we still have something in the partial cluster and we have removed
3026 * even the first extent, then we should free the blocks in the partial
3027 * cluster as well. (This code will only run when there are no leaves
3028 * to the immediate left of the truncated/punched region.)
3030 if (partial_cluster
> 0 && err
== 0) {
3031 /* don't zero partial_cluster since it's not used afterwards */
3032 ext4_free_blocks(handle
, inode
, NULL
,
3033 EXT4_C2B(sbi
, partial_cluster
),
3034 sbi
->s_cluster_ratio
,
3035 get_default_free_blocks_flags(inode
));
3038 /* TODO: flexible tree reduction should be here */
3039 if (path
->p_hdr
->eh_entries
== 0) {
3041 * truncate to zero freed all the tree,
3042 * so we need to correct eh_depth
3044 err
= ext4_ext_get_access(handle
, inode
, path
);
3046 ext_inode_hdr(inode
)->eh_depth
= 0;
3047 ext_inode_hdr(inode
)->eh_max
=
3048 cpu_to_le16(ext4_ext_space_root(inode
, 0));
3049 err
= ext4_ext_dirty(handle
, inode
, path
);
3053 ext4_ext_drop_refs(path
);
3058 ext4_journal_stop(handle
);
3064 * called at mount time
3066 void ext4_ext_init(struct super_block
*sb
)
3069 * possible initialization would be here
3072 if (ext4_has_feature_extents(sb
)) {
3073 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3074 printk(KERN_INFO
"EXT4-fs: file extents enabled"
3075 #ifdef AGGRESSIVE_TEST
3076 ", aggressive tests"
3078 #ifdef CHECK_BINSEARCH
3081 #ifdef EXTENTS_STATS
3086 #ifdef EXTENTS_STATS
3087 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
3088 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
3089 EXT4_SB(sb
)->s_ext_max
= 0;
3095 * called at umount time
3097 void ext4_ext_release(struct super_block
*sb
)
3099 if (!ext4_has_feature_extents(sb
))
3102 #ifdef EXTENTS_STATS
3103 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
3104 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3105 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3106 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
3107 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
3108 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3109 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
3114 static int ext4_zeroout_es(struct inode
*inode
, struct ext4_extent
*ex
)
3116 ext4_lblk_t ee_block
;
3117 ext4_fsblk_t ee_pblock
;
3118 unsigned int ee_len
;
3120 ee_block
= le32_to_cpu(ex
->ee_block
);
3121 ee_len
= ext4_ext_get_actual_len(ex
);
3122 ee_pblock
= ext4_ext_pblock(ex
);
3127 return ext4_es_insert_extent(inode
, ee_block
, ee_len
, ee_pblock
,
3128 EXTENT_STATUS_WRITTEN
);
3131 /* FIXME!! we need to try to merge to left or right after zero-out */
3132 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
3134 ext4_fsblk_t ee_pblock
;
3135 unsigned int ee_len
;
3138 ee_len
= ext4_ext_get_actual_len(ex
);
3139 ee_pblock
= ext4_ext_pblock(ex
);
3141 if (ext4_encrypted_inode(inode
))
3142 return ext4_encrypted_zeroout(inode
, ex
);
3144 ret
= sb_issue_zeroout(inode
->i_sb
, ee_pblock
, ee_len
, GFP_NOFS
);
3152 * ext4_split_extent_at() splits an extent at given block.
3154 * @handle: the journal handle
3155 * @inode: the file inode
3156 * @path: the path to the extent
3157 * @split: the logical block where the extent is splitted.
3158 * @split_flags: indicates if the extent could be zeroout if split fails, and
3159 * the states(init or unwritten) of new extents.
3160 * @flags: flags used to insert new extent to extent tree.
3163 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3164 * of which are deterimined by split_flag.
3166 * There are two cases:
3167 * a> the extent are splitted into two extent.
3168 * b> split is not needed, and just mark the extent.
3170 * return 0 on success.
3172 static int ext4_split_extent_at(handle_t
*handle
,
3173 struct inode
*inode
,
3174 struct ext4_ext_path
**ppath
,
3179 struct ext4_ext_path
*path
= *ppath
;
3180 ext4_fsblk_t newblock
;
3181 ext4_lblk_t ee_block
;
3182 struct ext4_extent
*ex
, newex
, orig_ex
, zero_ex
;
3183 struct ext4_extent
*ex2
= NULL
;
3184 unsigned int ee_len
, depth
;
3187 BUG_ON((split_flag
& (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
)) ==
3188 (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
));
3190 ext_debug("ext4_split_extents_at: inode %lu, logical"
3191 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
3193 ext4_ext_show_leaf(inode
, path
);
3195 depth
= ext_depth(inode
);
3196 ex
= path
[depth
].p_ext
;
3197 ee_block
= le32_to_cpu(ex
->ee_block
);
3198 ee_len
= ext4_ext_get_actual_len(ex
);
3199 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
3201 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
3202 BUG_ON(!ext4_ext_is_unwritten(ex
) &&
3203 split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3204 EXT4_EXT_MARK_UNWRIT1
|
3205 EXT4_EXT_MARK_UNWRIT2
));
3207 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3211 if (split
== ee_block
) {
3213 * case b: block @split is the block that the extent begins with
3214 * then we just change the state of the extent, and splitting
3217 if (split_flag
& EXT4_EXT_MARK_UNWRIT2
)
3218 ext4_ext_mark_unwritten(ex
);
3220 ext4_ext_mark_initialized(ex
);
3222 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
3223 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3225 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3230 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
3231 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
3232 if (split_flag
& EXT4_EXT_MARK_UNWRIT1
)
3233 ext4_ext_mark_unwritten(ex
);
3236 * path may lead to new leaf, not to original leaf any more
3237 * after ext4_ext_insert_extent() returns,
3239 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3241 goto fix_extent_len
;
3244 ex2
->ee_block
= cpu_to_le32(split
);
3245 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
3246 ext4_ext_store_pblock(ex2
, newblock
);
3247 if (split_flag
& EXT4_EXT_MARK_UNWRIT2
)
3248 ext4_ext_mark_unwritten(ex2
);
3250 err
= ext4_ext_insert_extent(handle
, inode
, ppath
, &newex
, flags
);
3251 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3252 if (split_flag
& (EXT4_EXT_DATA_VALID1
|EXT4_EXT_DATA_VALID2
)) {
3253 if (split_flag
& EXT4_EXT_DATA_VALID1
) {
3254 err
= ext4_ext_zeroout(inode
, ex2
);
3255 zero_ex
.ee_block
= ex2
->ee_block
;
3256 zero_ex
.ee_len
= cpu_to_le16(
3257 ext4_ext_get_actual_len(ex2
));
3258 ext4_ext_store_pblock(&zero_ex
,
3259 ext4_ext_pblock(ex2
));
3261 err
= ext4_ext_zeroout(inode
, ex
);
3262 zero_ex
.ee_block
= ex
->ee_block
;
3263 zero_ex
.ee_len
= cpu_to_le16(
3264 ext4_ext_get_actual_len(ex
));
3265 ext4_ext_store_pblock(&zero_ex
,
3266 ext4_ext_pblock(ex
));
3269 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3270 zero_ex
.ee_block
= orig_ex
.ee_block
;
3271 zero_ex
.ee_len
= cpu_to_le16(
3272 ext4_ext_get_actual_len(&orig_ex
));
3273 ext4_ext_store_pblock(&zero_ex
,
3274 ext4_ext_pblock(&orig_ex
));
3278 goto fix_extent_len
;
3279 /* update the extent length and mark as initialized */
3280 ex
->ee_len
= cpu_to_le16(ee_len
);
3281 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3282 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3284 goto fix_extent_len
;
3286 /* update extent status tree */
3287 err
= ext4_zeroout_es(inode
, &zero_ex
);
3291 goto fix_extent_len
;
3294 ext4_ext_show_leaf(inode
, path
);
3298 ex
->ee_len
= orig_ex
.ee_len
;
3299 ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3304 * ext4_split_extents() splits an extent and mark extent which is covered
3305 * by @map as split_flags indicates
3307 * It may result in splitting the extent into multiple extents (up to three)
3308 * There are three possibilities:
3309 * a> There is no split required
3310 * b> Splits in two extents: Split is happening at either end of the extent
3311 * c> Splits in three extents: Somone is splitting in middle of the extent
3314 static int ext4_split_extent(handle_t
*handle
,
3315 struct inode
*inode
,
3316 struct ext4_ext_path
**ppath
,
3317 struct ext4_map_blocks
*map
,
3321 struct ext4_ext_path
*path
= *ppath
;
3322 ext4_lblk_t ee_block
;
3323 struct ext4_extent
*ex
;
3324 unsigned int ee_len
, depth
;
3327 int split_flag1
, flags1
;
3328 int allocated
= map
->m_len
;
3330 depth
= ext_depth(inode
);
3331 ex
= path
[depth
].p_ext
;
3332 ee_block
= le32_to_cpu(ex
->ee_block
);
3333 ee_len
= ext4_ext_get_actual_len(ex
);
3334 unwritten
= ext4_ext_is_unwritten(ex
);
3336 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
3337 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
;
3338 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
3340 split_flag1
|= EXT4_EXT_MARK_UNWRIT1
|
3341 EXT4_EXT_MARK_UNWRIT2
;
3342 if (split_flag
& EXT4_EXT_DATA_VALID2
)
3343 split_flag1
|= EXT4_EXT_DATA_VALID1
;
3344 err
= ext4_split_extent_at(handle
, inode
, ppath
,
3345 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
3349 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3352 * Update path is required because previous ext4_split_extent_at() may
3353 * result in split of original leaf or extent zeroout.
3355 path
= ext4_find_extent(inode
, map
->m_lblk
, ppath
, 0);
3357 return PTR_ERR(path
);
3358 depth
= ext_depth(inode
);
3359 ex
= path
[depth
].p_ext
;
3361 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
3362 (unsigned long) map
->m_lblk
);
3363 return -EFSCORRUPTED
;
3365 unwritten
= ext4_ext_is_unwritten(ex
);
3368 if (map
->m_lblk
>= ee_block
) {
3369 split_flag1
= split_flag
& EXT4_EXT_DATA_VALID2
;
3371 split_flag1
|= EXT4_EXT_MARK_UNWRIT1
;
3372 split_flag1
|= split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3373 EXT4_EXT_MARK_UNWRIT2
);
3375 err
= ext4_split_extent_at(handle
, inode
, ppath
,
3376 map
->m_lblk
, split_flag1
, flags
);
3381 ext4_ext_show_leaf(inode
, path
);
3383 return err
? err
: allocated
;
3387 * This function is called by ext4_ext_map_blocks() if someone tries to write
3388 * to an unwritten extent. It may result in splitting the unwritten
3389 * extent into multiple extents (up to three - one initialized and two
3391 * There are three possibilities:
3392 * a> There is no split required: Entire extent should be initialized
3393 * b> Splits in two extents: Write is happening at either end of the extent
3394 * c> Splits in three extents: Somone is writing in middle of the extent
3397 * - The extent pointed to by 'path' is unwritten.
3398 * - The extent pointed to by 'path' contains a superset
3399 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3401 * Post-conditions on success:
3402 * - the returned value is the number of blocks beyond map->l_lblk
3403 * that are allocated and initialized.
3404 * It is guaranteed to be >= map->m_len.
3406 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3407 struct inode
*inode
,
3408 struct ext4_map_blocks
*map
,
3409 struct ext4_ext_path
**ppath
,
3412 struct ext4_ext_path
*path
= *ppath
;
3413 struct ext4_sb_info
*sbi
;
3414 struct ext4_extent_header
*eh
;
3415 struct ext4_map_blocks split_map
;
3416 struct ext4_extent zero_ex
;
3417 struct ext4_extent
*ex
, *abut_ex
;
3418 ext4_lblk_t ee_block
, eof_block
;
3419 unsigned int ee_len
, depth
, map_len
= map
->m_len
;
3420 int allocated
= 0, max_zeroout
= 0;
3424 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3425 "block %llu, max_blocks %u\n", inode
->i_ino
,
3426 (unsigned long long)map
->m_lblk
, map_len
);
3428 sbi
= EXT4_SB(inode
->i_sb
);
3429 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3430 inode
->i_sb
->s_blocksize_bits
;
3431 if (eof_block
< map
->m_lblk
+ map_len
)
3432 eof_block
= map
->m_lblk
+ map_len
;
3434 depth
= ext_depth(inode
);
3435 eh
= path
[depth
].p_hdr
;
3436 ex
= path
[depth
].p_ext
;
3437 ee_block
= le32_to_cpu(ex
->ee_block
);
3438 ee_len
= ext4_ext_get_actual_len(ex
);
3441 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3443 /* Pre-conditions */
3444 BUG_ON(!ext4_ext_is_unwritten(ex
));
3445 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3448 * Attempt to transfer newly initialized blocks from the currently
3449 * unwritten extent to its neighbor. This is much cheaper
3450 * than an insertion followed by a merge as those involve costly
3451 * memmove() calls. Transferring to the left is the common case in
3452 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3453 * followed by append writes.
3455 * Limitations of the current logic:
3456 * - L1: we do not deal with writes covering the whole extent.
3457 * This would require removing the extent if the transfer
3459 * - L2: we only attempt to merge with an extent stored in the
3460 * same extent tree node.
3462 if ((map
->m_lblk
== ee_block
) &&
3463 /* See if we can merge left */
3464 (map_len
< ee_len
) && /*L1*/
3465 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L2*/
3466 ext4_lblk_t prev_lblk
;
3467 ext4_fsblk_t prev_pblk
, ee_pblk
;
3468 unsigned int prev_len
;
3471 prev_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3472 prev_len
= ext4_ext_get_actual_len(abut_ex
);
3473 prev_pblk
= ext4_ext_pblock(abut_ex
);
3474 ee_pblk
= ext4_ext_pblock(ex
);
3477 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3478 * upon those conditions:
3479 * - C1: abut_ex is initialized,
3480 * - C2: abut_ex is logically abutting ex,
3481 * - C3: abut_ex is physically abutting ex,
3482 * - C4: abut_ex can receive the additional blocks without
3483 * overflowing the (initialized) length limit.
3485 if ((!ext4_ext_is_unwritten(abut_ex
)) && /*C1*/
3486 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3487 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3488 (prev_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3489 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3493 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3496 /* Shift the start of ex by 'map_len' blocks */
3497 ex
->ee_block
= cpu_to_le32(ee_block
+ map_len
);
3498 ext4_ext_store_pblock(ex
, ee_pblk
+ map_len
);
3499 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3500 ext4_ext_mark_unwritten(ex
); /* Restore the flag */
3502 /* Extend abut_ex by 'map_len' blocks */
3503 abut_ex
->ee_len
= cpu_to_le16(prev_len
+ map_len
);
3505 /* Result: number of initialized blocks past m_lblk */
3506 allocated
= map_len
;
3508 } else if (((map
->m_lblk
+ map_len
) == (ee_block
+ ee_len
)) &&
3509 (map_len
< ee_len
) && /*L1*/
3510 ex
< EXT_LAST_EXTENT(eh
)) { /*L2*/
3511 /* See if we can merge right */
3512 ext4_lblk_t next_lblk
;
3513 ext4_fsblk_t next_pblk
, ee_pblk
;
3514 unsigned int next_len
;
3517 next_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3518 next_len
= ext4_ext_get_actual_len(abut_ex
);
3519 next_pblk
= ext4_ext_pblock(abut_ex
);
3520 ee_pblk
= ext4_ext_pblock(ex
);
3523 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3524 * upon those conditions:
3525 * - C1: abut_ex is initialized,
3526 * - C2: abut_ex is logically abutting ex,
3527 * - C3: abut_ex is physically abutting ex,
3528 * - C4: abut_ex can receive the additional blocks without
3529 * overflowing the (initialized) length limit.
3531 if ((!ext4_ext_is_unwritten(abut_ex
)) && /*C1*/
3532 ((map
->m_lblk
+ map_len
) == next_lblk
) && /*C2*/
3533 ((ee_pblk
+ ee_len
) == next_pblk
) && /*C3*/
3534 (next_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3535 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3539 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3542 /* Shift the start of abut_ex by 'map_len' blocks */
3543 abut_ex
->ee_block
= cpu_to_le32(next_lblk
- map_len
);
3544 ext4_ext_store_pblock(abut_ex
, next_pblk
- map_len
);
3545 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3546 ext4_ext_mark_unwritten(ex
); /* Restore the flag */
3548 /* Extend abut_ex by 'map_len' blocks */
3549 abut_ex
->ee_len
= cpu_to_le16(next_len
+ map_len
);
3551 /* Result: number of initialized blocks past m_lblk */
3552 allocated
= map_len
;
3556 /* Mark the block containing both extents as dirty */
3557 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3559 /* Update path to point to the right extent */
3560 path
[depth
].p_ext
= abut_ex
;
3563 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3565 WARN_ON(map
->m_lblk
< ee_block
);
3567 * It is safe to convert extent to initialized via explicit
3568 * zeroout only if extent is fully inside i_size or new_size.
3570 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3572 if (EXT4_EXT_MAY_ZEROOUT
& split_flag
)
3573 max_zeroout
= sbi
->s_extent_max_zeroout_kb
>>
3574 (inode
->i_sb
->s_blocksize_bits
- 10);
3576 if (ext4_encrypted_inode(inode
))
3579 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3580 if (max_zeroout
&& (ee_len
<= max_zeroout
)) {
3581 err
= ext4_ext_zeroout(inode
, ex
);
3584 zero_ex
.ee_block
= ex
->ee_block
;
3585 zero_ex
.ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
));
3586 ext4_ext_store_pblock(&zero_ex
, ext4_ext_pblock(ex
));
3588 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3591 ext4_ext_mark_initialized(ex
);
3592 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3593 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3599 * 1. split the extent into three extents.
3600 * 2. split the extent into two extents, zeroout the first half.
3601 * 3. split the extent into two extents, zeroout the second half.
3602 * 4. split the extent into two extents with out zeroout.
3604 split_map
.m_lblk
= map
->m_lblk
;
3605 split_map
.m_len
= map
->m_len
;
3607 if (max_zeroout
&& (allocated
> map
->m_len
)) {
3608 if (allocated
<= max_zeroout
) {
3611 cpu_to_le32(map
->m_lblk
);
3612 zero_ex
.ee_len
= cpu_to_le16(allocated
);
3613 ext4_ext_store_pblock(&zero_ex
,
3614 ext4_ext_pblock(ex
) + map
->m_lblk
- ee_block
);
3615 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3618 split_map
.m_lblk
= map
->m_lblk
;
3619 split_map
.m_len
= allocated
;
3620 } else if (map
->m_lblk
- ee_block
+ map
->m_len
< max_zeroout
) {
3622 if (map
->m_lblk
!= ee_block
) {
3623 zero_ex
.ee_block
= ex
->ee_block
;
3624 zero_ex
.ee_len
= cpu_to_le16(map
->m_lblk
-
3626 ext4_ext_store_pblock(&zero_ex
,
3627 ext4_ext_pblock(ex
));
3628 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3633 split_map
.m_lblk
= ee_block
;
3634 split_map
.m_len
= map
->m_lblk
- ee_block
+ map
->m_len
;
3635 allocated
= map
->m_len
;
3639 err
= ext4_split_extent(handle
, inode
, ppath
, &split_map
, split_flag
,
3644 /* If we have gotten a failure, don't zero out status tree */
3646 err
= ext4_zeroout_es(inode
, &zero_ex
);
3647 return err
? err
: allocated
;
3651 * This function is called by ext4_ext_map_blocks() from
3652 * ext4_get_blocks_dio_write() when DIO to write
3653 * to an unwritten extent.
3655 * Writing to an unwritten extent may result in splitting the unwritten
3656 * extent into multiple initialized/unwritten extents (up to three)
3657 * There are three possibilities:
3658 * a> There is no split required: Entire extent should be unwritten
3659 * b> Splits in two extents: Write is happening at either end of the extent
3660 * c> Splits in three extents: Somone is writing in middle of the extent
3662 * This works the same way in the case of initialized -> unwritten conversion.
3664 * One of more index blocks maybe needed if the extent tree grow after
3665 * the unwritten extent split. To prevent ENOSPC occur at the IO
3666 * complete, we need to split the unwritten extent before DIO submit
3667 * the IO. The unwritten extent called at this time will be split
3668 * into three unwritten extent(at most). After IO complete, the part
3669 * being filled will be convert to initialized by the end_io callback function
3670 * via ext4_convert_unwritten_extents().
3672 * Returns the size of unwritten extent to be written on success.
3674 static int ext4_split_convert_extents(handle_t
*handle
,
3675 struct inode
*inode
,
3676 struct ext4_map_blocks
*map
,
3677 struct ext4_ext_path
**ppath
,
3680 struct ext4_ext_path
*path
= *ppath
;
3681 ext4_lblk_t eof_block
;
3682 ext4_lblk_t ee_block
;
3683 struct ext4_extent
*ex
;
3684 unsigned int ee_len
;
3685 int split_flag
= 0, depth
;
3687 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3688 __func__
, inode
->i_ino
,
3689 (unsigned long long)map
->m_lblk
, map
->m_len
);
3691 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3692 inode
->i_sb
->s_blocksize_bits
;
3693 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3694 eof_block
= map
->m_lblk
+ map
->m_len
;
3696 * It is safe to convert extent to initialized via explicit
3697 * zeroout only if extent is fully insde i_size or new_size.
3699 depth
= ext_depth(inode
);
3700 ex
= path
[depth
].p_ext
;
3701 ee_block
= le32_to_cpu(ex
->ee_block
);
3702 ee_len
= ext4_ext_get_actual_len(ex
);
3704 /* Convert to unwritten */
3705 if (flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
) {
3706 split_flag
|= EXT4_EXT_DATA_VALID1
;
3707 /* Convert to initialized */
3708 } else if (flags
& EXT4_GET_BLOCKS_CONVERT
) {
3709 split_flag
|= ee_block
+ ee_len
<= eof_block
?
3710 EXT4_EXT_MAY_ZEROOUT
: 0;
3711 split_flag
|= (EXT4_EXT_MARK_UNWRIT2
| EXT4_EXT_DATA_VALID2
);
3713 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3714 return ext4_split_extent(handle
, inode
, ppath
, map
, split_flag
, flags
);
3717 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3718 struct inode
*inode
,
3719 struct ext4_map_blocks
*map
,
3720 struct ext4_ext_path
**ppath
)
3722 struct ext4_ext_path
*path
= *ppath
;
3723 struct ext4_extent
*ex
;
3724 ext4_lblk_t ee_block
;
3725 unsigned int ee_len
;
3729 depth
= ext_depth(inode
);
3730 ex
= path
[depth
].p_ext
;
3731 ee_block
= le32_to_cpu(ex
->ee_block
);
3732 ee_len
= ext4_ext_get_actual_len(ex
);
3734 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3735 "block %llu, max_blocks %u\n", inode
->i_ino
,
3736 (unsigned long long)ee_block
, ee_len
);
3738 /* If extent is larger than requested it is a clear sign that we still
3739 * have some extent state machine issues left. So extent_split is still
3741 * TODO: Once all related issues will be fixed this situation should be
3744 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3746 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3747 " len %u; IO logical block %llu, len %u\n",
3748 inode
->i_ino
, (unsigned long long)ee_block
, ee_len
,
3749 (unsigned long long)map
->m_lblk
, map
->m_len
);
3751 err
= ext4_split_convert_extents(handle
, inode
, map
, ppath
,
3752 EXT4_GET_BLOCKS_CONVERT
);
3755 path
= ext4_find_extent(inode
, map
->m_lblk
, ppath
, 0);
3757 return PTR_ERR(path
);
3758 depth
= ext_depth(inode
);
3759 ex
= path
[depth
].p_ext
;
3762 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3765 /* first mark the extent as initialized */
3766 ext4_ext_mark_initialized(ex
);
3768 /* note: ext4_ext_correct_indexes() isn't needed here because
3769 * borders are not changed
3771 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3773 /* Mark modified extent as dirty */
3774 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3776 ext4_ext_show_leaf(inode
, path
);
3780 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3781 sector_t block
, int count
)
3784 for (i
= 0; i
< count
; i
++)
3785 unmap_underlying_metadata(bdev
, block
+ i
);
3789 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3791 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3793 struct ext4_ext_path
*path
,
3797 struct ext4_extent_header
*eh
;
3798 struct ext4_extent
*last_ex
;
3800 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3803 depth
= ext_depth(inode
);
3804 eh
= path
[depth
].p_hdr
;
3807 * We're going to remove EOFBLOCKS_FL entirely in future so we
3808 * do not care for this case anymore. Simply remove the flag
3809 * if there are no extents.
3811 if (unlikely(!eh
->eh_entries
))
3813 last_ex
= EXT_LAST_EXTENT(eh
);
3815 * We should clear the EOFBLOCKS_FL flag if we are writing the
3816 * last block in the last extent in the file. We test this by
3817 * first checking to see if the caller to
3818 * ext4_ext_get_blocks() was interested in the last block (or
3819 * a block beyond the last block) in the current extent. If
3820 * this turns out to be false, we can bail out from this
3821 * function immediately.
3823 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3824 ext4_ext_get_actual_len(last_ex
))
3827 * If the caller does appear to be planning to write at or
3828 * beyond the end of the current extent, we then test to see
3829 * if the current extent is the last extent in the file, by
3830 * checking to make sure it was reached via the rightmost node
3831 * at each level of the tree.
3833 for (i
= depth
-1; i
>= 0; i
--)
3834 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3837 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3838 return ext4_mark_inode_dirty(handle
, inode
);
3842 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3844 * Return 1 if there is a delalloc block in the range, otherwise 0.
3846 int ext4_find_delalloc_range(struct inode
*inode
,
3847 ext4_lblk_t lblk_start
,
3848 ext4_lblk_t lblk_end
)
3850 struct extent_status es
;
3852 ext4_es_find_delayed_extent_range(inode
, lblk_start
, lblk_end
, &es
);
3854 return 0; /* there is no delay extent in this tree */
3855 else if (es
.es_lblk
<= lblk_start
&&
3856 lblk_start
< es
.es_lblk
+ es
.es_len
)
3858 else if (lblk_start
<= es
.es_lblk
&& es
.es_lblk
<= lblk_end
)
3864 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
)
3866 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3867 ext4_lblk_t lblk_start
, lblk_end
;
3868 lblk_start
= EXT4_LBLK_CMASK(sbi
, lblk
);
3869 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3871 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
);
3875 * Determines how many complete clusters (out of those specified by the 'map')
3876 * are under delalloc and were reserved quota for.
3877 * This function is called when we are writing out the blocks that were
3878 * originally written with their allocation delayed, but then the space was
3879 * allocated using fallocate() before the delayed allocation could be resolved.
3880 * The cases to look for are:
3881 * ('=' indicated delayed allocated blocks
3882 * '-' indicates non-delayed allocated blocks)
3883 * (a) partial clusters towards beginning and/or end outside of allocated range
3884 * are not delalloc'ed.
3886 * |----c---=|====c====|====c====|===-c----|
3887 * |++++++ allocated ++++++|
3888 * ==> 4 complete clusters in above example
3890 * (b) partial cluster (outside of allocated range) towards either end is
3891 * marked for delayed allocation. In this case, we will exclude that
3894 * |----====c========|========c========|
3895 * |++++++ allocated ++++++|
3896 * ==> 1 complete clusters in above example
3899 * |================c================|
3900 * |++++++ allocated ++++++|
3901 * ==> 0 complete clusters in above example
3903 * The ext4_da_update_reserve_space will be called only if we
3904 * determine here that there were some "entire" clusters that span
3905 * this 'allocated' range.
3906 * In the non-bigalloc case, this function will just end up returning num_blks
3907 * without ever calling ext4_find_delalloc_range.
3910 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3911 unsigned int num_blks
)
3913 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3914 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3915 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3916 unsigned int allocated_clusters
= 0;
3918 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3919 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3921 /* max possible clusters for this allocation */
3922 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3924 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3926 /* Check towards left side */
3927 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
);
3929 lblk_from
= EXT4_LBLK_CMASK(sbi
, lblk_start
);
3930 lblk_to
= lblk_from
+ c_offset
- 1;
3932 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3933 allocated_clusters
--;
3936 /* Now check towards right. */
3937 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
+ num_blks
);
3938 if (allocated_clusters
&& c_offset
) {
3939 lblk_from
= lblk_start
+ num_blks
;
3940 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3942 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3943 allocated_clusters
--;
3946 return allocated_clusters
;
3950 convert_initialized_extent(handle_t
*handle
, struct inode
*inode
,
3951 struct ext4_map_blocks
*map
,
3952 struct ext4_ext_path
**ppath
, int flags
,
3953 unsigned int allocated
, ext4_fsblk_t newblock
)
3955 struct ext4_ext_path
*path
= *ppath
;
3956 struct ext4_extent
*ex
;
3957 ext4_lblk_t ee_block
;
3958 unsigned int ee_len
;
3963 * Make sure that the extent is no bigger than we support with
3966 if (map
->m_len
> EXT_UNWRITTEN_MAX_LEN
)
3967 map
->m_len
= EXT_UNWRITTEN_MAX_LEN
/ 2;
3969 depth
= ext_depth(inode
);
3970 ex
= path
[depth
].p_ext
;
3971 ee_block
= le32_to_cpu(ex
->ee_block
);
3972 ee_len
= ext4_ext_get_actual_len(ex
);
3974 ext_debug("%s: inode %lu, logical"
3975 "block %llu, max_blocks %u\n", __func__
, inode
->i_ino
,
3976 (unsigned long long)ee_block
, ee_len
);
3978 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3979 err
= ext4_split_convert_extents(handle
, inode
, map
, ppath
,
3980 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
);
3983 path
= ext4_find_extent(inode
, map
->m_lblk
, ppath
, 0);
3985 return PTR_ERR(path
);
3986 depth
= ext_depth(inode
);
3987 ex
= path
[depth
].p_ext
;
3989 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
3990 (unsigned long) map
->m_lblk
);
3991 return -EFSCORRUPTED
;
3995 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3998 /* first mark the extent as unwritten */
3999 ext4_ext_mark_unwritten(ex
);
4001 /* note: ext4_ext_correct_indexes() isn't needed here because
4002 * borders are not changed
4004 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
4006 /* Mark modified extent as dirty */
4007 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
4010 ext4_ext_show_leaf(inode
, path
);
4012 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4013 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
, map
->m_len
);
4016 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4017 if (allocated
> map
->m_len
)
4018 allocated
= map
->m_len
;
4019 map
->m_len
= allocated
;
4024 ext4_ext_handle_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
4025 struct ext4_map_blocks
*map
,
4026 struct ext4_ext_path
**ppath
, int flags
,
4027 unsigned int allocated
, ext4_fsblk_t newblock
)
4029 struct ext4_ext_path
*path
= *ppath
;
4032 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
4034 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4035 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4036 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
4038 ext4_ext_show_leaf(inode
, path
);
4041 * When writing into unwritten space, we should not fail to
4042 * allocate metadata blocks for the new extent block if needed.
4044 flags
|= EXT4_GET_BLOCKS_METADATA_NOFAIL
;
4046 trace_ext4_ext_handle_unwritten_extents(inode
, map
, flags
,
4047 allocated
, newblock
);
4049 /* get_block() before submit the IO, split the extent */
4050 if (flags
& EXT4_GET_BLOCKS_PRE_IO
) {
4051 ret
= ext4_split_convert_extents(handle
, inode
, map
, ppath
,
4052 flags
| EXT4_GET_BLOCKS_CONVERT
);
4056 * Flag the inode(non aio case) or end_io struct (aio case)
4057 * that this IO needs to conversion to written when IO is
4061 ext4_set_io_unwritten_flag(inode
, io
);
4063 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
4064 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4067 /* IO end_io complete, convert the filled extent to written */
4068 if (flags
& EXT4_GET_BLOCKS_CONVERT
) {
4069 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
, map
,
4072 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4073 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4077 map
->m_flags
|= EXT4_MAP_MAPPED
;
4078 map
->m_pblk
= newblock
;
4079 if (allocated
> map
->m_len
)
4080 allocated
= map
->m_len
;
4081 map
->m_len
= allocated
;
4084 /* buffered IO case */
4086 * repeat fallocate creation request
4087 * we already have an unwritten extent
4089 if (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
) {
4090 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4094 /* buffered READ or buffered write_begin() lookup */
4095 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4097 * We have blocks reserved already. We
4098 * return allocated blocks so that delalloc
4099 * won't do block reservation for us. But
4100 * the buffer head will be unmapped so that
4101 * a read from the block returns 0s.
4103 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4107 /* buffered write, writepage time, convert*/
4108 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, ppath
, flags
);
4110 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4117 map
->m_flags
|= EXT4_MAP_NEW
;
4119 * if we allocated more blocks than requested
4120 * we need to make sure we unmap the extra block
4121 * allocated. The actual needed block will get
4122 * unmapped later when we find the buffer_head marked
4125 if (allocated
> map
->m_len
) {
4126 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
4127 newblock
+ map
->m_len
,
4128 allocated
- map
->m_len
);
4129 allocated
= map
->m_len
;
4131 map
->m_len
= allocated
;
4134 * If we have done fallocate with the offset that is already
4135 * delayed allocated, we would have block reservation
4136 * and quota reservation done in the delayed write path.
4137 * But fallocate would have already updated quota and block
4138 * count for this offset. So cancel these reservation
4140 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4141 unsigned int reserved_clusters
;
4142 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4143 map
->m_lblk
, map
->m_len
);
4144 if (reserved_clusters
)
4145 ext4_da_update_reserve_space(inode
,
4151 map
->m_flags
|= EXT4_MAP_MAPPED
;
4152 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
4153 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
4159 if (allocated
> map
->m_len
)
4160 allocated
= map
->m_len
;
4161 ext4_ext_show_leaf(inode
, path
);
4162 map
->m_pblk
= newblock
;
4163 map
->m_len
= allocated
;
4165 return err
? err
: allocated
;
4169 * get_implied_cluster_alloc - check to see if the requested
4170 * allocation (in the map structure) overlaps with a cluster already
4171 * allocated in an extent.
4172 * @sb The filesystem superblock structure
4173 * @map The requested lblk->pblk mapping
4174 * @ex The extent structure which might contain an implied
4175 * cluster allocation
4177 * This function is called by ext4_ext_map_blocks() after we failed to
4178 * find blocks that were already in the inode's extent tree. Hence,
4179 * we know that the beginning of the requested region cannot overlap
4180 * the extent from the inode's extent tree. There are three cases we
4181 * want to catch. The first is this case:
4183 * |--- cluster # N--|
4184 * |--- extent ---| |---- requested region ---|
4187 * The second case that we need to test for is this one:
4189 * |--------- cluster # N ----------------|
4190 * |--- requested region --| |------- extent ----|
4191 * |=======================|
4193 * The third case is when the requested region lies between two extents
4194 * within the same cluster:
4195 * |------------- cluster # N-------------|
4196 * |----- ex -----| |---- ex_right ----|
4197 * |------ requested region ------|
4198 * |================|
4200 * In each of the above cases, we need to set the map->m_pblk and
4201 * map->m_len so it corresponds to the return the extent labelled as
4202 * "|====|" from cluster #N, since it is already in use for data in
4203 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4204 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4205 * as a new "allocated" block region. Otherwise, we will return 0 and
4206 * ext4_ext_map_blocks() will then allocate one or more new clusters
4207 * by calling ext4_mb_new_blocks().
4209 static int get_implied_cluster_alloc(struct super_block
*sb
,
4210 struct ext4_map_blocks
*map
,
4211 struct ext4_extent
*ex
,
4212 struct ext4_ext_path
*path
)
4214 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
4215 ext4_lblk_t c_offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4216 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
4217 ext4_lblk_t rr_cluster_start
;
4218 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4219 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4220 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
4222 /* The extent passed in that we are trying to match */
4223 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
4224 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
4226 /* The requested region passed into ext4_map_blocks() */
4227 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
4229 if ((rr_cluster_start
== ex_cluster_end
) ||
4230 (rr_cluster_start
== ex_cluster_start
)) {
4231 if (rr_cluster_start
== ex_cluster_end
)
4232 ee_start
+= ee_len
- 1;
4233 map
->m_pblk
= EXT4_PBLK_CMASK(sbi
, ee_start
) + c_offset
;
4234 map
->m_len
= min(map
->m_len
,
4235 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
4237 * Check for and handle this case:
4239 * |--------- cluster # N-------------|
4240 * |------- extent ----|
4241 * |--- requested region ---|
4245 if (map
->m_lblk
< ee_block
)
4246 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
4249 * Check for the case where there is already another allocated
4250 * block to the right of 'ex' but before the end of the cluster.
4252 * |------------- cluster # N-------------|
4253 * |----- ex -----| |---- ex_right ----|
4254 * |------ requested region ------|
4255 * |================|
4257 if (map
->m_lblk
> ee_block
) {
4258 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
4259 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
4262 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
4266 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
4272 * Block allocation/map/preallocation routine for extents based files
4275 * Need to be called with
4276 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4277 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4279 * return > 0, number of of blocks already mapped/allocated
4280 * if create == 0 and these are pre-allocated blocks
4281 * buffer head is unmapped
4282 * otherwise blocks are mapped
4284 * return = 0, if plain look up failed (blocks have not been allocated)
4285 * buffer head is unmapped
4287 * return < 0, error case.
4289 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
4290 struct ext4_map_blocks
*map
, int flags
)
4292 struct ext4_ext_path
*path
= NULL
;
4293 struct ext4_extent newex
, *ex
, *ex2
;
4294 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4295 ext4_fsblk_t newblock
= 0;
4296 int free_on_err
= 0, err
= 0, depth
, ret
;
4297 unsigned int allocated
= 0, offset
= 0;
4298 unsigned int allocated_clusters
= 0;
4299 struct ext4_allocation_request ar
;
4300 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
4301 ext4_lblk_t cluster_offset
;
4302 int set_unwritten
= 0;
4303 bool map_from_cluster
= false;
4305 ext_debug("blocks %u/%u requested for inode %lu\n",
4306 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
4307 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
4309 /* find extent for this block */
4310 path
= ext4_find_extent(inode
, map
->m_lblk
, NULL
, 0);
4312 err
= PTR_ERR(path
);
4317 depth
= ext_depth(inode
);
4320 * consistent leaf must not be empty;
4321 * this situation is possible, though, _during_ tree modification;
4322 * this is why assert can't be put in ext4_find_extent()
4324 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
4325 EXT4_ERROR_INODE(inode
, "bad extent address "
4326 "lblock: %lu, depth: %d pblock %lld",
4327 (unsigned long) map
->m_lblk
, depth
,
4328 path
[depth
].p_block
);
4329 err
= -EFSCORRUPTED
;
4333 ex
= path
[depth
].p_ext
;
4335 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4336 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4337 unsigned short ee_len
;
4341 * unwritten extents are treated as holes, except that
4342 * we split out initialized portions during a write.
4344 ee_len
= ext4_ext_get_actual_len(ex
);
4346 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
4348 /* if found extent covers block, simply return it */
4349 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
4350 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
4351 /* number of remaining blocks in the extent */
4352 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
4353 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
4354 ee_block
, ee_len
, newblock
);
4357 * If the extent is initialized check whether the
4358 * caller wants to convert it to unwritten.
4360 if ((!ext4_ext_is_unwritten(ex
)) &&
4361 (flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
)) {
4362 allocated
= convert_initialized_extent(
4363 handle
, inode
, map
, &path
,
4364 flags
, allocated
, newblock
);
4366 } else if (!ext4_ext_is_unwritten(ex
))
4369 ret
= ext4_ext_handle_unwritten_extents(
4370 handle
, inode
, map
, &path
, flags
,
4371 allocated
, newblock
);
4381 * requested block isn't allocated yet;
4382 * we couldn't try to create block if create flag is zero
4384 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4386 * put just found gap into cache to speed up
4387 * subsequent requests
4389 ext4_ext_put_gap_in_cache(inode
, path
, map
->m_lblk
);
4394 * Okay, we need to do block allocation.
4396 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
4397 cluster_offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4400 * If we are doing bigalloc, check to see if the extent returned
4401 * by ext4_find_extent() implies a cluster we can use.
4403 if (cluster_offset
&& ex
&&
4404 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
4405 ar
.len
= allocated
= map
->m_len
;
4406 newblock
= map
->m_pblk
;
4407 map_from_cluster
= true;
4408 goto got_allocated_blocks
;
4411 /* find neighbour allocated blocks */
4412 ar
.lleft
= map
->m_lblk
;
4413 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
4416 ar
.lright
= map
->m_lblk
;
4418 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
4422 /* Check if the extent after searching to the right implies a
4423 * cluster we can use. */
4424 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
4425 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
4426 ar
.len
= allocated
= map
->m_len
;
4427 newblock
= map
->m_pblk
;
4428 map_from_cluster
= true;
4429 goto got_allocated_blocks
;
4433 * See if request is beyond maximum number of blocks we can have in
4434 * a single extent. For an initialized extent this limit is
4435 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4436 * EXT_UNWRITTEN_MAX_LEN.
4438 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
4439 !(flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
))
4440 map
->m_len
= EXT_INIT_MAX_LEN
;
4441 else if (map
->m_len
> EXT_UNWRITTEN_MAX_LEN
&&
4442 (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
))
4443 map
->m_len
= EXT_UNWRITTEN_MAX_LEN
;
4445 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4446 newex
.ee_len
= cpu_to_le16(map
->m_len
);
4447 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
4449 allocated
= ext4_ext_get_actual_len(&newex
);
4451 allocated
= map
->m_len
;
4453 /* allocate new block */
4455 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
4456 ar
.logical
= map
->m_lblk
;
4458 * We calculate the offset from the beginning of the cluster
4459 * for the logical block number, since when we allocate a
4460 * physical cluster, the physical block should start at the
4461 * same offset from the beginning of the cluster. This is
4462 * needed so that future calls to get_implied_cluster_alloc()
4465 offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4466 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4468 ar
.logical
-= offset
;
4469 if (S_ISREG(inode
->i_mode
))
4470 ar
.flags
= EXT4_MB_HINT_DATA
;
4472 /* disable in-core preallocation for non-regular files */
4474 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4475 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4476 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
4477 ar
.flags
|= EXT4_MB_DELALLOC_RESERVED
;
4478 if (flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
4479 ar
.flags
|= EXT4_MB_USE_RESERVED
;
4480 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4483 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4484 ar
.goal
, newblock
, allocated
);
4486 allocated_clusters
= ar
.len
;
4487 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4488 if (ar
.len
> allocated
)
4491 got_allocated_blocks
:
4492 /* try to insert new extent into found leaf and return */
4493 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4494 newex
.ee_len
= cpu_to_le16(ar
.len
);
4495 /* Mark unwritten */
4496 if (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
){
4497 ext4_ext_mark_unwritten(&newex
);
4498 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4500 * io_end structure was created for every IO write to an
4501 * unwritten extent. To avoid unnecessary conversion,
4502 * here we flag the IO that really needs the conversion.
4503 * For non asycn direct IO case, flag the inode state
4504 * that we need to perform conversion when IO is done.
4506 if (flags
& EXT4_GET_BLOCKS_PRE_IO
)
4511 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4512 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4515 err
= ext4_ext_insert_extent(handle
, inode
, &path
,
4518 if (!err
&& set_unwritten
) {
4520 ext4_set_io_unwritten_flag(inode
, io
);
4522 ext4_set_inode_state(inode
,
4523 EXT4_STATE_DIO_UNWRITTEN
);
4526 if (err
&& free_on_err
) {
4527 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4528 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4529 /* free data blocks we just allocated */
4530 /* not a good idea to call discard here directly,
4531 * but otherwise we'd need to call it every free() */
4532 ext4_discard_preallocations(inode
);
4533 ext4_free_blocks(handle
, inode
, NULL
, newblock
,
4534 EXT4_C2B(sbi
, allocated_clusters
), fb_flags
);
4538 /* previous routine could use block we allocated */
4539 newblock
= ext4_ext_pblock(&newex
);
4540 allocated
= ext4_ext_get_actual_len(&newex
);
4541 if (allocated
> map
->m_len
)
4542 allocated
= map
->m_len
;
4543 map
->m_flags
|= EXT4_MAP_NEW
;
4546 * Update reserved blocks/metadata blocks after successful
4547 * block allocation which had been deferred till now.
4549 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4550 unsigned int reserved_clusters
;
4552 * Check how many clusters we had reserved this allocated range
4554 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4555 map
->m_lblk
, allocated
);
4556 if (!map_from_cluster
) {
4557 BUG_ON(allocated_clusters
< reserved_clusters
);
4558 if (reserved_clusters
< allocated_clusters
) {
4559 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4560 int reservation
= allocated_clusters
-
4563 * It seems we claimed few clusters outside of
4564 * the range of this allocation. We should give
4565 * it back to the reservation pool. This can
4566 * happen in the following case:
4568 * * Suppose s_cluster_ratio is 4 (i.e., each
4569 * cluster has 4 blocks. Thus, the clusters
4570 * are [0-3],[4-7],[8-11]...
4571 * * First comes delayed allocation write for
4572 * logical blocks 10 & 11. Since there were no
4573 * previous delayed allocated blocks in the
4574 * range [8-11], we would reserve 1 cluster
4576 * * Next comes write for logical blocks 3 to 8.
4577 * In this case, we will reserve 2 clusters
4578 * (for [0-3] and [4-7]; and not for [8-11] as
4579 * that range has a delayed allocated blocks.
4580 * Thus total reserved clusters now becomes 3.
4581 * * Now, during the delayed allocation writeout
4582 * time, we will first write blocks [3-8] and
4583 * allocate 3 clusters for writing these
4584 * blocks. Also, we would claim all these
4585 * three clusters above.
4586 * * Now when we come here to writeout the
4587 * blocks [10-11], we would expect to claim
4588 * the reservation of 1 cluster we had made
4589 * (and we would claim it since there are no
4590 * more delayed allocated blocks in the range
4591 * [8-11]. But our reserved cluster count had
4592 * already gone to 0.
4594 * Thus, at the step 4 above when we determine
4595 * that there are still some unwritten delayed
4596 * allocated blocks outside of our current
4597 * block range, we should increment the
4598 * reserved clusters count so that when the
4599 * remaining blocks finally gets written, we
4602 dquot_reserve_block(inode
,
4603 EXT4_C2B(sbi
, reservation
));
4604 spin_lock(&ei
->i_block_reservation_lock
);
4605 ei
->i_reserved_data_blocks
+= reservation
;
4606 spin_unlock(&ei
->i_block_reservation_lock
);
4609 * We will claim quota for all newly allocated blocks.
4610 * We're updating the reserved space *after* the
4611 * correction above so we do not accidentally free
4612 * all the metadata reservation because we might
4613 * actually need it later on.
4615 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4621 * Cache the extent and update transaction to commit on fdatasync only
4622 * when it is _not_ an unwritten extent.
4624 if ((flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
) == 0)
4625 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4627 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4629 if (allocated
> map
->m_len
)
4630 allocated
= map
->m_len
;
4631 ext4_ext_show_leaf(inode
, path
);
4632 map
->m_flags
|= EXT4_MAP_MAPPED
;
4633 map
->m_pblk
= newblock
;
4634 map
->m_len
= allocated
;
4636 ext4_ext_drop_refs(path
);
4639 trace_ext4_ext_map_blocks_exit(inode
, flags
, map
,
4640 err
? err
: allocated
);
4641 return err
? err
: allocated
;
4644 void ext4_ext_truncate(handle_t
*handle
, struct inode
*inode
)
4646 struct super_block
*sb
= inode
->i_sb
;
4647 ext4_lblk_t last_block
;
4651 * TODO: optimization is possible here.
4652 * Probably we need not scan at all,
4653 * because page truncation is enough.
4656 /* we have to know where to truncate from in crash case */
4657 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4658 ext4_mark_inode_dirty(handle
, inode
);
4660 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4661 >> EXT4_BLOCK_SIZE_BITS(sb
);
4663 err
= ext4_es_remove_extent(inode
, last_block
,
4664 EXT_MAX_BLOCKS
- last_block
);
4665 if (err
== -ENOMEM
) {
4667 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
4671 ext4_std_error(inode
->i_sb
, err
);
4674 err
= ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4675 ext4_std_error(inode
->i_sb
, err
);
4678 static int ext4_alloc_file_blocks(struct file
*file
, ext4_lblk_t offset
,
4679 ext4_lblk_t len
, loff_t new_size
,
4680 int flags
, int mode
)
4682 struct inode
*inode
= file_inode(file
);
4688 struct ext4_map_blocks map
;
4689 unsigned int credits
;
4692 map
.m_lblk
= offset
;
4695 * Don't normalize the request if it can fit in one extent so
4696 * that it doesn't get unnecessarily split into multiple
4699 if (len
<= EXT_UNWRITTEN_MAX_LEN
)
4700 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4703 * credits to insert 1 extent into extent tree
4705 credits
= ext4_chunk_trans_blocks(inode
, len
);
4707 * We can only call ext_depth() on extent based inodes
4709 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4710 depth
= ext_depth(inode
);
4715 while (ret
>= 0 && len
) {
4717 * Recalculate credits when extent tree depth changes.
4719 if (depth
>= 0 && depth
!= ext_depth(inode
)) {
4720 credits
= ext4_chunk_trans_blocks(inode
, len
);
4721 depth
= ext_depth(inode
);
4724 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
4726 if (IS_ERR(handle
)) {
4727 ret
= PTR_ERR(handle
);
4730 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4732 ext4_debug("inode #%lu: block %u: len %u: "
4733 "ext4_ext_map_blocks returned %d",
4734 inode
->i_ino
, map
.m_lblk
,
4736 ext4_mark_inode_dirty(handle
, inode
);
4737 ret2
= ext4_journal_stop(handle
);
4741 map
.m_len
= len
= len
- ret
;
4742 epos
= (loff_t
)map
.m_lblk
<< inode
->i_blkbits
;
4743 inode
->i_ctime
= ext4_current_time(inode
);
4745 if (epos
> new_size
)
4747 if (ext4_update_inode_size(inode
, epos
) & 0x1)
4748 inode
->i_mtime
= inode
->i_ctime
;
4750 if (epos
> inode
->i_size
)
4751 ext4_set_inode_flag(inode
,
4752 EXT4_INODE_EOFBLOCKS
);
4754 ext4_mark_inode_dirty(handle
, inode
);
4755 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4756 ret2
= ext4_journal_stop(handle
);
4760 if (ret
== -ENOSPC
&&
4761 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4766 return ret
> 0 ? ret2
: ret
;
4769 static long ext4_zero_range(struct file
*file
, loff_t offset
,
4770 loff_t len
, int mode
)
4772 struct inode
*inode
= file_inode(file
);
4773 handle_t
*handle
= NULL
;
4774 unsigned int max_blocks
;
4775 loff_t new_size
= 0;
4779 int partial_begin
, partial_end
;
4782 unsigned int blkbits
= inode
->i_blkbits
;
4784 trace_ext4_zero_range(inode
, offset
, len
, mode
);
4786 if (!S_ISREG(inode
->i_mode
))
4789 /* Call ext4_force_commit to flush all data in case of data=journal. */
4790 if (ext4_should_journal_data(inode
)) {
4791 ret
= ext4_force_commit(inode
->i_sb
);
4797 * Round up offset. This is not fallocate, we neet to zero out
4798 * blocks, so convert interior block aligned part of the range to
4799 * unwritten and possibly manually zero out unaligned parts of the
4802 start
= round_up(offset
, 1 << blkbits
);
4803 end
= round_down((offset
+ len
), 1 << blkbits
);
4805 if (start
< offset
|| end
> offset
+ len
)
4807 partial_begin
= offset
& ((1 << blkbits
) - 1);
4808 partial_end
= (offset
+ len
) & ((1 << blkbits
) - 1);
4810 lblk
= start
>> blkbits
;
4811 max_blocks
= (end
>> blkbits
);
4812 if (max_blocks
< lblk
)
4817 mutex_lock(&inode
->i_mutex
);
4820 * Indirect files do not support unwritten extnets
4822 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4827 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4828 (offset
+ len
> i_size_read(inode
) ||
4829 offset
+ len
> EXT4_I(inode
)->i_disksize
)) {
4830 new_size
= offset
+ len
;
4831 ret
= inode_newsize_ok(inode
, new_size
);
4836 flags
= EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT
;
4837 if (mode
& FALLOC_FL_KEEP_SIZE
)
4838 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4840 /* Wait all existing dio workers, newcomers will block on i_mutex */
4841 ext4_inode_block_unlocked_dio(inode
);
4842 inode_dio_wait(inode
);
4844 /* Preallocate the range including the unaligned edges */
4845 if (partial_begin
|| partial_end
) {
4846 ret
= ext4_alloc_file_blocks(file
,
4847 round_down(offset
, 1 << blkbits
) >> blkbits
,
4848 (round_up((offset
+ len
), 1 << blkbits
) -
4849 round_down(offset
, 1 << blkbits
)) >> blkbits
,
4850 new_size
, flags
, mode
);
4856 /* Zero range excluding the unaligned edges */
4857 if (max_blocks
> 0) {
4858 flags
|= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
|
4862 * Prevent page faults from reinstantiating pages we have
4863 * released from page cache.
4865 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4866 ret
= ext4_update_disksize_before_punch(inode
, offset
, len
);
4868 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4871 /* Now release the pages and zero block aligned part of pages */
4872 truncate_pagecache_range(inode
, start
, end
- 1);
4873 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4875 ret
= ext4_alloc_file_blocks(file
, lblk
, max_blocks
, new_size
,
4877 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4881 if (!partial_begin
&& !partial_end
)
4885 * In worst case we have to writeout two nonadjacent unwritten
4886 * blocks and update the inode
4888 credits
= (2 * ext4_ext_index_trans_blocks(inode
, 2)) + 1;
4889 if (ext4_should_journal_data(inode
))
4891 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, credits
);
4892 if (IS_ERR(handle
)) {
4893 ret
= PTR_ERR(handle
);
4894 ext4_std_error(inode
->i_sb
, ret
);
4898 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4900 ext4_update_inode_size(inode
, new_size
);
4903 * Mark that we allocate beyond EOF so the subsequent truncate
4904 * can proceed even if the new size is the same as i_size.
4906 if ((offset
+ len
) > i_size_read(inode
))
4907 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4909 ext4_mark_inode_dirty(handle
, inode
);
4911 /* Zero out partial block at the edges of the range */
4912 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
, len
);
4914 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4916 if (file
->f_flags
& O_SYNC
)
4917 ext4_handle_sync(handle
);
4919 ext4_journal_stop(handle
);
4921 ext4_inode_resume_unlocked_dio(inode
);
4923 mutex_unlock(&inode
->i_mutex
);
4928 * preallocate space for a file. This implements ext4's fallocate file
4929 * operation, which gets called from sys_fallocate system call.
4930 * For block-mapped files, posix_fallocate should fall back to the method
4931 * of writing zeroes to the required new blocks (the same behavior which is
4932 * expected for file systems which do not support fallocate() system call).
4934 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4936 struct inode
*inode
= file_inode(file
);
4937 loff_t new_size
= 0;
4938 unsigned int max_blocks
;
4942 unsigned int blkbits
= inode
->i_blkbits
;
4945 * Encrypted inodes can't handle collapse range or insert
4946 * range since we would need to re-encrypt blocks with a
4947 * different IV or XTS tweak (which are based on the logical
4950 * XXX It's not clear why zero range isn't working, but we'll
4951 * leave it disabled for encrypted inodes for now. This is a
4952 * bug we should fix....
4954 if (ext4_encrypted_inode(inode
) &&
4955 (mode
& (FALLOC_FL_COLLAPSE_RANGE
| FALLOC_FL_INSERT_RANGE
|
4956 FALLOC_FL_ZERO_RANGE
)))
4959 /* Return error if mode is not supported */
4960 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
|
4961 FALLOC_FL_COLLAPSE_RANGE
| FALLOC_FL_ZERO_RANGE
|
4962 FALLOC_FL_INSERT_RANGE
))
4965 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4966 return ext4_punch_hole(inode
, offset
, len
);
4968 ret
= ext4_convert_inline_data(inode
);
4972 if (mode
& FALLOC_FL_COLLAPSE_RANGE
)
4973 return ext4_collapse_range(inode
, offset
, len
);
4975 if (mode
& FALLOC_FL_INSERT_RANGE
)
4976 return ext4_insert_range(inode
, offset
, len
);
4978 if (mode
& FALLOC_FL_ZERO_RANGE
)
4979 return ext4_zero_range(file
, offset
, len
, mode
);
4981 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4982 lblk
= offset
>> blkbits
;
4984 * We can't just convert len to max_blocks because
4985 * If blocksize = 4096 offset = 3072 and len = 2048
4987 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
4990 flags
= EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT
;
4991 if (mode
& FALLOC_FL_KEEP_SIZE
)
4992 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4994 mutex_lock(&inode
->i_mutex
);
4997 * We only support preallocation for extent-based files only
4999 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5004 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
5005 (offset
+ len
> i_size_read(inode
) ||
5006 offset
+ len
> EXT4_I(inode
)->i_disksize
)) {
5007 new_size
= offset
+ len
;
5008 ret
= inode_newsize_ok(inode
, new_size
);
5013 /* Wait all existing dio workers, newcomers will block on i_mutex */
5014 ext4_inode_block_unlocked_dio(inode
);
5015 inode_dio_wait(inode
);
5017 ret
= ext4_alloc_file_blocks(file
, lblk
, max_blocks
, new_size
,
5019 ext4_inode_resume_unlocked_dio(inode
);
5023 if (file
->f_flags
& O_SYNC
&& EXT4_SB(inode
->i_sb
)->s_journal
) {
5024 ret
= jbd2_complete_transaction(EXT4_SB(inode
->i_sb
)->s_journal
,
5025 EXT4_I(inode
)->i_sync_tid
);
5028 mutex_unlock(&inode
->i_mutex
);
5029 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
5034 * This function convert a range of blocks to written extents
5035 * The caller of this function will pass the start offset and the size.
5036 * all unwritten extents within this range will be converted to
5039 * This function is called from the direct IO end io call back
5040 * function, to convert the fallocated extents after IO is completed.
5041 * Returns 0 on success.
5043 int ext4_convert_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
5044 loff_t offset
, ssize_t len
)
5046 unsigned int max_blocks
;
5049 struct ext4_map_blocks map
;
5050 unsigned int credits
, blkbits
= inode
->i_blkbits
;
5052 map
.m_lblk
= offset
>> blkbits
;
5054 * We can't just convert len to max_blocks because
5055 * If blocksize = 4096 offset = 3072 and len = 2048
5057 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
5060 * This is somewhat ugly but the idea is clear: When transaction is
5061 * reserved, everything goes into it. Otherwise we rather start several
5062 * smaller transactions for conversion of each extent separately.
5065 handle
= ext4_journal_start_reserved(handle
,
5066 EXT4_HT_EXT_CONVERT
);
5068 return PTR_ERR(handle
);
5072 * credits to insert 1 extent into extent tree
5074 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
5076 while (ret
>= 0 && ret
< max_blocks
) {
5078 map
.m_len
= (max_blocks
-= ret
);
5080 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
5082 if (IS_ERR(handle
)) {
5083 ret
= PTR_ERR(handle
);
5087 ret
= ext4_map_blocks(handle
, inode
, &map
,
5088 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
5090 ext4_warning(inode
->i_sb
,
5091 "inode #%lu: block %u: len %u: "
5092 "ext4_ext_map_blocks returned %d",
5093 inode
->i_ino
, map
.m_lblk
,
5095 ext4_mark_inode_dirty(handle
, inode
);
5097 ret2
= ext4_journal_stop(handle
);
5098 if (ret
<= 0 || ret2
)
5102 ret2
= ext4_journal_stop(handle
);
5103 return ret
> 0 ? ret2
: ret
;
5107 * If newes is not existing extent (newes->ec_pblk equals zero) find
5108 * delayed extent at start of newes and update newes accordingly and
5109 * return start of the next delayed extent.
5111 * If newes is existing extent (newes->ec_pblk is not equal zero)
5112 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5113 * extent found. Leave newes unmodified.
5115 static int ext4_find_delayed_extent(struct inode
*inode
,
5116 struct extent_status
*newes
)
5118 struct extent_status es
;
5119 ext4_lblk_t block
, next_del
;
5121 if (newes
->es_pblk
== 0) {
5122 ext4_es_find_delayed_extent_range(inode
, newes
->es_lblk
,
5123 newes
->es_lblk
+ newes
->es_len
- 1, &es
);
5126 * No extent in extent-tree contains block @newes->es_pblk,
5127 * then the block may stay in 1)a hole or 2)delayed-extent.
5133 if (es
.es_lblk
> newes
->es_lblk
) {
5135 newes
->es_len
= min(es
.es_lblk
- newes
->es_lblk
,
5140 newes
->es_len
= es
.es_lblk
+ es
.es_len
- newes
->es_lblk
;
5143 block
= newes
->es_lblk
+ newes
->es_len
;
5144 ext4_es_find_delayed_extent_range(inode
, block
, EXT_MAX_BLOCKS
, &es
);
5146 next_del
= EXT_MAX_BLOCKS
;
5148 next_del
= es
.es_lblk
;
5152 /* fiemap flags we can handle specified here */
5153 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5155 static int ext4_xattr_fiemap(struct inode
*inode
,
5156 struct fiemap_extent_info
*fieinfo
)
5160 __u32 flags
= FIEMAP_EXTENT_LAST
;
5161 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
5165 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
5166 struct ext4_iloc iloc
;
5167 int offset
; /* offset of xattr in inode */
5169 error
= ext4_get_inode_loc(inode
, &iloc
);
5172 physical
= (__u64
)iloc
.bh
->b_blocknr
<< blockbits
;
5173 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
5174 EXT4_I(inode
)->i_extra_isize
;
5176 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
5177 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
5179 } else { /* external block */
5180 physical
= (__u64
)EXT4_I(inode
)->i_file_acl
<< blockbits
;
5181 length
= inode
->i_sb
->s_blocksize
;
5185 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
5187 return (error
< 0 ? error
: 0);
5190 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
5191 __u64 start
, __u64 len
)
5193 ext4_lblk_t start_blk
;
5196 if (ext4_has_inline_data(inode
)) {
5199 error
= ext4_inline_data_fiemap(inode
, fieinfo
, &has_inline
,
5206 if (fieinfo
->fi_flags
& FIEMAP_FLAG_CACHE
) {
5207 error
= ext4_ext_precache(inode
);
5212 /* fallback to generic here if not in extents fmt */
5213 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5214 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
5217 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
5220 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
5221 error
= ext4_xattr_fiemap(inode
, fieinfo
);
5223 ext4_lblk_t len_blks
;
5226 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
5227 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
5228 if (last_blk
>= EXT_MAX_BLOCKS
)
5229 last_blk
= EXT_MAX_BLOCKS
-1;
5230 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
5233 * Walk the extent tree gathering extent information
5234 * and pushing extents back to the user.
5236 error
= ext4_fill_fiemap_extents(inode
, start_blk
,
5244 * Function to access the path buffer for marking it dirty.
5245 * It also checks if there are sufficient credits left in the journal handle
5249 ext4_access_path(handle_t
*handle
, struct inode
*inode
,
5250 struct ext4_ext_path
*path
)
5254 if (!ext4_handle_valid(handle
))
5258 * Check if need to extend journal credits
5259 * 3 for leaf, sb, and inode plus 2 (bmap and group
5260 * descriptor) for each block group; assume two block
5263 if (handle
->h_buffer_credits
< 7) {
5264 credits
= ext4_writepage_trans_blocks(inode
);
5265 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
5266 /* EAGAIN is success */
5267 if (err
&& err
!= -EAGAIN
)
5271 err
= ext4_ext_get_access(handle
, inode
, path
);
5276 * ext4_ext_shift_path_extents:
5277 * Shift the extents of a path structure lying between path[depth].p_ext
5278 * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5279 * if it is right shift or left shift operation.
5282 ext4_ext_shift_path_extents(struct ext4_ext_path
*path
, ext4_lblk_t shift
,
5283 struct inode
*inode
, handle_t
*handle
,
5284 enum SHIFT_DIRECTION SHIFT
)
5287 struct ext4_extent
*ex_start
, *ex_last
;
5289 depth
= path
->p_depth
;
5291 while (depth
>= 0) {
5292 if (depth
== path
->p_depth
) {
5293 ex_start
= path
[depth
].p_ext
;
5295 return -EFSCORRUPTED
;
5297 ex_last
= EXT_LAST_EXTENT(path
[depth
].p_hdr
);
5299 err
= ext4_access_path(handle
, inode
, path
+ depth
);
5303 if (ex_start
== EXT_FIRST_EXTENT(path
[depth
].p_hdr
))
5306 while (ex_start
<= ex_last
) {
5307 if (SHIFT
== SHIFT_LEFT
) {
5308 le32_add_cpu(&ex_start
->ee_block
,
5310 /* Try to merge to the left. */
5312 EXT_FIRST_EXTENT(path
[depth
].p_hdr
))
5314 ext4_ext_try_to_merge_right(inode
,
5315 path
, ex_start
- 1))
5320 le32_add_cpu(&ex_last
->ee_block
, shift
);
5321 ext4_ext_try_to_merge_right(inode
, path
,
5326 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
5330 if (--depth
< 0 || !update
)
5334 /* Update index too */
5335 err
= ext4_access_path(handle
, inode
, path
+ depth
);
5339 if (SHIFT
== SHIFT_LEFT
)
5340 le32_add_cpu(&path
[depth
].p_idx
->ei_block
, -shift
);
5342 le32_add_cpu(&path
[depth
].p_idx
->ei_block
, shift
);
5343 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
5347 /* we are done if current index is not a starting index */
5348 if (path
[depth
].p_idx
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))
5359 * ext4_ext_shift_extents:
5360 * All the extents which lies in the range from @start to the last allocated
5361 * block for the @inode are shifted either towards left or right (depending
5362 * upon @SHIFT) by @shift blocks.
5363 * On success, 0 is returned, error otherwise.
5366 ext4_ext_shift_extents(struct inode
*inode
, handle_t
*handle
,
5367 ext4_lblk_t start
, ext4_lblk_t shift
,
5368 enum SHIFT_DIRECTION SHIFT
)
5370 struct ext4_ext_path
*path
;
5372 struct ext4_extent
*extent
;
5373 ext4_lblk_t stop
, *iterator
, ex_start
, ex_end
;
5375 /* Let path point to the last extent */
5376 path
= ext4_find_extent(inode
, EXT_MAX_BLOCKS
- 1, NULL
,
5379 return PTR_ERR(path
);
5381 depth
= path
->p_depth
;
5382 extent
= path
[depth
].p_ext
;
5386 stop
= le32_to_cpu(extent
->ee_block
);
5389 * For left shifts, make sure the hole on the left is big enough to
5390 * accommodate the shift. For right shifts, make sure the last extent
5391 * won't be shifted beyond EXT_MAX_BLOCKS.
5393 if (SHIFT
== SHIFT_LEFT
) {
5394 path
= ext4_find_extent(inode
, start
- 1, &path
,
5397 return PTR_ERR(path
);
5398 depth
= path
->p_depth
;
5399 extent
= path
[depth
].p_ext
;
5401 ex_start
= le32_to_cpu(extent
->ee_block
);
5402 ex_end
= le32_to_cpu(extent
->ee_block
) +
5403 ext4_ext_get_actual_len(extent
);
5409 if ((start
== ex_start
&& shift
> ex_start
) ||
5410 (shift
> start
- ex_end
)) {
5415 if (shift
> EXT_MAX_BLOCKS
-
5416 (stop
+ ext4_ext_get_actual_len(extent
))) {
5423 * In case of left shift, iterator points to start and it is increased
5424 * till we reach stop. In case of right shift, iterator points to stop
5425 * and it is decreased till we reach start.
5427 if (SHIFT
== SHIFT_LEFT
)
5433 * Its safe to start updating extents. Start and stop are unsigned, so
5434 * in case of right shift if extent with 0 block is reached, iterator
5435 * becomes NULL to indicate the end of the loop.
5437 while (iterator
&& start
<= stop
) {
5438 path
= ext4_find_extent(inode
, *iterator
, &path
,
5441 return PTR_ERR(path
);
5442 depth
= path
->p_depth
;
5443 extent
= path
[depth
].p_ext
;
5445 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
5446 (unsigned long) *iterator
);
5447 return -EFSCORRUPTED
;
5449 if (SHIFT
== SHIFT_LEFT
&& *iterator
>
5450 le32_to_cpu(extent
->ee_block
)) {
5451 /* Hole, move to the next extent */
5452 if (extent
< EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
5453 path
[depth
].p_ext
++;
5455 *iterator
= ext4_ext_next_allocated_block(path
);
5460 if (SHIFT
== SHIFT_LEFT
) {
5461 extent
= EXT_LAST_EXTENT(path
[depth
].p_hdr
);
5462 *iterator
= le32_to_cpu(extent
->ee_block
) +
5463 ext4_ext_get_actual_len(extent
);
5465 extent
= EXT_FIRST_EXTENT(path
[depth
].p_hdr
);
5466 if (le32_to_cpu(extent
->ee_block
) > 0)
5467 *iterator
= le32_to_cpu(extent
->ee_block
) - 1;
5469 /* Beginning is reached, end of the loop */
5471 /* Update path extent in case we need to stop */
5472 while (le32_to_cpu(extent
->ee_block
) < start
)
5474 path
[depth
].p_ext
= extent
;
5476 ret
= ext4_ext_shift_path_extents(path
, shift
, inode
,
5482 ext4_ext_drop_refs(path
);
5488 * ext4_collapse_range:
5489 * This implements the fallocate's collapse range functionality for ext4
5490 * Returns: 0 and non-zero on error.
5492 int ext4_collapse_range(struct inode
*inode
, loff_t offset
, loff_t len
)
5494 struct super_block
*sb
= inode
->i_sb
;
5495 ext4_lblk_t punch_start
, punch_stop
;
5497 unsigned int credits
;
5498 loff_t new_size
, ioffset
;
5502 * We need to test this early because xfstests assumes that a
5503 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5504 * system does not support collapse range.
5506 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
5509 /* Collapse range works only on fs block size aligned offsets. */
5510 if (offset
& (EXT4_CLUSTER_SIZE(sb
) - 1) ||
5511 len
& (EXT4_CLUSTER_SIZE(sb
) - 1))
5514 if (!S_ISREG(inode
->i_mode
))
5517 trace_ext4_collapse_range(inode
, offset
, len
);
5519 punch_start
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
5520 punch_stop
= (offset
+ len
) >> EXT4_BLOCK_SIZE_BITS(sb
);
5522 /* Call ext4_force_commit to flush all data in case of data=journal. */
5523 if (ext4_should_journal_data(inode
)) {
5524 ret
= ext4_force_commit(inode
->i_sb
);
5529 mutex_lock(&inode
->i_mutex
);
5531 * There is no need to overlap collapse range with EOF, in which case
5532 * it is effectively a truncate operation
5534 if (offset
+ len
>= i_size_read(inode
)) {
5539 /* Currently just for extent based files */
5540 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5545 /* Wait for existing dio to complete */
5546 ext4_inode_block_unlocked_dio(inode
);
5547 inode_dio_wait(inode
);
5550 * Prevent page faults from reinstantiating pages we have released from
5553 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5555 * Need to round down offset to be aligned with page size boundary
5556 * for page size > block size.
5558 ioffset
= round_down(offset
, PAGE_SIZE
);
5560 * Write tail of the last page before removed range since it will get
5561 * removed from the page cache below.
5563 ret
= filemap_write_and_wait_range(inode
->i_mapping
, ioffset
, offset
);
5567 * Write data that will be shifted to preserve them when discarding
5568 * page cache below. We are also protected from pages becoming dirty
5571 ret
= filemap_write_and_wait_range(inode
->i_mapping
, offset
+ len
,
5575 truncate_pagecache(inode
, ioffset
);
5577 credits
= ext4_writepage_trans_blocks(inode
);
5578 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
5579 if (IS_ERR(handle
)) {
5580 ret
= PTR_ERR(handle
);
5584 down_write(&EXT4_I(inode
)->i_data_sem
);
5585 ext4_discard_preallocations(inode
);
5587 ret
= ext4_es_remove_extent(inode
, punch_start
,
5588 EXT_MAX_BLOCKS
- punch_start
);
5590 up_write(&EXT4_I(inode
)->i_data_sem
);
5594 ret
= ext4_ext_remove_space(inode
, punch_start
, punch_stop
- 1);
5596 up_write(&EXT4_I(inode
)->i_data_sem
);
5599 ext4_discard_preallocations(inode
);
5601 ret
= ext4_ext_shift_extents(inode
, handle
, punch_stop
,
5602 punch_stop
- punch_start
, SHIFT_LEFT
);
5604 up_write(&EXT4_I(inode
)->i_data_sem
);
5608 new_size
= i_size_read(inode
) - len
;
5609 i_size_write(inode
, new_size
);
5610 EXT4_I(inode
)->i_disksize
= new_size
;
5612 up_write(&EXT4_I(inode
)->i_data_sem
);
5614 ext4_handle_sync(handle
);
5615 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
5616 ext4_mark_inode_dirty(handle
, inode
);
5617 ext4_update_inode_fsync_trans(handle
, inode
, 1);
5620 ext4_journal_stop(handle
);
5622 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5623 ext4_inode_resume_unlocked_dio(inode
);
5625 mutex_unlock(&inode
->i_mutex
);
5630 * ext4_insert_range:
5631 * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5632 * The data blocks starting from @offset to the EOF are shifted by @len
5633 * towards right to create a hole in the @inode. Inode size is increased
5635 * Returns 0 on success, error otherwise.
5637 int ext4_insert_range(struct inode
*inode
, loff_t offset
, loff_t len
)
5639 struct super_block
*sb
= inode
->i_sb
;
5641 struct ext4_ext_path
*path
;
5642 struct ext4_extent
*extent
;
5643 ext4_lblk_t offset_lblk
, len_lblk
, ee_start_lblk
= 0;
5644 unsigned int credits
, ee_len
;
5645 int ret
= 0, depth
, split_flag
= 0;
5649 * We need to test this early because xfstests assumes that an
5650 * insert range of (0, 1) will return EOPNOTSUPP if the file
5651 * system does not support insert range.
5653 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
5656 /* Insert range works only on fs block size aligned offsets. */
5657 if (offset
& (EXT4_CLUSTER_SIZE(sb
) - 1) ||
5658 len
& (EXT4_CLUSTER_SIZE(sb
) - 1))
5661 if (!S_ISREG(inode
->i_mode
))
5664 trace_ext4_insert_range(inode
, offset
, len
);
5666 offset_lblk
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
5667 len_lblk
= len
>> EXT4_BLOCK_SIZE_BITS(sb
);
5669 /* Call ext4_force_commit to flush all data in case of data=journal */
5670 if (ext4_should_journal_data(inode
)) {
5671 ret
= ext4_force_commit(inode
->i_sb
);
5676 mutex_lock(&inode
->i_mutex
);
5677 /* Currently just for extent based files */
5678 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5683 /* Check for wrap through zero */
5684 if (inode
->i_size
+ len
> inode
->i_sb
->s_maxbytes
) {
5689 /* Offset should be less than i_size */
5690 if (offset
>= i_size_read(inode
)) {
5695 /* Wait for existing dio to complete */
5696 ext4_inode_block_unlocked_dio(inode
);
5697 inode_dio_wait(inode
);
5700 * Prevent page faults from reinstantiating pages we have released from
5703 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5705 * Need to round down to align start offset to page size boundary
5706 * for page size > block size.
5708 ioffset
= round_down(offset
, PAGE_SIZE
);
5709 /* Write out all dirty pages */
5710 ret
= filemap_write_and_wait_range(inode
->i_mapping
, ioffset
,
5714 truncate_pagecache(inode
, ioffset
);
5716 credits
= ext4_writepage_trans_blocks(inode
);
5717 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
5718 if (IS_ERR(handle
)) {
5719 ret
= PTR_ERR(handle
);
5723 /* Expand file to avoid data loss if there is error while shifting */
5724 inode
->i_size
+= len
;
5725 EXT4_I(inode
)->i_disksize
+= len
;
5726 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
5727 ret
= ext4_mark_inode_dirty(handle
, inode
);
5731 down_write(&EXT4_I(inode
)->i_data_sem
);
5732 ext4_discard_preallocations(inode
);
5734 path
= ext4_find_extent(inode
, offset_lblk
, NULL
, 0);
5736 up_write(&EXT4_I(inode
)->i_data_sem
);
5740 depth
= ext_depth(inode
);
5741 extent
= path
[depth
].p_ext
;
5743 ee_start_lblk
= le32_to_cpu(extent
->ee_block
);
5744 ee_len
= ext4_ext_get_actual_len(extent
);
5747 * If offset_lblk is not the starting block of extent, split
5748 * the extent @offset_lblk
5750 if ((offset_lblk
> ee_start_lblk
) &&
5751 (offset_lblk
< (ee_start_lblk
+ ee_len
))) {
5752 if (ext4_ext_is_unwritten(extent
))
5753 split_flag
= EXT4_EXT_MARK_UNWRIT1
|
5754 EXT4_EXT_MARK_UNWRIT2
;
5755 ret
= ext4_split_extent_at(handle
, inode
, &path
,
5756 offset_lblk
, split_flag
,
5758 EXT4_GET_BLOCKS_PRE_IO
|
5759 EXT4_GET_BLOCKS_METADATA_NOFAIL
);
5762 ext4_ext_drop_refs(path
);
5765 up_write(&EXT4_I(inode
)->i_data_sem
);
5769 ext4_ext_drop_refs(path
);
5773 ret
= ext4_es_remove_extent(inode
, offset_lblk
,
5774 EXT_MAX_BLOCKS
- offset_lblk
);
5776 up_write(&EXT4_I(inode
)->i_data_sem
);
5781 * if offset_lblk lies in a hole which is at start of file, use
5782 * ee_start_lblk to shift extents
5784 ret
= ext4_ext_shift_extents(inode
, handle
,
5785 ee_start_lblk
> offset_lblk
? ee_start_lblk
: offset_lblk
,
5786 len_lblk
, SHIFT_RIGHT
);
5788 up_write(&EXT4_I(inode
)->i_data_sem
);
5790 ext4_handle_sync(handle
);
5792 ext4_update_inode_fsync_trans(handle
, inode
, 1);
5795 ext4_journal_stop(handle
);
5797 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5798 ext4_inode_resume_unlocked_dio(inode
);
5800 mutex_unlock(&inode
->i_mutex
);
5805 * ext4_swap_extents - Swap extents between two inodes
5807 * @inode1: First inode
5808 * @inode2: Second inode
5809 * @lblk1: Start block for first inode
5810 * @lblk2: Start block for second inode
5811 * @count: Number of blocks to swap
5812 * @mark_unwritten: Mark second inode's extents as unwritten after swap
5813 * @erp: Pointer to save error value
5815 * This helper routine does exactly what is promise "swap extents". All other
5816 * stuff such as page-cache locking consistency, bh mapping consistency or
5817 * extent's data copying must be performed by caller.
5819 * i_mutex is held for both inodes
5820 * i_data_sem is locked for write for both inodes
5822 * All pages from requested range are locked for both inodes
5825 ext4_swap_extents(handle_t
*handle
, struct inode
*inode1
,
5826 struct inode
*inode2
, ext4_lblk_t lblk1
, ext4_lblk_t lblk2
,
5827 ext4_lblk_t count
, int unwritten
, int *erp
)
5829 struct ext4_ext_path
*path1
= NULL
;
5830 struct ext4_ext_path
*path2
= NULL
;
5831 int replaced_count
= 0;
5833 BUG_ON(!rwsem_is_locked(&EXT4_I(inode1
)->i_data_sem
));
5834 BUG_ON(!rwsem_is_locked(&EXT4_I(inode2
)->i_data_sem
));
5835 BUG_ON(!mutex_is_locked(&inode1
->i_mutex
));
5836 BUG_ON(!mutex_is_locked(&inode2
->i_mutex
));
5838 *erp
= ext4_es_remove_extent(inode1
, lblk1
, count
);
5841 *erp
= ext4_es_remove_extent(inode2
, lblk2
, count
);
5846 struct ext4_extent
*ex1
, *ex2
, tmp_ex
;
5847 ext4_lblk_t e1_blk
, e2_blk
;
5848 int e1_len
, e2_len
, len
;
5851 path1
= ext4_find_extent(inode1
, lblk1
, NULL
, EXT4_EX_NOCACHE
);
5852 if (IS_ERR(path1
)) {
5853 *erp
= PTR_ERR(path1
);
5859 path2
= ext4_find_extent(inode2
, lblk2
, NULL
, EXT4_EX_NOCACHE
);
5860 if (IS_ERR(path2
)) {
5861 *erp
= PTR_ERR(path2
);
5865 ex1
= path1
[path1
->p_depth
].p_ext
;
5866 ex2
= path2
[path2
->p_depth
].p_ext
;
5867 /* Do we have somthing to swap ? */
5868 if (unlikely(!ex2
|| !ex1
))
5871 e1_blk
= le32_to_cpu(ex1
->ee_block
);
5872 e2_blk
= le32_to_cpu(ex2
->ee_block
);
5873 e1_len
= ext4_ext_get_actual_len(ex1
);
5874 e2_len
= ext4_ext_get_actual_len(ex2
);
5877 if (!in_range(lblk1
, e1_blk
, e1_len
) ||
5878 !in_range(lblk2
, e2_blk
, e2_len
)) {
5879 ext4_lblk_t next1
, next2
;
5881 /* if hole after extent, then go to next extent */
5882 next1
= ext4_ext_next_allocated_block(path1
);
5883 next2
= ext4_ext_next_allocated_block(path2
);
5884 /* If hole before extent, then shift to that extent */
5889 /* Do we have something to swap */
5890 if (next1
== EXT_MAX_BLOCKS
|| next2
== EXT_MAX_BLOCKS
)
5892 /* Move to the rightest boundary */
5893 len
= next1
- lblk1
;
5894 if (len
< next2
- lblk2
)
5895 len
= next2
- lblk2
;
5904 /* Prepare left boundary */
5905 if (e1_blk
< lblk1
) {
5907 *erp
= ext4_force_split_extent_at(handle
, inode1
,
5912 if (e2_blk
< lblk2
) {
5914 *erp
= ext4_force_split_extent_at(handle
, inode2
,
5919 /* ext4_split_extent_at() may result in leaf extent split,
5920 * path must to be revalidated. */
5924 /* Prepare right boundary */
5926 if (len
> e1_blk
+ e1_len
- lblk1
)
5927 len
= e1_blk
+ e1_len
- lblk1
;
5928 if (len
> e2_blk
+ e2_len
- lblk2
)
5929 len
= e2_blk
+ e2_len
- lblk2
;
5931 if (len
!= e1_len
) {
5933 *erp
= ext4_force_split_extent_at(handle
, inode1
,
5934 &path1
, lblk1
+ len
, 0);
5938 if (len
!= e2_len
) {
5940 *erp
= ext4_force_split_extent_at(handle
, inode2
,
5941 &path2
, lblk2
+ len
, 0);
5945 /* ext4_split_extent_at() may result in leaf extent split,
5946 * path must to be revalidated. */
5950 BUG_ON(e2_len
!= e1_len
);
5951 *erp
= ext4_ext_get_access(handle
, inode1
, path1
+ path1
->p_depth
);
5954 *erp
= ext4_ext_get_access(handle
, inode2
, path2
+ path2
->p_depth
);
5958 /* Both extents are fully inside boundaries. Swap it now */
5960 ext4_ext_store_pblock(ex1
, ext4_ext_pblock(ex2
));
5961 ext4_ext_store_pblock(ex2
, ext4_ext_pblock(&tmp_ex
));
5962 ex1
->ee_len
= cpu_to_le16(e2_len
);
5963 ex2
->ee_len
= cpu_to_le16(e1_len
);
5965 ext4_ext_mark_unwritten(ex2
);
5966 if (ext4_ext_is_unwritten(&tmp_ex
))
5967 ext4_ext_mark_unwritten(ex1
);
5969 ext4_ext_try_to_merge(handle
, inode2
, path2
, ex2
);
5970 ext4_ext_try_to_merge(handle
, inode1
, path1
, ex1
);
5971 *erp
= ext4_ext_dirty(handle
, inode2
, path2
+
5975 *erp
= ext4_ext_dirty(handle
, inode1
, path1
+
5978 * Looks scarry ah..? second inode already points to new blocks,
5979 * and it was successfully dirtied. But luckily error may happen
5980 * only due to journal error, so full transaction will be
5987 replaced_count
+= len
;
5991 ext4_ext_drop_refs(path1
);
5993 ext4_ext_drop_refs(path2
);
5995 path1
= path2
= NULL
;
5997 return replaced_count
;