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 <linux/falloc.h>
41 #include <asm/uaccess.h>
42 #include <linux/fiemap.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_UNINIT1 0x2 /* mark first half uninitialized */
55 #define EXT4_EXT_MARK_UNINIT2 0x4 /* mark second half uninitialized */
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_RO_COMPAT_FEATURE(inode
->i_sb
,
78 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
81 et
= find_ext4_extent_tail(eh
);
82 if (et
->et_checksum
!= ext4_extent_block_csum(inode
, eh
))
87 static void ext4_extent_block_csum_set(struct inode
*inode
,
88 struct ext4_extent_header
*eh
)
90 struct ext4_extent_tail
*et
;
92 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
93 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
96 et
= find_ext4_extent_tail(eh
);
97 et
->et_checksum
= ext4_extent_block_csum(inode
, eh
);
100 static int ext4_split_extent(handle_t
*handle
,
102 struct ext4_ext_path
*path
,
103 struct ext4_map_blocks
*map
,
107 static int ext4_split_extent_at(handle_t
*handle
,
109 struct ext4_ext_path
*path
,
114 static int ext4_find_delayed_extent(struct inode
*inode
,
115 struct extent_status
*newes
);
117 static int ext4_ext_truncate_extend_restart(handle_t
*handle
,
123 if (!ext4_handle_valid(handle
))
125 if (handle
->h_buffer_credits
> needed
)
127 err
= ext4_journal_extend(handle
, needed
);
130 err
= ext4_truncate_restart_trans(handle
, inode
, needed
);
142 static int ext4_ext_get_access(handle_t
*handle
, struct inode
*inode
,
143 struct ext4_ext_path
*path
)
146 /* path points to block */
147 return ext4_journal_get_write_access(handle
, path
->p_bh
);
149 /* path points to leaf/index in inode body */
150 /* we use in-core data, no need to protect them */
160 int __ext4_ext_dirty(const char *where
, unsigned int line
, handle_t
*handle
,
161 struct inode
*inode
, struct ext4_ext_path
*path
)
165 ext4_extent_block_csum_set(inode
, ext_block_hdr(path
->p_bh
));
166 /* path points to block */
167 err
= __ext4_handle_dirty_metadata(where
, line
, handle
,
170 /* path points to leaf/index in inode body */
171 err
= ext4_mark_inode_dirty(handle
, inode
);
176 static ext4_fsblk_t
ext4_ext_find_goal(struct inode
*inode
,
177 struct ext4_ext_path
*path
,
181 int depth
= path
->p_depth
;
182 struct ext4_extent
*ex
;
185 * Try to predict block placement assuming that we are
186 * filling in a file which will eventually be
187 * non-sparse --- i.e., in the case of libbfd writing
188 * an ELF object sections out-of-order but in a way
189 * the eventually results in a contiguous object or
190 * executable file, or some database extending a table
191 * space file. However, this is actually somewhat
192 * non-ideal if we are writing a sparse file such as
193 * qemu or KVM writing a raw image file that is going
194 * to stay fairly sparse, since it will end up
195 * fragmenting the file system's free space. Maybe we
196 * should have some hueristics or some way to allow
197 * userspace to pass a hint to file system,
198 * especially if the latter case turns out to be
201 ex
= path
[depth
].p_ext
;
203 ext4_fsblk_t ext_pblk
= ext4_ext_pblock(ex
);
204 ext4_lblk_t ext_block
= le32_to_cpu(ex
->ee_block
);
206 if (block
> ext_block
)
207 return ext_pblk
+ (block
- ext_block
);
209 return ext_pblk
- (ext_block
- block
);
212 /* it looks like index is empty;
213 * try to find starting block from index itself */
214 if (path
[depth
].p_bh
)
215 return path
[depth
].p_bh
->b_blocknr
;
218 /* OK. use inode's group */
219 return ext4_inode_to_goal_block(inode
);
223 * Allocation for a meta data block
226 ext4_ext_new_meta_block(handle_t
*handle
, struct inode
*inode
,
227 struct ext4_ext_path
*path
,
228 struct ext4_extent
*ex
, int *err
, unsigned int flags
)
230 ext4_fsblk_t goal
, newblock
;
232 goal
= ext4_ext_find_goal(inode
, path
, le32_to_cpu(ex
->ee_block
));
233 newblock
= ext4_new_meta_blocks(handle
, inode
, goal
, flags
,
238 static inline int ext4_ext_space_block(struct inode
*inode
, int check
)
242 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
243 / sizeof(struct ext4_extent
);
244 #ifdef AGGRESSIVE_TEST
245 if (!check
&& size
> 6)
251 static inline int ext4_ext_space_block_idx(struct inode
*inode
, int check
)
255 size
= (inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
256 / sizeof(struct ext4_extent_idx
);
257 #ifdef AGGRESSIVE_TEST
258 if (!check
&& size
> 5)
264 static inline int ext4_ext_space_root(struct inode
*inode
, int check
)
268 size
= sizeof(EXT4_I(inode
)->i_data
);
269 size
-= sizeof(struct ext4_extent_header
);
270 size
/= sizeof(struct ext4_extent
);
271 #ifdef AGGRESSIVE_TEST
272 if (!check
&& size
> 3)
278 static inline int ext4_ext_space_root_idx(struct inode
*inode
, int check
)
282 size
= sizeof(EXT4_I(inode
)->i_data
);
283 size
-= sizeof(struct ext4_extent_header
);
284 size
/= sizeof(struct ext4_extent_idx
);
285 #ifdef AGGRESSIVE_TEST
286 if (!check
&& size
> 4)
293 * Calculate the number of metadata blocks needed
294 * to allocate @blocks
295 * Worse case is one block per extent
297 int ext4_ext_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
299 struct ext4_inode_info
*ei
= EXT4_I(inode
);
302 idxs
= ((inode
->i_sb
->s_blocksize
- sizeof(struct ext4_extent_header
))
303 / sizeof(struct ext4_extent_idx
));
306 * If the new delayed allocation block is contiguous with the
307 * previous da block, it can share index blocks with the
308 * previous block, so we only need to allocate a new index
309 * block every idxs leaf blocks. At ldxs**2 blocks, we need
310 * an additional index block, and at ldxs**3 blocks, yet
311 * another index blocks.
313 if (ei
->i_da_metadata_calc_len
&&
314 ei
->i_da_metadata_calc_last_lblock
+1 == lblock
) {
317 if ((ei
->i_da_metadata_calc_len
% idxs
) == 0)
319 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
)) == 0)
321 if ((ei
->i_da_metadata_calc_len
% (idxs
*idxs
*idxs
)) == 0) {
323 ei
->i_da_metadata_calc_len
= 0;
325 ei
->i_da_metadata_calc_len
++;
326 ei
->i_da_metadata_calc_last_lblock
++;
331 * In the worst case we need a new set of index blocks at
332 * every level of the inode's extent tree.
334 ei
->i_da_metadata_calc_len
= 1;
335 ei
->i_da_metadata_calc_last_lblock
= lblock
;
336 return ext_depth(inode
) + 1;
340 ext4_ext_max_entries(struct inode
*inode
, int depth
)
344 if (depth
== ext_depth(inode
)) {
346 max
= ext4_ext_space_root(inode
, 1);
348 max
= ext4_ext_space_root_idx(inode
, 1);
351 max
= ext4_ext_space_block(inode
, 1);
353 max
= ext4_ext_space_block_idx(inode
, 1);
359 static int ext4_valid_extent(struct inode
*inode
, struct ext4_extent
*ext
)
361 ext4_fsblk_t block
= ext4_ext_pblock(ext
);
362 int len
= ext4_ext_get_actual_len(ext
);
366 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
369 static int ext4_valid_extent_idx(struct inode
*inode
,
370 struct ext4_extent_idx
*ext_idx
)
372 ext4_fsblk_t block
= ext4_idx_pblock(ext_idx
);
374 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
377 static int ext4_valid_extent_entries(struct inode
*inode
,
378 struct ext4_extent_header
*eh
,
381 unsigned short entries
;
382 if (eh
->eh_entries
== 0)
385 entries
= le16_to_cpu(eh
->eh_entries
);
389 struct ext4_extent
*ext
= EXT_FIRST_EXTENT(eh
);
391 if (!ext4_valid_extent(inode
, ext
))
397 struct ext4_extent_idx
*ext_idx
= EXT_FIRST_INDEX(eh
);
399 if (!ext4_valid_extent_idx(inode
, ext_idx
))
408 static int __ext4_ext_check(const char *function
, unsigned int line
,
409 struct inode
*inode
, struct ext4_extent_header
*eh
,
410 int depth
, ext4_fsblk_t pblk
)
412 const char *error_msg
;
415 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
416 error_msg
= "invalid magic";
419 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
420 error_msg
= "unexpected eh_depth";
423 if (unlikely(eh
->eh_max
== 0)) {
424 error_msg
= "invalid eh_max";
427 max
= ext4_ext_max_entries(inode
, depth
);
428 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
429 error_msg
= "too large eh_max";
432 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
433 error_msg
= "invalid eh_entries";
436 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
437 error_msg
= "invalid extent entries";
440 /* Verify checksum on non-root extent tree nodes */
441 if (ext_depth(inode
) != depth
&&
442 !ext4_extent_block_csum_verify(inode
, eh
)) {
443 error_msg
= "extent tree corrupted";
449 ext4_error_inode(inode
, function
, line
, 0,
450 "pblk %llu bad header/extent: %s - magic %x, "
451 "entries %u, max %u(%u), depth %u(%u)",
452 (unsigned long long) pblk
, error_msg
,
453 le16_to_cpu(eh
->eh_magic
),
454 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
455 max
, le16_to_cpu(eh
->eh_depth
), depth
);
459 #define ext4_ext_check(inode, eh, depth, pblk) \
460 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
462 int ext4_ext_check_inode(struct inode
*inode
)
464 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
), 0);
467 static struct buffer_head
*
468 __read_extent_tree_block(const char *function
, unsigned int line
,
469 struct inode
*inode
, ext4_fsblk_t pblk
, int depth
,
472 struct buffer_head
*bh
;
475 bh
= sb_getblk(inode
->i_sb
, pblk
);
477 return ERR_PTR(-ENOMEM
);
479 if (!bh_uptodate_or_lock(bh
)) {
480 trace_ext4_ext_load_extent(inode
, pblk
, _RET_IP_
);
481 err
= bh_submit_read(bh
);
485 if (buffer_verified(bh
))
487 err
= __ext4_ext_check(function
, line
, inode
,
488 ext_block_hdr(bh
), depth
, pblk
);
491 set_buffer_verified(bh
);
493 * If this is a leaf block, cache all of its entries
495 if (!(flags
& EXT4_EX_NOCACHE
) && depth
== 0) {
496 struct ext4_extent_header
*eh
= ext_block_hdr(bh
);
497 struct ext4_extent
*ex
= EXT_FIRST_EXTENT(eh
);
498 ext4_lblk_t prev
= 0;
501 for (i
= le16_to_cpu(eh
->eh_entries
); i
> 0; i
--, ex
++) {
502 unsigned int status
= EXTENT_STATUS_WRITTEN
;
503 ext4_lblk_t lblk
= le32_to_cpu(ex
->ee_block
);
504 int len
= ext4_ext_get_actual_len(ex
);
506 if (prev
&& (prev
!= lblk
))
507 ext4_es_cache_extent(inode
, prev
,
511 if (ext4_ext_is_uninitialized(ex
))
512 status
= EXTENT_STATUS_UNWRITTEN
;
513 ext4_es_cache_extent(inode
, lblk
, len
,
514 ext4_ext_pblock(ex
), status
);
525 #define read_extent_tree_block(inode, pblk, depth, flags) \
526 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
530 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
532 int k
, l
= path
->p_depth
;
535 for (k
= 0; k
<= l
; k
++, path
++) {
537 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
538 ext4_idx_pblock(path
->p_idx
));
539 } else if (path
->p_ext
) {
540 ext_debug(" %d:[%d]%d:%llu ",
541 le32_to_cpu(path
->p_ext
->ee_block
),
542 ext4_ext_is_uninitialized(path
->p_ext
),
543 ext4_ext_get_actual_len(path
->p_ext
),
544 ext4_ext_pblock(path
->p_ext
));
551 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
553 int depth
= ext_depth(inode
);
554 struct ext4_extent_header
*eh
;
555 struct ext4_extent
*ex
;
561 eh
= path
[depth
].p_hdr
;
562 ex
= EXT_FIRST_EXTENT(eh
);
564 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
566 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
567 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
568 ext4_ext_is_uninitialized(ex
),
569 ext4_ext_get_actual_len(ex
), ext4_ext_pblock(ex
));
574 static void ext4_ext_show_move(struct inode
*inode
, struct ext4_ext_path
*path
,
575 ext4_fsblk_t newblock
, int level
)
577 int depth
= ext_depth(inode
);
578 struct ext4_extent
*ex
;
580 if (depth
!= level
) {
581 struct ext4_extent_idx
*idx
;
582 idx
= path
[level
].p_idx
;
583 while (idx
<= EXT_MAX_INDEX(path
[level
].p_hdr
)) {
584 ext_debug("%d: move %d:%llu in new index %llu\n", level
,
585 le32_to_cpu(idx
->ei_block
),
586 ext4_idx_pblock(idx
),
594 ex
= path
[depth
].p_ext
;
595 while (ex
<= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
596 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
597 le32_to_cpu(ex
->ee_block
),
599 ext4_ext_is_uninitialized(ex
),
600 ext4_ext_get_actual_len(ex
),
607 #define ext4_ext_show_path(inode, path)
608 #define ext4_ext_show_leaf(inode, path)
609 #define ext4_ext_show_move(inode, path, newblock, level)
612 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
614 int depth
= path
->p_depth
;
617 for (i
= 0; i
<= depth
; i
++, path
++)
625 * ext4_ext_binsearch_idx:
626 * binary search for the closest index of the given block
627 * the header must be checked before calling this
630 ext4_ext_binsearch_idx(struct inode
*inode
,
631 struct ext4_ext_path
*path
, ext4_lblk_t block
)
633 struct ext4_extent_header
*eh
= path
->p_hdr
;
634 struct ext4_extent_idx
*r
, *l
, *m
;
637 ext_debug("binsearch for %u(idx): ", block
);
639 l
= EXT_FIRST_INDEX(eh
) + 1;
640 r
= EXT_LAST_INDEX(eh
);
643 if (block
< le32_to_cpu(m
->ei_block
))
647 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
648 m
, le32_to_cpu(m
->ei_block
),
649 r
, le32_to_cpu(r
->ei_block
));
653 ext_debug(" -> %u->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
654 ext4_idx_pblock(path
->p_idx
));
656 #ifdef CHECK_BINSEARCH
658 struct ext4_extent_idx
*chix
, *ix
;
661 chix
= ix
= EXT_FIRST_INDEX(eh
);
662 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
664 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
665 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
667 ix
, EXT_FIRST_INDEX(eh
));
668 printk(KERN_DEBUG
"%u <= %u\n",
669 le32_to_cpu(ix
->ei_block
),
670 le32_to_cpu(ix
[-1].ei_block
));
672 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
673 <= le32_to_cpu(ix
[-1].ei_block
));
674 if (block
< le32_to_cpu(ix
->ei_block
))
678 BUG_ON(chix
!= path
->p_idx
);
685 * ext4_ext_binsearch:
686 * binary search for closest extent of the given block
687 * the header must be checked before calling this
690 ext4_ext_binsearch(struct inode
*inode
,
691 struct ext4_ext_path
*path
, ext4_lblk_t block
)
693 struct ext4_extent_header
*eh
= path
->p_hdr
;
694 struct ext4_extent
*r
, *l
, *m
;
696 if (eh
->eh_entries
== 0) {
698 * this leaf is empty:
699 * we get such a leaf in split/add case
704 ext_debug("binsearch for %u: ", block
);
706 l
= EXT_FIRST_EXTENT(eh
) + 1;
707 r
= EXT_LAST_EXTENT(eh
);
711 if (block
< le32_to_cpu(m
->ee_block
))
715 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
716 m
, le32_to_cpu(m
->ee_block
),
717 r
, le32_to_cpu(r
->ee_block
));
721 ext_debug(" -> %d:%llu:[%d]%d ",
722 le32_to_cpu(path
->p_ext
->ee_block
),
723 ext4_ext_pblock(path
->p_ext
),
724 ext4_ext_is_uninitialized(path
->p_ext
),
725 ext4_ext_get_actual_len(path
->p_ext
));
727 #ifdef CHECK_BINSEARCH
729 struct ext4_extent
*chex
, *ex
;
732 chex
= ex
= EXT_FIRST_EXTENT(eh
);
733 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
734 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
735 <= le32_to_cpu(ex
[-1].ee_block
));
736 if (block
< le32_to_cpu(ex
->ee_block
))
740 BUG_ON(chex
!= path
->p_ext
);
746 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
748 struct ext4_extent_header
*eh
;
750 eh
= ext_inode_hdr(inode
);
753 eh
->eh_magic
= EXT4_EXT_MAGIC
;
754 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
755 ext4_mark_inode_dirty(handle
, inode
);
759 struct ext4_ext_path
*
760 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
761 struct ext4_ext_path
*path
, int flags
)
763 struct ext4_extent_header
*eh
;
764 struct buffer_head
*bh
;
765 short int depth
, i
, ppos
= 0, alloc
= 0;
768 eh
= ext_inode_hdr(inode
);
769 depth
= ext_depth(inode
);
771 /* account possible depth increase */
773 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
776 return ERR_PTR(-ENOMEM
);
783 /* walk through the tree */
785 ext_debug("depth %d: num %d, max %d\n",
786 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
788 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
789 path
[ppos
].p_block
= ext4_idx_pblock(path
[ppos
].p_idx
);
790 path
[ppos
].p_depth
= i
;
791 path
[ppos
].p_ext
= NULL
;
793 bh
= read_extent_tree_block(inode
, path
[ppos
].p_block
, --i
,
800 eh
= ext_block_hdr(bh
);
802 if (unlikely(ppos
> depth
)) {
804 EXT4_ERROR_INODE(inode
,
805 "ppos %d > depth %d", ppos
, depth
);
809 path
[ppos
].p_bh
= bh
;
810 path
[ppos
].p_hdr
= eh
;
813 path
[ppos
].p_depth
= i
;
814 path
[ppos
].p_ext
= NULL
;
815 path
[ppos
].p_idx
= NULL
;
818 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
819 /* if not an empty leaf */
820 if (path
[ppos
].p_ext
)
821 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
823 ext4_ext_show_path(inode
, path
);
828 ext4_ext_drop_refs(path
);
835 * ext4_ext_insert_index:
836 * insert new index [@logical;@ptr] into the block at @curp;
837 * check where to insert: before @curp or after @curp
839 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
840 struct ext4_ext_path
*curp
,
841 int logical
, ext4_fsblk_t ptr
)
843 struct ext4_extent_idx
*ix
;
846 err
= ext4_ext_get_access(handle
, inode
, curp
);
850 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
851 EXT4_ERROR_INODE(inode
,
852 "logical %d == ei_block %d!",
853 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
857 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
858 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
859 EXT4_ERROR_INODE(inode
,
860 "eh_entries %d >= eh_max %d!",
861 le16_to_cpu(curp
->p_hdr
->eh_entries
),
862 le16_to_cpu(curp
->p_hdr
->eh_max
));
866 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
868 ext_debug("insert new index %d after: %llu\n", logical
, ptr
);
869 ix
= curp
->p_idx
+ 1;
872 ext_debug("insert new index %d before: %llu\n", logical
, ptr
);
876 len
= EXT_LAST_INDEX(curp
->p_hdr
) - ix
+ 1;
879 ext_debug("insert new index %d: "
880 "move %d indices from 0x%p to 0x%p\n",
881 logical
, len
, ix
, ix
+ 1);
882 memmove(ix
+ 1, ix
, len
* sizeof(struct ext4_extent_idx
));
885 if (unlikely(ix
> EXT_MAX_INDEX(curp
->p_hdr
))) {
886 EXT4_ERROR_INODE(inode
, "ix > EXT_MAX_INDEX!");
890 ix
->ei_block
= cpu_to_le32(logical
);
891 ext4_idx_store_pblock(ix
, ptr
);
892 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
894 if (unlikely(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
895 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
899 err
= ext4_ext_dirty(handle
, inode
, curp
);
900 ext4_std_error(inode
->i_sb
, err
);
907 * inserts new subtree into the path, using free index entry
909 * - allocates all needed blocks (new leaf and all intermediate index blocks)
910 * - makes decision where to split
911 * - moves remaining extents and index entries (right to the split point)
912 * into the newly allocated blocks
913 * - initializes subtree
915 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
917 struct ext4_ext_path
*path
,
918 struct ext4_extent
*newext
, int at
)
920 struct buffer_head
*bh
= NULL
;
921 int depth
= ext_depth(inode
);
922 struct ext4_extent_header
*neh
;
923 struct ext4_extent_idx
*fidx
;
925 ext4_fsblk_t newblock
, oldblock
;
927 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
930 /* make decision: where to split? */
931 /* FIXME: now decision is simplest: at current extent */
933 /* if current leaf will be split, then we should use
934 * border from split point */
935 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
936 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
939 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
940 border
= path
[depth
].p_ext
[1].ee_block
;
941 ext_debug("leaf will be split."
942 " next leaf starts at %d\n",
943 le32_to_cpu(border
));
945 border
= newext
->ee_block
;
946 ext_debug("leaf will be added."
947 " next leaf starts at %d\n",
948 le32_to_cpu(border
));
952 * If error occurs, then we break processing
953 * and mark filesystem read-only. index won't
954 * be inserted and tree will be in consistent
955 * state. Next mount will repair buffers too.
959 * Get array to track all allocated blocks.
960 * We need this to handle errors and free blocks
963 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
967 /* allocate all needed blocks */
968 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
969 for (a
= 0; a
< depth
- at
; a
++) {
970 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
971 newext
, &err
, flags
);
974 ablocks
[a
] = newblock
;
977 /* initialize new leaf */
978 newblock
= ablocks
[--a
];
979 if (unlikely(newblock
== 0)) {
980 EXT4_ERROR_INODE(inode
, "newblock == 0!");
984 bh
= sb_getblk(inode
->i_sb
, newblock
);
991 err
= ext4_journal_get_create_access(handle
, bh
);
995 neh
= ext_block_hdr(bh
);
997 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
998 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1001 /* move remainder of path[depth] to the new leaf */
1002 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
1003 path
[depth
].p_hdr
->eh_max
)) {
1004 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
1005 path
[depth
].p_hdr
->eh_entries
,
1006 path
[depth
].p_hdr
->eh_max
);
1010 /* start copy from next extent */
1011 m
= EXT_MAX_EXTENT(path
[depth
].p_hdr
) - path
[depth
].p_ext
++;
1012 ext4_ext_show_move(inode
, path
, newblock
, depth
);
1014 struct ext4_extent
*ex
;
1015 ex
= EXT_FIRST_EXTENT(neh
);
1016 memmove(ex
, path
[depth
].p_ext
, sizeof(struct ext4_extent
) * m
);
1017 le16_add_cpu(&neh
->eh_entries
, m
);
1020 ext4_extent_block_csum_set(inode
, neh
);
1021 set_buffer_uptodate(bh
);
1024 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1030 /* correct old leaf */
1032 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1035 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
1036 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1042 /* create intermediate indexes */
1044 if (unlikely(k
< 0)) {
1045 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
1050 ext_debug("create %d intermediate indices\n", k
);
1051 /* insert new index into current index block */
1052 /* current depth stored in i var */
1055 oldblock
= newblock
;
1056 newblock
= ablocks
[--a
];
1057 bh
= sb_getblk(inode
->i_sb
, newblock
);
1058 if (unlikely(!bh
)) {
1064 err
= ext4_journal_get_create_access(handle
, bh
);
1068 neh
= ext_block_hdr(bh
);
1069 neh
->eh_entries
= cpu_to_le16(1);
1070 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1071 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1072 neh
->eh_depth
= cpu_to_le16(depth
- i
);
1073 fidx
= EXT_FIRST_INDEX(neh
);
1074 fidx
->ei_block
= border
;
1075 ext4_idx_store_pblock(fidx
, oldblock
);
1077 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1078 i
, newblock
, le32_to_cpu(border
), oldblock
);
1080 /* move remainder of path[i] to the new index block */
1081 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
1082 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
1083 EXT4_ERROR_INODE(inode
,
1084 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1085 le32_to_cpu(path
[i
].p_ext
->ee_block
));
1089 /* start copy indexes */
1090 m
= EXT_MAX_INDEX(path
[i
].p_hdr
) - path
[i
].p_idx
++;
1091 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
1092 EXT_MAX_INDEX(path
[i
].p_hdr
));
1093 ext4_ext_show_move(inode
, path
, newblock
, i
);
1095 memmove(++fidx
, path
[i
].p_idx
,
1096 sizeof(struct ext4_extent_idx
) * m
);
1097 le16_add_cpu(&neh
->eh_entries
, m
);
1099 ext4_extent_block_csum_set(inode
, neh
);
1100 set_buffer_uptodate(bh
);
1103 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1109 /* correct old index */
1111 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1114 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1115 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1123 /* insert new index */
1124 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1125 le32_to_cpu(border
), newblock
);
1129 if (buffer_locked(bh
))
1135 /* free all allocated blocks in error case */
1136 for (i
= 0; i
< depth
; i
++) {
1139 ext4_free_blocks(handle
, inode
, NULL
, ablocks
[i
], 1,
1140 EXT4_FREE_BLOCKS_METADATA
);
1149 * ext4_ext_grow_indepth:
1150 * implements tree growing procedure:
1151 * - allocates new block
1152 * - moves top-level data (index block or leaf) into the new block
1153 * - initializes new top-level, creating index that points to the
1154 * just created block
1156 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1158 struct ext4_extent
*newext
)
1160 struct ext4_extent_header
*neh
;
1161 struct buffer_head
*bh
;
1162 ext4_fsblk_t newblock
;
1165 newblock
= ext4_ext_new_meta_block(handle
, inode
, NULL
,
1166 newext
, &err
, flags
);
1170 bh
= sb_getblk(inode
->i_sb
, newblock
);
1175 err
= ext4_journal_get_create_access(handle
, bh
);
1181 /* move top-level index/leaf into new block */
1182 memmove(bh
->b_data
, EXT4_I(inode
)->i_data
,
1183 sizeof(EXT4_I(inode
)->i_data
));
1185 /* set size of new block */
1186 neh
= ext_block_hdr(bh
);
1187 /* old root could have indexes or leaves
1188 * so calculate e_max right way */
1189 if (ext_depth(inode
))
1190 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1192 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1193 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1194 ext4_extent_block_csum_set(inode
, neh
);
1195 set_buffer_uptodate(bh
);
1198 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1202 /* Update top-level index: num,max,pointer */
1203 neh
= ext_inode_hdr(inode
);
1204 neh
->eh_entries
= cpu_to_le16(1);
1205 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh
), newblock
);
1206 if (neh
->eh_depth
== 0) {
1207 /* Root extent block becomes index block */
1208 neh
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1209 EXT_FIRST_INDEX(neh
)->ei_block
=
1210 EXT_FIRST_EXTENT(neh
)->ee_block
;
1212 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1213 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1214 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1215 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1217 le16_add_cpu(&neh
->eh_depth
, 1);
1218 ext4_mark_inode_dirty(handle
, inode
);
1226 * ext4_ext_create_new_leaf:
1227 * finds empty index and adds new leaf.
1228 * if no free index is found, then it requests in-depth growing.
1230 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1231 unsigned int mb_flags
,
1232 unsigned int gb_flags
,
1233 struct ext4_ext_path
*path
,
1234 struct ext4_extent
*newext
)
1236 struct ext4_ext_path
*curp
;
1237 int depth
, i
, err
= 0;
1240 i
= depth
= ext_depth(inode
);
1242 /* walk up to the tree and look for free index entry */
1243 curp
= path
+ depth
;
1244 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1249 /* we use already allocated block for index block,
1250 * so subsequent data blocks should be contiguous */
1251 if (EXT_HAS_FREE_INDEX(curp
)) {
1252 /* if we found index with free entry, then use that
1253 * entry: create all needed subtree and add new leaf */
1254 err
= ext4_ext_split(handle
, inode
, mb_flags
, path
, newext
, i
);
1259 ext4_ext_drop_refs(path
);
1260 path
= ext4_ext_find_extent(inode
,
1261 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1264 err
= PTR_ERR(path
);
1266 /* tree is full, time to grow in depth */
1267 err
= ext4_ext_grow_indepth(handle
, inode
, mb_flags
, newext
);
1272 ext4_ext_drop_refs(path
);
1273 path
= ext4_ext_find_extent(inode
,
1274 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1277 err
= PTR_ERR(path
);
1282 * only first (depth 0 -> 1) produces free space;
1283 * in all other cases we have to split the grown tree
1285 depth
= ext_depth(inode
);
1286 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1287 /* now we need to split */
1297 * search the closest allocated block to the left for *logical
1298 * and returns it at @logical + it's physical address at @phys
1299 * if *logical is the smallest allocated block, the function
1300 * returns 0 at @phys
1301 * return value contains 0 (success) or error code
1303 static int ext4_ext_search_left(struct inode
*inode
,
1304 struct ext4_ext_path
*path
,
1305 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1307 struct ext4_extent_idx
*ix
;
1308 struct ext4_extent
*ex
;
1311 if (unlikely(path
== NULL
)) {
1312 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1315 depth
= path
->p_depth
;
1318 if (depth
== 0 && path
->p_ext
== NULL
)
1321 /* usually extent in the path covers blocks smaller
1322 * then *logical, but it can be that extent is the
1323 * first one in the file */
1325 ex
= path
[depth
].p_ext
;
1326 ee_len
= ext4_ext_get_actual_len(ex
);
1327 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1328 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1329 EXT4_ERROR_INODE(inode
,
1330 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1331 *logical
, le32_to_cpu(ex
->ee_block
));
1334 while (--depth
>= 0) {
1335 ix
= path
[depth
].p_idx
;
1336 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1337 EXT4_ERROR_INODE(inode
,
1338 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1339 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1340 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1341 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1349 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1350 EXT4_ERROR_INODE(inode
,
1351 "logical %d < ee_block %d + ee_len %d!",
1352 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1356 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1357 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1362 * search the closest allocated block to the right for *logical
1363 * and returns it at @logical + it's physical address at @phys
1364 * if *logical is the largest allocated block, the function
1365 * returns 0 at @phys
1366 * return value contains 0 (success) or error code
1368 static int ext4_ext_search_right(struct inode
*inode
,
1369 struct ext4_ext_path
*path
,
1370 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1371 struct ext4_extent
**ret_ex
)
1373 struct buffer_head
*bh
= NULL
;
1374 struct ext4_extent_header
*eh
;
1375 struct ext4_extent_idx
*ix
;
1376 struct ext4_extent
*ex
;
1378 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1381 if (unlikely(path
== NULL
)) {
1382 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1385 depth
= path
->p_depth
;
1388 if (depth
== 0 && path
->p_ext
== NULL
)
1391 /* usually extent in the path covers blocks smaller
1392 * then *logical, but it can be that extent is the
1393 * first one in the file */
1395 ex
= path
[depth
].p_ext
;
1396 ee_len
= ext4_ext_get_actual_len(ex
);
1397 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1398 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1399 EXT4_ERROR_INODE(inode
,
1400 "first_extent(path[%d].p_hdr) != ex",
1404 while (--depth
>= 0) {
1405 ix
= path
[depth
].p_idx
;
1406 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1407 EXT4_ERROR_INODE(inode
,
1408 "ix != EXT_FIRST_INDEX *logical %d!",
1416 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1417 EXT4_ERROR_INODE(inode
,
1418 "logical %d < ee_block %d + ee_len %d!",
1419 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1423 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1424 /* next allocated block in this leaf */
1429 /* go up and search for index to the right */
1430 while (--depth
>= 0) {
1431 ix
= path
[depth
].p_idx
;
1432 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1436 /* we've gone up to the root and found no index to the right */
1440 /* we've found index to the right, let's
1441 * follow it and find the closest allocated
1442 * block to the right */
1444 block
= ext4_idx_pblock(ix
);
1445 while (++depth
< path
->p_depth
) {
1446 /* subtract from p_depth to get proper eh_depth */
1447 bh
= read_extent_tree_block(inode
, block
,
1448 path
->p_depth
- depth
, 0);
1451 eh
= ext_block_hdr(bh
);
1452 ix
= EXT_FIRST_INDEX(eh
);
1453 block
= ext4_idx_pblock(ix
);
1457 bh
= read_extent_tree_block(inode
, block
, path
->p_depth
- depth
, 0);
1460 eh
= ext_block_hdr(bh
);
1461 ex
= EXT_FIRST_EXTENT(eh
);
1463 *logical
= le32_to_cpu(ex
->ee_block
);
1464 *phys
= ext4_ext_pblock(ex
);
1472 * ext4_ext_next_allocated_block:
1473 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1474 * NOTE: it considers block number from index entry as
1475 * allocated block. Thus, index entries have to be consistent
1479 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1483 BUG_ON(path
== NULL
);
1484 depth
= path
->p_depth
;
1486 if (depth
== 0 && path
->p_ext
== NULL
)
1487 return EXT_MAX_BLOCKS
;
1489 while (depth
>= 0) {
1490 if (depth
== path
->p_depth
) {
1492 if (path
[depth
].p_ext
&&
1493 path
[depth
].p_ext
!=
1494 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1495 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1498 if (path
[depth
].p_idx
!=
1499 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1500 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1505 return EXT_MAX_BLOCKS
;
1509 * ext4_ext_next_leaf_block:
1510 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1512 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1516 BUG_ON(path
== NULL
);
1517 depth
= path
->p_depth
;
1519 /* zero-tree has no leaf blocks at all */
1521 return EXT_MAX_BLOCKS
;
1523 /* go to index block */
1526 while (depth
>= 0) {
1527 if (path
[depth
].p_idx
!=
1528 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1529 return (ext4_lblk_t
)
1530 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1534 return EXT_MAX_BLOCKS
;
1538 * ext4_ext_correct_indexes:
1539 * if leaf gets modified and modified extent is first in the leaf,
1540 * then we have to correct all indexes above.
1541 * TODO: do we need to correct tree in all cases?
1543 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1544 struct ext4_ext_path
*path
)
1546 struct ext4_extent_header
*eh
;
1547 int depth
= ext_depth(inode
);
1548 struct ext4_extent
*ex
;
1552 eh
= path
[depth
].p_hdr
;
1553 ex
= path
[depth
].p_ext
;
1555 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1556 EXT4_ERROR_INODE(inode
,
1557 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1562 /* there is no tree at all */
1566 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1567 /* we correct tree if first leaf got modified only */
1572 * TODO: we need correction if border is smaller than current one
1575 border
= path
[depth
].p_ext
->ee_block
;
1576 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1579 path
[k
].p_idx
->ei_block
= border
;
1580 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1585 /* change all left-side indexes */
1586 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1588 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1591 path
[k
].p_idx
->ei_block
= border
;
1592 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1601 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1602 struct ext4_extent
*ex2
)
1604 unsigned short ext1_ee_len
, ext2_ee_len
, max_len
;
1607 * Make sure that both extents are initialized. We don't merge
1608 * uninitialized extents so that we can be sure that end_io code has
1609 * the extent that was written properly split out and conversion to
1610 * initialized is trivial.
1612 if (ext4_ext_is_uninitialized(ex1
) || ext4_ext_is_uninitialized(ex2
))
1615 if (ext4_ext_is_uninitialized(ex1
))
1616 max_len
= EXT_UNINIT_MAX_LEN
;
1618 max_len
= EXT_INIT_MAX_LEN
;
1620 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1621 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1623 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1624 le32_to_cpu(ex2
->ee_block
))
1628 * To allow future support for preallocated extents to be added
1629 * as an RO_COMPAT feature, refuse to merge to extents if
1630 * this can result in the top bit of ee_len being set.
1632 if (ext1_ee_len
+ ext2_ee_len
> max_len
)
1634 #ifdef AGGRESSIVE_TEST
1635 if (ext1_ee_len
>= 4)
1639 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1645 * This function tries to merge the "ex" extent to the next extent in the tree.
1646 * It always tries to merge towards right. If you want to merge towards
1647 * left, pass "ex - 1" as argument instead of "ex".
1648 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1649 * 1 if they got merged.
1651 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1652 struct ext4_ext_path
*path
,
1653 struct ext4_extent
*ex
)
1655 struct ext4_extent_header
*eh
;
1656 unsigned int depth
, len
;
1658 int uninitialized
= 0;
1660 depth
= ext_depth(inode
);
1661 BUG_ON(path
[depth
].p_hdr
== NULL
);
1662 eh
= path
[depth
].p_hdr
;
1664 while (ex
< EXT_LAST_EXTENT(eh
)) {
1665 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1667 /* merge with next extent! */
1668 if (ext4_ext_is_uninitialized(ex
))
1670 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1671 + ext4_ext_get_actual_len(ex
+ 1));
1673 ext4_ext_mark_uninitialized(ex
);
1675 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1676 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1677 * sizeof(struct ext4_extent
);
1678 memmove(ex
+ 1, ex
+ 2, len
);
1680 le16_add_cpu(&eh
->eh_entries
, -1);
1682 WARN_ON(eh
->eh_entries
== 0);
1683 if (!eh
->eh_entries
)
1684 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1691 * This function does a very simple check to see if we can collapse
1692 * an extent tree with a single extent tree leaf block into the inode.
1694 static void ext4_ext_try_to_merge_up(handle_t
*handle
,
1695 struct inode
*inode
,
1696 struct ext4_ext_path
*path
)
1699 unsigned max_root
= ext4_ext_space_root(inode
, 0);
1702 if ((path
[0].p_depth
!= 1) ||
1703 (le16_to_cpu(path
[0].p_hdr
->eh_entries
) != 1) ||
1704 (le16_to_cpu(path
[1].p_hdr
->eh_entries
) > max_root
))
1708 * We need to modify the block allocation bitmap and the block
1709 * group descriptor to release the extent tree block. If we
1710 * can't get the journal credits, give up.
1712 if (ext4_journal_extend(handle
, 2))
1716 * Copy the extent data up to the inode
1718 blk
= ext4_idx_pblock(path
[0].p_idx
);
1719 s
= le16_to_cpu(path
[1].p_hdr
->eh_entries
) *
1720 sizeof(struct ext4_extent_idx
);
1721 s
+= sizeof(struct ext4_extent_header
);
1723 memcpy(path
[0].p_hdr
, path
[1].p_hdr
, s
);
1724 path
[0].p_depth
= 0;
1725 path
[0].p_ext
= EXT_FIRST_EXTENT(path
[0].p_hdr
) +
1726 (path
[1].p_ext
- EXT_FIRST_EXTENT(path
[1].p_hdr
));
1727 path
[0].p_hdr
->eh_max
= cpu_to_le16(max_root
);
1729 brelse(path
[1].p_bh
);
1730 ext4_free_blocks(handle
, inode
, NULL
, blk
, 1,
1731 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
1735 * This function tries to merge the @ex extent to neighbours in the tree.
1736 * return 1 if merge left else 0.
1738 static void ext4_ext_try_to_merge(handle_t
*handle
,
1739 struct inode
*inode
,
1740 struct ext4_ext_path
*path
,
1741 struct ext4_extent
*ex
) {
1742 struct ext4_extent_header
*eh
;
1746 depth
= ext_depth(inode
);
1747 BUG_ON(path
[depth
].p_hdr
== NULL
);
1748 eh
= path
[depth
].p_hdr
;
1750 if (ex
> EXT_FIRST_EXTENT(eh
))
1751 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1754 (void) ext4_ext_try_to_merge_right(inode
, path
, ex
);
1756 ext4_ext_try_to_merge_up(handle
, inode
, path
);
1760 * check if a portion of the "newext" extent overlaps with an
1763 * If there is an overlap discovered, it updates the length of the newext
1764 * such that there will be no overlap, and then returns 1.
1765 * If there is no overlap found, it returns 0.
1767 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1768 struct inode
*inode
,
1769 struct ext4_extent
*newext
,
1770 struct ext4_ext_path
*path
)
1773 unsigned int depth
, len1
;
1774 unsigned int ret
= 0;
1776 b1
= le32_to_cpu(newext
->ee_block
);
1777 len1
= ext4_ext_get_actual_len(newext
);
1778 depth
= ext_depth(inode
);
1779 if (!path
[depth
].p_ext
)
1781 b2
= le32_to_cpu(path
[depth
].p_ext
->ee_block
);
1782 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1785 * get the next allocated block if the extent in the path
1786 * is before the requested block(s)
1789 b2
= ext4_ext_next_allocated_block(path
);
1790 if (b2
== EXT_MAX_BLOCKS
)
1792 b2
&= ~(sbi
->s_cluster_ratio
- 1);
1795 /* check for wrap through zero on extent logical start block*/
1796 if (b1
+ len1
< b1
) {
1797 len1
= EXT_MAX_BLOCKS
- b1
;
1798 newext
->ee_len
= cpu_to_le16(len1
);
1802 /* check for overlap */
1803 if (b1
+ len1
> b2
) {
1804 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1812 * ext4_ext_insert_extent:
1813 * tries to merge requsted extent into the existing extent or
1814 * inserts requested extent as new one into the tree,
1815 * creating new leaf in the no-space case.
1817 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1818 struct ext4_ext_path
*path
,
1819 struct ext4_extent
*newext
, int gb_flags
)
1821 struct ext4_extent_header
*eh
;
1822 struct ext4_extent
*ex
, *fex
;
1823 struct ext4_extent
*nearex
; /* nearest extent */
1824 struct ext4_ext_path
*npath
= NULL
;
1825 int depth
, len
, err
;
1827 unsigned uninitialized
= 0;
1830 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1831 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1834 depth
= ext_depth(inode
);
1835 ex
= path
[depth
].p_ext
;
1836 eh
= path
[depth
].p_hdr
;
1837 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1838 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1842 /* try to insert block into found extent and return */
1843 if (ex
&& !(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
1846 * Try to see whether we should rather test the extent on
1847 * right from ex, or from the left of ex. This is because
1848 * ext4_ext_find_extent() can return either extent on the
1849 * left, or on the right from the searched position. This
1850 * will make merging more effective.
1852 if (ex
< EXT_LAST_EXTENT(eh
) &&
1853 (le32_to_cpu(ex
->ee_block
) +
1854 ext4_ext_get_actual_len(ex
) <
1855 le32_to_cpu(newext
->ee_block
))) {
1858 } else if ((ex
> EXT_FIRST_EXTENT(eh
)) &&
1859 (le32_to_cpu(newext
->ee_block
) +
1860 ext4_ext_get_actual_len(newext
) <
1861 le32_to_cpu(ex
->ee_block
)))
1864 /* Try to append newex to the ex */
1865 if (ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1866 ext_debug("append [%d]%d block to %u:[%d]%d"
1868 ext4_ext_is_uninitialized(newext
),
1869 ext4_ext_get_actual_len(newext
),
1870 le32_to_cpu(ex
->ee_block
),
1871 ext4_ext_is_uninitialized(ex
),
1872 ext4_ext_get_actual_len(ex
),
1873 ext4_ext_pblock(ex
));
1874 err
= ext4_ext_get_access(handle
, inode
,
1880 * ext4_can_extents_be_merged should have checked
1881 * that either both extents are uninitialized, or
1882 * both aren't. Thus we need to check only one of
1885 if (ext4_ext_is_uninitialized(ex
))
1887 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1888 + ext4_ext_get_actual_len(newext
));
1890 ext4_ext_mark_uninitialized(ex
);
1891 eh
= path
[depth
].p_hdr
;
1897 /* Try to prepend newex to the ex */
1898 if (ext4_can_extents_be_merged(inode
, newext
, ex
)) {
1899 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1901 le32_to_cpu(newext
->ee_block
),
1902 ext4_ext_is_uninitialized(newext
),
1903 ext4_ext_get_actual_len(newext
),
1904 le32_to_cpu(ex
->ee_block
),
1905 ext4_ext_is_uninitialized(ex
),
1906 ext4_ext_get_actual_len(ex
),
1907 ext4_ext_pblock(ex
));
1908 err
= ext4_ext_get_access(handle
, inode
,
1914 * ext4_can_extents_be_merged should have checked
1915 * that either both extents are uninitialized, or
1916 * both aren't. Thus we need to check only one of
1919 if (ext4_ext_is_uninitialized(ex
))
1921 ex
->ee_block
= newext
->ee_block
;
1922 ext4_ext_store_pblock(ex
, ext4_ext_pblock(newext
));
1923 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1924 + ext4_ext_get_actual_len(newext
));
1926 ext4_ext_mark_uninitialized(ex
);
1927 eh
= path
[depth
].p_hdr
;
1933 depth
= ext_depth(inode
);
1934 eh
= path
[depth
].p_hdr
;
1935 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1938 /* probably next leaf has space for us? */
1939 fex
= EXT_LAST_EXTENT(eh
);
1940 next
= EXT_MAX_BLOCKS
;
1941 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
1942 next
= ext4_ext_next_leaf_block(path
);
1943 if (next
!= EXT_MAX_BLOCKS
) {
1944 ext_debug("next leaf block - %u\n", next
);
1945 BUG_ON(npath
!= NULL
);
1946 npath
= ext4_ext_find_extent(inode
, next
, NULL
, 0);
1948 return PTR_ERR(npath
);
1949 BUG_ON(npath
->p_depth
!= path
->p_depth
);
1950 eh
= npath
[depth
].p_hdr
;
1951 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
1952 ext_debug("next leaf isn't full(%d)\n",
1953 le16_to_cpu(eh
->eh_entries
));
1957 ext_debug("next leaf has no free space(%d,%d)\n",
1958 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
1962 * There is no free space in the found leaf.
1963 * We're gonna add a new leaf in the tree.
1965 if (gb_flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
1966 mb_flags
= EXT4_MB_USE_RESERVED
;
1967 err
= ext4_ext_create_new_leaf(handle
, inode
, mb_flags
, gb_flags
,
1971 depth
= ext_depth(inode
);
1972 eh
= path
[depth
].p_hdr
;
1975 nearex
= path
[depth
].p_ext
;
1977 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1982 /* there is no extent in this leaf, create first one */
1983 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
1984 le32_to_cpu(newext
->ee_block
),
1985 ext4_ext_pblock(newext
),
1986 ext4_ext_is_uninitialized(newext
),
1987 ext4_ext_get_actual_len(newext
));
1988 nearex
= EXT_FIRST_EXTENT(eh
);
1990 if (le32_to_cpu(newext
->ee_block
)
1991 > le32_to_cpu(nearex
->ee_block
)) {
1993 ext_debug("insert %u:%llu:[%d]%d before: "
1995 le32_to_cpu(newext
->ee_block
),
1996 ext4_ext_pblock(newext
),
1997 ext4_ext_is_uninitialized(newext
),
1998 ext4_ext_get_actual_len(newext
),
2003 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
2004 ext_debug("insert %u:%llu:[%d]%d after: "
2006 le32_to_cpu(newext
->ee_block
),
2007 ext4_ext_pblock(newext
),
2008 ext4_ext_is_uninitialized(newext
),
2009 ext4_ext_get_actual_len(newext
),
2012 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
2014 ext_debug("insert %u:%llu:[%d]%d: "
2015 "move %d extents from 0x%p to 0x%p\n",
2016 le32_to_cpu(newext
->ee_block
),
2017 ext4_ext_pblock(newext
),
2018 ext4_ext_is_uninitialized(newext
),
2019 ext4_ext_get_actual_len(newext
),
2020 len
, nearex
, nearex
+ 1);
2021 memmove(nearex
+ 1, nearex
,
2022 len
* sizeof(struct ext4_extent
));
2026 le16_add_cpu(&eh
->eh_entries
, 1);
2027 path
[depth
].p_ext
= nearex
;
2028 nearex
->ee_block
= newext
->ee_block
;
2029 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
2030 nearex
->ee_len
= newext
->ee_len
;
2033 /* try to merge extents */
2034 if (!(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
))
2035 ext4_ext_try_to_merge(handle
, inode
, path
, nearex
);
2038 /* time to correct all indexes above */
2039 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2043 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
2047 ext4_ext_drop_refs(npath
);
2053 static int ext4_fill_fiemap_extents(struct inode
*inode
,
2054 ext4_lblk_t block
, ext4_lblk_t num
,
2055 struct fiemap_extent_info
*fieinfo
)
2057 struct ext4_ext_path
*path
= NULL
;
2058 struct ext4_extent
*ex
;
2059 struct extent_status es
;
2060 ext4_lblk_t next
, next_del
, start
= 0, end
= 0;
2061 ext4_lblk_t last
= block
+ num
;
2062 int exists
, depth
= 0, err
= 0;
2063 unsigned int flags
= 0;
2064 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
2066 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
2068 /* find extent for this block */
2069 down_read(&EXT4_I(inode
)->i_data_sem
);
2071 if (path
&& ext_depth(inode
) != depth
) {
2072 /* depth was changed. we have to realloc path */
2077 path
= ext4_ext_find_extent(inode
, block
, path
, 0);
2079 up_read(&EXT4_I(inode
)->i_data_sem
);
2080 err
= PTR_ERR(path
);
2085 depth
= ext_depth(inode
);
2086 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2087 up_read(&EXT4_I(inode
)->i_data_sem
);
2088 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2092 ex
= path
[depth
].p_ext
;
2093 next
= ext4_ext_next_allocated_block(path
);
2094 ext4_ext_drop_refs(path
);
2099 /* there is no extent yet, so try to allocate
2100 * all requested space */
2103 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
2104 /* need to allocate space before found extent */
2106 end
= le32_to_cpu(ex
->ee_block
);
2107 if (block
+ num
< end
)
2109 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2110 + ext4_ext_get_actual_len(ex
)) {
2111 /* need to allocate space after found extent */
2116 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
2118 * some part of requested space is covered
2122 end
= le32_to_cpu(ex
->ee_block
)
2123 + ext4_ext_get_actual_len(ex
);
2124 if (block
+ num
< end
)
2130 BUG_ON(end
<= start
);
2134 es
.es_len
= end
- start
;
2137 es
.es_lblk
= le32_to_cpu(ex
->ee_block
);
2138 es
.es_len
= ext4_ext_get_actual_len(ex
);
2139 es
.es_pblk
= ext4_ext_pblock(ex
);
2140 if (ext4_ext_is_uninitialized(ex
))
2141 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
2145 * Find delayed extent and update es accordingly. We call
2146 * it even in !exists case to find out whether es is the
2147 * last existing extent or not.
2149 next_del
= ext4_find_delayed_extent(inode
, &es
);
2150 if (!exists
&& next_del
) {
2152 flags
|= (FIEMAP_EXTENT_DELALLOC
|
2153 FIEMAP_EXTENT_UNKNOWN
);
2155 up_read(&EXT4_I(inode
)->i_data_sem
);
2157 if (unlikely(es
.es_len
== 0)) {
2158 EXT4_ERROR_INODE(inode
, "es.es_len == 0");
2164 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2165 * we need to check next == EXT_MAX_BLOCKS because it is
2166 * possible that an extent is with unwritten and delayed
2167 * status due to when an extent is delayed allocated and
2168 * is allocated by fallocate status tree will track both of
2171 * So we could return a unwritten and delayed extent, and
2172 * its block is equal to 'next'.
2174 if (next
== next_del
&& next
== EXT_MAX_BLOCKS
) {
2175 flags
|= FIEMAP_EXTENT_LAST
;
2176 if (unlikely(next_del
!= EXT_MAX_BLOCKS
||
2177 next
!= EXT_MAX_BLOCKS
)) {
2178 EXT4_ERROR_INODE(inode
,
2179 "next extent == %u, next "
2180 "delalloc extent = %u",
2188 err
= fiemap_fill_next_extent(fieinfo
,
2189 (__u64
)es
.es_lblk
<< blksize_bits
,
2190 (__u64
)es
.es_pblk
<< blksize_bits
,
2191 (__u64
)es
.es_len
<< blksize_bits
,
2201 block
= es
.es_lblk
+ es
.es_len
;
2205 ext4_ext_drop_refs(path
);
2213 * ext4_ext_put_gap_in_cache:
2214 * calculate boundaries of the gap that the requested block fits into
2215 * and cache this gap
2218 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
2221 int depth
= ext_depth(inode
);
2224 struct ext4_extent
*ex
;
2226 ex
= path
[depth
].p_ext
;
2229 * there is no extent yet, so gap is [0;-] and we
2232 ext_debug("cache gap(whole file):");
2233 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
2235 len
= le32_to_cpu(ex
->ee_block
) - block
;
2236 ext_debug("cache gap(before): %u [%u:%u]",
2238 le32_to_cpu(ex
->ee_block
),
2239 ext4_ext_get_actual_len(ex
));
2240 if (!ext4_find_delalloc_range(inode
, lblock
, lblock
+ len
- 1))
2241 ext4_es_insert_extent(inode
, lblock
, len
, ~0,
2242 EXTENT_STATUS_HOLE
);
2243 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2244 + ext4_ext_get_actual_len(ex
)) {
2246 lblock
= le32_to_cpu(ex
->ee_block
)
2247 + ext4_ext_get_actual_len(ex
);
2249 next
= ext4_ext_next_allocated_block(path
);
2250 ext_debug("cache gap(after): [%u:%u] %u",
2251 le32_to_cpu(ex
->ee_block
),
2252 ext4_ext_get_actual_len(ex
),
2254 BUG_ON(next
== lblock
);
2255 len
= next
- lblock
;
2256 if (!ext4_find_delalloc_range(inode
, lblock
, lblock
+ len
- 1))
2257 ext4_es_insert_extent(inode
, lblock
, len
, ~0,
2258 EXTENT_STATUS_HOLE
);
2264 ext_debug(" -> %u:%lu\n", lblock
, len
);
2269 * removes index from the index block.
2271 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2272 struct ext4_ext_path
*path
, int depth
)
2277 /* free index block */
2279 path
= path
+ depth
;
2280 leaf
= ext4_idx_pblock(path
->p_idx
);
2281 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2282 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2285 err
= ext4_ext_get_access(handle
, inode
, path
);
2289 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2290 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2291 len
*= sizeof(struct ext4_extent_idx
);
2292 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2295 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2296 err
= ext4_ext_dirty(handle
, inode
, path
);
2299 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2300 trace_ext4_ext_rm_idx(inode
, leaf
);
2302 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2303 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2305 while (--depth
>= 0) {
2306 if (path
->p_idx
!= EXT_FIRST_INDEX(path
->p_hdr
))
2309 err
= ext4_ext_get_access(handle
, inode
, path
);
2312 path
->p_idx
->ei_block
= (path
+1)->p_idx
->ei_block
;
2313 err
= ext4_ext_dirty(handle
, inode
, path
);
2321 * ext4_ext_calc_credits_for_single_extent:
2322 * This routine returns max. credits that needed to insert an extent
2323 * to the extent tree.
2324 * When pass the actual path, the caller should calculate credits
2327 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2328 struct ext4_ext_path
*path
)
2331 int depth
= ext_depth(inode
);
2334 /* probably there is space in leaf? */
2335 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2336 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2339 * There are some space in the leaf tree, no
2340 * need to account for leaf block credit
2342 * bitmaps and block group descriptor blocks
2343 * and other metadata blocks still need to be
2346 /* 1 bitmap, 1 block group descriptor */
2347 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2352 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2356 * How many index/leaf blocks need to change/allocate to add @extents extents?
2358 * If we add a single extent, then in the worse case, each tree level
2359 * index/leaf need to be changed in case of the tree split.
2361 * If more extents are inserted, they could cause the whole tree split more
2362 * than once, but this is really rare.
2364 int ext4_ext_index_trans_blocks(struct inode
*inode
, int extents
)
2369 /* If we are converting the inline data, only one is needed here. */
2370 if (ext4_has_inline_data(inode
))
2373 depth
= ext_depth(inode
);
2383 static inline int get_default_free_blocks_flags(struct inode
*inode
)
2385 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2386 return EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
;
2387 else if (ext4_should_journal_data(inode
))
2388 return EXT4_FREE_BLOCKS_FORGET
;
2392 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2393 struct ext4_extent
*ex
,
2394 long long *partial_cluster
,
2395 ext4_lblk_t from
, ext4_lblk_t to
)
2397 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2398 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2400 int flags
= get_default_free_blocks_flags(inode
);
2403 * For bigalloc file systems, we never free a partial cluster
2404 * at the beginning of the extent. Instead, we make a note
2405 * that we tried freeing the cluster, and check to see if we
2406 * need to free it on a subsequent call to ext4_remove_blocks,
2407 * or at the end of the ext4_truncate() operation.
2409 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2411 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2413 * If we have a partial cluster, and it's different from the
2414 * cluster of the last block, we need to explicitly free the
2415 * partial cluster here.
2417 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2418 if ((*partial_cluster
> 0) &&
2419 (EXT4_B2C(sbi
, pblk
) != *partial_cluster
)) {
2420 ext4_free_blocks(handle
, inode
, NULL
,
2421 EXT4_C2B(sbi
, *partial_cluster
),
2422 sbi
->s_cluster_ratio
, flags
);
2423 *partial_cluster
= 0;
2426 #ifdef EXTENTS_STATS
2428 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2429 spin_lock(&sbi
->s_ext_stats_lock
);
2430 sbi
->s_ext_blocks
+= ee_len
;
2431 sbi
->s_ext_extents
++;
2432 if (ee_len
< sbi
->s_ext_min
)
2433 sbi
->s_ext_min
= ee_len
;
2434 if (ee_len
> sbi
->s_ext_max
)
2435 sbi
->s_ext_max
= ee_len
;
2436 if (ext_depth(inode
) > sbi
->s_depth_max
)
2437 sbi
->s_depth_max
= ext_depth(inode
);
2438 spin_unlock(&sbi
->s_ext_stats_lock
);
2441 if (from
>= le32_to_cpu(ex
->ee_block
)
2442 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2445 unsigned int unaligned
;
2447 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2448 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2450 * Usually we want to free partial cluster at the end of the
2451 * extent, except for the situation when the cluster is still
2452 * used by any other extent (partial_cluster is negative).
2454 if (*partial_cluster
< 0 &&
2455 -(*partial_cluster
) == EXT4_B2C(sbi
, pblk
+ num
- 1))
2456 flags
|= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER
;
2458 ext_debug("free last %u blocks starting %llu partial %lld\n",
2459 num
, pblk
, *partial_cluster
);
2460 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2462 * If the block range to be freed didn't start at the
2463 * beginning of a cluster, and we removed the entire
2464 * extent and the cluster is not used by any other extent,
2465 * save the partial cluster here, since we might need to
2466 * delete if we determine that the truncate operation has
2467 * removed all of the blocks in the cluster.
2469 * On the other hand, if we did not manage to free the whole
2470 * extent, we have to mark the cluster as used (store negative
2471 * cluster number in partial_cluster).
2473 unaligned
= pblk
& (sbi
->s_cluster_ratio
- 1);
2474 if (unaligned
&& (ee_len
== num
) &&
2475 (*partial_cluster
!= -((long long)EXT4_B2C(sbi
, pblk
))))
2476 *partial_cluster
= EXT4_B2C(sbi
, pblk
);
2478 *partial_cluster
= -((long long)EXT4_B2C(sbi
, pblk
));
2479 else if (*partial_cluster
> 0)
2480 *partial_cluster
= 0;
2482 ext4_error(sbi
->s_sb
, "strange request: removal(2) "
2483 "%u-%u from %u:%u\n",
2484 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2490 * ext4_ext_rm_leaf() Removes the extents associated with the
2491 * blocks appearing between "start" and "end", and splits the extents
2492 * if "start" and "end" appear in the same extent
2494 * @handle: The journal handle
2495 * @inode: The files inode
2496 * @path: The path to the leaf
2497 * @partial_cluster: The cluster which we'll have to free if all extents
2498 * has been released from it. It gets negative in case
2499 * that the cluster is still used.
2500 * @start: The first block to remove
2501 * @end: The last block to remove
2504 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2505 struct ext4_ext_path
*path
,
2506 long long *partial_cluster
,
2507 ext4_lblk_t start
, ext4_lblk_t end
)
2509 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2510 int err
= 0, correct_index
= 0;
2511 int depth
= ext_depth(inode
), credits
;
2512 struct ext4_extent_header
*eh
;
2515 ext4_lblk_t ex_ee_block
;
2516 unsigned short ex_ee_len
;
2517 unsigned uninitialized
= 0;
2518 struct ext4_extent
*ex
;
2521 /* the header must be checked already in ext4_ext_remove_space() */
2522 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2523 if (!path
[depth
].p_hdr
)
2524 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2525 eh
= path
[depth
].p_hdr
;
2526 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2527 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2530 /* find where to start removing */
2531 ex
= path
[depth
].p_ext
;
2533 ex
= EXT_LAST_EXTENT(eh
);
2535 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2536 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2538 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2540 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2541 ex_ee_block
+ ex_ee_len
> start
) {
2543 if (ext4_ext_is_uninitialized(ex
))
2548 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2549 uninitialized
, ex_ee_len
);
2550 path
[depth
].p_ext
= ex
;
2552 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2553 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2554 ex_ee_block
+ex_ee_len
- 1 : end
;
2556 ext_debug(" border %u:%u\n", a
, b
);
2558 /* If this extent is beyond the end of the hole, skip it */
2559 if (end
< ex_ee_block
) {
2561 * We're going to skip this extent and move to another,
2562 * so if this extent is not cluster aligned we have
2563 * to mark the current cluster as used to avoid
2564 * accidentally freeing it later on
2566 pblk
= ext4_ext_pblock(ex
);
2567 if (pblk
& (sbi
->s_cluster_ratio
- 1))
2569 -((long long)EXT4_B2C(sbi
, pblk
));
2571 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2572 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2574 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2575 EXT4_ERROR_INODE(inode
,
2576 "can not handle truncate %u:%u "
2578 start
, end
, ex_ee_block
,
2579 ex_ee_block
+ ex_ee_len
- 1);
2582 } else if (a
!= ex_ee_block
) {
2583 /* remove tail of the extent */
2584 num
= a
- ex_ee_block
;
2586 /* remove whole extent: excellent! */
2590 * 3 for leaf, sb, and inode plus 2 (bmap and group
2591 * descriptor) for each block group; assume two block
2592 * groups plus ex_ee_len/blocks_per_block_group for
2595 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2596 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2598 credits
+= (ext_depth(inode
)) + 1;
2600 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2602 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2606 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2610 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2616 /* this extent is removed; mark slot entirely unused */
2617 ext4_ext_store_pblock(ex
, 0);
2619 ex
->ee_len
= cpu_to_le16(num
);
2621 * Do not mark uninitialized if all the blocks in the
2622 * extent have been removed.
2624 if (uninitialized
&& num
)
2625 ext4_ext_mark_uninitialized(ex
);
2627 * If the extent was completely released,
2628 * we need to remove it from the leaf
2631 if (end
!= EXT_MAX_BLOCKS
- 1) {
2633 * For hole punching, we need to scoot all the
2634 * extents up when an extent is removed so that
2635 * we dont have blank extents in the middle
2637 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2638 sizeof(struct ext4_extent
));
2640 /* Now get rid of the one at the end */
2641 memset(EXT_LAST_EXTENT(eh
), 0,
2642 sizeof(struct ext4_extent
));
2644 le16_add_cpu(&eh
->eh_entries
, -1);
2645 } else if (*partial_cluster
> 0)
2646 *partial_cluster
= 0;
2648 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2652 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2653 ext4_ext_pblock(ex
));
2655 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2656 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2659 if (correct_index
&& eh
->eh_entries
)
2660 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2663 * Free the partial cluster only if the current extent does not
2664 * reference it. Otherwise we might free used cluster.
2666 if (*partial_cluster
> 0 &&
2667 (EXT4_B2C(sbi
, ext4_ext_pblock(ex
) + ex_ee_len
- 1) !=
2668 *partial_cluster
)) {
2669 int flags
= get_default_free_blocks_flags(inode
);
2671 ext4_free_blocks(handle
, inode
, NULL
,
2672 EXT4_C2B(sbi
, *partial_cluster
),
2673 sbi
->s_cluster_ratio
, flags
);
2674 *partial_cluster
= 0;
2677 /* if this leaf is free, then we should
2678 * remove it from index block above */
2679 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2680 err
= ext4_ext_rm_idx(handle
, inode
, path
, depth
);
2687 * ext4_ext_more_to_rm:
2688 * returns 1 if current index has to be freed (even partial)
2691 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2693 BUG_ON(path
->p_idx
== NULL
);
2695 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2699 * if truncate on deeper level happened, it wasn't partial,
2700 * so we have to consider current index for truncation
2702 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2707 int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2710 struct super_block
*sb
= inode
->i_sb
;
2711 int depth
= ext_depth(inode
);
2712 struct ext4_ext_path
*path
= NULL
;
2713 long long partial_cluster
= 0;
2717 ext_debug("truncate since %u to %u\n", start
, end
);
2719 /* probably first extent we're gonna free will be last in block */
2720 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, depth
+ 1);
2722 return PTR_ERR(handle
);
2725 trace_ext4_ext_remove_space(inode
, start
, end
, depth
);
2728 * Check if we are removing extents inside the extent tree. If that
2729 * is the case, we are going to punch a hole inside the extent tree
2730 * so we have to check whether we need to split the extent covering
2731 * the last block to remove so we can easily remove the part of it
2732 * in ext4_ext_rm_leaf().
2734 if (end
< EXT_MAX_BLOCKS
- 1) {
2735 struct ext4_extent
*ex
;
2736 ext4_lblk_t ee_block
;
2738 /* find extent for this block */
2739 path
= ext4_ext_find_extent(inode
, end
, NULL
, EXT4_EX_NOCACHE
);
2741 ext4_journal_stop(handle
);
2742 return PTR_ERR(path
);
2744 depth
= ext_depth(inode
);
2745 /* Leaf not may not exist only if inode has no blocks at all */
2746 ex
= path
[depth
].p_ext
;
2749 EXT4_ERROR_INODE(inode
,
2750 "path[%d].p_hdr == NULL",
2757 ee_block
= le32_to_cpu(ex
->ee_block
);
2760 * See if the last block is inside the extent, if so split
2761 * the extent at 'end' block so we can easily remove the
2762 * tail of the first part of the split extent in
2763 * ext4_ext_rm_leaf().
2765 if (end
>= ee_block
&&
2766 end
< ee_block
+ ext4_ext_get_actual_len(ex
) - 1) {
2769 if (ext4_ext_is_uninitialized(ex
))
2770 split_flag
= EXT4_EXT_MARK_UNINIT1
|
2771 EXT4_EXT_MARK_UNINIT2
;
2774 * Split the extent in two so that 'end' is the last
2775 * block in the first new extent. Also we should not
2776 * fail removing space due to ENOSPC so try to use
2777 * reserved block if that happens.
2779 err
= ext4_split_extent_at(handle
, inode
, path
,
2780 end
+ 1, split_flag
,
2782 EXT4_GET_BLOCKS_PRE_IO
|
2783 EXT4_GET_BLOCKS_METADATA_NOFAIL
);
2790 * We start scanning from right side, freeing all the blocks
2791 * after i_size and walking into the tree depth-wise.
2793 depth
= ext_depth(inode
);
2798 le16_to_cpu(path
[k
].p_hdr
->eh_entries
)+1;
2800 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
2803 ext4_journal_stop(handle
);
2806 path
[0].p_depth
= depth
;
2807 path
[0].p_hdr
= ext_inode_hdr(inode
);
2810 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0)) {
2817 while (i
>= 0 && err
== 0) {
2819 /* this is leaf block */
2820 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2821 &partial_cluster
, start
,
2823 /* root level has p_bh == NULL, brelse() eats this */
2824 brelse(path
[i
].p_bh
);
2825 path
[i
].p_bh
= NULL
;
2830 /* this is index block */
2831 if (!path
[i
].p_hdr
) {
2832 ext_debug("initialize header\n");
2833 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2836 if (!path
[i
].p_idx
) {
2837 /* this level hasn't been touched yet */
2838 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2839 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2840 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2842 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2844 /* we were already here, see at next index */
2848 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2849 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2851 if (ext4_ext_more_to_rm(path
+ i
)) {
2852 struct buffer_head
*bh
;
2853 /* go to the next level */
2854 ext_debug("move to level %d (block %llu)\n",
2855 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
2856 memset(path
+ i
+ 1, 0, sizeof(*path
));
2857 bh
= read_extent_tree_block(inode
,
2858 ext4_idx_pblock(path
[i
].p_idx
), depth
- i
- 1,
2861 /* should we reset i_size? */
2865 /* Yield here to deal with large extent trees.
2866 * Should be a no-op if we did IO above. */
2868 if (WARN_ON(i
+ 1 > depth
)) {
2872 path
[i
+ 1].p_bh
= bh
;
2874 /* save actual number of indexes since this
2875 * number is changed at the next iteration */
2876 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2879 /* we finished processing this index, go up */
2880 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2881 /* index is empty, remove it;
2882 * handle must be already prepared by the
2883 * truncatei_leaf() */
2884 err
= ext4_ext_rm_idx(handle
, inode
, path
, i
);
2886 /* root level has p_bh == NULL, brelse() eats this */
2887 brelse(path
[i
].p_bh
);
2888 path
[i
].p_bh
= NULL
;
2890 ext_debug("return to level %d\n", i
);
2894 trace_ext4_ext_remove_space_done(inode
, start
, end
, depth
,
2895 partial_cluster
, path
->p_hdr
->eh_entries
);
2897 /* If we still have something in the partial cluster and we have removed
2898 * even the first extent, then we should free the blocks in the partial
2899 * cluster as well. */
2900 if (partial_cluster
> 0 && path
->p_hdr
->eh_entries
== 0) {
2901 int flags
= get_default_free_blocks_flags(inode
);
2903 ext4_free_blocks(handle
, inode
, NULL
,
2904 EXT4_C2B(EXT4_SB(sb
), partial_cluster
),
2905 EXT4_SB(sb
)->s_cluster_ratio
, flags
);
2906 partial_cluster
= 0;
2909 /* TODO: flexible tree reduction should be here */
2910 if (path
->p_hdr
->eh_entries
== 0) {
2912 * truncate to zero freed all the tree,
2913 * so we need to correct eh_depth
2915 err
= ext4_ext_get_access(handle
, inode
, path
);
2917 ext_inode_hdr(inode
)->eh_depth
= 0;
2918 ext_inode_hdr(inode
)->eh_max
=
2919 cpu_to_le16(ext4_ext_space_root(inode
, 0));
2920 err
= ext4_ext_dirty(handle
, inode
, path
);
2924 ext4_ext_drop_refs(path
);
2926 if (err
== -EAGAIN
) {
2930 ext4_journal_stop(handle
);
2936 * called at mount time
2938 void ext4_ext_init(struct super_block
*sb
)
2941 * possible initialization would be here
2944 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
2945 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2946 printk(KERN_INFO
"EXT4-fs: file extents enabled"
2947 #ifdef AGGRESSIVE_TEST
2948 ", aggressive tests"
2950 #ifdef CHECK_BINSEARCH
2953 #ifdef EXTENTS_STATS
2958 #ifdef EXTENTS_STATS
2959 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
2960 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
2961 EXT4_SB(sb
)->s_ext_max
= 0;
2967 * called at umount time
2969 void ext4_ext_release(struct super_block
*sb
)
2971 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
2974 #ifdef EXTENTS_STATS
2975 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
2976 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2977 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2978 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
2979 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
2980 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2981 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
2986 /* FIXME!! we need to try to merge to left or right after zero-out */
2987 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
2989 ext4_fsblk_t ee_pblock
;
2990 unsigned int ee_len
;
2993 ee_len
= ext4_ext_get_actual_len(ex
);
2994 ee_pblock
= ext4_ext_pblock(ex
);
2996 ret
= sb_issue_zeroout(inode
->i_sb
, ee_pblock
, ee_len
, GFP_NOFS
);
3004 * ext4_split_extent_at() splits an extent at given block.
3006 * @handle: the journal handle
3007 * @inode: the file inode
3008 * @path: the path to the extent
3009 * @split: the logical block where the extent is splitted.
3010 * @split_flags: indicates if the extent could be zeroout if split fails, and
3011 * the states(init or uninit) of new extents.
3012 * @flags: flags used to insert new extent to extent tree.
3015 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3016 * of which are deterimined by split_flag.
3018 * There are two cases:
3019 * a> the extent are splitted into two extent.
3020 * b> split is not needed, and just mark the extent.
3022 * return 0 on success.
3024 static int ext4_split_extent_at(handle_t
*handle
,
3025 struct inode
*inode
,
3026 struct ext4_ext_path
*path
,
3031 ext4_fsblk_t newblock
;
3032 ext4_lblk_t ee_block
;
3033 struct ext4_extent
*ex
, newex
, orig_ex
, zero_ex
;
3034 struct ext4_extent
*ex2
= NULL
;
3035 unsigned int ee_len
, depth
;
3038 BUG_ON((split_flag
& (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
)) ==
3039 (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
));
3041 ext_debug("ext4_split_extents_at: inode %lu, logical"
3042 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
3044 ext4_ext_show_leaf(inode
, path
);
3046 depth
= ext_depth(inode
);
3047 ex
= path
[depth
].p_ext
;
3048 ee_block
= le32_to_cpu(ex
->ee_block
);
3049 ee_len
= ext4_ext_get_actual_len(ex
);
3050 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
3052 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
3053 BUG_ON(!ext4_ext_is_uninitialized(ex
) &&
3054 split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3055 EXT4_EXT_MARK_UNINIT1
|
3056 EXT4_EXT_MARK_UNINIT2
));
3058 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3062 if (split
== ee_block
) {
3064 * case b: block @split is the block that the extent begins with
3065 * then we just change the state of the extent, and splitting
3068 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3069 ext4_ext_mark_uninitialized(ex
);
3071 ext4_ext_mark_initialized(ex
);
3073 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
3074 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3076 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3081 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
3082 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
3083 if (split_flag
& EXT4_EXT_MARK_UNINIT1
)
3084 ext4_ext_mark_uninitialized(ex
);
3087 * path may lead to new leaf, not to original leaf any more
3088 * after ext4_ext_insert_extent() returns,
3090 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3092 goto fix_extent_len
;
3095 ex2
->ee_block
= cpu_to_le32(split
);
3096 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
3097 ext4_ext_store_pblock(ex2
, newblock
);
3098 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3099 ext4_ext_mark_uninitialized(ex2
);
3101 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
3102 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3103 if (split_flag
& (EXT4_EXT_DATA_VALID1
|EXT4_EXT_DATA_VALID2
)) {
3104 if (split_flag
& EXT4_EXT_DATA_VALID1
) {
3105 err
= ext4_ext_zeroout(inode
, ex2
);
3106 zero_ex
.ee_block
= ex2
->ee_block
;
3107 zero_ex
.ee_len
= cpu_to_le16(
3108 ext4_ext_get_actual_len(ex2
));
3109 ext4_ext_store_pblock(&zero_ex
,
3110 ext4_ext_pblock(ex2
));
3112 err
= ext4_ext_zeroout(inode
, ex
);
3113 zero_ex
.ee_block
= ex
->ee_block
;
3114 zero_ex
.ee_len
= cpu_to_le16(
3115 ext4_ext_get_actual_len(ex
));
3116 ext4_ext_store_pblock(&zero_ex
,
3117 ext4_ext_pblock(ex
));
3120 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3121 zero_ex
.ee_block
= orig_ex
.ee_block
;
3122 zero_ex
.ee_len
= cpu_to_le16(
3123 ext4_ext_get_actual_len(&orig_ex
));
3124 ext4_ext_store_pblock(&zero_ex
,
3125 ext4_ext_pblock(&orig_ex
));
3129 goto fix_extent_len
;
3130 /* update the extent length and mark as initialized */
3131 ex
->ee_len
= cpu_to_le16(ee_len
);
3132 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3133 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3135 goto fix_extent_len
;
3137 /* update extent status tree */
3138 err
= ext4_es_zeroout(inode
, &zero_ex
);
3142 goto fix_extent_len
;
3145 ext4_ext_show_leaf(inode
, path
);
3149 ex
->ee_len
= orig_ex
.ee_len
;
3150 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3155 * ext4_split_extents() splits an extent and mark extent which is covered
3156 * by @map as split_flags indicates
3158 * It may result in splitting the extent into multiple extents (upto three)
3159 * There are three possibilities:
3160 * a> There is no split required
3161 * b> Splits in two extents: Split is happening at either end of the extent
3162 * c> Splits in three extents: Somone is splitting in middle of the extent
3165 static int ext4_split_extent(handle_t
*handle
,
3166 struct inode
*inode
,
3167 struct ext4_ext_path
*path
,
3168 struct ext4_map_blocks
*map
,
3172 ext4_lblk_t ee_block
;
3173 struct ext4_extent
*ex
;
3174 unsigned int ee_len
, depth
;
3177 int split_flag1
, flags1
;
3178 int allocated
= map
->m_len
;
3180 depth
= ext_depth(inode
);
3181 ex
= path
[depth
].p_ext
;
3182 ee_block
= le32_to_cpu(ex
->ee_block
);
3183 ee_len
= ext4_ext_get_actual_len(ex
);
3184 uninitialized
= ext4_ext_is_uninitialized(ex
);
3186 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
3187 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
;
3188 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
3190 split_flag1
|= EXT4_EXT_MARK_UNINIT1
|
3191 EXT4_EXT_MARK_UNINIT2
;
3192 if (split_flag
& EXT4_EXT_DATA_VALID2
)
3193 split_flag1
|= EXT4_EXT_DATA_VALID1
;
3194 err
= ext4_split_extent_at(handle
, inode
, path
,
3195 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
3199 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3202 * Update path is required because previous ext4_split_extent_at() may
3203 * result in split of original leaf or extent zeroout.
3205 ext4_ext_drop_refs(path
);
3206 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
, 0);
3208 return PTR_ERR(path
);
3209 depth
= ext_depth(inode
);
3210 ex
= path
[depth
].p_ext
;
3211 uninitialized
= ext4_ext_is_uninitialized(ex
);
3214 if (map
->m_lblk
>= ee_block
) {
3215 split_flag1
= split_flag
& EXT4_EXT_DATA_VALID2
;
3216 if (uninitialized
) {
3217 split_flag1
|= EXT4_EXT_MARK_UNINIT1
;
3218 split_flag1
|= split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3219 EXT4_EXT_MARK_UNINIT2
);
3221 err
= ext4_split_extent_at(handle
, inode
, path
,
3222 map
->m_lblk
, split_flag1
, flags
);
3227 ext4_ext_show_leaf(inode
, path
);
3229 return err
? err
: allocated
;
3233 * This function is called by ext4_ext_map_blocks() if someone tries to write
3234 * to an uninitialized extent. It may result in splitting the uninitialized
3235 * extent into multiple extents (up to three - one initialized and two
3237 * There are three possibilities:
3238 * a> There is no split required: Entire extent should be initialized
3239 * b> Splits in two extents: Write is happening at either end of the extent
3240 * c> Splits in three extents: Somone is writing in middle of the extent
3243 * - The extent pointed to by 'path' is uninitialized.
3244 * - The extent pointed to by 'path' contains a superset
3245 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3247 * Post-conditions on success:
3248 * - the returned value is the number of blocks beyond map->l_lblk
3249 * that are allocated and initialized.
3250 * It is guaranteed to be >= map->m_len.
3252 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3253 struct inode
*inode
,
3254 struct ext4_map_blocks
*map
,
3255 struct ext4_ext_path
*path
,
3258 struct ext4_sb_info
*sbi
;
3259 struct ext4_extent_header
*eh
;
3260 struct ext4_map_blocks split_map
;
3261 struct ext4_extent zero_ex
;
3262 struct ext4_extent
*ex
, *abut_ex
;
3263 ext4_lblk_t ee_block
, eof_block
;
3264 unsigned int ee_len
, depth
, map_len
= map
->m_len
;
3265 int allocated
= 0, max_zeroout
= 0;
3269 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3270 "block %llu, max_blocks %u\n", inode
->i_ino
,
3271 (unsigned long long)map
->m_lblk
, map_len
);
3273 sbi
= EXT4_SB(inode
->i_sb
);
3274 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3275 inode
->i_sb
->s_blocksize_bits
;
3276 if (eof_block
< map
->m_lblk
+ map_len
)
3277 eof_block
= map
->m_lblk
+ map_len
;
3279 depth
= ext_depth(inode
);
3280 eh
= path
[depth
].p_hdr
;
3281 ex
= path
[depth
].p_ext
;
3282 ee_block
= le32_to_cpu(ex
->ee_block
);
3283 ee_len
= ext4_ext_get_actual_len(ex
);
3286 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3288 /* Pre-conditions */
3289 BUG_ON(!ext4_ext_is_uninitialized(ex
));
3290 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3293 * Attempt to transfer newly initialized blocks from the currently
3294 * uninitialized extent to its neighbor. This is much cheaper
3295 * than an insertion followed by a merge as those involve costly
3296 * memmove() calls. Transferring to the left is the common case in
3297 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3298 * followed by append writes.
3300 * Limitations of the current logic:
3301 * - L1: we do not deal with writes covering the whole extent.
3302 * This would require removing the extent if the transfer
3304 * - L2: we only attempt to merge with an extent stored in the
3305 * same extent tree node.
3307 if ((map
->m_lblk
== ee_block
) &&
3308 /* See if we can merge left */
3309 (map_len
< ee_len
) && /*L1*/
3310 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L2*/
3311 ext4_lblk_t prev_lblk
;
3312 ext4_fsblk_t prev_pblk
, ee_pblk
;
3313 unsigned int prev_len
;
3316 prev_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3317 prev_len
= ext4_ext_get_actual_len(abut_ex
);
3318 prev_pblk
= ext4_ext_pblock(abut_ex
);
3319 ee_pblk
= ext4_ext_pblock(ex
);
3322 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3323 * upon those conditions:
3324 * - C1: abut_ex is initialized,
3325 * - C2: abut_ex is logically abutting ex,
3326 * - C3: abut_ex is physically abutting ex,
3327 * - C4: abut_ex can receive the additional blocks without
3328 * overflowing the (initialized) length limit.
3330 if ((!ext4_ext_is_uninitialized(abut_ex
)) && /*C1*/
3331 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3332 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3333 (prev_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3334 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3338 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3341 /* Shift the start of ex by 'map_len' blocks */
3342 ex
->ee_block
= cpu_to_le32(ee_block
+ map_len
);
3343 ext4_ext_store_pblock(ex
, ee_pblk
+ map_len
);
3344 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3345 ext4_ext_mark_uninitialized(ex
); /* Restore the flag */
3347 /* Extend abut_ex by 'map_len' blocks */
3348 abut_ex
->ee_len
= cpu_to_le16(prev_len
+ map_len
);
3350 /* Result: number of initialized blocks past m_lblk */
3351 allocated
= map_len
;
3353 } else if (((map
->m_lblk
+ map_len
) == (ee_block
+ ee_len
)) &&
3354 (map_len
< ee_len
) && /*L1*/
3355 ex
< EXT_LAST_EXTENT(eh
)) { /*L2*/
3356 /* See if we can merge right */
3357 ext4_lblk_t next_lblk
;
3358 ext4_fsblk_t next_pblk
, ee_pblk
;
3359 unsigned int next_len
;
3362 next_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3363 next_len
= ext4_ext_get_actual_len(abut_ex
);
3364 next_pblk
= ext4_ext_pblock(abut_ex
);
3365 ee_pblk
= ext4_ext_pblock(ex
);
3368 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3369 * upon those conditions:
3370 * - C1: abut_ex is initialized,
3371 * - C2: abut_ex is logically abutting ex,
3372 * - C3: abut_ex is physically abutting ex,
3373 * - C4: abut_ex can receive the additional blocks without
3374 * overflowing the (initialized) length limit.
3376 if ((!ext4_ext_is_uninitialized(abut_ex
)) && /*C1*/
3377 ((map
->m_lblk
+ map_len
) == next_lblk
) && /*C2*/
3378 ((ee_pblk
+ ee_len
) == next_pblk
) && /*C3*/
3379 (next_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3380 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3384 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3387 /* Shift the start of abut_ex by 'map_len' blocks */
3388 abut_ex
->ee_block
= cpu_to_le32(next_lblk
- map_len
);
3389 ext4_ext_store_pblock(abut_ex
, next_pblk
- map_len
);
3390 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3391 ext4_ext_mark_uninitialized(ex
); /* Restore the flag */
3393 /* Extend abut_ex by 'map_len' blocks */
3394 abut_ex
->ee_len
= cpu_to_le16(next_len
+ map_len
);
3396 /* Result: number of initialized blocks past m_lblk */
3397 allocated
= map_len
;
3401 /* Mark the block containing both extents as dirty */
3402 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3404 /* Update path to point to the right extent */
3405 path
[depth
].p_ext
= abut_ex
;
3408 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3410 WARN_ON(map
->m_lblk
< ee_block
);
3412 * It is safe to convert extent to initialized via explicit
3413 * zeroout only if extent is fully insde i_size or new_size.
3415 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3417 if (EXT4_EXT_MAY_ZEROOUT
& split_flag
)
3418 max_zeroout
= sbi
->s_extent_max_zeroout_kb
>>
3419 (inode
->i_sb
->s_blocksize_bits
- 10);
3421 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3422 if (max_zeroout
&& (ee_len
<= max_zeroout
)) {
3423 err
= ext4_ext_zeroout(inode
, ex
);
3426 zero_ex
.ee_block
= ex
->ee_block
;
3427 zero_ex
.ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
));
3428 ext4_ext_store_pblock(&zero_ex
, ext4_ext_pblock(ex
));
3430 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3433 ext4_ext_mark_initialized(ex
);
3434 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3435 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3441 * 1. split the extent into three extents.
3442 * 2. split the extent into two extents, zeroout the first half.
3443 * 3. split the extent into two extents, zeroout the second half.
3444 * 4. split the extent into two extents with out zeroout.
3446 split_map
.m_lblk
= map
->m_lblk
;
3447 split_map
.m_len
= map
->m_len
;
3449 if (max_zeroout
&& (allocated
> map
->m_len
)) {
3450 if (allocated
<= max_zeroout
) {
3453 cpu_to_le32(map
->m_lblk
);
3454 zero_ex
.ee_len
= cpu_to_le16(allocated
);
3455 ext4_ext_store_pblock(&zero_ex
,
3456 ext4_ext_pblock(ex
) + map
->m_lblk
- ee_block
);
3457 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3460 split_map
.m_lblk
= map
->m_lblk
;
3461 split_map
.m_len
= allocated
;
3462 } else if (map
->m_lblk
- ee_block
+ map
->m_len
< max_zeroout
) {
3464 if (map
->m_lblk
!= ee_block
) {
3465 zero_ex
.ee_block
= ex
->ee_block
;
3466 zero_ex
.ee_len
= cpu_to_le16(map
->m_lblk
-
3468 ext4_ext_store_pblock(&zero_ex
,
3469 ext4_ext_pblock(ex
));
3470 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3475 split_map
.m_lblk
= ee_block
;
3476 split_map
.m_len
= map
->m_lblk
- ee_block
+ map
->m_len
;
3477 allocated
= map
->m_len
;
3481 allocated
= ext4_split_extent(handle
, inode
, path
,
3482 &split_map
, split_flag
, flags
);
3487 /* If we have gotten a failure, don't zero out status tree */
3489 err
= ext4_es_zeroout(inode
, &zero_ex
);
3490 return err
? err
: allocated
;
3494 * This function is called by ext4_ext_map_blocks() from
3495 * ext4_get_blocks_dio_write() when DIO to write
3496 * to an uninitialized extent.
3498 * Writing to an uninitialized extent may result in splitting the uninitialized
3499 * extent into multiple initialized/uninitialized extents (up to three)
3500 * There are three possibilities:
3501 * a> There is no split required: Entire extent should be uninitialized
3502 * b> Splits in two extents: Write is happening at either end of the extent
3503 * c> Splits in three extents: Somone is writing in middle of the extent
3505 * One of more index blocks maybe needed if the extent tree grow after
3506 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3507 * complete, we need to split the uninitialized extent before DIO submit
3508 * the IO. The uninitialized extent called at this time will be split
3509 * into three uninitialized extent(at most). After IO complete, the part
3510 * being filled will be convert to initialized by the end_io callback function
3511 * via ext4_convert_unwritten_extents().
3513 * Returns the size of uninitialized extent to be written on success.
3515 static int ext4_split_unwritten_extents(handle_t
*handle
,
3516 struct inode
*inode
,
3517 struct ext4_map_blocks
*map
,
3518 struct ext4_ext_path
*path
,
3521 ext4_lblk_t eof_block
;
3522 ext4_lblk_t ee_block
;
3523 struct ext4_extent
*ex
;
3524 unsigned int ee_len
;
3525 int split_flag
= 0, depth
;
3527 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3528 "block %llu, max_blocks %u\n", inode
->i_ino
,
3529 (unsigned long long)map
->m_lblk
, map
->m_len
);
3531 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3532 inode
->i_sb
->s_blocksize_bits
;
3533 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3534 eof_block
= map
->m_lblk
+ map
->m_len
;
3536 * It is safe to convert extent to initialized via explicit
3537 * zeroout only if extent is fully insde i_size or new_size.
3539 depth
= ext_depth(inode
);
3540 ex
= path
[depth
].p_ext
;
3541 ee_block
= le32_to_cpu(ex
->ee_block
);
3542 ee_len
= ext4_ext_get_actual_len(ex
);
3544 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3545 split_flag
|= EXT4_EXT_MARK_UNINIT2
;
3546 if (flags
& EXT4_GET_BLOCKS_CONVERT
)
3547 split_flag
|= EXT4_EXT_DATA_VALID2
;
3548 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3549 return ext4_split_extent(handle
, inode
, path
, map
, split_flag
, flags
);
3552 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3553 struct inode
*inode
,
3554 struct ext4_map_blocks
*map
,
3555 struct ext4_ext_path
*path
)
3557 struct ext4_extent
*ex
;
3558 ext4_lblk_t ee_block
;
3559 unsigned int ee_len
;
3563 depth
= ext_depth(inode
);
3564 ex
= path
[depth
].p_ext
;
3565 ee_block
= le32_to_cpu(ex
->ee_block
);
3566 ee_len
= ext4_ext_get_actual_len(ex
);
3568 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3569 "block %llu, max_blocks %u\n", inode
->i_ino
,
3570 (unsigned long long)ee_block
, ee_len
);
3572 /* If extent is larger than requested it is a clear sign that we still
3573 * have some extent state machine issues left. So extent_split is still
3575 * TODO: Once all related issues will be fixed this situation should be
3578 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3580 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3581 " len %u; IO logical block %llu, len %u\n",
3582 inode
->i_ino
, (unsigned long long)ee_block
, ee_len
,
3583 (unsigned long long)map
->m_lblk
, map
->m_len
);
3585 err
= ext4_split_unwritten_extents(handle
, inode
, map
, path
,
3586 EXT4_GET_BLOCKS_CONVERT
);
3589 ext4_ext_drop_refs(path
);
3590 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
, 0);
3592 err
= PTR_ERR(path
);
3595 depth
= ext_depth(inode
);
3596 ex
= path
[depth
].p_ext
;
3599 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3602 /* first mark the extent as initialized */
3603 ext4_ext_mark_initialized(ex
);
3605 /* note: ext4_ext_correct_indexes() isn't needed here because
3606 * borders are not changed
3608 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3610 /* Mark modified extent as dirty */
3611 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3613 ext4_ext_show_leaf(inode
, path
);
3617 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3618 sector_t block
, int count
)
3621 for (i
= 0; i
< count
; i
++)
3622 unmap_underlying_metadata(bdev
, block
+ i
);
3626 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3628 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3630 struct ext4_ext_path
*path
,
3634 struct ext4_extent_header
*eh
;
3635 struct ext4_extent
*last_ex
;
3637 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3640 depth
= ext_depth(inode
);
3641 eh
= path
[depth
].p_hdr
;
3644 * We're going to remove EOFBLOCKS_FL entirely in future so we
3645 * do not care for this case anymore. Simply remove the flag
3646 * if there are no extents.
3648 if (unlikely(!eh
->eh_entries
))
3650 last_ex
= EXT_LAST_EXTENT(eh
);
3652 * We should clear the EOFBLOCKS_FL flag if we are writing the
3653 * last block in the last extent in the file. We test this by
3654 * first checking to see if the caller to
3655 * ext4_ext_get_blocks() was interested in the last block (or
3656 * a block beyond the last block) in the current extent. If
3657 * this turns out to be false, we can bail out from this
3658 * function immediately.
3660 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3661 ext4_ext_get_actual_len(last_ex
))
3664 * If the caller does appear to be planning to write at or
3665 * beyond the end of the current extent, we then test to see
3666 * if the current extent is the last extent in the file, by
3667 * checking to make sure it was reached via the rightmost node
3668 * at each level of the tree.
3670 for (i
= depth
-1; i
>= 0; i
--)
3671 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3674 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3675 return ext4_mark_inode_dirty(handle
, inode
);
3679 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3681 * Return 1 if there is a delalloc block in the range, otherwise 0.
3683 int ext4_find_delalloc_range(struct inode
*inode
,
3684 ext4_lblk_t lblk_start
,
3685 ext4_lblk_t lblk_end
)
3687 struct extent_status es
;
3689 ext4_es_find_delayed_extent_range(inode
, lblk_start
, lblk_end
, &es
);
3691 return 0; /* there is no delay extent in this tree */
3692 else if (es
.es_lblk
<= lblk_start
&&
3693 lblk_start
< es
.es_lblk
+ es
.es_len
)
3695 else if (lblk_start
<= es
.es_lblk
&& es
.es_lblk
<= lblk_end
)
3701 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
)
3703 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3704 ext4_lblk_t lblk_start
, lblk_end
;
3705 lblk_start
= lblk
& (~(sbi
->s_cluster_ratio
- 1));
3706 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3708 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
);
3712 * Determines how many complete clusters (out of those specified by the 'map')
3713 * are under delalloc and were reserved quota for.
3714 * This function is called when we are writing out the blocks that were
3715 * originally written with their allocation delayed, but then the space was
3716 * allocated using fallocate() before the delayed allocation could be resolved.
3717 * The cases to look for are:
3718 * ('=' indicated delayed allocated blocks
3719 * '-' indicates non-delayed allocated blocks)
3720 * (a) partial clusters towards beginning and/or end outside of allocated range
3721 * are not delalloc'ed.
3723 * |----c---=|====c====|====c====|===-c----|
3724 * |++++++ allocated ++++++|
3725 * ==> 4 complete clusters in above example
3727 * (b) partial cluster (outside of allocated range) towards either end is
3728 * marked for delayed allocation. In this case, we will exclude that
3731 * |----====c========|========c========|
3732 * |++++++ allocated ++++++|
3733 * ==> 1 complete clusters in above example
3736 * |================c================|
3737 * |++++++ allocated ++++++|
3738 * ==> 0 complete clusters in above example
3740 * The ext4_da_update_reserve_space will be called only if we
3741 * determine here that there were some "entire" clusters that span
3742 * this 'allocated' range.
3743 * In the non-bigalloc case, this function will just end up returning num_blks
3744 * without ever calling ext4_find_delalloc_range.
3747 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3748 unsigned int num_blks
)
3750 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3751 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3752 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3753 unsigned int allocated_clusters
= 0;
3755 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3756 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3758 /* max possible clusters for this allocation */
3759 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3761 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3763 /* Check towards left side */
3764 c_offset
= lblk_start
& (sbi
->s_cluster_ratio
- 1);
3766 lblk_from
= lblk_start
& (~(sbi
->s_cluster_ratio
- 1));
3767 lblk_to
= lblk_from
+ c_offset
- 1;
3769 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3770 allocated_clusters
--;
3773 /* Now check towards right. */
3774 c_offset
= (lblk_start
+ num_blks
) & (sbi
->s_cluster_ratio
- 1);
3775 if (allocated_clusters
&& c_offset
) {
3776 lblk_from
= lblk_start
+ num_blks
;
3777 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3779 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3780 allocated_clusters
--;
3783 return allocated_clusters
;
3787 ext4_ext_handle_uninitialized_extents(handle_t
*handle
, struct inode
*inode
,
3788 struct ext4_map_blocks
*map
,
3789 struct ext4_ext_path
*path
, int flags
,
3790 unsigned int allocated
, ext4_fsblk_t newblock
)
3794 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
3796 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3797 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3798 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
3800 ext4_ext_show_leaf(inode
, path
);
3803 * When writing into uninitialized space, we should not fail to
3804 * allocate metadata blocks for the new extent block if needed.
3806 flags
|= EXT4_GET_BLOCKS_METADATA_NOFAIL
;
3808 trace_ext4_ext_handle_uninitialized_extents(inode
, map
, flags
,
3809 allocated
, newblock
);
3811 /* get_block() before submit the IO, split the extent */
3812 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
3813 ret
= ext4_split_unwritten_extents(handle
, inode
, map
,
3818 * Flag the inode(non aio case) or end_io struct (aio case)
3819 * that this IO needs to conversion to written when IO is
3823 ext4_set_io_unwritten_flag(inode
, io
);
3825 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3826 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3827 if (ext4_should_dioread_nolock(inode
))
3828 map
->m_flags
|= EXT4_MAP_UNINIT
;
3831 /* IO end_io complete, convert the filled extent to written */
3832 if ((flags
& EXT4_GET_BLOCKS_CONVERT
)) {
3833 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
, map
,
3836 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3837 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
3841 map
->m_flags
|= EXT4_MAP_MAPPED
;
3842 if (allocated
> map
->m_len
)
3843 allocated
= map
->m_len
;
3844 map
->m_len
= allocated
;
3847 /* buffered IO case */
3849 * repeat fallocate creation request
3850 * we already have an unwritten extent
3852 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) {
3853 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3857 /* buffered READ or buffered write_begin() lookup */
3858 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3860 * We have blocks reserved already. We
3861 * return allocated blocks so that delalloc
3862 * won't do block reservation for us. But
3863 * the buffer head will be unmapped so that
3864 * a read from the block returns 0s.
3866 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3870 /* buffered write, writepage time, convert*/
3871 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, path
, flags
);
3873 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3880 map
->m_flags
|= EXT4_MAP_NEW
;
3882 * if we allocated more blocks than requested
3883 * we need to make sure we unmap the extra block
3884 * allocated. The actual needed block will get
3885 * unmapped later when we find the buffer_head marked
3888 if (allocated
> map
->m_len
) {
3889 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
3890 newblock
+ map
->m_len
,
3891 allocated
- map
->m_len
);
3892 allocated
= map
->m_len
;
3894 map
->m_len
= allocated
;
3897 * If we have done fallocate with the offset that is already
3898 * delayed allocated, we would have block reservation
3899 * and quota reservation done in the delayed write path.
3900 * But fallocate would have already updated quota and block
3901 * count for this offset. So cancel these reservation
3903 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
3904 unsigned int reserved_clusters
;
3905 reserved_clusters
= get_reserved_cluster_alloc(inode
,
3906 map
->m_lblk
, map
->m_len
);
3907 if (reserved_clusters
)
3908 ext4_da_update_reserve_space(inode
,
3914 map
->m_flags
|= EXT4_MAP_MAPPED
;
3915 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
3916 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
3922 if (allocated
> map
->m_len
)
3923 allocated
= map
->m_len
;
3924 ext4_ext_show_leaf(inode
, path
);
3925 map
->m_pblk
= newblock
;
3926 map
->m_len
= allocated
;
3929 ext4_ext_drop_refs(path
);
3932 return err
? err
: allocated
;
3936 * get_implied_cluster_alloc - check to see if the requested
3937 * allocation (in the map structure) overlaps with a cluster already
3938 * allocated in an extent.
3939 * @sb The filesystem superblock structure
3940 * @map The requested lblk->pblk mapping
3941 * @ex The extent structure which might contain an implied
3942 * cluster allocation
3944 * This function is called by ext4_ext_map_blocks() after we failed to
3945 * find blocks that were already in the inode's extent tree. Hence,
3946 * we know that the beginning of the requested region cannot overlap
3947 * the extent from the inode's extent tree. There are three cases we
3948 * want to catch. The first is this case:
3950 * |--- cluster # N--|
3951 * |--- extent ---| |---- requested region ---|
3954 * The second case that we need to test for is this one:
3956 * |--------- cluster # N ----------------|
3957 * |--- requested region --| |------- extent ----|
3958 * |=======================|
3960 * The third case is when the requested region lies between two extents
3961 * within the same cluster:
3962 * |------------- cluster # N-------------|
3963 * |----- ex -----| |---- ex_right ----|
3964 * |------ requested region ------|
3965 * |================|
3967 * In each of the above cases, we need to set the map->m_pblk and
3968 * map->m_len so it corresponds to the return the extent labelled as
3969 * "|====|" from cluster #N, since it is already in use for data in
3970 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
3971 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
3972 * as a new "allocated" block region. Otherwise, we will return 0 and
3973 * ext4_ext_map_blocks() will then allocate one or more new clusters
3974 * by calling ext4_mb_new_blocks().
3976 static int get_implied_cluster_alloc(struct super_block
*sb
,
3977 struct ext4_map_blocks
*map
,
3978 struct ext4_extent
*ex
,
3979 struct ext4_ext_path
*path
)
3981 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3982 ext4_lblk_t c_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
3983 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
3984 ext4_lblk_t rr_cluster_start
;
3985 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
3986 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
3987 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
3989 /* The extent passed in that we are trying to match */
3990 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
3991 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
3993 /* The requested region passed into ext4_map_blocks() */
3994 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
3996 if ((rr_cluster_start
== ex_cluster_end
) ||
3997 (rr_cluster_start
== ex_cluster_start
)) {
3998 if (rr_cluster_start
== ex_cluster_end
)
3999 ee_start
+= ee_len
- 1;
4000 map
->m_pblk
= (ee_start
& ~(sbi
->s_cluster_ratio
- 1)) +
4002 map
->m_len
= min(map
->m_len
,
4003 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
4005 * Check for and handle this case:
4007 * |--------- cluster # N-------------|
4008 * |------- extent ----|
4009 * |--- requested region ---|
4013 if (map
->m_lblk
< ee_block
)
4014 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
4017 * Check for the case where there is already another allocated
4018 * block to the right of 'ex' but before the end of the cluster.
4020 * |------------- cluster # N-------------|
4021 * |----- ex -----| |---- ex_right ----|
4022 * |------ requested region ------|
4023 * |================|
4025 if (map
->m_lblk
> ee_block
) {
4026 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
4027 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
4030 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
4034 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
4040 * Block allocation/map/preallocation routine for extents based files
4043 * Need to be called with
4044 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4045 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4047 * return > 0, number of of blocks already mapped/allocated
4048 * if create == 0 and these are pre-allocated blocks
4049 * buffer head is unmapped
4050 * otherwise blocks are mapped
4052 * return = 0, if plain look up failed (blocks have not been allocated)
4053 * buffer head is unmapped
4055 * return < 0, error case.
4057 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
4058 struct ext4_map_blocks
*map
, int flags
)
4060 struct ext4_ext_path
*path
= NULL
;
4061 struct ext4_extent newex
, *ex
, *ex2
;
4062 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4063 ext4_fsblk_t newblock
= 0;
4064 int free_on_err
= 0, err
= 0, depth
;
4065 unsigned int allocated
= 0, offset
= 0;
4066 unsigned int allocated_clusters
= 0;
4067 struct ext4_allocation_request ar
;
4068 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
4069 ext4_lblk_t cluster_offset
;
4070 int set_unwritten
= 0;
4072 ext_debug("blocks %u/%u requested for inode %lu\n",
4073 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
4074 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
4076 /* find extent for this block */
4077 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, NULL
, 0);
4079 err
= PTR_ERR(path
);
4084 depth
= ext_depth(inode
);
4087 * consistent leaf must not be empty;
4088 * this situation is possible, though, _during_ tree modification;
4089 * this is why assert can't be put in ext4_ext_find_extent()
4091 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
4092 EXT4_ERROR_INODE(inode
, "bad extent address "
4093 "lblock: %lu, depth: %d pblock %lld",
4094 (unsigned long) map
->m_lblk
, depth
,
4095 path
[depth
].p_block
);
4100 ex
= path
[depth
].p_ext
;
4102 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4103 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4104 unsigned short ee_len
;
4107 * Uninitialized extents are treated as holes, except that
4108 * we split out initialized portions during a write.
4110 ee_len
= ext4_ext_get_actual_len(ex
);
4112 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
4114 /* if found extent covers block, simply return it */
4115 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
4116 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
4117 /* number of remaining blocks in the extent */
4118 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
4119 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
4120 ee_block
, ee_len
, newblock
);
4122 if (!ext4_ext_is_uninitialized(ex
))
4125 allocated
= ext4_ext_handle_uninitialized_extents(
4126 handle
, inode
, map
, path
, flags
,
4127 allocated
, newblock
);
4132 if ((sbi
->s_cluster_ratio
> 1) &&
4133 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
4134 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4137 * requested block isn't allocated yet;
4138 * we couldn't try to create block if create flag is zero
4140 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4142 * put just found gap into cache to speed up
4143 * subsequent requests
4145 if ((flags
& EXT4_GET_BLOCKS_NO_PUT_HOLE
) == 0)
4146 ext4_ext_put_gap_in_cache(inode
, path
, map
->m_lblk
);
4151 * Okay, we need to do block allocation.
4153 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
4154 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
4155 cluster_offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
-1);
4158 * If we are doing bigalloc, check to see if the extent returned
4159 * by ext4_ext_find_extent() implies a cluster we can use.
4161 if (cluster_offset
&& ex
&&
4162 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
4163 ar
.len
= allocated
= map
->m_len
;
4164 newblock
= map
->m_pblk
;
4165 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4166 goto got_allocated_blocks
;
4169 /* find neighbour allocated blocks */
4170 ar
.lleft
= map
->m_lblk
;
4171 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
4174 ar
.lright
= map
->m_lblk
;
4176 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
4180 /* Check if the extent after searching to the right implies a
4181 * cluster we can use. */
4182 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
4183 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
4184 ar
.len
= allocated
= map
->m_len
;
4185 newblock
= map
->m_pblk
;
4186 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4187 goto got_allocated_blocks
;
4191 * See if request is beyond maximum number of blocks we can have in
4192 * a single extent. For an initialized extent this limit is
4193 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4194 * EXT_UNINIT_MAX_LEN.
4196 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
4197 !(flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4198 map
->m_len
= EXT_INIT_MAX_LEN
;
4199 else if (map
->m_len
> EXT_UNINIT_MAX_LEN
&&
4200 (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4201 map
->m_len
= EXT_UNINIT_MAX_LEN
;
4203 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4204 newex
.ee_len
= cpu_to_le16(map
->m_len
);
4205 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
4207 allocated
= ext4_ext_get_actual_len(&newex
);
4209 allocated
= map
->m_len
;
4211 /* allocate new block */
4213 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
4214 ar
.logical
= map
->m_lblk
;
4216 * We calculate the offset from the beginning of the cluster
4217 * for the logical block number, since when we allocate a
4218 * physical cluster, the physical block should start at the
4219 * same offset from the beginning of the cluster. This is
4220 * needed so that future calls to get_implied_cluster_alloc()
4223 offset
= map
->m_lblk
& (sbi
->s_cluster_ratio
- 1);
4224 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4226 ar
.logical
-= offset
;
4227 if (S_ISREG(inode
->i_mode
))
4228 ar
.flags
= EXT4_MB_HINT_DATA
;
4230 /* disable in-core preallocation for non-regular files */
4232 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4233 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4234 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4237 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4238 ar
.goal
, newblock
, allocated
);
4240 allocated_clusters
= ar
.len
;
4241 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4242 if (ar
.len
> allocated
)
4245 got_allocated_blocks
:
4246 /* try to insert new extent into found leaf and return */
4247 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4248 newex
.ee_len
= cpu_to_le16(ar
.len
);
4249 /* Mark uninitialized */
4250 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
){
4251 ext4_ext_mark_uninitialized(&newex
);
4252 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4254 * io_end structure was created for every IO write to an
4255 * uninitialized extent. To avoid unnecessary conversion,
4256 * here we flag the IO that really needs the conversion.
4257 * For non asycn direct IO case, flag the inode state
4258 * that we need to perform conversion when IO is done.
4260 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
))
4262 if (ext4_should_dioread_nolock(inode
))
4263 map
->m_flags
|= EXT4_MAP_UNINIT
;
4267 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4268 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4271 err
= ext4_ext_insert_extent(handle
, inode
, path
,
4274 if (!err
&& set_unwritten
) {
4276 ext4_set_io_unwritten_flag(inode
, io
);
4278 ext4_set_inode_state(inode
,
4279 EXT4_STATE_DIO_UNWRITTEN
);
4282 if (err
&& free_on_err
) {
4283 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4284 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4285 /* free data blocks we just allocated */
4286 /* not a good idea to call discard here directly,
4287 * but otherwise we'd need to call it every free() */
4288 ext4_discard_preallocations(inode
);
4289 ext4_free_blocks(handle
, inode
, NULL
, newblock
,
4290 EXT4_C2B(sbi
, allocated_clusters
), fb_flags
);
4294 /* previous routine could use block we allocated */
4295 newblock
= ext4_ext_pblock(&newex
);
4296 allocated
= ext4_ext_get_actual_len(&newex
);
4297 if (allocated
> map
->m_len
)
4298 allocated
= map
->m_len
;
4299 map
->m_flags
|= EXT4_MAP_NEW
;
4302 * Update reserved blocks/metadata blocks after successful
4303 * block allocation which had been deferred till now.
4305 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4306 unsigned int reserved_clusters
;
4308 * Check how many clusters we had reserved this allocated range
4310 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4311 map
->m_lblk
, allocated
);
4312 if (map
->m_flags
& EXT4_MAP_FROM_CLUSTER
) {
4313 if (reserved_clusters
) {
4315 * We have clusters reserved for this range.
4316 * But since we are not doing actual allocation
4317 * and are simply using blocks from previously
4318 * allocated cluster, we should release the
4319 * reservation and not claim quota.
4321 ext4_da_update_reserve_space(inode
,
4322 reserved_clusters
, 0);
4325 BUG_ON(allocated_clusters
< reserved_clusters
);
4326 if (reserved_clusters
< allocated_clusters
) {
4327 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4328 int reservation
= allocated_clusters
-
4331 * It seems we claimed few clusters outside of
4332 * the range of this allocation. We should give
4333 * it back to the reservation pool. This can
4334 * happen in the following case:
4336 * * Suppose s_cluster_ratio is 4 (i.e., each
4337 * cluster has 4 blocks. Thus, the clusters
4338 * are [0-3],[4-7],[8-11]...
4339 * * First comes delayed allocation write for
4340 * logical blocks 10 & 11. Since there were no
4341 * previous delayed allocated blocks in the
4342 * range [8-11], we would reserve 1 cluster
4344 * * Next comes write for logical blocks 3 to 8.
4345 * In this case, we will reserve 2 clusters
4346 * (for [0-3] and [4-7]; and not for [8-11] as
4347 * that range has a delayed allocated blocks.
4348 * Thus total reserved clusters now becomes 3.
4349 * * Now, during the delayed allocation writeout
4350 * time, we will first write blocks [3-8] and
4351 * allocate 3 clusters for writing these
4352 * blocks. Also, we would claim all these
4353 * three clusters above.
4354 * * Now when we come here to writeout the
4355 * blocks [10-11], we would expect to claim
4356 * the reservation of 1 cluster we had made
4357 * (and we would claim it since there are no
4358 * more delayed allocated blocks in the range
4359 * [8-11]. But our reserved cluster count had
4360 * already gone to 0.
4362 * Thus, at the step 4 above when we determine
4363 * that there are still some unwritten delayed
4364 * allocated blocks outside of our current
4365 * block range, we should increment the
4366 * reserved clusters count so that when the
4367 * remaining blocks finally gets written, we
4370 dquot_reserve_block(inode
,
4371 EXT4_C2B(sbi
, reservation
));
4372 spin_lock(&ei
->i_block_reservation_lock
);
4373 ei
->i_reserved_data_blocks
+= reservation
;
4374 spin_unlock(&ei
->i_block_reservation_lock
);
4377 * We will claim quota for all newly allocated blocks.
4378 * We're updating the reserved space *after* the
4379 * correction above so we do not accidentally free
4380 * all the metadata reservation because we might
4381 * actually need it later on.
4383 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4389 * Cache the extent and update transaction to commit on fdatasync only
4390 * when it is _not_ an uninitialized extent.
4392 if ((flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) == 0)
4393 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4395 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4397 if (allocated
> map
->m_len
)
4398 allocated
= map
->m_len
;
4399 ext4_ext_show_leaf(inode
, path
);
4400 map
->m_flags
|= EXT4_MAP_MAPPED
;
4401 map
->m_pblk
= newblock
;
4402 map
->m_len
= allocated
;
4405 ext4_ext_drop_refs(path
);
4410 trace_ext4_ext_map_blocks_exit(inode
, flags
, map
,
4411 err
? err
: allocated
);
4412 ext4_es_lru_add(inode
);
4413 return err
? err
: allocated
;
4416 void ext4_ext_truncate(handle_t
*handle
, struct inode
*inode
)
4418 struct super_block
*sb
= inode
->i_sb
;
4419 ext4_lblk_t last_block
;
4423 * TODO: optimization is possible here.
4424 * Probably we need not scan at all,
4425 * because page truncation is enough.
4428 /* we have to know where to truncate from in crash case */
4429 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4430 ext4_mark_inode_dirty(handle
, inode
);
4432 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4433 >> EXT4_BLOCK_SIZE_BITS(sb
);
4435 err
= ext4_es_remove_extent(inode
, last_block
,
4436 EXT_MAX_BLOCKS
- last_block
);
4437 if (err
== -ENOMEM
) {
4439 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
4443 ext4_std_error(inode
->i_sb
, err
);
4446 err
= ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4447 ext4_std_error(inode
->i_sb
, err
);
4450 static void ext4_falloc_update_inode(struct inode
*inode
,
4451 int mode
, loff_t new_size
, int update_ctime
)
4453 struct timespec now
;
4456 now
= current_fs_time(inode
->i_sb
);
4457 if (!timespec_equal(&inode
->i_ctime
, &now
))
4458 inode
->i_ctime
= now
;
4461 * Update only when preallocation was requested beyond
4464 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
4465 if (new_size
> i_size_read(inode
))
4466 i_size_write(inode
, new_size
);
4467 if (new_size
> EXT4_I(inode
)->i_disksize
)
4468 ext4_update_i_disksize(inode
, new_size
);
4471 * Mark that we allocate beyond EOF so the subsequent truncate
4472 * can proceed even if the new size is the same as i_size.
4474 if (new_size
> i_size_read(inode
))
4475 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4481 * preallocate space for a file. This implements ext4's fallocate file
4482 * operation, which gets called from sys_fallocate system call.
4483 * For block-mapped files, posix_fallocate should fall back to the method
4484 * of writing zeroes to the required new blocks (the same behavior which is
4485 * expected for file systems which do not support fallocate() system call).
4487 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4489 struct inode
*inode
= file_inode(file
);
4492 unsigned int max_blocks
;
4497 struct ext4_map_blocks map
;
4498 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4500 /* Return error if mode is not supported */
4501 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
4504 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4505 return ext4_punch_hole(inode
, offset
, len
);
4507 ret
= ext4_convert_inline_data(inode
);
4512 * currently supporting (pre)allocate mode for extent-based
4515 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4518 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4519 map
.m_lblk
= offset
>> blkbits
;
4521 * We can't just convert len to max_blocks because
4522 * If blocksize = 4096 offset = 3072 and len = 2048
4524 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
4527 * credits to insert 1 extent into extent tree
4529 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4530 mutex_lock(&inode
->i_mutex
);
4531 ret
= inode_newsize_ok(inode
, (len
+ offset
));
4533 mutex_unlock(&inode
->i_mutex
);
4534 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
4537 flags
= EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
;
4538 if (mode
& FALLOC_FL_KEEP_SIZE
)
4539 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4541 * Don't normalize the request if it can fit in one extent so
4542 * that it doesn't get unnecessarily split into multiple
4545 if (len
<= EXT_UNINIT_MAX_LEN
<< blkbits
)
4546 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4549 while (ret
>= 0 && ret
< max_blocks
) {
4550 map
.m_lblk
= map
.m_lblk
+ ret
;
4551 map
.m_len
= max_blocks
= max_blocks
- ret
;
4552 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
4554 if (IS_ERR(handle
)) {
4555 ret
= PTR_ERR(handle
);
4558 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4561 ext4_warning(inode
->i_sb
,
4562 "inode #%lu: block %u: len %u: "
4563 "ext4_ext_map_blocks returned %d",
4564 inode
->i_ino
, map
.m_lblk
,
4567 ext4_mark_inode_dirty(handle
, inode
);
4568 ret2
= ext4_journal_stop(handle
);
4571 if ((map
.m_lblk
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
4572 blkbits
) >> blkbits
))
4573 new_size
= offset
+ len
;
4575 new_size
= ((loff_t
) map
.m_lblk
+ ret
) << blkbits
;
4577 ext4_falloc_update_inode(inode
, mode
, new_size
,
4578 (map
.m_flags
& EXT4_MAP_NEW
));
4579 ext4_mark_inode_dirty(handle
, inode
);
4580 if ((file
->f_flags
& O_SYNC
) && ret
>= max_blocks
)
4581 ext4_handle_sync(handle
);
4582 ret2
= ext4_journal_stop(handle
);
4586 if (ret
== -ENOSPC
&&
4587 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4591 mutex_unlock(&inode
->i_mutex
);
4592 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
,
4593 ret
> 0 ? ret2
: ret
);
4594 return ret
> 0 ? ret2
: ret
;
4598 * This function convert a range of blocks to written extents
4599 * The caller of this function will pass the start offset and the size.
4600 * all unwritten extents within this range will be converted to
4603 * This function is called from the direct IO end io call back
4604 * function, to convert the fallocated extents after IO is completed.
4605 * Returns 0 on success.
4607 int ext4_convert_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
4608 loff_t offset
, ssize_t len
)
4610 unsigned int max_blocks
;
4613 struct ext4_map_blocks map
;
4614 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4616 map
.m_lblk
= offset
>> blkbits
;
4618 * We can't just convert len to max_blocks because
4619 * If blocksize = 4096 offset = 3072 and len = 2048
4621 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
4624 * This is somewhat ugly but the idea is clear: When transaction is
4625 * reserved, everything goes into it. Otherwise we rather start several
4626 * smaller transactions for conversion of each extent separately.
4629 handle
= ext4_journal_start_reserved(handle
,
4630 EXT4_HT_EXT_CONVERT
);
4632 return PTR_ERR(handle
);
4636 * credits to insert 1 extent into extent tree
4638 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4640 while (ret
>= 0 && ret
< max_blocks
) {
4642 map
.m_len
= (max_blocks
-= ret
);
4644 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
4646 if (IS_ERR(handle
)) {
4647 ret
= PTR_ERR(handle
);
4651 ret
= ext4_map_blocks(handle
, inode
, &map
,
4652 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
4654 ext4_warning(inode
->i_sb
,
4655 "inode #%lu: block %u: len %u: "
4656 "ext4_ext_map_blocks returned %d",
4657 inode
->i_ino
, map
.m_lblk
,
4659 ext4_mark_inode_dirty(handle
, inode
);
4661 ret2
= ext4_journal_stop(handle
);
4662 if (ret
<= 0 || ret2
)
4666 ret2
= ext4_journal_stop(handle
);
4667 return ret
> 0 ? ret2
: ret
;
4671 * If newes is not existing extent (newes->ec_pblk equals zero) find
4672 * delayed extent at start of newes and update newes accordingly and
4673 * return start of the next delayed extent.
4675 * If newes is existing extent (newes->ec_pblk is not equal zero)
4676 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
4677 * extent found. Leave newes unmodified.
4679 static int ext4_find_delayed_extent(struct inode
*inode
,
4680 struct extent_status
*newes
)
4682 struct extent_status es
;
4683 ext4_lblk_t block
, next_del
;
4685 if (newes
->es_pblk
== 0) {
4686 ext4_es_find_delayed_extent_range(inode
, newes
->es_lblk
,
4687 newes
->es_lblk
+ newes
->es_len
- 1, &es
);
4690 * No extent in extent-tree contains block @newes->es_pblk,
4691 * then the block may stay in 1)a hole or 2)delayed-extent.
4697 if (es
.es_lblk
> newes
->es_lblk
) {
4699 newes
->es_len
= min(es
.es_lblk
- newes
->es_lblk
,
4704 newes
->es_len
= es
.es_lblk
+ es
.es_len
- newes
->es_lblk
;
4707 block
= newes
->es_lblk
+ newes
->es_len
;
4708 ext4_es_find_delayed_extent_range(inode
, block
, EXT_MAX_BLOCKS
, &es
);
4710 next_del
= EXT_MAX_BLOCKS
;
4712 next_del
= es
.es_lblk
;
4716 /* fiemap flags we can handle specified here */
4717 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4719 static int ext4_xattr_fiemap(struct inode
*inode
,
4720 struct fiemap_extent_info
*fieinfo
)
4724 __u32 flags
= FIEMAP_EXTENT_LAST
;
4725 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
4729 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
4730 struct ext4_iloc iloc
;
4731 int offset
; /* offset of xattr in inode */
4733 error
= ext4_get_inode_loc(inode
, &iloc
);
4736 physical
= (__u64
)iloc
.bh
->b_blocknr
<< blockbits
;
4737 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
4738 EXT4_I(inode
)->i_extra_isize
;
4740 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
4741 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
4743 } else { /* external block */
4744 physical
= (__u64
)EXT4_I(inode
)->i_file_acl
<< blockbits
;
4745 length
= inode
->i_sb
->s_blocksize
;
4749 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
4751 return (error
< 0 ? error
: 0);
4754 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4755 __u64 start
, __u64 len
)
4757 ext4_lblk_t start_blk
;
4760 if (ext4_has_inline_data(inode
)) {
4763 error
= ext4_inline_data_fiemap(inode
, fieinfo
, &has_inline
);
4769 /* fallback to generic here if not in extents fmt */
4770 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4771 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
4774 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
4777 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
4778 error
= ext4_xattr_fiemap(inode
, fieinfo
);
4780 ext4_lblk_t len_blks
;
4783 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
4784 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
4785 if (last_blk
>= EXT_MAX_BLOCKS
)
4786 last_blk
= EXT_MAX_BLOCKS
-1;
4787 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
4790 * Walk the extent tree gathering extent information
4791 * and pushing extents back to the user.
4793 error
= ext4_fill_fiemap_extents(inode
, start_blk
,
4796 ext4_es_lru_add(inode
);