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
);
363 ext4_lblk_t lblock
= le32_to_cpu(ext
->ee_block
);
364 ext4_lblk_t last
= lblock
+ len
- 1;
368 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, len
);
371 static int ext4_valid_extent_idx(struct inode
*inode
,
372 struct ext4_extent_idx
*ext_idx
)
374 ext4_fsblk_t block
= ext4_idx_pblock(ext_idx
);
376 return ext4_data_block_valid(EXT4_SB(inode
->i_sb
), block
, 1);
379 static int ext4_valid_extent_entries(struct inode
*inode
,
380 struct ext4_extent_header
*eh
,
383 unsigned short entries
;
384 if (eh
->eh_entries
== 0)
387 entries
= le16_to_cpu(eh
->eh_entries
);
391 struct ext4_extent
*ext
= EXT_FIRST_EXTENT(eh
);
392 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
393 ext4_fsblk_t pblock
= 0;
394 ext4_lblk_t lblock
= 0;
395 ext4_lblk_t prev
= 0;
398 if (!ext4_valid_extent(inode
, ext
))
401 /* Check for overlapping extents */
402 lblock
= le32_to_cpu(ext
->ee_block
);
403 len
= ext4_ext_get_actual_len(ext
);
404 if ((lblock
<= prev
) && prev
) {
405 pblock
= ext4_ext_pblock(ext
);
406 es
->s_last_error_block
= cpu_to_le64(pblock
);
411 prev
= lblock
+ len
- 1;
414 struct ext4_extent_idx
*ext_idx
= EXT_FIRST_INDEX(eh
);
416 if (!ext4_valid_extent_idx(inode
, ext_idx
))
425 static int __ext4_ext_check(const char *function
, unsigned int line
,
426 struct inode
*inode
, struct ext4_extent_header
*eh
,
427 int depth
, ext4_fsblk_t pblk
)
429 const char *error_msg
;
432 if (unlikely(eh
->eh_magic
!= EXT4_EXT_MAGIC
)) {
433 error_msg
= "invalid magic";
436 if (unlikely(le16_to_cpu(eh
->eh_depth
) != depth
)) {
437 error_msg
= "unexpected eh_depth";
440 if (unlikely(eh
->eh_max
== 0)) {
441 error_msg
= "invalid eh_max";
444 max
= ext4_ext_max_entries(inode
, depth
);
445 if (unlikely(le16_to_cpu(eh
->eh_max
) > max
)) {
446 error_msg
= "too large eh_max";
449 if (unlikely(le16_to_cpu(eh
->eh_entries
) > le16_to_cpu(eh
->eh_max
))) {
450 error_msg
= "invalid eh_entries";
453 if (!ext4_valid_extent_entries(inode
, eh
, depth
)) {
454 error_msg
= "invalid extent entries";
457 /* Verify checksum on non-root extent tree nodes */
458 if (ext_depth(inode
) != depth
&&
459 !ext4_extent_block_csum_verify(inode
, eh
)) {
460 error_msg
= "extent tree corrupted";
466 ext4_error_inode(inode
, function
, line
, 0,
467 "pblk %llu bad header/extent: %s - magic %x, "
468 "entries %u, max %u(%u), depth %u(%u)",
469 (unsigned long long) pblk
, error_msg
,
470 le16_to_cpu(eh
->eh_magic
),
471 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
),
472 max
, le16_to_cpu(eh
->eh_depth
), depth
);
476 #define ext4_ext_check(inode, eh, depth, pblk) \
477 __ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
479 int ext4_ext_check_inode(struct inode
*inode
)
481 return ext4_ext_check(inode
, ext_inode_hdr(inode
), ext_depth(inode
), 0);
484 static struct buffer_head
*
485 __read_extent_tree_block(const char *function
, unsigned int line
,
486 struct inode
*inode
, ext4_fsblk_t pblk
, int depth
,
489 struct buffer_head
*bh
;
492 bh
= sb_getblk(inode
->i_sb
, pblk
);
494 return ERR_PTR(-ENOMEM
);
496 if (!bh_uptodate_or_lock(bh
)) {
497 trace_ext4_ext_load_extent(inode
, pblk
, _RET_IP_
);
498 err
= bh_submit_read(bh
);
502 if (buffer_verified(bh
) && !(flags
& EXT4_EX_FORCE_CACHE
))
504 err
= __ext4_ext_check(function
, line
, inode
,
505 ext_block_hdr(bh
), depth
, pblk
);
508 set_buffer_verified(bh
);
510 * If this is a leaf block, cache all of its entries
512 if (!(flags
& EXT4_EX_NOCACHE
) && depth
== 0) {
513 struct ext4_extent_header
*eh
= ext_block_hdr(bh
);
514 struct ext4_extent
*ex
= EXT_FIRST_EXTENT(eh
);
515 ext4_lblk_t prev
= 0;
518 for (i
= le16_to_cpu(eh
->eh_entries
); i
> 0; i
--, ex
++) {
519 unsigned int status
= EXTENT_STATUS_WRITTEN
;
520 ext4_lblk_t lblk
= le32_to_cpu(ex
->ee_block
);
521 int len
= ext4_ext_get_actual_len(ex
);
523 if (prev
&& (prev
!= lblk
))
524 ext4_es_cache_extent(inode
, prev
,
528 if (ext4_ext_is_uninitialized(ex
))
529 status
= EXTENT_STATUS_UNWRITTEN
;
530 ext4_es_cache_extent(inode
, lblk
, len
,
531 ext4_ext_pblock(ex
), status
);
542 #define read_extent_tree_block(inode, pblk, depth, flags) \
543 __read_extent_tree_block(__func__, __LINE__, (inode), (pblk), \
547 * This function is called to cache a file's extent information in the
550 int ext4_ext_precache(struct inode
*inode
)
552 struct ext4_inode_info
*ei
= EXT4_I(inode
);
553 struct ext4_ext_path
*path
= NULL
;
554 struct buffer_head
*bh
;
555 int i
= 0, depth
, ret
= 0;
557 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
558 return 0; /* not an extent-mapped inode */
560 down_read(&ei
->i_data_sem
);
561 depth
= ext_depth(inode
);
563 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
566 up_read(&ei
->i_data_sem
);
570 /* Don't cache anything if there are no external extent blocks */
573 path
[0].p_hdr
= ext_inode_hdr(inode
);
574 ret
= ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0);
577 path
[0].p_idx
= EXT_FIRST_INDEX(path
[0].p_hdr
);
580 * If this is a leaf block or we've reached the end of
581 * the index block, go up
584 path
[i
].p_idx
> EXT_LAST_INDEX(path
[i
].p_hdr
)) {
585 brelse(path
[i
].p_bh
);
590 bh
= read_extent_tree_block(inode
,
591 ext4_idx_pblock(path
[i
].p_idx
++),
593 EXT4_EX_FORCE_CACHE
);
600 path
[i
].p_hdr
= ext_block_hdr(bh
);
601 path
[i
].p_idx
= EXT_FIRST_INDEX(path
[i
].p_hdr
);
603 ext4_set_inode_state(inode
, EXT4_STATE_EXT_PRECACHED
);
605 up_read(&ei
->i_data_sem
);
606 ext4_ext_drop_refs(path
);
612 static void ext4_ext_show_path(struct inode
*inode
, struct ext4_ext_path
*path
)
614 int k
, l
= path
->p_depth
;
617 for (k
= 0; k
<= l
; k
++, path
++) {
619 ext_debug(" %d->%llu", le32_to_cpu(path
->p_idx
->ei_block
),
620 ext4_idx_pblock(path
->p_idx
));
621 } else if (path
->p_ext
) {
622 ext_debug(" %d:[%d]%d:%llu ",
623 le32_to_cpu(path
->p_ext
->ee_block
),
624 ext4_ext_is_uninitialized(path
->p_ext
),
625 ext4_ext_get_actual_len(path
->p_ext
),
626 ext4_ext_pblock(path
->p_ext
));
633 static void ext4_ext_show_leaf(struct inode
*inode
, struct ext4_ext_path
*path
)
635 int depth
= ext_depth(inode
);
636 struct ext4_extent_header
*eh
;
637 struct ext4_extent
*ex
;
643 eh
= path
[depth
].p_hdr
;
644 ex
= EXT_FIRST_EXTENT(eh
);
646 ext_debug("Displaying leaf extents for inode %lu\n", inode
->i_ino
);
648 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ex
++) {
649 ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex
->ee_block
),
650 ext4_ext_is_uninitialized(ex
),
651 ext4_ext_get_actual_len(ex
), ext4_ext_pblock(ex
));
656 static void ext4_ext_show_move(struct inode
*inode
, struct ext4_ext_path
*path
,
657 ext4_fsblk_t newblock
, int level
)
659 int depth
= ext_depth(inode
);
660 struct ext4_extent
*ex
;
662 if (depth
!= level
) {
663 struct ext4_extent_idx
*idx
;
664 idx
= path
[level
].p_idx
;
665 while (idx
<= EXT_MAX_INDEX(path
[level
].p_hdr
)) {
666 ext_debug("%d: move %d:%llu in new index %llu\n", level
,
667 le32_to_cpu(idx
->ei_block
),
668 ext4_idx_pblock(idx
),
676 ex
= path
[depth
].p_ext
;
677 while (ex
<= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
678 ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
679 le32_to_cpu(ex
->ee_block
),
681 ext4_ext_is_uninitialized(ex
),
682 ext4_ext_get_actual_len(ex
),
689 #define ext4_ext_show_path(inode, path)
690 #define ext4_ext_show_leaf(inode, path)
691 #define ext4_ext_show_move(inode, path, newblock, level)
694 void ext4_ext_drop_refs(struct ext4_ext_path
*path
)
696 int depth
= path
->p_depth
;
699 for (i
= 0; i
<= depth
; i
++, path
++)
707 * ext4_ext_binsearch_idx:
708 * binary search for the closest index of the given block
709 * the header must be checked before calling this
712 ext4_ext_binsearch_idx(struct inode
*inode
,
713 struct ext4_ext_path
*path
, ext4_lblk_t block
)
715 struct ext4_extent_header
*eh
= path
->p_hdr
;
716 struct ext4_extent_idx
*r
, *l
, *m
;
719 ext_debug("binsearch for %u(idx): ", block
);
721 l
= EXT_FIRST_INDEX(eh
) + 1;
722 r
= EXT_LAST_INDEX(eh
);
725 if (block
< le32_to_cpu(m
->ei_block
))
729 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ei_block
),
730 m
, le32_to_cpu(m
->ei_block
),
731 r
, le32_to_cpu(r
->ei_block
));
735 ext_debug(" -> %u->%lld ", le32_to_cpu(path
->p_idx
->ei_block
),
736 ext4_idx_pblock(path
->p_idx
));
738 #ifdef CHECK_BINSEARCH
740 struct ext4_extent_idx
*chix
, *ix
;
743 chix
= ix
= EXT_FIRST_INDEX(eh
);
744 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ix
++) {
746 le32_to_cpu(ix
->ei_block
) <= le32_to_cpu(ix
[-1].ei_block
)) {
747 printk(KERN_DEBUG
"k=%d, ix=0x%p, "
749 ix
, EXT_FIRST_INDEX(eh
));
750 printk(KERN_DEBUG
"%u <= %u\n",
751 le32_to_cpu(ix
->ei_block
),
752 le32_to_cpu(ix
[-1].ei_block
));
754 BUG_ON(k
&& le32_to_cpu(ix
->ei_block
)
755 <= le32_to_cpu(ix
[-1].ei_block
));
756 if (block
< le32_to_cpu(ix
->ei_block
))
760 BUG_ON(chix
!= path
->p_idx
);
767 * ext4_ext_binsearch:
768 * binary search for closest extent of the given block
769 * the header must be checked before calling this
772 ext4_ext_binsearch(struct inode
*inode
,
773 struct ext4_ext_path
*path
, ext4_lblk_t block
)
775 struct ext4_extent_header
*eh
= path
->p_hdr
;
776 struct ext4_extent
*r
, *l
, *m
;
778 if (eh
->eh_entries
== 0) {
780 * this leaf is empty:
781 * we get such a leaf in split/add case
786 ext_debug("binsearch for %u: ", block
);
788 l
= EXT_FIRST_EXTENT(eh
) + 1;
789 r
= EXT_LAST_EXTENT(eh
);
793 if (block
< le32_to_cpu(m
->ee_block
))
797 ext_debug("%p(%u):%p(%u):%p(%u) ", l
, le32_to_cpu(l
->ee_block
),
798 m
, le32_to_cpu(m
->ee_block
),
799 r
, le32_to_cpu(r
->ee_block
));
803 ext_debug(" -> %d:%llu:[%d]%d ",
804 le32_to_cpu(path
->p_ext
->ee_block
),
805 ext4_ext_pblock(path
->p_ext
),
806 ext4_ext_is_uninitialized(path
->p_ext
),
807 ext4_ext_get_actual_len(path
->p_ext
));
809 #ifdef CHECK_BINSEARCH
811 struct ext4_extent
*chex
, *ex
;
814 chex
= ex
= EXT_FIRST_EXTENT(eh
);
815 for (k
= 0; k
< le16_to_cpu(eh
->eh_entries
); k
++, ex
++) {
816 BUG_ON(k
&& le32_to_cpu(ex
->ee_block
)
817 <= le32_to_cpu(ex
[-1].ee_block
));
818 if (block
< le32_to_cpu(ex
->ee_block
))
822 BUG_ON(chex
!= path
->p_ext
);
828 int ext4_ext_tree_init(handle_t
*handle
, struct inode
*inode
)
830 struct ext4_extent_header
*eh
;
832 eh
= ext_inode_hdr(inode
);
835 eh
->eh_magic
= EXT4_EXT_MAGIC
;
836 eh
->eh_max
= cpu_to_le16(ext4_ext_space_root(inode
, 0));
837 ext4_mark_inode_dirty(handle
, inode
);
841 struct ext4_ext_path
*
842 ext4_ext_find_extent(struct inode
*inode
, ext4_lblk_t block
,
843 struct ext4_ext_path
*path
, int flags
)
845 struct ext4_extent_header
*eh
;
846 struct buffer_head
*bh
;
847 short int depth
, i
, ppos
= 0, alloc
= 0;
850 eh
= ext_inode_hdr(inode
);
851 depth
= ext_depth(inode
);
853 /* account possible depth increase */
855 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 2),
858 return ERR_PTR(-ENOMEM
);
865 /* walk through the tree */
867 ext_debug("depth %d: num %d, max %d\n",
868 ppos
, le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
870 ext4_ext_binsearch_idx(inode
, path
+ ppos
, block
);
871 path
[ppos
].p_block
= ext4_idx_pblock(path
[ppos
].p_idx
);
872 path
[ppos
].p_depth
= i
;
873 path
[ppos
].p_ext
= NULL
;
875 bh
= read_extent_tree_block(inode
, path
[ppos
].p_block
, --i
,
882 eh
= ext_block_hdr(bh
);
884 if (unlikely(ppos
> depth
)) {
886 EXT4_ERROR_INODE(inode
,
887 "ppos %d > depth %d", ppos
, depth
);
891 path
[ppos
].p_bh
= bh
;
892 path
[ppos
].p_hdr
= eh
;
895 path
[ppos
].p_depth
= i
;
896 path
[ppos
].p_ext
= NULL
;
897 path
[ppos
].p_idx
= NULL
;
900 ext4_ext_binsearch(inode
, path
+ ppos
, block
);
901 /* if not an empty leaf */
902 if (path
[ppos
].p_ext
)
903 path
[ppos
].p_block
= ext4_ext_pblock(path
[ppos
].p_ext
);
905 ext4_ext_show_path(inode
, path
);
910 ext4_ext_drop_refs(path
);
917 * ext4_ext_insert_index:
918 * insert new index [@logical;@ptr] into the block at @curp;
919 * check where to insert: before @curp or after @curp
921 static int ext4_ext_insert_index(handle_t
*handle
, struct inode
*inode
,
922 struct ext4_ext_path
*curp
,
923 int logical
, ext4_fsblk_t ptr
)
925 struct ext4_extent_idx
*ix
;
928 err
= ext4_ext_get_access(handle
, inode
, curp
);
932 if (unlikely(logical
== le32_to_cpu(curp
->p_idx
->ei_block
))) {
933 EXT4_ERROR_INODE(inode
,
934 "logical %d == ei_block %d!",
935 logical
, le32_to_cpu(curp
->p_idx
->ei_block
));
939 if (unlikely(le16_to_cpu(curp
->p_hdr
->eh_entries
)
940 >= le16_to_cpu(curp
->p_hdr
->eh_max
))) {
941 EXT4_ERROR_INODE(inode
,
942 "eh_entries %d >= eh_max %d!",
943 le16_to_cpu(curp
->p_hdr
->eh_entries
),
944 le16_to_cpu(curp
->p_hdr
->eh_max
));
948 if (logical
> le32_to_cpu(curp
->p_idx
->ei_block
)) {
950 ext_debug("insert new index %d after: %llu\n", logical
, ptr
);
951 ix
= curp
->p_idx
+ 1;
954 ext_debug("insert new index %d before: %llu\n", logical
, ptr
);
958 len
= EXT_LAST_INDEX(curp
->p_hdr
) - ix
+ 1;
961 ext_debug("insert new index %d: "
962 "move %d indices from 0x%p to 0x%p\n",
963 logical
, len
, ix
, ix
+ 1);
964 memmove(ix
+ 1, ix
, len
* sizeof(struct ext4_extent_idx
));
967 if (unlikely(ix
> EXT_MAX_INDEX(curp
->p_hdr
))) {
968 EXT4_ERROR_INODE(inode
, "ix > EXT_MAX_INDEX!");
972 ix
->ei_block
= cpu_to_le32(logical
);
973 ext4_idx_store_pblock(ix
, ptr
);
974 le16_add_cpu(&curp
->p_hdr
->eh_entries
, 1);
976 if (unlikely(ix
> EXT_LAST_INDEX(curp
->p_hdr
))) {
977 EXT4_ERROR_INODE(inode
, "ix > EXT_LAST_INDEX!");
981 err
= ext4_ext_dirty(handle
, inode
, curp
);
982 ext4_std_error(inode
->i_sb
, err
);
989 * inserts new subtree into the path, using free index entry
991 * - allocates all needed blocks (new leaf and all intermediate index blocks)
992 * - makes decision where to split
993 * - moves remaining extents and index entries (right to the split point)
994 * into the newly allocated blocks
995 * - initializes subtree
997 static int ext4_ext_split(handle_t
*handle
, struct inode
*inode
,
999 struct ext4_ext_path
*path
,
1000 struct ext4_extent
*newext
, int at
)
1002 struct buffer_head
*bh
= NULL
;
1003 int depth
= ext_depth(inode
);
1004 struct ext4_extent_header
*neh
;
1005 struct ext4_extent_idx
*fidx
;
1006 int i
= at
, k
, m
, a
;
1007 ext4_fsblk_t newblock
, oldblock
;
1009 ext4_fsblk_t
*ablocks
= NULL
; /* array of allocated blocks */
1012 /* make decision: where to split? */
1013 /* FIXME: now decision is simplest: at current extent */
1015 /* if current leaf will be split, then we should use
1016 * border from split point */
1017 if (unlikely(path
[depth
].p_ext
> EXT_MAX_EXTENT(path
[depth
].p_hdr
))) {
1018 EXT4_ERROR_INODE(inode
, "p_ext > EXT_MAX_EXTENT!");
1021 if (path
[depth
].p_ext
!= EXT_MAX_EXTENT(path
[depth
].p_hdr
)) {
1022 border
= path
[depth
].p_ext
[1].ee_block
;
1023 ext_debug("leaf will be split."
1024 " next leaf starts at %d\n",
1025 le32_to_cpu(border
));
1027 border
= newext
->ee_block
;
1028 ext_debug("leaf will be added."
1029 " next leaf starts at %d\n",
1030 le32_to_cpu(border
));
1034 * If error occurs, then we break processing
1035 * and mark filesystem read-only. index won't
1036 * be inserted and tree will be in consistent
1037 * state. Next mount will repair buffers too.
1041 * Get array to track all allocated blocks.
1042 * We need this to handle errors and free blocks
1045 ablocks
= kzalloc(sizeof(ext4_fsblk_t
) * depth
, GFP_NOFS
);
1049 /* allocate all needed blocks */
1050 ext_debug("allocate %d blocks for indexes/leaf\n", depth
- at
);
1051 for (a
= 0; a
< depth
- at
; a
++) {
1052 newblock
= ext4_ext_new_meta_block(handle
, inode
, path
,
1053 newext
, &err
, flags
);
1056 ablocks
[a
] = newblock
;
1059 /* initialize new leaf */
1060 newblock
= ablocks
[--a
];
1061 if (unlikely(newblock
== 0)) {
1062 EXT4_ERROR_INODE(inode
, "newblock == 0!");
1066 bh
= sb_getblk(inode
->i_sb
, newblock
);
1067 if (unlikely(!bh
)) {
1073 err
= ext4_journal_get_create_access(handle
, bh
);
1077 neh
= ext_block_hdr(bh
);
1078 neh
->eh_entries
= 0;
1079 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1080 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1083 /* move remainder of path[depth] to the new leaf */
1084 if (unlikely(path
[depth
].p_hdr
->eh_entries
!=
1085 path
[depth
].p_hdr
->eh_max
)) {
1086 EXT4_ERROR_INODE(inode
, "eh_entries %d != eh_max %d!",
1087 path
[depth
].p_hdr
->eh_entries
,
1088 path
[depth
].p_hdr
->eh_max
);
1092 /* start copy from next extent */
1093 m
= EXT_MAX_EXTENT(path
[depth
].p_hdr
) - path
[depth
].p_ext
++;
1094 ext4_ext_show_move(inode
, path
, newblock
, depth
);
1096 struct ext4_extent
*ex
;
1097 ex
= EXT_FIRST_EXTENT(neh
);
1098 memmove(ex
, path
[depth
].p_ext
, sizeof(struct ext4_extent
) * m
);
1099 le16_add_cpu(&neh
->eh_entries
, m
);
1102 ext4_extent_block_csum_set(inode
, neh
);
1103 set_buffer_uptodate(bh
);
1106 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1112 /* correct old leaf */
1114 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
1117 le16_add_cpu(&path
[depth
].p_hdr
->eh_entries
, -m
);
1118 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
1124 /* create intermediate indexes */
1126 if (unlikely(k
< 0)) {
1127 EXT4_ERROR_INODE(inode
, "k %d < 0!", k
);
1132 ext_debug("create %d intermediate indices\n", k
);
1133 /* insert new index into current index block */
1134 /* current depth stored in i var */
1137 oldblock
= newblock
;
1138 newblock
= ablocks
[--a
];
1139 bh
= sb_getblk(inode
->i_sb
, newblock
);
1140 if (unlikely(!bh
)) {
1146 err
= ext4_journal_get_create_access(handle
, bh
);
1150 neh
= ext_block_hdr(bh
);
1151 neh
->eh_entries
= cpu_to_le16(1);
1152 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1153 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1154 neh
->eh_depth
= cpu_to_le16(depth
- i
);
1155 fidx
= EXT_FIRST_INDEX(neh
);
1156 fidx
->ei_block
= border
;
1157 ext4_idx_store_pblock(fidx
, oldblock
);
1159 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1160 i
, newblock
, le32_to_cpu(border
), oldblock
);
1162 /* move remainder of path[i] to the new index block */
1163 if (unlikely(EXT_MAX_INDEX(path
[i
].p_hdr
) !=
1164 EXT_LAST_INDEX(path
[i
].p_hdr
))) {
1165 EXT4_ERROR_INODE(inode
,
1166 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1167 le32_to_cpu(path
[i
].p_ext
->ee_block
));
1171 /* start copy indexes */
1172 m
= EXT_MAX_INDEX(path
[i
].p_hdr
) - path
[i
].p_idx
++;
1173 ext_debug("cur 0x%p, last 0x%p\n", path
[i
].p_idx
,
1174 EXT_MAX_INDEX(path
[i
].p_hdr
));
1175 ext4_ext_show_move(inode
, path
, newblock
, i
);
1177 memmove(++fidx
, path
[i
].p_idx
,
1178 sizeof(struct ext4_extent_idx
) * m
);
1179 le16_add_cpu(&neh
->eh_entries
, m
);
1181 ext4_extent_block_csum_set(inode
, neh
);
1182 set_buffer_uptodate(bh
);
1185 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1191 /* correct old index */
1193 err
= ext4_ext_get_access(handle
, inode
, path
+ i
);
1196 le16_add_cpu(&path
[i
].p_hdr
->eh_entries
, -m
);
1197 err
= ext4_ext_dirty(handle
, inode
, path
+ i
);
1205 /* insert new index */
1206 err
= ext4_ext_insert_index(handle
, inode
, path
+ at
,
1207 le32_to_cpu(border
), newblock
);
1211 if (buffer_locked(bh
))
1217 /* free all allocated blocks in error case */
1218 for (i
= 0; i
< depth
; i
++) {
1221 ext4_free_blocks(handle
, inode
, NULL
, ablocks
[i
], 1,
1222 EXT4_FREE_BLOCKS_METADATA
);
1231 * ext4_ext_grow_indepth:
1232 * implements tree growing procedure:
1233 * - allocates new block
1234 * - moves top-level data (index block or leaf) into the new block
1235 * - initializes new top-level, creating index that points to the
1236 * just created block
1238 static int ext4_ext_grow_indepth(handle_t
*handle
, struct inode
*inode
,
1240 struct ext4_extent
*newext
)
1242 struct ext4_extent_header
*neh
;
1243 struct buffer_head
*bh
;
1244 ext4_fsblk_t newblock
;
1247 newblock
= ext4_ext_new_meta_block(handle
, inode
, NULL
,
1248 newext
, &err
, flags
);
1252 bh
= sb_getblk(inode
->i_sb
, newblock
);
1257 err
= ext4_journal_get_create_access(handle
, bh
);
1263 /* move top-level index/leaf into new block */
1264 memmove(bh
->b_data
, EXT4_I(inode
)->i_data
,
1265 sizeof(EXT4_I(inode
)->i_data
));
1267 /* set size of new block */
1268 neh
= ext_block_hdr(bh
);
1269 /* old root could have indexes or leaves
1270 * so calculate e_max right way */
1271 if (ext_depth(inode
))
1272 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block_idx(inode
, 0));
1274 neh
->eh_max
= cpu_to_le16(ext4_ext_space_block(inode
, 0));
1275 neh
->eh_magic
= EXT4_EXT_MAGIC
;
1276 ext4_extent_block_csum_set(inode
, neh
);
1277 set_buffer_uptodate(bh
);
1280 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1284 /* Update top-level index: num,max,pointer */
1285 neh
= ext_inode_hdr(inode
);
1286 neh
->eh_entries
= cpu_to_le16(1);
1287 ext4_idx_store_pblock(EXT_FIRST_INDEX(neh
), newblock
);
1288 if (neh
->eh_depth
== 0) {
1289 /* Root extent block becomes index block */
1290 neh
->eh_max
= cpu_to_le16(ext4_ext_space_root_idx(inode
, 0));
1291 EXT_FIRST_INDEX(neh
)->ei_block
=
1292 EXT_FIRST_EXTENT(neh
)->ee_block
;
1294 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1295 le16_to_cpu(neh
->eh_entries
), le16_to_cpu(neh
->eh_max
),
1296 le32_to_cpu(EXT_FIRST_INDEX(neh
)->ei_block
),
1297 ext4_idx_pblock(EXT_FIRST_INDEX(neh
)));
1299 le16_add_cpu(&neh
->eh_depth
, 1);
1300 ext4_mark_inode_dirty(handle
, inode
);
1308 * ext4_ext_create_new_leaf:
1309 * finds empty index and adds new leaf.
1310 * if no free index is found, then it requests in-depth growing.
1312 static int ext4_ext_create_new_leaf(handle_t
*handle
, struct inode
*inode
,
1313 unsigned int mb_flags
,
1314 unsigned int gb_flags
,
1315 struct ext4_ext_path
*path
,
1316 struct ext4_extent
*newext
)
1318 struct ext4_ext_path
*curp
;
1319 int depth
, i
, err
= 0;
1322 i
= depth
= ext_depth(inode
);
1324 /* walk up to the tree and look for free index entry */
1325 curp
= path
+ depth
;
1326 while (i
> 0 && !EXT_HAS_FREE_INDEX(curp
)) {
1331 /* we use already allocated block for index block,
1332 * so subsequent data blocks should be contiguous */
1333 if (EXT_HAS_FREE_INDEX(curp
)) {
1334 /* if we found index with free entry, then use that
1335 * entry: create all needed subtree and add new leaf */
1336 err
= ext4_ext_split(handle
, inode
, mb_flags
, path
, newext
, i
);
1341 ext4_ext_drop_refs(path
);
1342 path
= ext4_ext_find_extent(inode
,
1343 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1346 err
= PTR_ERR(path
);
1348 /* tree is full, time to grow in depth */
1349 err
= ext4_ext_grow_indepth(handle
, inode
, mb_flags
, newext
);
1354 ext4_ext_drop_refs(path
);
1355 path
= ext4_ext_find_extent(inode
,
1356 (ext4_lblk_t
)le32_to_cpu(newext
->ee_block
),
1359 err
= PTR_ERR(path
);
1364 * only first (depth 0 -> 1) produces free space;
1365 * in all other cases we have to split the grown tree
1367 depth
= ext_depth(inode
);
1368 if (path
[depth
].p_hdr
->eh_entries
== path
[depth
].p_hdr
->eh_max
) {
1369 /* now we need to split */
1379 * search the closest allocated block to the left for *logical
1380 * and returns it at @logical + it's physical address at @phys
1381 * if *logical is the smallest allocated block, the function
1382 * returns 0 at @phys
1383 * return value contains 0 (success) or error code
1385 static int ext4_ext_search_left(struct inode
*inode
,
1386 struct ext4_ext_path
*path
,
1387 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
)
1389 struct ext4_extent_idx
*ix
;
1390 struct ext4_extent
*ex
;
1393 if (unlikely(path
== NULL
)) {
1394 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1397 depth
= path
->p_depth
;
1400 if (depth
== 0 && path
->p_ext
== NULL
)
1403 /* usually extent in the path covers blocks smaller
1404 * then *logical, but it can be that extent is the
1405 * first one in the file */
1407 ex
= path
[depth
].p_ext
;
1408 ee_len
= ext4_ext_get_actual_len(ex
);
1409 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1410 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1411 EXT4_ERROR_INODE(inode
,
1412 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1413 *logical
, le32_to_cpu(ex
->ee_block
));
1416 while (--depth
>= 0) {
1417 ix
= path
[depth
].p_idx
;
1418 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1419 EXT4_ERROR_INODE(inode
,
1420 "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1421 ix
!= NULL
? le32_to_cpu(ix
->ei_block
) : 0,
1422 EXT_FIRST_INDEX(path
[depth
].p_hdr
) != NULL
?
1423 le32_to_cpu(EXT_FIRST_INDEX(path
[depth
].p_hdr
)->ei_block
) : 0,
1431 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1432 EXT4_ERROR_INODE(inode
,
1433 "logical %d < ee_block %d + ee_len %d!",
1434 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1438 *logical
= le32_to_cpu(ex
->ee_block
) + ee_len
- 1;
1439 *phys
= ext4_ext_pblock(ex
) + ee_len
- 1;
1444 * search the closest allocated block to the right for *logical
1445 * and returns it at @logical + it's physical address at @phys
1446 * if *logical is the largest allocated block, the function
1447 * returns 0 at @phys
1448 * return value contains 0 (success) or error code
1450 static int ext4_ext_search_right(struct inode
*inode
,
1451 struct ext4_ext_path
*path
,
1452 ext4_lblk_t
*logical
, ext4_fsblk_t
*phys
,
1453 struct ext4_extent
**ret_ex
)
1455 struct buffer_head
*bh
= NULL
;
1456 struct ext4_extent_header
*eh
;
1457 struct ext4_extent_idx
*ix
;
1458 struct ext4_extent
*ex
;
1460 int depth
; /* Note, NOT eh_depth; depth from top of tree */
1463 if (unlikely(path
== NULL
)) {
1464 EXT4_ERROR_INODE(inode
, "path == NULL *logical %d!", *logical
);
1467 depth
= path
->p_depth
;
1470 if (depth
== 0 && path
->p_ext
== NULL
)
1473 /* usually extent in the path covers blocks smaller
1474 * then *logical, but it can be that extent is the
1475 * first one in the file */
1477 ex
= path
[depth
].p_ext
;
1478 ee_len
= ext4_ext_get_actual_len(ex
);
1479 if (*logical
< le32_to_cpu(ex
->ee_block
)) {
1480 if (unlikely(EXT_FIRST_EXTENT(path
[depth
].p_hdr
) != ex
)) {
1481 EXT4_ERROR_INODE(inode
,
1482 "first_extent(path[%d].p_hdr) != ex",
1486 while (--depth
>= 0) {
1487 ix
= path
[depth
].p_idx
;
1488 if (unlikely(ix
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))) {
1489 EXT4_ERROR_INODE(inode
,
1490 "ix != EXT_FIRST_INDEX *logical %d!",
1498 if (unlikely(*logical
< (le32_to_cpu(ex
->ee_block
) + ee_len
))) {
1499 EXT4_ERROR_INODE(inode
,
1500 "logical %d < ee_block %d + ee_len %d!",
1501 *logical
, le32_to_cpu(ex
->ee_block
), ee_len
);
1505 if (ex
!= EXT_LAST_EXTENT(path
[depth
].p_hdr
)) {
1506 /* next allocated block in this leaf */
1511 /* go up and search for index to the right */
1512 while (--depth
>= 0) {
1513 ix
= path
[depth
].p_idx
;
1514 if (ix
!= EXT_LAST_INDEX(path
[depth
].p_hdr
))
1518 /* we've gone up to the root and found no index to the right */
1522 /* we've found index to the right, let's
1523 * follow it and find the closest allocated
1524 * block to the right */
1526 block
= ext4_idx_pblock(ix
);
1527 while (++depth
< path
->p_depth
) {
1528 /* subtract from p_depth to get proper eh_depth */
1529 bh
= read_extent_tree_block(inode
, block
,
1530 path
->p_depth
- depth
, 0);
1533 eh
= ext_block_hdr(bh
);
1534 ix
= EXT_FIRST_INDEX(eh
);
1535 block
= ext4_idx_pblock(ix
);
1539 bh
= read_extent_tree_block(inode
, block
, path
->p_depth
- depth
, 0);
1542 eh
= ext_block_hdr(bh
);
1543 ex
= EXT_FIRST_EXTENT(eh
);
1545 *logical
= le32_to_cpu(ex
->ee_block
);
1546 *phys
= ext4_ext_pblock(ex
);
1554 * ext4_ext_next_allocated_block:
1555 * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1556 * NOTE: it considers block number from index entry as
1557 * allocated block. Thus, index entries have to be consistent
1561 ext4_ext_next_allocated_block(struct ext4_ext_path
*path
)
1565 BUG_ON(path
== NULL
);
1566 depth
= path
->p_depth
;
1568 if (depth
== 0 && path
->p_ext
== NULL
)
1569 return EXT_MAX_BLOCKS
;
1571 while (depth
>= 0) {
1572 if (depth
== path
->p_depth
) {
1574 if (path
[depth
].p_ext
&&
1575 path
[depth
].p_ext
!=
1576 EXT_LAST_EXTENT(path
[depth
].p_hdr
))
1577 return le32_to_cpu(path
[depth
].p_ext
[1].ee_block
);
1580 if (path
[depth
].p_idx
!=
1581 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1582 return le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1587 return EXT_MAX_BLOCKS
;
1591 * ext4_ext_next_leaf_block:
1592 * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1594 static ext4_lblk_t
ext4_ext_next_leaf_block(struct ext4_ext_path
*path
)
1598 BUG_ON(path
== NULL
);
1599 depth
= path
->p_depth
;
1601 /* zero-tree has no leaf blocks at all */
1603 return EXT_MAX_BLOCKS
;
1605 /* go to index block */
1608 while (depth
>= 0) {
1609 if (path
[depth
].p_idx
!=
1610 EXT_LAST_INDEX(path
[depth
].p_hdr
))
1611 return (ext4_lblk_t
)
1612 le32_to_cpu(path
[depth
].p_idx
[1].ei_block
);
1616 return EXT_MAX_BLOCKS
;
1620 * ext4_ext_correct_indexes:
1621 * if leaf gets modified and modified extent is first in the leaf,
1622 * then we have to correct all indexes above.
1623 * TODO: do we need to correct tree in all cases?
1625 static int ext4_ext_correct_indexes(handle_t
*handle
, struct inode
*inode
,
1626 struct ext4_ext_path
*path
)
1628 struct ext4_extent_header
*eh
;
1629 int depth
= ext_depth(inode
);
1630 struct ext4_extent
*ex
;
1634 eh
= path
[depth
].p_hdr
;
1635 ex
= path
[depth
].p_ext
;
1637 if (unlikely(ex
== NULL
|| eh
== NULL
)) {
1638 EXT4_ERROR_INODE(inode
,
1639 "ex %p == NULL or eh %p == NULL", ex
, eh
);
1644 /* there is no tree at all */
1648 if (ex
!= EXT_FIRST_EXTENT(eh
)) {
1649 /* we correct tree if first leaf got modified only */
1654 * TODO: we need correction if border is smaller than current one
1657 border
= path
[depth
].p_ext
->ee_block
;
1658 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1661 path
[k
].p_idx
->ei_block
= border
;
1662 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1667 /* change all left-side indexes */
1668 if (path
[k
+1].p_idx
!= EXT_FIRST_INDEX(path
[k
+1].p_hdr
))
1670 err
= ext4_ext_get_access(handle
, inode
, path
+ k
);
1673 path
[k
].p_idx
->ei_block
= border
;
1674 err
= ext4_ext_dirty(handle
, inode
, path
+ k
);
1683 ext4_can_extents_be_merged(struct inode
*inode
, struct ext4_extent
*ex1
,
1684 struct ext4_extent
*ex2
)
1686 unsigned short ext1_ee_len
, ext2_ee_len
;
1689 * Make sure that both extents are initialized. We don't merge
1690 * uninitialized extents so that we can be sure that end_io code has
1691 * the extent that was written properly split out and conversion to
1692 * initialized is trivial.
1694 if (ext4_ext_is_uninitialized(ex1
) || ext4_ext_is_uninitialized(ex2
))
1697 ext1_ee_len
= ext4_ext_get_actual_len(ex1
);
1698 ext2_ee_len
= ext4_ext_get_actual_len(ex2
);
1700 if (le32_to_cpu(ex1
->ee_block
) + ext1_ee_len
!=
1701 le32_to_cpu(ex2
->ee_block
))
1705 * To allow future support for preallocated extents to be added
1706 * as an RO_COMPAT feature, refuse to merge to extents if
1707 * this can result in the top bit of ee_len being set.
1709 if (ext1_ee_len
+ ext2_ee_len
> EXT_INIT_MAX_LEN
)
1711 #ifdef AGGRESSIVE_TEST
1712 if (ext1_ee_len
>= 4)
1716 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1722 * This function tries to merge the "ex" extent to the next extent in the tree.
1723 * It always tries to merge towards right. If you want to merge towards
1724 * left, pass "ex - 1" as argument instead of "ex".
1725 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1726 * 1 if they got merged.
1728 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1729 struct ext4_ext_path
*path
,
1730 struct ext4_extent
*ex
)
1732 struct ext4_extent_header
*eh
;
1733 unsigned int depth
, len
;
1736 depth
= ext_depth(inode
);
1737 BUG_ON(path
[depth
].p_hdr
== NULL
);
1738 eh
= path
[depth
].p_hdr
;
1740 while (ex
< EXT_LAST_EXTENT(eh
)) {
1741 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1743 /* merge with next extent! */
1744 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1745 + ext4_ext_get_actual_len(ex
+ 1));
1747 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1748 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1749 * sizeof(struct ext4_extent
);
1750 memmove(ex
+ 1, ex
+ 2, len
);
1752 le16_add_cpu(&eh
->eh_entries
, -1);
1754 WARN_ON(eh
->eh_entries
== 0);
1755 if (!eh
->eh_entries
)
1756 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1763 * This function does a very simple check to see if we can collapse
1764 * an extent tree with a single extent tree leaf block into the inode.
1766 static void ext4_ext_try_to_merge_up(handle_t
*handle
,
1767 struct inode
*inode
,
1768 struct ext4_ext_path
*path
)
1771 unsigned max_root
= ext4_ext_space_root(inode
, 0);
1774 if ((path
[0].p_depth
!= 1) ||
1775 (le16_to_cpu(path
[0].p_hdr
->eh_entries
) != 1) ||
1776 (le16_to_cpu(path
[1].p_hdr
->eh_entries
) > max_root
))
1780 * We need to modify the block allocation bitmap and the block
1781 * group descriptor to release the extent tree block. If we
1782 * can't get the journal credits, give up.
1784 if (ext4_journal_extend(handle
, 2))
1788 * Copy the extent data up to the inode
1790 blk
= ext4_idx_pblock(path
[0].p_idx
);
1791 s
= le16_to_cpu(path
[1].p_hdr
->eh_entries
) *
1792 sizeof(struct ext4_extent_idx
);
1793 s
+= sizeof(struct ext4_extent_header
);
1795 memcpy(path
[0].p_hdr
, path
[1].p_hdr
, s
);
1796 path
[0].p_depth
= 0;
1797 path
[0].p_ext
= EXT_FIRST_EXTENT(path
[0].p_hdr
) +
1798 (path
[1].p_ext
- EXT_FIRST_EXTENT(path
[1].p_hdr
));
1799 path
[0].p_hdr
->eh_max
= cpu_to_le16(max_root
);
1801 brelse(path
[1].p_bh
);
1802 ext4_free_blocks(handle
, inode
, NULL
, blk
, 1,
1803 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
|
1804 EXT4_FREE_BLOCKS_RESERVE
);
1808 * This function tries to merge the @ex extent to neighbours in the tree.
1809 * return 1 if merge left else 0.
1811 static void ext4_ext_try_to_merge(handle_t
*handle
,
1812 struct inode
*inode
,
1813 struct ext4_ext_path
*path
,
1814 struct ext4_extent
*ex
) {
1815 struct ext4_extent_header
*eh
;
1819 depth
= ext_depth(inode
);
1820 BUG_ON(path
[depth
].p_hdr
== NULL
);
1821 eh
= path
[depth
].p_hdr
;
1823 if (ex
> EXT_FIRST_EXTENT(eh
))
1824 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1827 (void) ext4_ext_try_to_merge_right(inode
, path
, ex
);
1829 ext4_ext_try_to_merge_up(handle
, inode
, path
);
1833 * check if a portion of the "newext" extent overlaps with an
1836 * If there is an overlap discovered, it updates the length of the newext
1837 * such that there will be no overlap, and then returns 1.
1838 * If there is no overlap found, it returns 0.
1840 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1841 struct inode
*inode
,
1842 struct ext4_extent
*newext
,
1843 struct ext4_ext_path
*path
)
1846 unsigned int depth
, len1
;
1847 unsigned int ret
= 0;
1849 b1
= le32_to_cpu(newext
->ee_block
);
1850 len1
= ext4_ext_get_actual_len(newext
);
1851 depth
= ext_depth(inode
);
1852 if (!path
[depth
].p_ext
)
1854 b2
= EXT4_LBLK_CMASK(sbi
, le32_to_cpu(path
[depth
].p_ext
->ee_block
));
1857 * get the next allocated block if the extent in the path
1858 * is before the requested block(s)
1861 b2
= ext4_ext_next_allocated_block(path
);
1862 if (b2
== EXT_MAX_BLOCKS
)
1864 b2
= EXT4_LBLK_CMASK(sbi
, b2
);
1867 /* check for wrap through zero on extent logical start block*/
1868 if (b1
+ len1
< b1
) {
1869 len1
= EXT_MAX_BLOCKS
- b1
;
1870 newext
->ee_len
= cpu_to_le16(len1
);
1874 /* check for overlap */
1875 if (b1
+ len1
> b2
) {
1876 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1884 * ext4_ext_insert_extent:
1885 * tries to merge requsted extent into the existing extent or
1886 * inserts requested extent as new one into the tree,
1887 * creating new leaf in the no-space case.
1889 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1890 struct ext4_ext_path
*path
,
1891 struct ext4_extent
*newext
, int gb_flags
)
1893 struct ext4_extent_header
*eh
;
1894 struct ext4_extent
*ex
, *fex
;
1895 struct ext4_extent
*nearex
; /* nearest extent */
1896 struct ext4_ext_path
*npath
= NULL
;
1897 int depth
, len
, err
;
1901 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1902 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1905 depth
= ext_depth(inode
);
1906 ex
= path
[depth
].p_ext
;
1907 eh
= path
[depth
].p_hdr
;
1908 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1909 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1913 /* try to insert block into found extent and return */
1914 if (ex
&& !(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
1917 * Try to see whether we should rather test the extent on
1918 * right from ex, or from the left of ex. This is because
1919 * ext4_ext_find_extent() can return either extent on the
1920 * left, or on the right from the searched position. This
1921 * will make merging more effective.
1923 if (ex
< EXT_LAST_EXTENT(eh
) &&
1924 (le32_to_cpu(ex
->ee_block
) +
1925 ext4_ext_get_actual_len(ex
) <
1926 le32_to_cpu(newext
->ee_block
))) {
1929 } else if ((ex
> EXT_FIRST_EXTENT(eh
)) &&
1930 (le32_to_cpu(newext
->ee_block
) +
1931 ext4_ext_get_actual_len(newext
) <
1932 le32_to_cpu(ex
->ee_block
)))
1935 /* Try to append newex to the ex */
1936 if (ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1937 ext_debug("append [%d]%d block to %u:[%d]%d"
1939 ext4_ext_is_uninitialized(newext
),
1940 ext4_ext_get_actual_len(newext
),
1941 le32_to_cpu(ex
->ee_block
),
1942 ext4_ext_is_uninitialized(ex
),
1943 ext4_ext_get_actual_len(ex
),
1944 ext4_ext_pblock(ex
));
1945 err
= ext4_ext_get_access(handle
, inode
,
1950 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1951 + ext4_ext_get_actual_len(newext
));
1952 eh
= path
[depth
].p_hdr
;
1958 /* Try to prepend newex to the ex */
1959 if (ext4_can_extents_be_merged(inode
, newext
, ex
)) {
1960 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1962 le32_to_cpu(newext
->ee_block
),
1963 ext4_ext_is_uninitialized(newext
),
1964 ext4_ext_get_actual_len(newext
),
1965 le32_to_cpu(ex
->ee_block
),
1966 ext4_ext_is_uninitialized(ex
),
1967 ext4_ext_get_actual_len(ex
),
1968 ext4_ext_pblock(ex
));
1969 err
= ext4_ext_get_access(handle
, inode
,
1974 ex
->ee_block
= newext
->ee_block
;
1975 ext4_ext_store_pblock(ex
, ext4_ext_pblock(newext
));
1976 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1977 + ext4_ext_get_actual_len(newext
));
1978 eh
= path
[depth
].p_hdr
;
1984 depth
= ext_depth(inode
);
1985 eh
= path
[depth
].p_hdr
;
1986 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
1989 /* probably next leaf has space for us? */
1990 fex
= EXT_LAST_EXTENT(eh
);
1991 next
= EXT_MAX_BLOCKS
;
1992 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
1993 next
= ext4_ext_next_leaf_block(path
);
1994 if (next
!= EXT_MAX_BLOCKS
) {
1995 ext_debug("next leaf block - %u\n", next
);
1996 BUG_ON(npath
!= NULL
);
1997 npath
= ext4_ext_find_extent(inode
, next
, NULL
, 0);
1999 return PTR_ERR(npath
);
2000 BUG_ON(npath
->p_depth
!= path
->p_depth
);
2001 eh
= npath
[depth
].p_hdr
;
2002 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
2003 ext_debug("next leaf isn't full(%d)\n",
2004 le16_to_cpu(eh
->eh_entries
));
2008 ext_debug("next leaf has no free space(%d,%d)\n",
2009 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
2013 * There is no free space in the found leaf.
2014 * We're gonna add a new leaf in the tree.
2016 if (gb_flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
2017 mb_flags
= EXT4_MB_USE_RESERVED
;
2018 err
= ext4_ext_create_new_leaf(handle
, inode
, mb_flags
, gb_flags
,
2022 depth
= ext_depth(inode
);
2023 eh
= path
[depth
].p_hdr
;
2026 nearex
= path
[depth
].p_ext
;
2028 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2033 /* there is no extent in this leaf, create first one */
2034 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2035 le32_to_cpu(newext
->ee_block
),
2036 ext4_ext_pblock(newext
),
2037 ext4_ext_is_uninitialized(newext
),
2038 ext4_ext_get_actual_len(newext
));
2039 nearex
= EXT_FIRST_EXTENT(eh
);
2041 if (le32_to_cpu(newext
->ee_block
)
2042 > le32_to_cpu(nearex
->ee_block
)) {
2044 ext_debug("insert %u:%llu:[%d]%d before: "
2046 le32_to_cpu(newext
->ee_block
),
2047 ext4_ext_pblock(newext
),
2048 ext4_ext_is_uninitialized(newext
),
2049 ext4_ext_get_actual_len(newext
),
2054 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
2055 ext_debug("insert %u:%llu:[%d]%d after: "
2057 le32_to_cpu(newext
->ee_block
),
2058 ext4_ext_pblock(newext
),
2059 ext4_ext_is_uninitialized(newext
),
2060 ext4_ext_get_actual_len(newext
),
2063 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
2065 ext_debug("insert %u:%llu:[%d]%d: "
2066 "move %d extents from 0x%p to 0x%p\n",
2067 le32_to_cpu(newext
->ee_block
),
2068 ext4_ext_pblock(newext
),
2069 ext4_ext_is_uninitialized(newext
),
2070 ext4_ext_get_actual_len(newext
),
2071 len
, nearex
, nearex
+ 1);
2072 memmove(nearex
+ 1, nearex
,
2073 len
* sizeof(struct ext4_extent
));
2077 le16_add_cpu(&eh
->eh_entries
, 1);
2078 path
[depth
].p_ext
= nearex
;
2079 nearex
->ee_block
= newext
->ee_block
;
2080 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
2081 nearex
->ee_len
= newext
->ee_len
;
2084 /* try to merge extents */
2085 if (!(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
))
2086 ext4_ext_try_to_merge(handle
, inode
, path
, nearex
);
2089 /* time to correct all indexes above */
2090 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2094 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
2098 ext4_ext_drop_refs(npath
);
2104 static int ext4_fill_fiemap_extents(struct inode
*inode
,
2105 ext4_lblk_t block
, ext4_lblk_t num
,
2106 struct fiemap_extent_info
*fieinfo
)
2108 struct ext4_ext_path
*path
= NULL
;
2109 struct ext4_extent
*ex
;
2110 struct extent_status es
;
2111 ext4_lblk_t next
, next_del
, start
= 0, end
= 0;
2112 ext4_lblk_t last
= block
+ num
;
2113 int exists
, depth
= 0, err
= 0;
2114 unsigned int flags
= 0;
2115 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
2117 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
2119 /* find extent for this block */
2120 down_read(&EXT4_I(inode
)->i_data_sem
);
2122 if (path
&& ext_depth(inode
) != depth
) {
2123 /* depth was changed. we have to realloc path */
2128 path
= ext4_ext_find_extent(inode
, block
, path
, 0);
2130 up_read(&EXT4_I(inode
)->i_data_sem
);
2131 err
= PTR_ERR(path
);
2136 depth
= ext_depth(inode
);
2137 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2138 up_read(&EXT4_I(inode
)->i_data_sem
);
2139 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2143 ex
= path
[depth
].p_ext
;
2144 next
= ext4_ext_next_allocated_block(path
);
2145 ext4_ext_drop_refs(path
);
2150 /* there is no extent yet, so try to allocate
2151 * all requested space */
2154 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
2155 /* need to allocate space before found extent */
2157 end
= le32_to_cpu(ex
->ee_block
);
2158 if (block
+ num
< end
)
2160 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2161 + ext4_ext_get_actual_len(ex
)) {
2162 /* need to allocate space after found extent */
2167 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
2169 * some part of requested space is covered
2173 end
= le32_to_cpu(ex
->ee_block
)
2174 + ext4_ext_get_actual_len(ex
);
2175 if (block
+ num
< end
)
2181 BUG_ON(end
<= start
);
2185 es
.es_len
= end
- start
;
2188 es
.es_lblk
= le32_to_cpu(ex
->ee_block
);
2189 es
.es_len
= ext4_ext_get_actual_len(ex
);
2190 es
.es_pblk
= ext4_ext_pblock(ex
);
2191 if (ext4_ext_is_uninitialized(ex
))
2192 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
2196 * Find delayed extent and update es accordingly. We call
2197 * it even in !exists case to find out whether es is the
2198 * last existing extent or not.
2200 next_del
= ext4_find_delayed_extent(inode
, &es
);
2201 if (!exists
&& next_del
) {
2203 flags
|= (FIEMAP_EXTENT_DELALLOC
|
2204 FIEMAP_EXTENT_UNKNOWN
);
2206 up_read(&EXT4_I(inode
)->i_data_sem
);
2208 if (unlikely(es
.es_len
== 0)) {
2209 EXT4_ERROR_INODE(inode
, "es.es_len == 0");
2215 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2216 * we need to check next == EXT_MAX_BLOCKS because it is
2217 * possible that an extent is with unwritten and delayed
2218 * status due to when an extent is delayed allocated and
2219 * is allocated by fallocate status tree will track both of
2222 * So we could return a unwritten and delayed extent, and
2223 * its block is equal to 'next'.
2225 if (next
== next_del
&& next
== EXT_MAX_BLOCKS
) {
2226 flags
|= FIEMAP_EXTENT_LAST
;
2227 if (unlikely(next_del
!= EXT_MAX_BLOCKS
||
2228 next
!= EXT_MAX_BLOCKS
)) {
2229 EXT4_ERROR_INODE(inode
,
2230 "next extent == %u, next "
2231 "delalloc extent = %u",
2239 err
= fiemap_fill_next_extent(fieinfo
,
2240 (__u64
)es
.es_lblk
<< blksize_bits
,
2241 (__u64
)es
.es_pblk
<< blksize_bits
,
2242 (__u64
)es
.es_len
<< blksize_bits
,
2252 block
= es
.es_lblk
+ es
.es_len
;
2256 ext4_ext_drop_refs(path
);
2264 * ext4_ext_put_gap_in_cache:
2265 * calculate boundaries of the gap that the requested block fits into
2266 * and cache this gap
2269 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
2272 int depth
= ext_depth(inode
);
2273 unsigned long len
= 0;
2274 ext4_lblk_t lblock
= 0;
2275 struct ext4_extent
*ex
;
2277 ex
= path
[depth
].p_ext
;
2280 * there is no extent yet, so gap is [0;-] and we
2283 ext_debug("cache gap(whole file):");
2284 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
2286 len
= le32_to_cpu(ex
->ee_block
) - block
;
2287 ext_debug("cache gap(before): %u [%u:%u]",
2289 le32_to_cpu(ex
->ee_block
),
2290 ext4_ext_get_actual_len(ex
));
2291 if (!ext4_find_delalloc_range(inode
, lblock
, lblock
+ len
- 1))
2292 ext4_es_insert_extent(inode
, lblock
, len
, ~0,
2293 EXTENT_STATUS_HOLE
);
2294 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2295 + ext4_ext_get_actual_len(ex
)) {
2297 lblock
= le32_to_cpu(ex
->ee_block
)
2298 + ext4_ext_get_actual_len(ex
);
2300 next
= ext4_ext_next_allocated_block(path
);
2301 ext_debug("cache gap(after): [%u:%u] %u",
2302 le32_to_cpu(ex
->ee_block
),
2303 ext4_ext_get_actual_len(ex
),
2305 BUG_ON(next
== lblock
);
2306 len
= next
- lblock
;
2307 if (!ext4_find_delalloc_range(inode
, lblock
, lblock
+ len
- 1))
2308 ext4_es_insert_extent(inode
, lblock
, len
, ~0,
2309 EXTENT_STATUS_HOLE
);
2314 ext_debug(" -> %u:%lu\n", lblock
, len
);
2319 * removes index from the index block.
2321 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2322 struct ext4_ext_path
*path
, int depth
)
2327 /* free index block */
2329 path
= path
+ depth
;
2330 leaf
= ext4_idx_pblock(path
->p_idx
);
2331 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2332 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2335 err
= ext4_ext_get_access(handle
, inode
, path
);
2339 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2340 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2341 len
*= sizeof(struct ext4_extent_idx
);
2342 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2345 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2346 err
= ext4_ext_dirty(handle
, inode
, path
);
2349 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2350 trace_ext4_ext_rm_idx(inode
, leaf
);
2352 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2353 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2355 while (--depth
>= 0) {
2356 if (path
->p_idx
!= EXT_FIRST_INDEX(path
->p_hdr
))
2359 err
= ext4_ext_get_access(handle
, inode
, path
);
2362 path
->p_idx
->ei_block
= (path
+1)->p_idx
->ei_block
;
2363 err
= ext4_ext_dirty(handle
, inode
, path
);
2371 * ext4_ext_calc_credits_for_single_extent:
2372 * This routine returns max. credits that needed to insert an extent
2373 * to the extent tree.
2374 * When pass the actual path, the caller should calculate credits
2377 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2378 struct ext4_ext_path
*path
)
2381 int depth
= ext_depth(inode
);
2384 /* probably there is space in leaf? */
2385 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2386 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2389 * There are some space in the leaf tree, no
2390 * need to account for leaf block credit
2392 * bitmaps and block group descriptor blocks
2393 * and other metadata blocks still need to be
2396 /* 1 bitmap, 1 block group descriptor */
2397 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2402 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2406 * How many index/leaf blocks need to change/allocate to add @extents extents?
2408 * If we add a single extent, then in the worse case, each tree level
2409 * index/leaf need to be changed in case of the tree split.
2411 * If more extents are inserted, they could cause the whole tree split more
2412 * than once, but this is really rare.
2414 int ext4_ext_index_trans_blocks(struct inode
*inode
, int extents
)
2419 /* If we are converting the inline data, only one is needed here. */
2420 if (ext4_has_inline_data(inode
))
2423 depth
= ext_depth(inode
);
2433 static inline int get_default_free_blocks_flags(struct inode
*inode
)
2435 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2436 return EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
;
2437 else if (ext4_should_journal_data(inode
))
2438 return EXT4_FREE_BLOCKS_FORGET
;
2442 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2443 struct ext4_extent
*ex
,
2444 long long *partial_cluster
,
2445 ext4_lblk_t from
, ext4_lblk_t to
)
2447 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2448 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2450 int flags
= get_default_free_blocks_flags(inode
);
2453 * For bigalloc file systems, we never free a partial cluster
2454 * at the beginning of the extent. Instead, we make a note
2455 * that we tried freeing the cluster, and check to see if we
2456 * need to free it on a subsequent call to ext4_remove_blocks,
2457 * or at the end of the ext4_truncate() operation.
2459 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2461 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2463 * If we have a partial cluster, and it's different from the
2464 * cluster of the last block, we need to explicitly free the
2465 * partial cluster here.
2467 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2468 if ((*partial_cluster
> 0) &&
2469 (EXT4_B2C(sbi
, pblk
) != *partial_cluster
)) {
2470 ext4_free_blocks(handle
, inode
, NULL
,
2471 EXT4_C2B(sbi
, *partial_cluster
),
2472 sbi
->s_cluster_ratio
, flags
);
2473 *partial_cluster
= 0;
2476 #ifdef EXTENTS_STATS
2478 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2479 spin_lock(&sbi
->s_ext_stats_lock
);
2480 sbi
->s_ext_blocks
+= ee_len
;
2481 sbi
->s_ext_extents
++;
2482 if (ee_len
< sbi
->s_ext_min
)
2483 sbi
->s_ext_min
= ee_len
;
2484 if (ee_len
> sbi
->s_ext_max
)
2485 sbi
->s_ext_max
= ee_len
;
2486 if (ext_depth(inode
) > sbi
->s_depth_max
)
2487 sbi
->s_depth_max
= ext_depth(inode
);
2488 spin_unlock(&sbi
->s_ext_stats_lock
);
2491 if (from
>= le32_to_cpu(ex
->ee_block
)
2492 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2495 unsigned int unaligned
;
2497 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2498 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2500 * Usually we want to free partial cluster at the end of the
2501 * extent, except for the situation when the cluster is still
2502 * used by any other extent (partial_cluster is negative).
2504 if (*partial_cluster
< 0 &&
2505 -(*partial_cluster
) == EXT4_B2C(sbi
, pblk
+ num
- 1))
2506 flags
|= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER
;
2508 ext_debug("free last %u blocks starting %llu partial %lld\n",
2509 num
, pblk
, *partial_cluster
);
2510 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2512 * If the block range to be freed didn't start at the
2513 * beginning of a cluster, and we removed the entire
2514 * extent and the cluster is not used by any other extent,
2515 * save the partial cluster here, since we might need to
2516 * delete if we determine that the truncate operation has
2517 * removed all of the blocks in the cluster.
2519 * On the other hand, if we did not manage to free the whole
2520 * extent, we have to mark the cluster as used (store negative
2521 * cluster number in partial_cluster).
2523 unaligned
= EXT4_PBLK_COFF(sbi
, pblk
);
2524 if (unaligned
&& (ee_len
== num
) &&
2525 (*partial_cluster
!= -((long long)EXT4_B2C(sbi
, pblk
))))
2526 *partial_cluster
= EXT4_B2C(sbi
, pblk
);
2528 *partial_cluster
= -((long long)EXT4_B2C(sbi
, pblk
));
2529 else if (*partial_cluster
> 0)
2530 *partial_cluster
= 0;
2532 ext4_error(sbi
->s_sb
, "strange request: removal(2) "
2533 "%u-%u from %u:%u\n",
2534 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2540 * ext4_ext_rm_leaf() Removes the extents associated with the
2541 * blocks appearing between "start" and "end", and splits the extents
2542 * if "start" and "end" appear in the same extent
2544 * @handle: The journal handle
2545 * @inode: The files inode
2546 * @path: The path to the leaf
2547 * @partial_cluster: The cluster which we'll have to free if all extents
2548 * has been released from it. It gets negative in case
2549 * that the cluster is still used.
2550 * @start: The first block to remove
2551 * @end: The last block to remove
2554 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2555 struct ext4_ext_path
*path
,
2556 long long *partial_cluster
,
2557 ext4_lblk_t start
, ext4_lblk_t end
)
2559 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2560 int err
= 0, correct_index
= 0;
2561 int depth
= ext_depth(inode
), credits
;
2562 struct ext4_extent_header
*eh
;
2565 ext4_lblk_t ex_ee_block
;
2566 unsigned short ex_ee_len
;
2567 unsigned uninitialized
= 0;
2568 struct ext4_extent
*ex
;
2571 /* the header must be checked already in ext4_ext_remove_space() */
2572 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2573 if (!path
[depth
].p_hdr
)
2574 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2575 eh
= path
[depth
].p_hdr
;
2576 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2577 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2580 /* find where to start removing */
2581 ex
= path
[depth
].p_ext
;
2583 ex
= EXT_LAST_EXTENT(eh
);
2585 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2586 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2588 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2590 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2591 ex_ee_block
+ ex_ee_len
> start
) {
2593 if (ext4_ext_is_uninitialized(ex
))
2598 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2599 uninitialized
, ex_ee_len
);
2600 path
[depth
].p_ext
= ex
;
2602 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2603 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2604 ex_ee_block
+ex_ee_len
- 1 : end
;
2606 ext_debug(" border %u:%u\n", a
, b
);
2608 /* If this extent is beyond the end of the hole, skip it */
2609 if (end
< ex_ee_block
) {
2611 * We're going to skip this extent and move to another,
2612 * so if this extent is not cluster aligned we have
2613 * to mark the current cluster as used to avoid
2614 * accidentally freeing it later on
2616 pblk
= ext4_ext_pblock(ex
);
2617 if (EXT4_PBLK_COFF(sbi
, pblk
))
2619 -((long long)EXT4_B2C(sbi
, pblk
));
2621 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2622 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2624 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2625 EXT4_ERROR_INODE(inode
,
2626 "can not handle truncate %u:%u "
2628 start
, end
, ex_ee_block
,
2629 ex_ee_block
+ ex_ee_len
- 1);
2632 } else if (a
!= ex_ee_block
) {
2633 /* remove tail of the extent */
2634 num
= a
- ex_ee_block
;
2636 /* remove whole extent: excellent! */
2640 * 3 for leaf, sb, and inode plus 2 (bmap and group
2641 * descriptor) for each block group; assume two block
2642 * groups plus ex_ee_len/blocks_per_block_group for
2645 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2646 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2648 credits
+= (ext_depth(inode
)) + 1;
2650 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2652 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2656 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2660 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2666 /* this extent is removed; mark slot entirely unused */
2667 ext4_ext_store_pblock(ex
, 0);
2669 ex
->ee_len
= cpu_to_le16(num
);
2671 * Do not mark uninitialized if all the blocks in the
2672 * extent have been removed.
2674 if (uninitialized
&& num
)
2675 ext4_ext_mark_uninitialized(ex
);
2677 * If the extent was completely released,
2678 * we need to remove it from the leaf
2681 if (end
!= EXT_MAX_BLOCKS
- 1) {
2683 * For hole punching, we need to scoot all the
2684 * extents up when an extent is removed so that
2685 * we dont have blank extents in the middle
2687 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2688 sizeof(struct ext4_extent
));
2690 /* Now get rid of the one at the end */
2691 memset(EXT_LAST_EXTENT(eh
), 0,
2692 sizeof(struct ext4_extent
));
2694 le16_add_cpu(&eh
->eh_entries
, -1);
2695 } else if (*partial_cluster
> 0)
2696 *partial_cluster
= 0;
2698 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2702 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2703 ext4_ext_pblock(ex
));
2705 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2706 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2709 if (correct_index
&& eh
->eh_entries
)
2710 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2713 * Free the partial cluster only if the current extent does not
2714 * reference it. Otherwise we might free used cluster.
2716 if (*partial_cluster
> 0 &&
2717 (EXT4_B2C(sbi
, ext4_ext_pblock(ex
) + ex_ee_len
- 1) !=
2718 *partial_cluster
)) {
2719 int flags
= get_default_free_blocks_flags(inode
);
2721 ext4_free_blocks(handle
, inode
, NULL
,
2722 EXT4_C2B(sbi
, *partial_cluster
),
2723 sbi
->s_cluster_ratio
, flags
);
2724 *partial_cluster
= 0;
2727 /* if this leaf is free, then we should
2728 * remove it from index block above */
2729 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2730 err
= ext4_ext_rm_idx(handle
, inode
, path
, depth
);
2737 * ext4_ext_more_to_rm:
2738 * returns 1 if current index has to be freed (even partial)
2741 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2743 BUG_ON(path
->p_idx
== NULL
);
2745 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2749 * if truncate on deeper level happened, it wasn't partial,
2750 * so we have to consider current index for truncation
2752 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2757 int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2760 struct super_block
*sb
= inode
->i_sb
;
2761 int depth
= ext_depth(inode
);
2762 struct ext4_ext_path
*path
= NULL
;
2763 long long partial_cluster
= 0;
2767 ext_debug("truncate since %u to %u\n", start
, end
);
2769 /* probably first extent we're gonna free will be last in block */
2770 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, depth
+ 1);
2772 return PTR_ERR(handle
);
2775 trace_ext4_ext_remove_space(inode
, start
, end
, depth
);
2778 * Check if we are removing extents inside the extent tree. If that
2779 * is the case, we are going to punch a hole inside the extent tree
2780 * so we have to check whether we need to split the extent covering
2781 * the last block to remove so we can easily remove the part of it
2782 * in ext4_ext_rm_leaf().
2784 if (end
< EXT_MAX_BLOCKS
- 1) {
2785 struct ext4_extent
*ex
;
2786 ext4_lblk_t ee_block
;
2788 /* find extent for this block */
2789 path
= ext4_ext_find_extent(inode
, end
, NULL
, EXT4_EX_NOCACHE
);
2791 ext4_journal_stop(handle
);
2792 return PTR_ERR(path
);
2794 depth
= ext_depth(inode
);
2795 /* Leaf not may not exist only if inode has no blocks at all */
2796 ex
= path
[depth
].p_ext
;
2799 EXT4_ERROR_INODE(inode
,
2800 "path[%d].p_hdr == NULL",
2807 ee_block
= le32_to_cpu(ex
->ee_block
);
2810 * See if the last block is inside the extent, if so split
2811 * the extent at 'end' block so we can easily remove the
2812 * tail of the first part of the split extent in
2813 * ext4_ext_rm_leaf().
2815 if (end
>= ee_block
&&
2816 end
< ee_block
+ ext4_ext_get_actual_len(ex
) - 1) {
2819 if (ext4_ext_is_uninitialized(ex
))
2820 split_flag
= EXT4_EXT_MARK_UNINIT1
|
2821 EXT4_EXT_MARK_UNINIT2
;
2824 * Split the extent in two so that 'end' is the last
2825 * block in the first new extent. Also we should not
2826 * fail removing space due to ENOSPC so try to use
2827 * reserved block if that happens.
2829 err
= ext4_split_extent_at(handle
, inode
, path
,
2830 end
+ 1, split_flag
,
2832 EXT4_GET_BLOCKS_PRE_IO
|
2833 EXT4_GET_BLOCKS_METADATA_NOFAIL
);
2840 * We start scanning from right side, freeing all the blocks
2841 * after i_size and walking into the tree depth-wise.
2843 depth
= ext_depth(inode
);
2848 le16_to_cpu(path
[k
].p_hdr
->eh_entries
)+1;
2850 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
2853 ext4_journal_stop(handle
);
2856 path
[0].p_depth
= depth
;
2857 path
[0].p_hdr
= ext_inode_hdr(inode
);
2860 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0)) {
2867 while (i
>= 0 && err
== 0) {
2869 /* this is leaf block */
2870 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2871 &partial_cluster
, start
,
2873 /* root level has p_bh == NULL, brelse() eats this */
2874 brelse(path
[i
].p_bh
);
2875 path
[i
].p_bh
= NULL
;
2880 /* this is index block */
2881 if (!path
[i
].p_hdr
) {
2882 ext_debug("initialize header\n");
2883 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2886 if (!path
[i
].p_idx
) {
2887 /* this level hasn't been touched yet */
2888 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2889 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2890 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2892 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2894 /* we were already here, see at next index */
2898 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2899 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2901 if (ext4_ext_more_to_rm(path
+ i
)) {
2902 struct buffer_head
*bh
;
2903 /* go to the next level */
2904 ext_debug("move to level %d (block %llu)\n",
2905 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
2906 memset(path
+ i
+ 1, 0, sizeof(*path
));
2907 bh
= read_extent_tree_block(inode
,
2908 ext4_idx_pblock(path
[i
].p_idx
), depth
- i
- 1,
2911 /* should we reset i_size? */
2915 /* Yield here to deal with large extent trees.
2916 * Should be a no-op if we did IO above. */
2918 if (WARN_ON(i
+ 1 > depth
)) {
2922 path
[i
+ 1].p_bh
= bh
;
2924 /* save actual number of indexes since this
2925 * number is changed at the next iteration */
2926 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2929 /* we finished processing this index, go up */
2930 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2931 /* index is empty, remove it;
2932 * handle must be already prepared by the
2933 * truncatei_leaf() */
2934 err
= ext4_ext_rm_idx(handle
, inode
, path
, i
);
2936 /* root level has p_bh == NULL, brelse() eats this */
2937 brelse(path
[i
].p_bh
);
2938 path
[i
].p_bh
= NULL
;
2940 ext_debug("return to level %d\n", i
);
2944 trace_ext4_ext_remove_space_done(inode
, start
, end
, depth
,
2945 partial_cluster
, path
->p_hdr
->eh_entries
);
2947 /* If we still have something in the partial cluster and we have removed
2948 * even the first extent, then we should free the blocks in the partial
2949 * cluster as well. */
2950 if (partial_cluster
> 0 && path
->p_hdr
->eh_entries
== 0) {
2951 int flags
= get_default_free_blocks_flags(inode
);
2953 ext4_free_blocks(handle
, inode
, NULL
,
2954 EXT4_C2B(EXT4_SB(sb
), partial_cluster
),
2955 EXT4_SB(sb
)->s_cluster_ratio
, flags
);
2956 partial_cluster
= 0;
2959 /* TODO: flexible tree reduction should be here */
2960 if (path
->p_hdr
->eh_entries
== 0) {
2962 * truncate to zero freed all the tree,
2963 * so we need to correct eh_depth
2965 err
= ext4_ext_get_access(handle
, inode
, path
);
2967 ext_inode_hdr(inode
)->eh_depth
= 0;
2968 ext_inode_hdr(inode
)->eh_max
=
2969 cpu_to_le16(ext4_ext_space_root(inode
, 0));
2970 err
= ext4_ext_dirty(handle
, inode
, path
);
2974 ext4_ext_drop_refs(path
);
2976 if (err
== -EAGAIN
) {
2980 ext4_journal_stop(handle
);
2986 * called at mount time
2988 void ext4_ext_init(struct super_block
*sb
)
2991 * possible initialization would be here
2994 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
2995 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
2996 printk(KERN_INFO
"EXT4-fs: file extents enabled"
2997 #ifdef AGGRESSIVE_TEST
2998 ", aggressive tests"
3000 #ifdef CHECK_BINSEARCH
3003 #ifdef EXTENTS_STATS
3008 #ifdef EXTENTS_STATS
3009 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
3010 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
3011 EXT4_SB(sb
)->s_ext_max
= 0;
3017 * called at umount time
3019 void ext4_ext_release(struct super_block
*sb
)
3021 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
3024 #ifdef EXTENTS_STATS
3025 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
3026 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3027 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3028 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
3029 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
3030 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3031 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
3036 static int ext4_zeroout_es(struct inode
*inode
, struct ext4_extent
*ex
)
3038 ext4_lblk_t ee_block
;
3039 ext4_fsblk_t ee_pblock
;
3040 unsigned int ee_len
;
3042 ee_block
= le32_to_cpu(ex
->ee_block
);
3043 ee_len
= ext4_ext_get_actual_len(ex
);
3044 ee_pblock
= ext4_ext_pblock(ex
);
3049 return ext4_es_insert_extent(inode
, ee_block
, ee_len
, ee_pblock
,
3050 EXTENT_STATUS_WRITTEN
);
3053 /* FIXME!! we need to try to merge to left or right after zero-out */
3054 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
3056 ext4_fsblk_t ee_pblock
;
3057 unsigned int ee_len
;
3060 ee_len
= ext4_ext_get_actual_len(ex
);
3061 ee_pblock
= ext4_ext_pblock(ex
);
3063 ret
= sb_issue_zeroout(inode
->i_sb
, ee_pblock
, ee_len
, GFP_NOFS
);
3071 * ext4_split_extent_at() splits an extent at given block.
3073 * @handle: the journal handle
3074 * @inode: the file inode
3075 * @path: the path to the extent
3076 * @split: the logical block where the extent is splitted.
3077 * @split_flags: indicates if the extent could be zeroout if split fails, and
3078 * the states(init or uninit) of new extents.
3079 * @flags: flags used to insert new extent to extent tree.
3082 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3083 * of which are deterimined by split_flag.
3085 * There are two cases:
3086 * a> the extent are splitted into two extent.
3087 * b> split is not needed, and just mark the extent.
3089 * return 0 on success.
3091 static int ext4_split_extent_at(handle_t
*handle
,
3092 struct inode
*inode
,
3093 struct ext4_ext_path
*path
,
3098 ext4_fsblk_t newblock
;
3099 ext4_lblk_t ee_block
;
3100 struct ext4_extent
*ex
, newex
, orig_ex
, zero_ex
;
3101 struct ext4_extent
*ex2
= NULL
;
3102 unsigned int ee_len
, depth
;
3105 BUG_ON((split_flag
& (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
)) ==
3106 (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
));
3108 ext_debug("ext4_split_extents_at: inode %lu, logical"
3109 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
3111 ext4_ext_show_leaf(inode
, path
);
3113 depth
= ext_depth(inode
);
3114 ex
= path
[depth
].p_ext
;
3115 ee_block
= le32_to_cpu(ex
->ee_block
);
3116 ee_len
= ext4_ext_get_actual_len(ex
);
3117 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
3119 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
3120 BUG_ON(!ext4_ext_is_uninitialized(ex
) &&
3121 split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3122 EXT4_EXT_MARK_UNINIT1
|
3123 EXT4_EXT_MARK_UNINIT2
));
3125 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3129 if (split
== ee_block
) {
3131 * case b: block @split is the block that the extent begins with
3132 * then we just change the state of the extent, and splitting
3135 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3136 ext4_ext_mark_uninitialized(ex
);
3138 ext4_ext_mark_initialized(ex
);
3140 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
3141 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3143 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3148 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
3149 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
3150 if (split_flag
& EXT4_EXT_MARK_UNINIT1
)
3151 ext4_ext_mark_uninitialized(ex
);
3154 * path may lead to new leaf, not to original leaf any more
3155 * after ext4_ext_insert_extent() returns,
3157 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3159 goto fix_extent_len
;
3162 ex2
->ee_block
= cpu_to_le32(split
);
3163 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
3164 ext4_ext_store_pblock(ex2
, newblock
);
3165 if (split_flag
& EXT4_EXT_MARK_UNINIT2
)
3166 ext4_ext_mark_uninitialized(ex2
);
3168 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
3169 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3170 if (split_flag
& (EXT4_EXT_DATA_VALID1
|EXT4_EXT_DATA_VALID2
)) {
3171 if (split_flag
& EXT4_EXT_DATA_VALID1
) {
3172 err
= ext4_ext_zeroout(inode
, ex2
);
3173 zero_ex
.ee_block
= ex2
->ee_block
;
3174 zero_ex
.ee_len
= cpu_to_le16(
3175 ext4_ext_get_actual_len(ex2
));
3176 ext4_ext_store_pblock(&zero_ex
,
3177 ext4_ext_pblock(ex2
));
3179 err
= ext4_ext_zeroout(inode
, ex
);
3180 zero_ex
.ee_block
= ex
->ee_block
;
3181 zero_ex
.ee_len
= cpu_to_le16(
3182 ext4_ext_get_actual_len(ex
));
3183 ext4_ext_store_pblock(&zero_ex
,
3184 ext4_ext_pblock(ex
));
3187 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3188 zero_ex
.ee_block
= orig_ex
.ee_block
;
3189 zero_ex
.ee_len
= cpu_to_le16(
3190 ext4_ext_get_actual_len(&orig_ex
));
3191 ext4_ext_store_pblock(&zero_ex
,
3192 ext4_ext_pblock(&orig_ex
));
3196 goto fix_extent_len
;
3197 /* update the extent length and mark as initialized */
3198 ex
->ee_len
= cpu_to_le16(ee_len
);
3199 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3200 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3202 goto fix_extent_len
;
3204 /* update extent status tree */
3205 err
= ext4_zeroout_es(inode
, &zero_ex
);
3209 goto fix_extent_len
;
3212 ext4_ext_show_leaf(inode
, path
);
3216 ex
->ee_len
= orig_ex
.ee_len
;
3217 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3222 * ext4_split_extents() splits an extent and mark extent which is covered
3223 * by @map as split_flags indicates
3225 * It may result in splitting the extent into multiple extents (up to three)
3226 * There are three possibilities:
3227 * a> There is no split required
3228 * b> Splits in two extents: Split is happening at either end of the extent
3229 * c> Splits in three extents: Somone is splitting in middle of the extent
3232 static int ext4_split_extent(handle_t
*handle
,
3233 struct inode
*inode
,
3234 struct ext4_ext_path
*path
,
3235 struct ext4_map_blocks
*map
,
3239 ext4_lblk_t ee_block
;
3240 struct ext4_extent
*ex
;
3241 unsigned int ee_len
, depth
;
3244 int split_flag1
, flags1
;
3245 int allocated
= map
->m_len
;
3247 depth
= ext_depth(inode
);
3248 ex
= path
[depth
].p_ext
;
3249 ee_block
= le32_to_cpu(ex
->ee_block
);
3250 ee_len
= ext4_ext_get_actual_len(ex
);
3251 uninitialized
= ext4_ext_is_uninitialized(ex
);
3253 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
3254 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
;
3255 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
3257 split_flag1
|= EXT4_EXT_MARK_UNINIT1
|
3258 EXT4_EXT_MARK_UNINIT2
;
3259 if (split_flag
& EXT4_EXT_DATA_VALID2
)
3260 split_flag1
|= EXT4_EXT_DATA_VALID1
;
3261 err
= ext4_split_extent_at(handle
, inode
, path
,
3262 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
3266 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3269 * Update path is required because previous ext4_split_extent_at() may
3270 * result in split of original leaf or extent zeroout.
3272 ext4_ext_drop_refs(path
);
3273 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
, 0);
3275 return PTR_ERR(path
);
3276 depth
= ext_depth(inode
);
3277 ex
= path
[depth
].p_ext
;
3278 uninitialized
= ext4_ext_is_uninitialized(ex
);
3281 if (map
->m_lblk
>= ee_block
) {
3282 split_flag1
= split_flag
& EXT4_EXT_DATA_VALID2
;
3283 if (uninitialized
) {
3284 split_flag1
|= EXT4_EXT_MARK_UNINIT1
;
3285 split_flag1
|= split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3286 EXT4_EXT_MARK_UNINIT2
);
3288 err
= ext4_split_extent_at(handle
, inode
, path
,
3289 map
->m_lblk
, split_flag1
, flags
);
3294 ext4_ext_show_leaf(inode
, path
);
3296 return err
? err
: allocated
;
3300 * This function is called by ext4_ext_map_blocks() if someone tries to write
3301 * to an uninitialized extent. It may result in splitting the uninitialized
3302 * extent into multiple extents (up to three - one initialized and two
3304 * There are three possibilities:
3305 * a> There is no split required: Entire extent should be initialized
3306 * b> Splits in two extents: Write is happening at either end of the extent
3307 * c> Splits in three extents: Somone is writing in middle of the extent
3310 * - The extent pointed to by 'path' is uninitialized.
3311 * - The extent pointed to by 'path' contains a superset
3312 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3314 * Post-conditions on success:
3315 * - the returned value is the number of blocks beyond map->l_lblk
3316 * that are allocated and initialized.
3317 * It is guaranteed to be >= map->m_len.
3319 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3320 struct inode
*inode
,
3321 struct ext4_map_blocks
*map
,
3322 struct ext4_ext_path
*path
,
3325 struct ext4_sb_info
*sbi
;
3326 struct ext4_extent_header
*eh
;
3327 struct ext4_map_blocks split_map
;
3328 struct ext4_extent zero_ex
;
3329 struct ext4_extent
*ex
, *abut_ex
;
3330 ext4_lblk_t ee_block
, eof_block
;
3331 unsigned int ee_len
, depth
, map_len
= map
->m_len
;
3332 int allocated
= 0, max_zeroout
= 0;
3336 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3337 "block %llu, max_blocks %u\n", inode
->i_ino
,
3338 (unsigned long long)map
->m_lblk
, map_len
);
3340 sbi
= EXT4_SB(inode
->i_sb
);
3341 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3342 inode
->i_sb
->s_blocksize_bits
;
3343 if (eof_block
< map
->m_lblk
+ map_len
)
3344 eof_block
= map
->m_lblk
+ map_len
;
3346 depth
= ext_depth(inode
);
3347 eh
= path
[depth
].p_hdr
;
3348 ex
= path
[depth
].p_ext
;
3349 ee_block
= le32_to_cpu(ex
->ee_block
);
3350 ee_len
= ext4_ext_get_actual_len(ex
);
3353 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3355 /* Pre-conditions */
3356 BUG_ON(!ext4_ext_is_uninitialized(ex
));
3357 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3360 * Attempt to transfer newly initialized blocks from the currently
3361 * uninitialized extent to its neighbor. This is much cheaper
3362 * than an insertion followed by a merge as those involve costly
3363 * memmove() calls. Transferring to the left is the common case in
3364 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3365 * followed by append writes.
3367 * Limitations of the current logic:
3368 * - L1: we do not deal with writes covering the whole extent.
3369 * This would require removing the extent if the transfer
3371 * - L2: we only attempt to merge with an extent stored in the
3372 * same extent tree node.
3374 if ((map
->m_lblk
== ee_block
) &&
3375 /* See if we can merge left */
3376 (map_len
< ee_len
) && /*L1*/
3377 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L2*/
3378 ext4_lblk_t prev_lblk
;
3379 ext4_fsblk_t prev_pblk
, ee_pblk
;
3380 unsigned int prev_len
;
3383 prev_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3384 prev_len
= ext4_ext_get_actual_len(abut_ex
);
3385 prev_pblk
= ext4_ext_pblock(abut_ex
);
3386 ee_pblk
= ext4_ext_pblock(ex
);
3389 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3390 * upon those conditions:
3391 * - C1: abut_ex is initialized,
3392 * - C2: abut_ex is logically abutting ex,
3393 * - C3: abut_ex is physically abutting ex,
3394 * - C4: abut_ex can receive the additional blocks without
3395 * overflowing the (initialized) length limit.
3397 if ((!ext4_ext_is_uninitialized(abut_ex
)) && /*C1*/
3398 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3399 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3400 (prev_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3401 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3405 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3408 /* Shift the start of ex by 'map_len' blocks */
3409 ex
->ee_block
= cpu_to_le32(ee_block
+ map_len
);
3410 ext4_ext_store_pblock(ex
, ee_pblk
+ map_len
);
3411 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3412 ext4_ext_mark_uninitialized(ex
); /* Restore the flag */
3414 /* Extend abut_ex by 'map_len' blocks */
3415 abut_ex
->ee_len
= cpu_to_le16(prev_len
+ map_len
);
3417 /* Result: number of initialized blocks past m_lblk */
3418 allocated
= map_len
;
3420 } else if (((map
->m_lblk
+ map_len
) == (ee_block
+ ee_len
)) &&
3421 (map_len
< ee_len
) && /*L1*/
3422 ex
< EXT_LAST_EXTENT(eh
)) { /*L2*/
3423 /* See if we can merge right */
3424 ext4_lblk_t next_lblk
;
3425 ext4_fsblk_t next_pblk
, ee_pblk
;
3426 unsigned int next_len
;
3429 next_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3430 next_len
= ext4_ext_get_actual_len(abut_ex
);
3431 next_pblk
= ext4_ext_pblock(abut_ex
);
3432 ee_pblk
= ext4_ext_pblock(ex
);
3435 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3436 * upon those conditions:
3437 * - C1: abut_ex is initialized,
3438 * - C2: abut_ex is logically abutting ex,
3439 * - C3: abut_ex is physically abutting ex,
3440 * - C4: abut_ex can receive the additional blocks without
3441 * overflowing the (initialized) length limit.
3443 if ((!ext4_ext_is_uninitialized(abut_ex
)) && /*C1*/
3444 ((map
->m_lblk
+ map_len
) == next_lblk
) && /*C2*/
3445 ((ee_pblk
+ ee_len
) == next_pblk
) && /*C3*/
3446 (next_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3447 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3451 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3454 /* Shift the start of abut_ex by 'map_len' blocks */
3455 abut_ex
->ee_block
= cpu_to_le32(next_lblk
- map_len
);
3456 ext4_ext_store_pblock(abut_ex
, next_pblk
- map_len
);
3457 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3458 ext4_ext_mark_uninitialized(ex
); /* Restore the flag */
3460 /* Extend abut_ex by 'map_len' blocks */
3461 abut_ex
->ee_len
= cpu_to_le16(next_len
+ map_len
);
3463 /* Result: number of initialized blocks past m_lblk */
3464 allocated
= map_len
;
3468 /* Mark the block containing both extents as dirty */
3469 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3471 /* Update path to point to the right extent */
3472 path
[depth
].p_ext
= abut_ex
;
3475 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3477 WARN_ON(map
->m_lblk
< ee_block
);
3479 * It is safe to convert extent to initialized via explicit
3480 * zeroout only if extent is fully insde i_size or new_size.
3482 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3484 if (EXT4_EXT_MAY_ZEROOUT
& split_flag
)
3485 max_zeroout
= sbi
->s_extent_max_zeroout_kb
>>
3486 (inode
->i_sb
->s_blocksize_bits
- 10);
3488 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3489 if (max_zeroout
&& (ee_len
<= max_zeroout
)) {
3490 err
= ext4_ext_zeroout(inode
, ex
);
3493 zero_ex
.ee_block
= ex
->ee_block
;
3494 zero_ex
.ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
));
3495 ext4_ext_store_pblock(&zero_ex
, ext4_ext_pblock(ex
));
3497 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3500 ext4_ext_mark_initialized(ex
);
3501 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3502 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3508 * 1. split the extent into three extents.
3509 * 2. split the extent into two extents, zeroout the first half.
3510 * 3. split the extent into two extents, zeroout the second half.
3511 * 4. split the extent into two extents with out zeroout.
3513 split_map
.m_lblk
= map
->m_lblk
;
3514 split_map
.m_len
= map
->m_len
;
3516 if (max_zeroout
&& (allocated
> map
->m_len
)) {
3517 if (allocated
<= max_zeroout
) {
3520 cpu_to_le32(map
->m_lblk
);
3521 zero_ex
.ee_len
= cpu_to_le16(allocated
);
3522 ext4_ext_store_pblock(&zero_ex
,
3523 ext4_ext_pblock(ex
) + map
->m_lblk
- ee_block
);
3524 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3527 split_map
.m_lblk
= map
->m_lblk
;
3528 split_map
.m_len
= allocated
;
3529 } else if (map
->m_lblk
- ee_block
+ map
->m_len
< max_zeroout
) {
3531 if (map
->m_lblk
!= ee_block
) {
3532 zero_ex
.ee_block
= ex
->ee_block
;
3533 zero_ex
.ee_len
= cpu_to_le16(map
->m_lblk
-
3535 ext4_ext_store_pblock(&zero_ex
,
3536 ext4_ext_pblock(ex
));
3537 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3542 split_map
.m_lblk
= ee_block
;
3543 split_map
.m_len
= map
->m_lblk
- ee_block
+ map
->m_len
;
3544 allocated
= map
->m_len
;
3548 allocated
= ext4_split_extent(handle
, inode
, path
,
3549 &split_map
, split_flag
, flags
);
3554 /* If we have gotten a failure, don't zero out status tree */
3556 err
= ext4_zeroout_es(inode
, &zero_ex
);
3557 return err
? err
: allocated
;
3561 * This function is called by ext4_ext_map_blocks() from
3562 * ext4_get_blocks_dio_write() when DIO to write
3563 * to an uninitialized extent.
3565 * Writing to an uninitialized extent may result in splitting the uninitialized
3566 * extent into multiple initialized/uninitialized extents (up to three)
3567 * There are three possibilities:
3568 * a> There is no split required: Entire extent should be uninitialized
3569 * b> Splits in two extents: Write is happening at either end of the extent
3570 * c> Splits in three extents: Somone is writing in middle of the extent
3572 * One of more index blocks maybe needed if the extent tree grow after
3573 * the uninitialized extent split. To prevent ENOSPC occur at the IO
3574 * complete, we need to split the uninitialized extent before DIO submit
3575 * the IO. The uninitialized extent called at this time will be split
3576 * into three uninitialized extent(at most). After IO complete, the part
3577 * being filled will be convert to initialized by the end_io callback function
3578 * via ext4_convert_unwritten_extents().
3580 * Returns the size of uninitialized extent to be written on success.
3582 static int ext4_split_unwritten_extents(handle_t
*handle
,
3583 struct inode
*inode
,
3584 struct ext4_map_blocks
*map
,
3585 struct ext4_ext_path
*path
,
3588 ext4_lblk_t eof_block
;
3589 ext4_lblk_t ee_block
;
3590 struct ext4_extent
*ex
;
3591 unsigned int ee_len
;
3592 int split_flag
= 0, depth
;
3594 ext_debug("ext4_split_unwritten_extents: inode %lu, logical"
3595 "block %llu, max_blocks %u\n", inode
->i_ino
,
3596 (unsigned long long)map
->m_lblk
, map
->m_len
);
3598 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3599 inode
->i_sb
->s_blocksize_bits
;
3600 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3601 eof_block
= map
->m_lblk
+ map
->m_len
;
3603 * It is safe to convert extent to initialized via explicit
3604 * zeroout only if extent is fully insde i_size or new_size.
3606 depth
= ext_depth(inode
);
3607 ex
= path
[depth
].p_ext
;
3608 ee_block
= le32_to_cpu(ex
->ee_block
);
3609 ee_len
= ext4_ext_get_actual_len(ex
);
3611 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3612 split_flag
|= EXT4_EXT_MARK_UNINIT2
;
3613 if (flags
& EXT4_GET_BLOCKS_CONVERT
)
3614 split_flag
|= EXT4_EXT_DATA_VALID2
;
3615 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3616 return ext4_split_extent(handle
, inode
, path
, map
, split_flag
, flags
);
3619 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3620 struct inode
*inode
,
3621 struct ext4_map_blocks
*map
,
3622 struct ext4_ext_path
*path
)
3624 struct ext4_extent
*ex
;
3625 ext4_lblk_t ee_block
;
3626 unsigned int ee_len
;
3630 depth
= ext_depth(inode
);
3631 ex
= path
[depth
].p_ext
;
3632 ee_block
= le32_to_cpu(ex
->ee_block
);
3633 ee_len
= ext4_ext_get_actual_len(ex
);
3635 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3636 "block %llu, max_blocks %u\n", inode
->i_ino
,
3637 (unsigned long long)ee_block
, ee_len
);
3639 /* If extent is larger than requested it is a clear sign that we still
3640 * have some extent state machine issues left. So extent_split is still
3642 * TODO: Once all related issues will be fixed this situation should be
3645 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3647 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3648 " len %u; IO logical block %llu, len %u\n",
3649 inode
->i_ino
, (unsigned long long)ee_block
, ee_len
,
3650 (unsigned long long)map
->m_lblk
, map
->m_len
);
3652 err
= ext4_split_unwritten_extents(handle
, inode
, map
, path
,
3653 EXT4_GET_BLOCKS_CONVERT
);
3656 ext4_ext_drop_refs(path
);
3657 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
, 0);
3659 err
= PTR_ERR(path
);
3662 depth
= ext_depth(inode
);
3663 ex
= path
[depth
].p_ext
;
3666 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3669 /* first mark the extent as initialized */
3670 ext4_ext_mark_initialized(ex
);
3672 /* note: ext4_ext_correct_indexes() isn't needed here because
3673 * borders are not changed
3675 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3677 /* Mark modified extent as dirty */
3678 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3680 ext4_ext_show_leaf(inode
, path
);
3684 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3685 sector_t block
, int count
)
3688 for (i
= 0; i
< count
; i
++)
3689 unmap_underlying_metadata(bdev
, block
+ i
);
3693 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3695 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3697 struct ext4_ext_path
*path
,
3701 struct ext4_extent_header
*eh
;
3702 struct ext4_extent
*last_ex
;
3704 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3707 depth
= ext_depth(inode
);
3708 eh
= path
[depth
].p_hdr
;
3711 * We're going to remove EOFBLOCKS_FL entirely in future so we
3712 * do not care for this case anymore. Simply remove the flag
3713 * if there are no extents.
3715 if (unlikely(!eh
->eh_entries
))
3717 last_ex
= EXT_LAST_EXTENT(eh
);
3719 * We should clear the EOFBLOCKS_FL flag if we are writing the
3720 * last block in the last extent in the file. We test this by
3721 * first checking to see if the caller to
3722 * ext4_ext_get_blocks() was interested in the last block (or
3723 * a block beyond the last block) in the current extent. If
3724 * this turns out to be false, we can bail out from this
3725 * function immediately.
3727 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3728 ext4_ext_get_actual_len(last_ex
))
3731 * If the caller does appear to be planning to write at or
3732 * beyond the end of the current extent, we then test to see
3733 * if the current extent is the last extent in the file, by
3734 * checking to make sure it was reached via the rightmost node
3735 * at each level of the tree.
3737 for (i
= depth
-1; i
>= 0; i
--)
3738 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3741 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3742 return ext4_mark_inode_dirty(handle
, inode
);
3746 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3748 * Return 1 if there is a delalloc block in the range, otherwise 0.
3750 int ext4_find_delalloc_range(struct inode
*inode
,
3751 ext4_lblk_t lblk_start
,
3752 ext4_lblk_t lblk_end
)
3754 struct extent_status es
;
3756 ext4_es_find_delayed_extent_range(inode
, lblk_start
, lblk_end
, &es
);
3758 return 0; /* there is no delay extent in this tree */
3759 else if (es
.es_lblk
<= lblk_start
&&
3760 lblk_start
< es
.es_lblk
+ es
.es_len
)
3762 else if (lblk_start
<= es
.es_lblk
&& es
.es_lblk
<= lblk_end
)
3768 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
)
3770 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3771 ext4_lblk_t lblk_start
, lblk_end
;
3772 lblk_start
= EXT4_LBLK_CMASK(sbi
, lblk
);
3773 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3775 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
);
3779 * Determines how many complete clusters (out of those specified by the 'map')
3780 * are under delalloc and were reserved quota for.
3781 * This function is called when we are writing out the blocks that were
3782 * originally written with their allocation delayed, but then the space was
3783 * allocated using fallocate() before the delayed allocation could be resolved.
3784 * The cases to look for are:
3785 * ('=' indicated delayed allocated blocks
3786 * '-' indicates non-delayed allocated blocks)
3787 * (a) partial clusters towards beginning and/or end outside of allocated range
3788 * are not delalloc'ed.
3790 * |----c---=|====c====|====c====|===-c----|
3791 * |++++++ allocated ++++++|
3792 * ==> 4 complete clusters in above example
3794 * (b) partial cluster (outside of allocated range) towards either end is
3795 * marked for delayed allocation. In this case, we will exclude that
3798 * |----====c========|========c========|
3799 * |++++++ allocated ++++++|
3800 * ==> 1 complete clusters in above example
3803 * |================c================|
3804 * |++++++ allocated ++++++|
3805 * ==> 0 complete clusters in above example
3807 * The ext4_da_update_reserve_space will be called only if we
3808 * determine here that there were some "entire" clusters that span
3809 * this 'allocated' range.
3810 * In the non-bigalloc case, this function will just end up returning num_blks
3811 * without ever calling ext4_find_delalloc_range.
3814 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3815 unsigned int num_blks
)
3817 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3818 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3819 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3820 unsigned int allocated_clusters
= 0;
3822 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3823 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3825 /* max possible clusters for this allocation */
3826 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3828 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3830 /* Check towards left side */
3831 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
);
3833 lblk_from
= EXT4_LBLK_CMASK(sbi
, lblk_start
);
3834 lblk_to
= lblk_from
+ c_offset
- 1;
3836 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3837 allocated_clusters
--;
3840 /* Now check towards right. */
3841 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
+ num_blks
);
3842 if (allocated_clusters
&& c_offset
) {
3843 lblk_from
= lblk_start
+ num_blks
;
3844 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3846 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3847 allocated_clusters
--;
3850 return allocated_clusters
;
3854 ext4_ext_handle_uninitialized_extents(handle_t
*handle
, struct inode
*inode
,
3855 struct ext4_map_blocks
*map
,
3856 struct ext4_ext_path
*path
, int flags
,
3857 unsigned int allocated
, ext4_fsblk_t newblock
)
3861 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
3863 ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical "
3864 "block %llu, max_blocks %u, flags %x, allocated %u\n",
3865 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
3867 ext4_ext_show_leaf(inode
, path
);
3870 * When writing into uninitialized space, we should not fail to
3871 * allocate metadata blocks for the new extent block if needed.
3873 flags
|= EXT4_GET_BLOCKS_METADATA_NOFAIL
;
3875 trace_ext4_ext_handle_uninitialized_extents(inode
, map
, flags
,
3876 allocated
, newblock
);
3878 /* get_block() before submit the IO, split the extent */
3879 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
3880 ret
= ext4_split_unwritten_extents(handle
, inode
, map
,
3885 * Flag the inode(non aio case) or end_io struct (aio case)
3886 * that this IO needs to conversion to written when IO is
3890 ext4_set_io_unwritten_flag(inode
, io
);
3892 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3893 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3894 if (ext4_should_dioread_nolock(inode
))
3895 map
->m_flags
|= EXT4_MAP_UNINIT
;
3898 /* IO end_io complete, convert the filled extent to written */
3899 if ((flags
& EXT4_GET_BLOCKS_CONVERT
)) {
3900 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
, map
,
3903 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3904 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
3908 map
->m_flags
|= EXT4_MAP_MAPPED
;
3909 if (allocated
> map
->m_len
)
3910 allocated
= map
->m_len
;
3911 map
->m_len
= allocated
;
3914 /* buffered IO case */
3916 * repeat fallocate creation request
3917 * we already have an unwritten extent
3919 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) {
3920 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3924 /* buffered READ or buffered write_begin() lookup */
3925 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
3927 * We have blocks reserved already. We
3928 * return allocated blocks so that delalloc
3929 * won't do block reservation for us. But
3930 * the buffer head will be unmapped so that
3931 * a read from the block returns 0s.
3933 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3937 /* buffered write, writepage time, convert*/
3938 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, path
, flags
);
3940 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3947 map
->m_flags
|= EXT4_MAP_NEW
;
3949 * if we allocated more blocks than requested
3950 * we need to make sure we unmap the extra block
3951 * allocated. The actual needed block will get
3952 * unmapped later when we find the buffer_head marked
3955 if (allocated
> map
->m_len
) {
3956 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
3957 newblock
+ map
->m_len
,
3958 allocated
- map
->m_len
);
3959 allocated
= map
->m_len
;
3961 map
->m_len
= allocated
;
3964 * If we have done fallocate with the offset that is already
3965 * delayed allocated, we would have block reservation
3966 * and quota reservation done in the delayed write path.
3967 * But fallocate would have already updated quota and block
3968 * count for this offset. So cancel these reservation
3970 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
3971 unsigned int reserved_clusters
;
3972 reserved_clusters
= get_reserved_cluster_alloc(inode
,
3973 map
->m_lblk
, map
->m_len
);
3974 if (reserved_clusters
)
3975 ext4_da_update_reserve_space(inode
,
3981 map
->m_flags
|= EXT4_MAP_MAPPED
;
3982 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
3983 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
3989 if (allocated
> map
->m_len
)
3990 allocated
= map
->m_len
;
3991 ext4_ext_show_leaf(inode
, path
);
3992 map
->m_pblk
= newblock
;
3993 map
->m_len
= allocated
;
3996 ext4_ext_drop_refs(path
);
3999 return err
? err
: allocated
;
4003 * get_implied_cluster_alloc - check to see if the requested
4004 * allocation (in the map structure) overlaps with a cluster already
4005 * allocated in an extent.
4006 * @sb The filesystem superblock structure
4007 * @map The requested lblk->pblk mapping
4008 * @ex The extent structure which might contain an implied
4009 * cluster allocation
4011 * This function is called by ext4_ext_map_blocks() after we failed to
4012 * find blocks that were already in the inode's extent tree. Hence,
4013 * we know that the beginning of the requested region cannot overlap
4014 * the extent from the inode's extent tree. There are three cases we
4015 * want to catch. The first is this case:
4017 * |--- cluster # N--|
4018 * |--- extent ---| |---- requested region ---|
4021 * The second case that we need to test for is this one:
4023 * |--------- cluster # N ----------------|
4024 * |--- requested region --| |------- extent ----|
4025 * |=======================|
4027 * The third case is when the requested region lies between two extents
4028 * within the same cluster:
4029 * |------------- cluster # N-------------|
4030 * |----- ex -----| |---- ex_right ----|
4031 * |------ requested region ------|
4032 * |================|
4034 * In each of the above cases, we need to set the map->m_pblk and
4035 * map->m_len so it corresponds to the return the extent labelled as
4036 * "|====|" from cluster #N, since it is already in use for data in
4037 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4038 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4039 * as a new "allocated" block region. Otherwise, we will return 0 and
4040 * ext4_ext_map_blocks() will then allocate one or more new clusters
4041 * by calling ext4_mb_new_blocks().
4043 static int get_implied_cluster_alloc(struct super_block
*sb
,
4044 struct ext4_map_blocks
*map
,
4045 struct ext4_extent
*ex
,
4046 struct ext4_ext_path
*path
)
4048 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
4049 ext4_lblk_t c_offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4050 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
4051 ext4_lblk_t rr_cluster_start
;
4052 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4053 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4054 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
4056 /* The extent passed in that we are trying to match */
4057 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
4058 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
4060 /* The requested region passed into ext4_map_blocks() */
4061 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
4063 if ((rr_cluster_start
== ex_cluster_end
) ||
4064 (rr_cluster_start
== ex_cluster_start
)) {
4065 if (rr_cluster_start
== ex_cluster_end
)
4066 ee_start
+= ee_len
- 1;
4067 map
->m_pblk
= EXT4_PBLK_CMASK(sbi
, ee_start
) + c_offset
;
4068 map
->m_len
= min(map
->m_len
,
4069 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
4071 * Check for and handle this case:
4073 * |--------- cluster # N-------------|
4074 * |------- extent ----|
4075 * |--- requested region ---|
4079 if (map
->m_lblk
< ee_block
)
4080 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
4083 * Check for the case where there is already another allocated
4084 * block to the right of 'ex' but before the end of the cluster.
4086 * |------------- cluster # N-------------|
4087 * |----- ex -----| |---- ex_right ----|
4088 * |------ requested region ------|
4089 * |================|
4091 if (map
->m_lblk
> ee_block
) {
4092 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
4093 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
4096 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
4100 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
4106 * Block allocation/map/preallocation routine for extents based files
4109 * Need to be called with
4110 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4111 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4113 * return > 0, number of of blocks already mapped/allocated
4114 * if create == 0 and these are pre-allocated blocks
4115 * buffer head is unmapped
4116 * otherwise blocks are mapped
4118 * return = 0, if plain look up failed (blocks have not been allocated)
4119 * buffer head is unmapped
4121 * return < 0, error case.
4123 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
4124 struct ext4_map_blocks
*map
, int flags
)
4126 struct ext4_ext_path
*path
= NULL
;
4127 struct ext4_extent newex
, *ex
, *ex2
;
4128 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4129 ext4_fsblk_t newblock
= 0;
4130 int free_on_err
= 0, err
= 0, depth
;
4131 unsigned int allocated
= 0, offset
= 0;
4132 unsigned int allocated_clusters
= 0;
4133 struct ext4_allocation_request ar
;
4134 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
4135 ext4_lblk_t cluster_offset
;
4136 int set_unwritten
= 0;
4138 ext_debug("blocks %u/%u requested for inode %lu\n",
4139 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
4140 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
4142 /* find extent for this block */
4143 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, NULL
, 0);
4145 err
= PTR_ERR(path
);
4150 depth
= ext_depth(inode
);
4153 * consistent leaf must not be empty;
4154 * this situation is possible, though, _during_ tree modification;
4155 * this is why assert can't be put in ext4_ext_find_extent()
4157 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
4158 EXT4_ERROR_INODE(inode
, "bad extent address "
4159 "lblock: %lu, depth: %d pblock %lld",
4160 (unsigned long) map
->m_lblk
, depth
,
4161 path
[depth
].p_block
);
4166 ex
= path
[depth
].p_ext
;
4168 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4169 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4170 unsigned short ee_len
;
4173 * Uninitialized extents are treated as holes, except that
4174 * we split out initialized portions during a write.
4176 ee_len
= ext4_ext_get_actual_len(ex
);
4178 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
4180 /* if found extent covers block, simply return it */
4181 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
4182 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
4183 /* number of remaining blocks in the extent */
4184 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
4185 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
4186 ee_block
, ee_len
, newblock
);
4188 if (!ext4_ext_is_uninitialized(ex
))
4191 allocated
= ext4_ext_handle_uninitialized_extents(
4192 handle
, inode
, map
, path
, flags
,
4193 allocated
, newblock
);
4198 if ((sbi
->s_cluster_ratio
> 1) &&
4199 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
4200 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4203 * requested block isn't allocated yet;
4204 * we couldn't try to create block if create flag is zero
4206 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4208 * put just found gap into cache to speed up
4209 * subsequent requests
4211 if ((flags
& EXT4_GET_BLOCKS_NO_PUT_HOLE
) == 0)
4212 ext4_ext_put_gap_in_cache(inode
, path
, map
->m_lblk
);
4217 * Okay, we need to do block allocation.
4219 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
4220 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
4221 cluster_offset
= EXT4_LBLK_CMASK(sbi
, map
->m_lblk
);
4224 * If we are doing bigalloc, check to see if the extent returned
4225 * by ext4_ext_find_extent() implies a cluster we can use.
4227 if (cluster_offset
&& ex
&&
4228 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
4229 ar
.len
= allocated
= map
->m_len
;
4230 newblock
= map
->m_pblk
;
4231 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4232 goto got_allocated_blocks
;
4235 /* find neighbour allocated blocks */
4236 ar
.lleft
= map
->m_lblk
;
4237 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
4240 ar
.lright
= map
->m_lblk
;
4242 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
4246 /* Check if the extent after searching to the right implies a
4247 * cluster we can use. */
4248 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
4249 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
4250 ar
.len
= allocated
= map
->m_len
;
4251 newblock
= map
->m_pblk
;
4252 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4253 goto got_allocated_blocks
;
4257 * See if request is beyond maximum number of blocks we can have in
4258 * a single extent. For an initialized extent this limit is
4259 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
4260 * EXT_UNINIT_MAX_LEN.
4262 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
4263 !(flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4264 map
->m_len
= EXT_INIT_MAX_LEN
;
4265 else if (map
->m_len
> EXT_UNINIT_MAX_LEN
&&
4266 (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
))
4267 map
->m_len
= EXT_UNINIT_MAX_LEN
;
4269 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4270 newex
.ee_len
= cpu_to_le16(map
->m_len
);
4271 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
4273 allocated
= ext4_ext_get_actual_len(&newex
);
4275 allocated
= map
->m_len
;
4277 /* allocate new block */
4279 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
4280 ar
.logical
= map
->m_lblk
;
4282 * We calculate the offset from the beginning of the cluster
4283 * for the logical block number, since when we allocate a
4284 * physical cluster, the physical block should start at the
4285 * same offset from the beginning of the cluster. This is
4286 * needed so that future calls to get_implied_cluster_alloc()
4289 offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4290 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4292 ar
.logical
-= offset
;
4293 if (S_ISREG(inode
->i_mode
))
4294 ar
.flags
= EXT4_MB_HINT_DATA
;
4296 /* disable in-core preallocation for non-regular files */
4298 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4299 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4300 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4303 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4304 ar
.goal
, newblock
, allocated
);
4306 allocated_clusters
= ar
.len
;
4307 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4308 if (ar
.len
> allocated
)
4311 got_allocated_blocks
:
4312 /* try to insert new extent into found leaf and return */
4313 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4314 newex
.ee_len
= cpu_to_le16(ar
.len
);
4315 /* Mark uninitialized */
4316 if (flags
& EXT4_GET_BLOCKS_UNINIT_EXT
){
4317 ext4_ext_mark_uninitialized(&newex
);
4318 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4320 * io_end structure was created for every IO write to an
4321 * uninitialized extent. To avoid unnecessary conversion,
4322 * here we flag the IO that really needs the conversion.
4323 * For non asycn direct IO case, flag the inode state
4324 * that we need to perform conversion when IO is done.
4326 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
))
4328 if (ext4_should_dioread_nolock(inode
))
4329 map
->m_flags
|= EXT4_MAP_UNINIT
;
4333 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4334 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4337 err
= ext4_ext_insert_extent(handle
, inode
, path
,
4340 if (!err
&& set_unwritten
) {
4342 ext4_set_io_unwritten_flag(inode
, io
);
4344 ext4_set_inode_state(inode
,
4345 EXT4_STATE_DIO_UNWRITTEN
);
4348 if (err
&& free_on_err
) {
4349 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4350 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4351 /* free data blocks we just allocated */
4352 /* not a good idea to call discard here directly,
4353 * but otherwise we'd need to call it every free() */
4354 ext4_discard_preallocations(inode
);
4355 ext4_free_blocks(handle
, inode
, NULL
, newblock
,
4356 EXT4_C2B(sbi
, allocated_clusters
), fb_flags
);
4360 /* previous routine could use block we allocated */
4361 newblock
= ext4_ext_pblock(&newex
);
4362 allocated
= ext4_ext_get_actual_len(&newex
);
4363 if (allocated
> map
->m_len
)
4364 allocated
= map
->m_len
;
4365 map
->m_flags
|= EXT4_MAP_NEW
;
4368 * Update reserved blocks/metadata blocks after successful
4369 * block allocation which had been deferred till now.
4371 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4372 unsigned int reserved_clusters
;
4374 * Check how many clusters we had reserved this allocated range
4376 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4377 map
->m_lblk
, allocated
);
4378 if (map
->m_flags
& EXT4_MAP_FROM_CLUSTER
) {
4379 if (reserved_clusters
) {
4381 * We have clusters reserved for this range.
4382 * But since we are not doing actual allocation
4383 * and are simply using blocks from previously
4384 * allocated cluster, we should release the
4385 * reservation and not claim quota.
4387 ext4_da_update_reserve_space(inode
,
4388 reserved_clusters
, 0);
4391 BUG_ON(allocated_clusters
< reserved_clusters
);
4392 if (reserved_clusters
< allocated_clusters
) {
4393 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4394 int reservation
= allocated_clusters
-
4397 * It seems we claimed few clusters outside of
4398 * the range of this allocation. We should give
4399 * it back to the reservation pool. This can
4400 * happen in the following case:
4402 * * Suppose s_cluster_ratio is 4 (i.e., each
4403 * cluster has 4 blocks. Thus, the clusters
4404 * are [0-3],[4-7],[8-11]...
4405 * * First comes delayed allocation write for
4406 * logical blocks 10 & 11. Since there were no
4407 * previous delayed allocated blocks in the
4408 * range [8-11], we would reserve 1 cluster
4410 * * Next comes write for logical blocks 3 to 8.
4411 * In this case, we will reserve 2 clusters
4412 * (for [0-3] and [4-7]; and not for [8-11] as
4413 * that range has a delayed allocated blocks.
4414 * Thus total reserved clusters now becomes 3.
4415 * * Now, during the delayed allocation writeout
4416 * time, we will first write blocks [3-8] and
4417 * allocate 3 clusters for writing these
4418 * blocks. Also, we would claim all these
4419 * three clusters above.
4420 * * Now when we come here to writeout the
4421 * blocks [10-11], we would expect to claim
4422 * the reservation of 1 cluster we had made
4423 * (and we would claim it since there are no
4424 * more delayed allocated blocks in the range
4425 * [8-11]. But our reserved cluster count had
4426 * already gone to 0.
4428 * Thus, at the step 4 above when we determine
4429 * that there are still some unwritten delayed
4430 * allocated blocks outside of our current
4431 * block range, we should increment the
4432 * reserved clusters count so that when the
4433 * remaining blocks finally gets written, we
4436 dquot_reserve_block(inode
,
4437 EXT4_C2B(sbi
, reservation
));
4438 spin_lock(&ei
->i_block_reservation_lock
);
4439 ei
->i_reserved_data_blocks
+= reservation
;
4440 spin_unlock(&ei
->i_block_reservation_lock
);
4443 * We will claim quota for all newly allocated blocks.
4444 * We're updating the reserved space *after* the
4445 * correction above so we do not accidentally free
4446 * all the metadata reservation because we might
4447 * actually need it later on.
4449 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4455 * Cache the extent and update transaction to commit on fdatasync only
4456 * when it is _not_ an uninitialized extent.
4458 if ((flags
& EXT4_GET_BLOCKS_UNINIT_EXT
) == 0)
4459 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4461 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4463 if (allocated
> map
->m_len
)
4464 allocated
= map
->m_len
;
4465 ext4_ext_show_leaf(inode
, path
);
4466 map
->m_flags
|= EXT4_MAP_MAPPED
;
4467 map
->m_pblk
= newblock
;
4468 map
->m_len
= allocated
;
4471 ext4_ext_drop_refs(path
);
4476 trace_ext4_ext_map_blocks_exit(inode
, flags
, map
,
4477 err
? err
: allocated
);
4478 ext4_es_lru_add(inode
);
4479 return err
? err
: allocated
;
4482 void ext4_ext_truncate(handle_t
*handle
, struct inode
*inode
)
4484 struct super_block
*sb
= inode
->i_sb
;
4485 ext4_lblk_t last_block
;
4489 * TODO: optimization is possible here.
4490 * Probably we need not scan at all,
4491 * because page truncation is enough.
4494 /* we have to know where to truncate from in crash case */
4495 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4496 ext4_mark_inode_dirty(handle
, inode
);
4498 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4499 >> EXT4_BLOCK_SIZE_BITS(sb
);
4501 err
= ext4_es_remove_extent(inode
, last_block
,
4502 EXT_MAX_BLOCKS
- last_block
);
4503 if (err
== -ENOMEM
) {
4505 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
4509 ext4_std_error(inode
->i_sb
, err
);
4512 err
= ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4513 ext4_std_error(inode
->i_sb
, err
);
4516 static void ext4_falloc_update_inode(struct inode
*inode
,
4517 int mode
, loff_t new_size
, int update_ctime
)
4519 struct timespec now
;
4522 now
= current_fs_time(inode
->i_sb
);
4523 if (!timespec_equal(&inode
->i_ctime
, &now
))
4524 inode
->i_ctime
= now
;
4527 * Update only when preallocation was requested beyond
4530 if (!(mode
& FALLOC_FL_KEEP_SIZE
)) {
4531 if (new_size
> i_size_read(inode
))
4532 i_size_write(inode
, new_size
);
4533 if (new_size
> EXT4_I(inode
)->i_disksize
)
4534 ext4_update_i_disksize(inode
, new_size
);
4537 * Mark that we allocate beyond EOF so the subsequent truncate
4538 * can proceed even if the new size is the same as i_size.
4540 if (new_size
> i_size_read(inode
))
4541 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4547 * preallocate space for a file. This implements ext4's fallocate file
4548 * operation, which gets called from sys_fallocate system call.
4549 * For block-mapped files, posix_fallocate should fall back to the method
4550 * of writing zeroes to the required new blocks (the same behavior which is
4551 * expected for file systems which do not support fallocate() system call).
4553 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4555 struct inode
*inode
= file_inode(file
);
4558 unsigned int max_blocks
;
4563 struct ext4_map_blocks map
;
4564 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4566 /* Return error if mode is not supported */
4567 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
))
4570 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4571 return ext4_punch_hole(inode
, offset
, len
);
4573 ret
= ext4_convert_inline_data(inode
);
4578 * currently supporting (pre)allocate mode for extent-based
4581 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4584 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4585 map
.m_lblk
= offset
>> blkbits
;
4587 * We can't just convert len to max_blocks because
4588 * If blocksize = 4096 offset = 3072 and len = 2048
4590 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
4593 * credits to insert 1 extent into extent tree
4595 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4596 mutex_lock(&inode
->i_mutex
);
4597 ret
= inode_newsize_ok(inode
, (len
+ offset
));
4599 mutex_unlock(&inode
->i_mutex
);
4600 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
4603 flags
= EXT4_GET_BLOCKS_CREATE_UNINIT_EXT
;
4604 if (mode
& FALLOC_FL_KEEP_SIZE
)
4605 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4607 * Don't normalize the request if it can fit in one extent so
4608 * that it doesn't get unnecessarily split into multiple
4611 if (len
<= EXT_UNINIT_MAX_LEN
<< blkbits
)
4612 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4615 while (ret
>= 0 && ret
< max_blocks
) {
4616 map
.m_lblk
= map
.m_lblk
+ ret
;
4617 map
.m_len
= max_blocks
= max_blocks
- ret
;
4618 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
4620 if (IS_ERR(handle
)) {
4621 ret
= PTR_ERR(handle
);
4624 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4627 ext4_warning(inode
->i_sb
,
4628 "inode #%lu: block %u: len %u: "
4629 "ext4_ext_map_blocks returned %d",
4630 inode
->i_ino
, map
.m_lblk
,
4633 ext4_mark_inode_dirty(handle
, inode
);
4634 ret2
= ext4_journal_stop(handle
);
4637 if ((map
.m_lblk
+ ret
) >= (EXT4_BLOCK_ALIGN(offset
+ len
,
4638 blkbits
) >> blkbits
))
4639 new_size
= offset
+ len
;
4641 new_size
= ((loff_t
) map
.m_lblk
+ ret
) << blkbits
;
4643 ext4_falloc_update_inode(inode
, mode
, new_size
,
4644 (map
.m_flags
& EXT4_MAP_NEW
));
4645 ext4_mark_inode_dirty(handle
, inode
);
4646 if ((file
->f_flags
& O_SYNC
) && ret
>= max_blocks
)
4647 ext4_handle_sync(handle
);
4648 ret2
= ext4_journal_stop(handle
);
4652 if (ret
== -ENOSPC
&&
4653 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4657 mutex_unlock(&inode
->i_mutex
);
4658 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
,
4659 ret
> 0 ? ret2
: ret
);
4660 return ret
> 0 ? ret2
: ret
;
4664 * This function convert a range of blocks to written extents
4665 * The caller of this function will pass the start offset and the size.
4666 * all unwritten extents within this range will be converted to
4669 * This function is called from the direct IO end io call back
4670 * function, to convert the fallocated extents after IO is completed.
4671 * Returns 0 on success.
4673 int ext4_convert_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
4674 loff_t offset
, ssize_t len
)
4676 unsigned int max_blocks
;
4679 struct ext4_map_blocks map
;
4680 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4682 map
.m_lblk
= offset
>> blkbits
;
4684 * We can't just convert len to max_blocks because
4685 * If blocksize = 4096 offset = 3072 and len = 2048
4687 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
4690 * This is somewhat ugly but the idea is clear: When transaction is
4691 * reserved, everything goes into it. Otherwise we rather start several
4692 * smaller transactions for conversion of each extent separately.
4695 handle
= ext4_journal_start_reserved(handle
,
4696 EXT4_HT_EXT_CONVERT
);
4698 return PTR_ERR(handle
);
4702 * credits to insert 1 extent into extent tree
4704 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
4706 while (ret
>= 0 && ret
< max_blocks
) {
4708 map
.m_len
= (max_blocks
-= ret
);
4710 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
4712 if (IS_ERR(handle
)) {
4713 ret
= PTR_ERR(handle
);
4717 ret
= ext4_map_blocks(handle
, inode
, &map
,
4718 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
4720 ext4_warning(inode
->i_sb
,
4721 "inode #%lu: block %u: len %u: "
4722 "ext4_ext_map_blocks returned %d",
4723 inode
->i_ino
, map
.m_lblk
,
4725 ext4_mark_inode_dirty(handle
, inode
);
4727 ret2
= ext4_journal_stop(handle
);
4728 if (ret
<= 0 || ret2
)
4732 ret2
= ext4_journal_stop(handle
);
4733 return ret
> 0 ? ret2
: ret
;
4737 * If newes is not existing extent (newes->ec_pblk equals zero) find
4738 * delayed extent at start of newes and update newes accordingly and
4739 * return start of the next delayed extent.
4741 * If newes is existing extent (newes->ec_pblk is not equal zero)
4742 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
4743 * extent found. Leave newes unmodified.
4745 static int ext4_find_delayed_extent(struct inode
*inode
,
4746 struct extent_status
*newes
)
4748 struct extent_status es
;
4749 ext4_lblk_t block
, next_del
;
4751 if (newes
->es_pblk
== 0) {
4752 ext4_es_find_delayed_extent_range(inode
, newes
->es_lblk
,
4753 newes
->es_lblk
+ newes
->es_len
- 1, &es
);
4756 * No extent in extent-tree contains block @newes->es_pblk,
4757 * then the block may stay in 1)a hole or 2)delayed-extent.
4763 if (es
.es_lblk
> newes
->es_lblk
) {
4765 newes
->es_len
= min(es
.es_lblk
- newes
->es_lblk
,
4770 newes
->es_len
= es
.es_lblk
+ es
.es_len
- newes
->es_lblk
;
4773 block
= newes
->es_lblk
+ newes
->es_len
;
4774 ext4_es_find_delayed_extent_range(inode
, block
, EXT_MAX_BLOCKS
, &es
);
4776 next_del
= EXT_MAX_BLOCKS
;
4778 next_del
= es
.es_lblk
;
4782 /* fiemap flags we can handle specified here */
4783 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
4785 static int ext4_xattr_fiemap(struct inode
*inode
,
4786 struct fiemap_extent_info
*fieinfo
)
4790 __u32 flags
= FIEMAP_EXTENT_LAST
;
4791 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
4795 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
4796 struct ext4_iloc iloc
;
4797 int offset
; /* offset of xattr in inode */
4799 error
= ext4_get_inode_loc(inode
, &iloc
);
4802 physical
= (__u64
)iloc
.bh
->b_blocknr
<< blockbits
;
4803 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
4804 EXT4_I(inode
)->i_extra_isize
;
4806 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
4807 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
4809 } else { /* external block */
4810 physical
= (__u64
)EXT4_I(inode
)->i_file_acl
<< blockbits
;
4811 length
= inode
->i_sb
->s_blocksize
;
4815 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
4817 return (error
< 0 ? error
: 0);
4820 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4821 __u64 start
, __u64 len
)
4823 ext4_lblk_t start_blk
;
4826 if (ext4_has_inline_data(inode
)) {
4829 error
= ext4_inline_data_fiemap(inode
, fieinfo
, &has_inline
);
4835 if (fieinfo
->fi_flags
& FIEMAP_FLAG_CACHE
) {
4836 error
= ext4_ext_precache(inode
);
4841 /* fallback to generic here if not in extents fmt */
4842 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4843 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
4846 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
4849 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
4850 error
= ext4_xattr_fiemap(inode
, fieinfo
);
4852 ext4_lblk_t len_blks
;
4855 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
4856 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
4857 if (last_blk
>= EXT_MAX_BLOCKS
)
4858 last_blk
= EXT_MAX_BLOCKS
-1;
4859 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
4862 * Walk the extent tree gathering extent information
4863 * and pushing extents back to the user.
4865 error
= ext4_fill_fiemap_extents(inode
, start_blk
,
4868 ext4_es_lru_add(inode
);