2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
24 * Extents support for EXT4
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
33 #include <linux/time.h>
34 #include <linux/jbd2.h>
35 #include <linux/highuid.h>
36 #include <linux/pagemap.h>
37 #include <linux/quotaops.h>
38 #include <linux/string.h>
39 #include <linux/slab.h>
40 #include <asm/uaccess.h>
41 #include <linux/fiemap.h>
42 #include "ext4_jbd2.h"
43 #include "ext4_extents.h"
46 #include <trace/events/ext4.h>
49 * used by extent splitting.
51 #define EXT4_EXT_MAY_ZEROOUT 0x1 /* safe to zeroout if split fails \
53 #define EXT4_EXT_MARK_UNWRIT1 0x2 /* mark first half unwritten */
54 #define EXT4_EXT_MARK_UNWRIT2 0x4 /* mark second half unwritten */
56 #define EXT4_EXT_DATA_VALID1 0x8 /* first half contains valid data */
57 #define EXT4_EXT_DATA_VALID2 0x10 /* second half contains valid data */
59 static __le32
ext4_extent_block_csum(struct inode
*inode
,
60 struct ext4_extent_header
*eh
)
62 struct ext4_inode_info
*ei
= EXT4_I(inode
);
63 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
66 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)eh
,
67 EXT4_EXTENT_TAIL_OFFSET(eh
));
68 return cpu_to_le32(csum
);
71 static int ext4_extent_block_csum_verify(struct inode
*inode
,
72 struct ext4_extent_header
*eh
)
74 struct ext4_extent_tail
*et
;
76 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
77 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
80 et
= find_ext4_extent_tail(eh
);
81 if (et
->et_checksum
!= ext4_extent_block_csum(inode
, eh
))
86 static void ext4_extent_block_csum_set(struct inode
*inode
,
87 struct ext4_extent_header
*eh
)
89 struct ext4_extent_tail
*et
;
91 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
92 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
95 et
= find_ext4_extent_tail(eh
);
96 et
->et_checksum
= ext4_extent_block_csum(inode
, eh
);
99 static int ext4_split_extent(handle_t
*handle
,
101 struct ext4_ext_path
*path
,
102 struct ext4_map_blocks
*map
,
106 static int ext4_split_extent_at(handle_t
*handle
,
108 struct ext4_ext_path
*path
,
113 static int ext4_find_delayed_extent(struct inode
*inode
,
114 struct extent_status
*newes
);
116 static int ext4_ext_truncate_extend_restart(handle_t
*handle
,
122 if (!ext4_handle_valid(handle
))
124 if (handle
->h_buffer_credits
> needed
)
126 err
= ext4_journal_extend(handle
, needed
);
129 err
= ext4_truncate_restart_trans(handle
, inode
, needed
);
141 static int ext4_ext_get_access(handle_t
*handle
, struct inode
*inode
,
142 struct ext4_ext_path
*path
)
145 /* path points to block */
146 BUFFER_TRACE(path
->p_bh
, "get_write_access");
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_unwritten(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_unwritten(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_unwritten(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_unwritten(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_unwritten(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 * unwritten 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_unwritten(ex1
) != ext4_ext_is_unwritten(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 if (ext4_ext_is_unwritten(ex1
) &&
1712 (ext4_test_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
) ||
1713 atomic_read(&EXT4_I(inode
)->i_unwritten
) ||
1714 (ext1_ee_len
+ ext2_ee_len
> EXT_UNWRITTEN_MAX_LEN
)))
1716 #ifdef AGGRESSIVE_TEST
1717 if (ext1_ee_len
>= 4)
1721 if (ext4_ext_pblock(ex1
) + ext1_ee_len
== ext4_ext_pblock(ex2
))
1727 * This function tries to merge the "ex" extent to the next extent in the tree.
1728 * It always tries to merge towards right. If you want to merge towards
1729 * left, pass "ex - 1" as argument instead of "ex".
1730 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1731 * 1 if they got merged.
1733 static int ext4_ext_try_to_merge_right(struct inode
*inode
,
1734 struct ext4_ext_path
*path
,
1735 struct ext4_extent
*ex
)
1737 struct ext4_extent_header
*eh
;
1738 unsigned int depth
, len
;
1739 int merge_done
= 0, unwritten
;
1741 depth
= ext_depth(inode
);
1742 BUG_ON(path
[depth
].p_hdr
== NULL
);
1743 eh
= path
[depth
].p_hdr
;
1745 while (ex
< EXT_LAST_EXTENT(eh
)) {
1746 if (!ext4_can_extents_be_merged(inode
, ex
, ex
+ 1))
1748 /* merge with next extent! */
1749 unwritten
= ext4_ext_is_unwritten(ex
);
1750 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1751 + ext4_ext_get_actual_len(ex
+ 1));
1753 ext4_ext_mark_unwritten(ex
);
1755 if (ex
+ 1 < EXT_LAST_EXTENT(eh
)) {
1756 len
= (EXT_LAST_EXTENT(eh
) - ex
- 1)
1757 * sizeof(struct ext4_extent
);
1758 memmove(ex
+ 1, ex
+ 2, len
);
1760 le16_add_cpu(&eh
->eh_entries
, -1);
1762 WARN_ON(eh
->eh_entries
== 0);
1763 if (!eh
->eh_entries
)
1764 EXT4_ERROR_INODE(inode
, "eh->eh_entries = 0!");
1771 * This function does a very simple check to see if we can collapse
1772 * an extent tree with a single extent tree leaf block into the inode.
1774 static void ext4_ext_try_to_merge_up(handle_t
*handle
,
1775 struct inode
*inode
,
1776 struct ext4_ext_path
*path
)
1779 unsigned max_root
= ext4_ext_space_root(inode
, 0);
1782 if ((path
[0].p_depth
!= 1) ||
1783 (le16_to_cpu(path
[0].p_hdr
->eh_entries
) != 1) ||
1784 (le16_to_cpu(path
[1].p_hdr
->eh_entries
) > max_root
))
1788 * We need to modify the block allocation bitmap and the block
1789 * group descriptor to release the extent tree block. If we
1790 * can't get the journal credits, give up.
1792 if (ext4_journal_extend(handle
, 2))
1796 * Copy the extent data up to the inode
1798 blk
= ext4_idx_pblock(path
[0].p_idx
);
1799 s
= le16_to_cpu(path
[1].p_hdr
->eh_entries
) *
1800 sizeof(struct ext4_extent_idx
);
1801 s
+= sizeof(struct ext4_extent_header
);
1803 memcpy(path
[0].p_hdr
, path
[1].p_hdr
, s
);
1804 path
[0].p_depth
= 0;
1805 path
[0].p_ext
= EXT_FIRST_EXTENT(path
[0].p_hdr
) +
1806 (path
[1].p_ext
- EXT_FIRST_EXTENT(path
[1].p_hdr
));
1807 path
[0].p_hdr
->eh_max
= cpu_to_le16(max_root
);
1809 brelse(path
[1].p_bh
);
1810 ext4_free_blocks(handle
, inode
, NULL
, blk
, 1,
1811 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
|
1812 EXT4_FREE_BLOCKS_RESERVE
);
1816 * This function tries to merge the @ex extent to neighbours in the tree.
1817 * return 1 if merge left else 0.
1819 static void ext4_ext_try_to_merge(handle_t
*handle
,
1820 struct inode
*inode
,
1821 struct ext4_ext_path
*path
,
1822 struct ext4_extent
*ex
) {
1823 struct ext4_extent_header
*eh
;
1827 depth
= ext_depth(inode
);
1828 BUG_ON(path
[depth
].p_hdr
== NULL
);
1829 eh
= path
[depth
].p_hdr
;
1831 if (ex
> EXT_FIRST_EXTENT(eh
))
1832 merge_done
= ext4_ext_try_to_merge_right(inode
, path
, ex
- 1);
1835 (void) ext4_ext_try_to_merge_right(inode
, path
, ex
);
1837 ext4_ext_try_to_merge_up(handle
, inode
, path
);
1841 * check if a portion of the "newext" extent overlaps with an
1844 * If there is an overlap discovered, it updates the length of the newext
1845 * such that there will be no overlap, and then returns 1.
1846 * If there is no overlap found, it returns 0.
1848 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info
*sbi
,
1849 struct inode
*inode
,
1850 struct ext4_extent
*newext
,
1851 struct ext4_ext_path
*path
)
1854 unsigned int depth
, len1
;
1855 unsigned int ret
= 0;
1857 b1
= le32_to_cpu(newext
->ee_block
);
1858 len1
= ext4_ext_get_actual_len(newext
);
1859 depth
= ext_depth(inode
);
1860 if (!path
[depth
].p_ext
)
1862 b2
= EXT4_LBLK_CMASK(sbi
, le32_to_cpu(path
[depth
].p_ext
->ee_block
));
1865 * get the next allocated block if the extent in the path
1866 * is before the requested block(s)
1869 b2
= ext4_ext_next_allocated_block(path
);
1870 if (b2
== EXT_MAX_BLOCKS
)
1872 b2
= EXT4_LBLK_CMASK(sbi
, b2
);
1875 /* check for wrap through zero on extent logical start block*/
1876 if (b1
+ len1
< b1
) {
1877 len1
= EXT_MAX_BLOCKS
- b1
;
1878 newext
->ee_len
= cpu_to_le16(len1
);
1882 /* check for overlap */
1883 if (b1
+ len1
> b2
) {
1884 newext
->ee_len
= cpu_to_le16(b2
- b1
);
1892 * ext4_ext_insert_extent:
1893 * tries to merge requsted extent into the existing extent or
1894 * inserts requested extent as new one into the tree,
1895 * creating new leaf in the no-space case.
1897 int ext4_ext_insert_extent(handle_t
*handle
, struct inode
*inode
,
1898 struct ext4_ext_path
*path
,
1899 struct ext4_extent
*newext
, int gb_flags
)
1901 struct ext4_extent_header
*eh
;
1902 struct ext4_extent
*ex
, *fex
;
1903 struct ext4_extent
*nearex
; /* nearest extent */
1904 struct ext4_ext_path
*npath
= NULL
;
1905 int depth
, len
, err
;
1907 int mb_flags
= 0, unwritten
;
1909 if (unlikely(ext4_ext_get_actual_len(newext
) == 0)) {
1910 EXT4_ERROR_INODE(inode
, "ext4_ext_get_actual_len(newext) == 0");
1913 depth
= ext_depth(inode
);
1914 ex
= path
[depth
].p_ext
;
1915 eh
= path
[depth
].p_hdr
;
1916 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
1917 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
1921 /* try to insert block into found extent and return */
1922 if (ex
&& !(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
)) {
1925 * Try to see whether we should rather test the extent on
1926 * right from ex, or from the left of ex. This is because
1927 * ext4_ext_find_extent() can return either extent on the
1928 * left, or on the right from the searched position. This
1929 * will make merging more effective.
1931 if (ex
< EXT_LAST_EXTENT(eh
) &&
1932 (le32_to_cpu(ex
->ee_block
) +
1933 ext4_ext_get_actual_len(ex
) <
1934 le32_to_cpu(newext
->ee_block
))) {
1937 } else if ((ex
> EXT_FIRST_EXTENT(eh
)) &&
1938 (le32_to_cpu(newext
->ee_block
) +
1939 ext4_ext_get_actual_len(newext
) <
1940 le32_to_cpu(ex
->ee_block
)))
1943 /* Try to append newex to the ex */
1944 if (ext4_can_extents_be_merged(inode
, ex
, newext
)) {
1945 ext_debug("append [%d]%d block to %u:[%d]%d"
1947 ext4_ext_is_unwritten(newext
),
1948 ext4_ext_get_actual_len(newext
),
1949 le32_to_cpu(ex
->ee_block
),
1950 ext4_ext_is_unwritten(ex
),
1951 ext4_ext_get_actual_len(ex
),
1952 ext4_ext_pblock(ex
));
1953 err
= ext4_ext_get_access(handle
, inode
,
1957 unwritten
= ext4_ext_is_unwritten(ex
);
1958 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1959 + ext4_ext_get_actual_len(newext
));
1961 ext4_ext_mark_unwritten(ex
);
1962 eh
= path
[depth
].p_hdr
;
1968 /* Try to prepend newex to the ex */
1969 if (ext4_can_extents_be_merged(inode
, newext
, ex
)) {
1970 ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1972 le32_to_cpu(newext
->ee_block
),
1973 ext4_ext_is_unwritten(newext
),
1974 ext4_ext_get_actual_len(newext
),
1975 le32_to_cpu(ex
->ee_block
),
1976 ext4_ext_is_unwritten(ex
),
1977 ext4_ext_get_actual_len(ex
),
1978 ext4_ext_pblock(ex
));
1979 err
= ext4_ext_get_access(handle
, inode
,
1984 unwritten
= ext4_ext_is_unwritten(ex
);
1985 ex
->ee_block
= newext
->ee_block
;
1986 ext4_ext_store_pblock(ex
, ext4_ext_pblock(newext
));
1987 ex
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
)
1988 + ext4_ext_get_actual_len(newext
));
1990 ext4_ext_mark_unwritten(ex
);
1991 eh
= path
[depth
].p_hdr
;
1997 depth
= ext_depth(inode
);
1998 eh
= path
[depth
].p_hdr
;
1999 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
))
2002 /* probably next leaf has space for us? */
2003 fex
= EXT_LAST_EXTENT(eh
);
2004 next
= EXT_MAX_BLOCKS
;
2005 if (le32_to_cpu(newext
->ee_block
) > le32_to_cpu(fex
->ee_block
))
2006 next
= ext4_ext_next_leaf_block(path
);
2007 if (next
!= EXT_MAX_BLOCKS
) {
2008 ext_debug("next leaf block - %u\n", next
);
2009 BUG_ON(npath
!= NULL
);
2010 npath
= ext4_ext_find_extent(inode
, next
, NULL
, 0);
2012 return PTR_ERR(npath
);
2013 BUG_ON(npath
->p_depth
!= path
->p_depth
);
2014 eh
= npath
[depth
].p_hdr
;
2015 if (le16_to_cpu(eh
->eh_entries
) < le16_to_cpu(eh
->eh_max
)) {
2016 ext_debug("next leaf isn't full(%d)\n",
2017 le16_to_cpu(eh
->eh_entries
));
2021 ext_debug("next leaf has no free space(%d,%d)\n",
2022 le16_to_cpu(eh
->eh_entries
), le16_to_cpu(eh
->eh_max
));
2026 * There is no free space in the found leaf.
2027 * We're gonna add a new leaf in the tree.
2029 if (gb_flags
& EXT4_GET_BLOCKS_METADATA_NOFAIL
)
2030 mb_flags
= EXT4_MB_USE_RESERVED
;
2031 err
= ext4_ext_create_new_leaf(handle
, inode
, mb_flags
, gb_flags
,
2035 depth
= ext_depth(inode
);
2036 eh
= path
[depth
].p_hdr
;
2039 nearex
= path
[depth
].p_ext
;
2041 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2046 /* there is no extent in this leaf, create first one */
2047 ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2048 le32_to_cpu(newext
->ee_block
),
2049 ext4_ext_pblock(newext
),
2050 ext4_ext_is_unwritten(newext
),
2051 ext4_ext_get_actual_len(newext
));
2052 nearex
= EXT_FIRST_EXTENT(eh
);
2054 if (le32_to_cpu(newext
->ee_block
)
2055 > le32_to_cpu(nearex
->ee_block
)) {
2057 ext_debug("insert %u:%llu:[%d]%d before: "
2059 le32_to_cpu(newext
->ee_block
),
2060 ext4_ext_pblock(newext
),
2061 ext4_ext_is_unwritten(newext
),
2062 ext4_ext_get_actual_len(newext
),
2067 BUG_ON(newext
->ee_block
== nearex
->ee_block
);
2068 ext_debug("insert %u:%llu:[%d]%d after: "
2070 le32_to_cpu(newext
->ee_block
),
2071 ext4_ext_pblock(newext
),
2072 ext4_ext_is_unwritten(newext
),
2073 ext4_ext_get_actual_len(newext
),
2076 len
= EXT_LAST_EXTENT(eh
) - nearex
+ 1;
2078 ext_debug("insert %u:%llu:[%d]%d: "
2079 "move %d extents from 0x%p to 0x%p\n",
2080 le32_to_cpu(newext
->ee_block
),
2081 ext4_ext_pblock(newext
),
2082 ext4_ext_is_unwritten(newext
),
2083 ext4_ext_get_actual_len(newext
),
2084 len
, nearex
, nearex
+ 1);
2085 memmove(nearex
+ 1, nearex
,
2086 len
* sizeof(struct ext4_extent
));
2090 le16_add_cpu(&eh
->eh_entries
, 1);
2091 path
[depth
].p_ext
= nearex
;
2092 nearex
->ee_block
= newext
->ee_block
;
2093 ext4_ext_store_pblock(nearex
, ext4_ext_pblock(newext
));
2094 nearex
->ee_len
= newext
->ee_len
;
2097 /* try to merge extents */
2098 if (!(gb_flags
& EXT4_GET_BLOCKS_PRE_IO
))
2099 ext4_ext_try_to_merge(handle
, inode
, path
, nearex
);
2102 /* time to correct all indexes above */
2103 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2107 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
2111 ext4_ext_drop_refs(npath
);
2117 static int ext4_fill_fiemap_extents(struct inode
*inode
,
2118 ext4_lblk_t block
, ext4_lblk_t num
,
2119 struct fiemap_extent_info
*fieinfo
)
2121 struct ext4_ext_path
*path
= NULL
;
2122 struct ext4_extent
*ex
;
2123 struct extent_status es
;
2124 ext4_lblk_t next
, next_del
, start
= 0, end
= 0;
2125 ext4_lblk_t last
= block
+ num
;
2126 int exists
, depth
= 0, err
= 0;
2127 unsigned int flags
= 0;
2128 unsigned char blksize_bits
= inode
->i_sb
->s_blocksize_bits
;
2130 while (block
< last
&& block
!= EXT_MAX_BLOCKS
) {
2132 /* find extent for this block */
2133 down_read(&EXT4_I(inode
)->i_data_sem
);
2135 if (path
&& ext_depth(inode
) != depth
) {
2136 /* depth was changed. we have to realloc path */
2141 path
= ext4_ext_find_extent(inode
, block
, path
, 0);
2143 up_read(&EXT4_I(inode
)->i_data_sem
);
2144 err
= PTR_ERR(path
);
2149 depth
= ext_depth(inode
);
2150 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2151 up_read(&EXT4_I(inode
)->i_data_sem
);
2152 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2156 ex
= path
[depth
].p_ext
;
2157 next
= ext4_ext_next_allocated_block(path
);
2158 ext4_ext_drop_refs(path
);
2163 /* there is no extent yet, so try to allocate
2164 * all requested space */
2167 } else if (le32_to_cpu(ex
->ee_block
) > block
) {
2168 /* need to allocate space before found extent */
2170 end
= le32_to_cpu(ex
->ee_block
);
2171 if (block
+ num
< end
)
2173 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2174 + ext4_ext_get_actual_len(ex
)) {
2175 /* need to allocate space after found extent */
2180 } else if (block
>= le32_to_cpu(ex
->ee_block
)) {
2182 * some part of requested space is covered
2186 end
= le32_to_cpu(ex
->ee_block
)
2187 + ext4_ext_get_actual_len(ex
);
2188 if (block
+ num
< end
)
2194 BUG_ON(end
<= start
);
2198 es
.es_len
= end
- start
;
2201 es
.es_lblk
= le32_to_cpu(ex
->ee_block
);
2202 es
.es_len
= ext4_ext_get_actual_len(ex
);
2203 es
.es_pblk
= ext4_ext_pblock(ex
);
2204 if (ext4_ext_is_unwritten(ex
))
2205 flags
|= FIEMAP_EXTENT_UNWRITTEN
;
2209 * Find delayed extent and update es accordingly. We call
2210 * it even in !exists case to find out whether es is the
2211 * last existing extent or not.
2213 next_del
= ext4_find_delayed_extent(inode
, &es
);
2214 if (!exists
&& next_del
) {
2216 flags
|= (FIEMAP_EXTENT_DELALLOC
|
2217 FIEMAP_EXTENT_UNKNOWN
);
2219 up_read(&EXT4_I(inode
)->i_data_sem
);
2221 if (unlikely(es
.es_len
== 0)) {
2222 EXT4_ERROR_INODE(inode
, "es.es_len == 0");
2228 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2229 * we need to check next == EXT_MAX_BLOCKS because it is
2230 * possible that an extent is with unwritten and delayed
2231 * status due to when an extent is delayed allocated and
2232 * is allocated by fallocate status tree will track both of
2235 * So we could return a unwritten and delayed extent, and
2236 * its block is equal to 'next'.
2238 if (next
== next_del
&& next
== EXT_MAX_BLOCKS
) {
2239 flags
|= FIEMAP_EXTENT_LAST
;
2240 if (unlikely(next_del
!= EXT_MAX_BLOCKS
||
2241 next
!= EXT_MAX_BLOCKS
)) {
2242 EXT4_ERROR_INODE(inode
,
2243 "next extent == %u, next "
2244 "delalloc extent = %u",
2252 err
= fiemap_fill_next_extent(fieinfo
,
2253 (__u64
)es
.es_lblk
<< blksize_bits
,
2254 (__u64
)es
.es_pblk
<< blksize_bits
,
2255 (__u64
)es
.es_len
<< blksize_bits
,
2265 block
= es
.es_lblk
+ es
.es_len
;
2269 ext4_ext_drop_refs(path
);
2277 * ext4_ext_put_gap_in_cache:
2278 * calculate boundaries of the gap that the requested block fits into
2279 * and cache this gap
2282 ext4_ext_put_gap_in_cache(struct inode
*inode
, struct ext4_ext_path
*path
,
2285 int depth
= ext_depth(inode
);
2286 unsigned long len
= 0;
2287 ext4_lblk_t lblock
= 0;
2288 struct ext4_extent
*ex
;
2290 ex
= path
[depth
].p_ext
;
2293 * there is no extent yet, so gap is [0;-] and we
2296 ext_debug("cache gap(whole file):");
2297 } else if (block
< le32_to_cpu(ex
->ee_block
)) {
2299 len
= le32_to_cpu(ex
->ee_block
) - block
;
2300 ext_debug("cache gap(before): %u [%u:%u]",
2302 le32_to_cpu(ex
->ee_block
),
2303 ext4_ext_get_actual_len(ex
));
2304 if (!ext4_find_delalloc_range(inode
, lblock
, lblock
+ len
- 1))
2305 ext4_es_insert_extent(inode
, lblock
, len
, ~0,
2306 EXTENT_STATUS_HOLE
);
2307 } else if (block
>= le32_to_cpu(ex
->ee_block
)
2308 + ext4_ext_get_actual_len(ex
)) {
2310 lblock
= le32_to_cpu(ex
->ee_block
)
2311 + ext4_ext_get_actual_len(ex
);
2313 next
= ext4_ext_next_allocated_block(path
);
2314 ext_debug("cache gap(after): [%u:%u] %u",
2315 le32_to_cpu(ex
->ee_block
),
2316 ext4_ext_get_actual_len(ex
),
2318 BUG_ON(next
== lblock
);
2319 len
= next
- lblock
;
2320 if (!ext4_find_delalloc_range(inode
, lblock
, lblock
+ len
- 1))
2321 ext4_es_insert_extent(inode
, lblock
, len
, ~0,
2322 EXTENT_STATUS_HOLE
);
2327 ext_debug(" -> %u:%lu\n", lblock
, len
);
2332 * removes index from the index block.
2334 static int ext4_ext_rm_idx(handle_t
*handle
, struct inode
*inode
,
2335 struct ext4_ext_path
*path
, int depth
)
2340 /* free index block */
2342 path
= path
+ depth
;
2343 leaf
= ext4_idx_pblock(path
->p_idx
);
2344 if (unlikely(path
->p_hdr
->eh_entries
== 0)) {
2345 EXT4_ERROR_INODE(inode
, "path->p_hdr->eh_entries == 0");
2348 err
= ext4_ext_get_access(handle
, inode
, path
);
2352 if (path
->p_idx
!= EXT_LAST_INDEX(path
->p_hdr
)) {
2353 int len
= EXT_LAST_INDEX(path
->p_hdr
) - path
->p_idx
;
2354 len
*= sizeof(struct ext4_extent_idx
);
2355 memmove(path
->p_idx
, path
->p_idx
+ 1, len
);
2358 le16_add_cpu(&path
->p_hdr
->eh_entries
, -1);
2359 err
= ext4_ext_dirty(handle
, inode
, path
);
2362 ext_debug("index is empty, remove it, free block %llu\n", leaf
);
2363 trace_ext4_ext_rm_idx(inode
, leaf
);
2365 ext4_free_blocks(handle
, inode
, NULL
, leaf
, 1,
2366 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
2368 while (--depth
>= 0) {
2369 if (path
->p_idx
!= EXT_FIRST_INDEX(path
->p_hdr
))
2372 err
= ext4_ext_get_access(handle
, inode
, path
);
2375 path
->p_idx
->ei_block
= (path
+1)->p_idx
->ei_block
;
2376 err
= ext4_ext_dirty(handle
, inode
, path
);
2384 * ext4_ext_calc_credits_for_single_extent:
2385 * This routine returns max. credits that needed to insert an extent
2386 * to the extent tree.
2387 * When pass the actual path, the caller should calculate credits
2390 int ext4_ext_calc_credits_for_single_extent(struct inode
*inode
, int nrblocks
,
2391 struct ext4_ext_path
*path
)
2394 int depth
= ext_depth(inode
);
2397 /* probably there is space in leaf? */
2398 if (le16_to_cpu(path
[depth
].p_hdr
->eh_entries
)
2399 < le16_to_cpu(path
[depth
].p_hdr
->eh_max
)) {
2402 * There are some space in the leaf tree, no
2403 * need to account for leaf block credit
2405 * bitmaps and block group descriptor blocks
2406 * and other metadata blocks still need to be
2409 /* 1 bitmap, 1 block group descriptor */
2410 ret
= 2 + EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
2415 return ext4_chunk_trans_blocks(inode
, nrblocks
);
2419 * How many index/leaf blocks need to change/allocate to add @extents extents?
2421 * If we add a single extent, then in the worse case, each tree level
2422 * index/leaf need to be changed in case of the tree split.
2424 * If more extents are inserted, they could cause the whole tree split more
2425 * than once, but this is really rare.
2427 int ext4_ext_index_trans_blocks(struct inode
*inode
, int extents
)
2432 /* If we are converting the inline data, only one is needed here. */
2433 if (ext4_has_inline_data(inode
))
2436 depth
= ext_depth(inode
);
2446 static inline int get_default_free_blocks_flags(struct inode
*inode
)
2448 if (S_ISDIR(inode
->i_mode
) || S_ISLNK(inode
->i_mode
))
2449 return EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
;
2450 else if (ext4_should_journal_data(inode
))
2451 return EXT4_FREE_BLOCKS_FORGET
;
2455 static int ext4_remove_blocks(handle_t
*handle
, struct inode
*inode
,
2456 struct ext4_extent
*ex
,
2457 long long *partial_cluster
,
2458 ext4_lblk_t from
, ext4_lblk_t to
)
2460 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2461 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
2463 int flags
= get_default_free_blocks_flags(inode
);
2466 * For bigalloc file systems, we never free a partial cluster
2467 * at the beginning of the extent. Instead, we make a note
2468 * that we tried freeing the cluster, and check to see if we
2469 * need to free it on a subsequent call to ext4_remove_blocks,
2470 * or at the end of the ext4_truncate() operation.
2472 flags
|= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER
;
2474 trace_ext4_remove_blocks(inode
, ex
, from
, to
, *partial_cluster
);
2476 * If we have a partial cluster, and it's different from the
2477 * cluster of the last block, we need to explicitly free the
2478 * partial cluster here.
2480 pblk
= ext4_ext_pblock(ex
) + ee_len
- 1;
2481 if ((*partial_cluster
> 0) &&
2482 (EXT4_B2C(sbi
, pblk
) != *partial_cluster
)) {
2483 ext4_free_blocks(handle
, inode
, NULL
,
2484 EXT4_C2B(sbi
, *partial_cluster
),
2485 sbi
->s_cluster_ratio
, flags
);
2486 *partial_cluster
= 0;
2489 #ifdef EXTENTS_STATS
2491 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2492 spin_lock(&sbi
->s_ext_stats_lock
);
2493 sbi
->s_ext_blocks
+= ee_len
;
2494 sbi
->s_ext_extents
++;
2495 if (ee_len
< sbi
->s_ext_min
)
2496 sbi
->s_ext_min
= ee_len
;
2497 if (ee_len
> sbi
->s_ext_max
)
2498 sbi
->s_ext_max
= ee_len
;
2499 if (ext_depth(inode
) > sbi
->s_depth_max
)
2500 sbi
->s_depth_max
= ext_depth(inode
);
2501 spin_unlock(&sbi
->s_ext_stats_lock
);
2504 if (from
>= le32_to_cpu(ex
->ee_block
)
2505 && to
== le32_to_cpu(ex
->ee_block
) + ee_len
- 1) {
2508 unsigned int unaligned
;
2510 num
= le32_to_cpu(ex
->ee_block
) + ee_len
- from
;
2511 pblk
= ext4_ext_pblock(ex
) + ee_len
- num
;
2513 * Usually we want to free partial cluster at the end of the
2514 * extent, except for the situation when the cluster is still
2515 * used by any other extent (partial_cluster is negative).
2517 if (*partial_cluster
< 0 &&
2518 -(*partial_cluster
) == EXT4_B2C(sbi
, pblk
+ num
- 1))
2519 flags
|= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER
;
2521 ext_debug("free last %u blocks starting %llu partial %lld\n",
2522 num
, pblk
, *partial_cluster
);
2523 ext4_free_blocks(handle
, inode
, NULL
, pblk
, num
, flags
);
2525 * If the block range to be freed didn't start at the
2526 * beginning of a cluster, and we removed the entire
2527 * extent and the cluster is not used by any other extent,
2528 * save the partial cluster here, since we might need to
2529 * delete if we determine that the truncate operation has
2530 * removed all of the blocks in the cluster.
2532 * On the other hand, if we did not manage to free the whole
2533 * extent, we have to mark the cluster as used (store negative
2534 * cluster number in partial_cluster).
2536 unaligned
= EXT4_PBLK_COFF(sbi
, pblk
);
2537 if (unaligned
&& (ee_len
== num
) &&
2538 (*partial_cluster
!= -((long long)EXT4_B2C(sbi
, pblk
))))
2539 *partial_cluster
= EXT4_B2C(sbi
, pblk
);
2541 *partial_cluster
= -((long long)EXT4_B2C(sbi
, pblk
));
2542 else if (*partial_cluster
> 0)
2543 *partial_cluster
= 0;
2545 ext4_error(sbi
->s_sb
, "strange request: removal(2) "
2546 "%u-%u from %u:%u\n",
2547 from
, to
, le32_to_cpu(ex
->ee_block
), ee_len
);
2553 * ext4_ext_rm_leaf() Removes the extents associated with the
2554 * blocks appearing between "start" and "end", and splits the extents
2555 * if "start" and "end" appear in the same extent
2557 * @handle: The journal handle
2558 * @inode: The files inode
2559 * @path: The path to the leaf
2560 * @partial_cluster: The cluster which we'll have to free if all extents
2561 * has been released from it. It gets negative in case
2562 * that the cluster is still used.
2563 * @start: The first block to remove
2564 * @end: The last block to remove
2567 ext4_ext_rm_leaf(handle_t
*handle
, struct inode
*inode
,
2568 struct ext4_ext_path
*path
,
2569 long long *partial_cluster
,
2570 ext4_lblk_t start
, ext4_lblk_t end
)
2572 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
2573 int err
= 0, correct_index
= 0;
2574 int depth
= ext_depth(inode
), credits
;
2575 struct ext4_extent_header
*eh
;
2578 ext4_lblk_t ex_ee_block
;
2579 unsigned short ex_ee_len
;
2580 unsigned unwritten
= 0;
2581 struct ext4_extent
*ex
;
2584 /* the header must be checked already in ext4_ext_remove_space() */
2585 ext_debug("truncate since %u in leaf to %u\n", start
, end
);
2586 if (!path
[depth
].p_hdr
)
2587 path
[depth
].p_hdr
= ext_block_hdr(path
[depth
].p_bh
);
2588 eh
= path
[depth
].p_hdr
;
2589 if (unlikely(path
[depth
].p_hdr
== NULL
)) {
2590 EXT4_ERROR_INODE(inode
, "path[%d].p_hdr == NULL", depth
);
2593 /* find where to start removing */
2594 ex
= path
[depth
].p_ext
;
2596 ex
= EXT_LAST_EXTENT(eh
);
2598 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2599 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2602 * If we're starting with an extent other than the last one in the
2603 * node, we need to see if it shares a cluster with the extent to
2604 * the right (towards the end of the file). If its leftmost cluster
2605 * is this extent's rightmost cluster and it is not cluster aligned,
2606 * we'll mark it as a partial that is not to be deallocated.
2609 if (ex
!= EXT_LAST_EXTENT(eh
)) {
2610 ext4_fsblk_t current_pblk
, right_pblk
;
2611 long long current_cluster
, right_cluster
;
2613 current_pblk
= ext4_ext_pblock(ex
) + ex_ee_len
- 1;
2614 current_cluster
= (long long)EXT4_B2C(sbi
, current_pblk
);
2615 right_pblk
= ext4_ext_pblock(ex
+ 1);
2616 right_cluster
= (long long)EXT4_B2C(sbi
, right_pblk
);
2617 if (current_cluster
== right_cluster
&&
2618 EXT4_PBLK_COFF(sbi
, right_pblk
))
2619 *partial_cluster
= -right_cluster
;
2622 trace_ext4_ext_rm_leaf(inode
, start
, ex
, *partial_cluster
);
2624 while (ex
>= EXT_FIRST_EXTENT(eh
) &&
2625 ex_ee_block
+ ex_ee_len
> start
) {
2627 if (ext4_ext_is_unwritten(ex
))
2632 ext_debug("remove ext %u:[%d]%d\n", ex_ee_block
,
2633 unwritten
, ex_ee_len
);
2634 path
[depth
].p_ext
= ex
;
2636 a
= ex_ee_block
> start
? ex_ee_block
: start
;
2637 b
= ex_ee_block
+ex_ee_len
- 1 < end
?
2638 ex_ee_block
+ex_ee_len
- 1 : end
;
2640 ext_debug(" border %u:%u\n", a
, b
);
2642 /* If this extent is beyond the end of the hole, skip it */
2643 if (end
< ex_ee_block
) {
2645 * We're going to skip this extent and move to another,
2646 * so if this extent is not cluster aligned we have
2647 * to mark the current cluster as used to avoid
2648 * accidentally freeing it later on
2650 pblk
= ext4_ext_pblock(ex
);
2651 if (EXT4_PBLK_COFF(sbi
, pblk
))
2653 -((long long)EXT4_B2C(sbi
, pblk
));
2655 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2656 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2658 } else if (b
!= ex_ee_block
+ ex_ee_len
- 1) {
2659 EXT4_ERROR_INODE(inode
,
2660 "can not handle truncate %u:%u "
2662 start
, end
, ex_ee_block
,
2663 ex_ee_block
+ ex_ee_len
- 1);
2666 } else if (a
!= ex_ee_block
) {
2667 /* remove tail of the extent */
2668 num
= a
- ex_ee_block
;
2670 /* remove whole extent: excellent! */
2674 * 3 for leaf, sb, and inode plus 2 (bmap and group
2675 * descriptor) for each block group; assume two block
2676 * groups plus ex_ee_len/blocks_per_block_group for
2679 credits
= 7 + 2*(ex_ee_len
/EXT4_BLOCKS_PER_GROUP(inode
->i_sb
));
2680 if (ex
== EXT_FIRST_EXTENT(eh
)) {
2682 credits
+= (ext_depth(inode
)) + 1;
2684 credits
+= EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
2686 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
2690 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
2694 err
= ext4_remove_blocks(handle
, inode
, ex
, partial_cluster
,
2700 /* this extent is removed; mark slot entirely unused */
2701 ext4_ext_store_pblock(ex
, 0);
2703 ex
->ee_len
= cpu_to_le16(num
);
2705 * Do not mark unwritten if all the blocks in the
2706 * extent have been removed.
2708 if (unwritten
&& num
)
2709 ext4_ext_mark_unwritten(ex
);
2711 * If the extent was completely released,
2712 * we need to remove it from the leaf
2715 if (end
!= EXT_MAX_BLOCKS
- 1) {
2717 * For hole punching, we need to scoot all the
2718 * extents up when an extent is removed so that
2719 * we dont have blank extents in the middle
2721 memmove(ex
, ex
+1, (EXT_LAST_EXTENT(eh
) - ex
) *
2722 sizeof(struct ext4_extent
));
2724 /* Now get rid of the one at the end */
2725 memset(EXT_LAST_EXTENT(eh
), 0,
2726 sizeof(struct ext4_extent
));
2728 le16_add_cpu(&eh
->eh_entries
, -1);
2729 } else if (*partial_cluster
> 0)
2730 *partial_cluster
= 0;
2732 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
2736 ext_debug("new extent: %u:%u:%llu\n", ex_ee_block
, num
,
2737 ext4_ext_pblock(ex
));
2739 ex_ee_block
= le32_to_cpu(ex
->ee_block
);
2740 ex_ee_len
= ext4_ext_get_actual_len(ex
);
2743 if (correct_index
&& eh
->eh_entries
)
2744 err
= ext4_ext_correct_indexes(handle
, inode
, path
);
2747 * If there's a partial cluster and at least one extent remains in
2748 * the leaf, free the partial cluster if it isn't shared with the
2749 * current extent. If there's a partial cluster and no extents
2750 * remain in the leaf, it can't be freed here. It can only be
2751 * freed when it's possible to determine if it's not shared with
2752 * any other extent - when the next leaf is processed or when space
2753 * removal is complete.
2755 if (*partial_cluster
> 0 && eh
->eh_entries
&&
2756 (EXT4_B2C(sbi
, ext4_ext_pblock(ex
) + ex_ee_len
- 1) !=
2757 *partial_cluster
)) {
2758 int flags
= get_default_free_blocks_flags(inode
);
2760 ext4_free_blocks(handle
, inode
, NULL
,
2761 EXT4_C2B(sbi
, *partial_cluster
),
2762 sbi
->s_cluster_ratio
, flags
);
2763 *partial_cluster
= 0;
2766 /* if this leaf is free, then we should
2767 * remove it from index block above */
2768 if (err
== 0 && eh
->eh_entries
== 0 && path
[depth
].p_bh
!= NULL
)
2769 err
= ext4_ext_rm_idx(handle
, inode
, path
, depth
);
2776 * ext4_ext_more_to_rm:
2777 * returns 1 if current index has to be freed (even partial)
2780 ext4_ext_more_to_rm(struct ext4_ext_path
*path
)
2782 BUG_ON(path
->p_idx
== NULL
);
2784 if (path
->p_idx
< EXT_FIRST_INDEX(path
->p_hdr
))
2788 * if truncate on deeper level happened, it wasn't partial,
2789 * so we have to consider current index for truncation
2791 if (le16_to_cpu(path
->p_hdr
->eh_entries
) == path
->p_block
)
2796 int ext4_ext_remove_space(struct inode
*inode
, ext4_lblk_t start
,
2799 struct super_block
*sb
= inode
->i_sb
;
2800 int depth
= ext_depth(inode
);
2801 struct ext4_ext_path
*path
= NULL
;
2802 long long partial_cluster
= 0;
2806 ext_debug("truncate since %u to %u\n", start
, end
);
2808 /* probably first extent we're gonna free will be last in block */
2809 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, depth
+ 1);
2811 return PTR_ERR(handle
);
2814 trace_ext4_ext_remove_space(inode
, start
, end
, depth
);
2817 * Check if we are removing extents inside the extent tree. If that
2818 * is the case, we are going to punch a hole inside the extent tree
2819 * so we have to check whether we need to split the extent covering
2820 * the last block to remove so we can easily remove the part of it
2821 * in ext4_ext_rm_leaf().
2823 if (end
< EXT_MAX_BLOCKS
- 1) {
2824 struct ext4_extent
*ex
;
2825 ext4_lblk_t ee_block
;
2827 /* find extent for this block */
2828 path
= ext4_ext_find_extent(inode
, end
, NULL
, EXT4_EX_NOCACHE
);
2830 ext4_journal_stop(handle
);
2831 return PTR_ERR(path
);
2833 depth
= ext_depth(inode
);
2834 /* Leaf not may not exist only if inode has no blocks at all */
2835 ex
= path
[depth
].p_ext
;
2838 EXT4_ERROR_INODE(inode
,
2839 "path[%d].p_hdr == NULL",
2846 ee_block
= le32_to_cpu(ex
->ee_block
);
2849 * See if the last block is inside the extent, if so split
2850 * the extent at 'end' block so we can easily remove the
2851 * tail of the first part of the split extent in
2852 * ext4_ext_rm_leaf().
2854 if (end
>= ee_block
&&
2855 end
< ee_block
+ ext4_ext_get_actual_len(ex
) - 1) {
2858 if (ext4_ext_is_unwritten(ex
))
2859 split_flag
= EXT4_EXT_MARK_UNWRIT1
|
2860 EXT4_EXT_MARK_UNWRIT2
;
2863 * Split the extent in two so that 'end' is the last
2864 * block in the first new extent. Also we should not
2865 * fail removing space due to ENOSPC so try to use
2866 * reserved block if that happens.
2868 err
= ext4_split_extent_at(handle
, inode
, path
,
2869 end
+ 1, split_flag
,
2871 EXT4_GET_BLOCKS_PRE_IO
|
2872 EXT4_GET_BLOCKS_METADATA_NOFAIL
);
2879 * We start scanning from right side, freeing all the blocks
2880 * after i_size and walking into the tree depth-wise.
2882 depth
= ext_depth(inode
);
2887 le16_to_cpu(path
[k
].p_hdr
->eh_entries
)+1;
2889 path
= kzalloc(sizeof(struct ext4_ext_path
) * (depth
+ 1),
2892 ext4_journal_stop(handle
);
2895 path
[0].p_depth
= depth
;
2896 path
[0].p_hdr
= ext_inode_hdr(inode
);
2899 if (ext4_ext_check(inode
, path
[0].p_hdr
, depth
, 0)) {
2906 while (i
>= 0 && err
== 0) {
2908 /* this is leaf block */
2909 err
= ext4_ext_rm_leaf(handle
, inode
, path
,
2910 &partial_cluster
, start
,
2912 /* root level has p_bh == NULL, brelse() eats this */
2913 brelse(path
[i
].p_bh
);
2914 path
[i
].p_bh
= NULL
;
2919 /* this is index block */
2920 if (!path
[i
].p_hdr
) {
2921 ext_debug("initialize header\n");
2922 path
[i
].p_hdr
= ext_block_hdr(path
[i
].p_bh
);
2925 if (!path
[i
].p_idx
) {
2926 /* this level hasn't been touched yet */
2927 path
[i
].p_idx
= EXT_LAST_INDEX(path
[i
].p_hdr
);
2928 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
)+1;
2929 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2931 le16_to_cpu(path
[i
].p_hdr
->eh_entries
));
2933 /* we were already here, see at next index */
2937 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2938 i
, EXT_FIRST_INDEX(path
[i
].p_hdr
),
2940 if (ext4_ext_more_to_rm(path
+ i
)) {
2941 struct buffer_head
*bh
;
2942 /* go to the next level */
2943 ext_debug("move to level %d (block %llu)\n",
2944 i
+ 1, ext4_idx_pblock(path
[i
].p_idx
));
2945 memset(path
+ i
+ 1, 0, sizeof(*path
));
2946 bh
= read_extent_tree_block(inode
,
2947 ext4_idx_pblock(path
[i
].p_idx
), depth
- i
- 1,
2950 /* should we reset i_size? */
2954 /* Yield here to deal with large extent trees.
2955 * Should be a no-op if we did IO above. */
2957 if (WARN_ON(i
+ 1 > depth
)) {
2961 path
[i
+ 1].p_bh
= bh
;
2963 /* save actual number of indexes since this
2964 * number is changed at the next iteration */
2965 path
[i
].p_block
= le16_to_cpu(path
[i
].p_hdr
->eh_entries
);
2968 /* we finished processing this index, go up */
2969 if (path
[i
].p_hdr
->eh_entries
== 0 && i
> 0) {
2970 /* index is empty, remove it;
2971 * handle must be already prepared by the
2972 * truncatei_leaf() */
2973 err
= ext4_ext_rm_idx(handle
, inode
, path
, i
);
2975 /* root level has p_bh == NULL, brelse() eats this */
2976 brelse(path
[i
].p_bh
);
2977 path
[i
].p_bh
= NULL
;
2979 ext_debug("return to level %d\n", i
);
2983 trace_ext4_ext_remove_space_done(inode
, start
, end
, depth
,
2984 partial_cluster
, path
->p_hdr
->eh_entries
);
2986 /* If we still have something in the partial cluster and we have removed
2987 * even the first extent, then we should free the blocks in the partial
2988 * cluster as well. */
2989 if (partial_cluster
> 0 && path
->p_hdr
->eh_entries
== 0) {
2990 int flags
= get_default_free_blocks_flags(inode
);
2992 ext4_free_blocks(handle
, inode
, NULL
,
2993 EXT4_C2B(EXT4_SB(sb
), partial_cluster
),
2994 EXT4_SB(sb
)->s_cluster_ratio
, flags
);
2995 partial_cluster
= 0;
2998 /* TODO: flexible tree reduction should be here */
2999 if (path
->p_hdr
->eh_entries
== 0) {
3001 * truncate to zero freed all the tree,
3002 * so we need to correct eh_depth
3004 err
= ext4_ext_get_access(handle
, inode
, path
);
3006 ext_inode_hdr(inode
)->eh_depth
= 0;
3007 ext_inode_hdr(inode
)->eh_max
=
3008 cpu_to_le16(ext4_ext_space_root(inode
, 0));
3009 err
= ext4_ext_dirty(handle
, inode
, path
);
3013 ext4_ext_drop_refs(path
);
3015 if (err
== -EAGAIN
) {
3019 ext4_journal_stop(handle
);
3025 * called at mount time
3027 void ext4_ext_init(struct super_block
*sb
)
3030 * possible initialization would be here
3033 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
)) {
3034 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3035 printk(KERN_INFO
"EXT4-fs: file extents enabled"
3036 #ifdef AGGRESSIVE_TEST
3037 ", aggressive tests"
3039 #ifdef CHECK_BINSEARCH
3042 #ifdef EXTENTS_STATS
3047 #ifdef EXTENTS_STATS
3048 spin_lock_init(&EXT4_SB(sb
)->s_ext_stats_lock
);
3049 EXT4_SB(sb
)->s_ext_min
= 1 << 30;
3050 EXT4_SB(sb
)->s_ext_max
= 0;
3056 * called at umount time
3058 void ext4_ext_release(struct super_block
*sb
)
3060 if (!EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_EXTENTS
))
3063 #ifdef EXTENTS_STATS
3064 if (EXT4_SB(sb
)->s_ext_blocks
&& EXT4_SB(sb
)->s_ext_extents
) {
3065 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
3066 printk(KERN_ERR
"EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3067 sbi
->s_ext_blocks
, sbi
->s_ext_extents
,
3068 sbi
->s_ext_blocks
/ sbi
->s_ext_extents
);
3069 printk(KERN_ERR
"EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3070 sbi
->s_ext_min
, sbi
->s_ext_max
, sbi
->s_depth_max
);
3075 static int ext4_zeroout_es(struct inode
*inode
, struct ext4_extent
*ex
)
3077 ext4_lblk_t ee_block
;
3078 ext4_fsblk_t ee_pblock
;
3079 unsigned int ee_len
;
3081 ee_block
= le32_to_cpu(ex
->ee_block
);
3082 ee_len
= ext4_ext_get_actual_len(ex
);
3083 ee_pblock
= ext4_ext_pblock(ex
);
3088 return ext4_es_insert_extent(inode
, ee_block
, ee_len
, ee_pblock
,
3089 EXTENT_STATUS_WRITTEN
);
3092 /* FIXME!! we need to try to merge to left or right after zero-out */
3093 static int ext4_ext_zeroout(struct inode
*inode
, struct ext4_extent
*ex
)
3095 ext4_fsblk_t ee_pblock
;
3096 unsigned int ee_len
;
3099 ee_len
= ext4_ext_get_actual_len(ex
);
3100 ee_pblock
= ext4_ext_pblock(ex
);
3102 ret
= sb_issue_zeroout(inode
->i_sb
, ee_pblock
, ee_len
, GFP_NOFS
);
3110 * ext4_split_extent_at() splits an extent at given block.
3112 * @handle: the journal handle
3113 * @inode: the file inode
3114 * @path: the path to the extent
3115 * @split: the logical block where the extent is splitted.
3116 * @split_flags: indicates if the extent could be zeroout if split fails, and
3117 * the states(init or unwritten) of new extents.
3118 * @flags: flags used to insert new extent to extent tree.
3121 * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3122 * of which are deterimined by split_flag.
3124 * There are two cases:
3125 * a> the extent are splitted into two extent.
3126 * b> split is not needed, and just mark the extent.
3128 * return 0 on success.
3130 static int ext4_split_extent_at(handle_t
*handle
,
3131 struct inode
*inode
,
3132 struct ext4_ext_path
*path
,
3137 ext4_fsblk_t newblock
;
3138 ext4_lblk_t ee_block
;
3139 struct ext4_extent
*ex
, newex
, orig_ex
, zero_ex
;
3140 struct ext4_extent
*ex2
= NULL
;
3141 unsigned int ee_len
, depth
;
3144 BUG_ON((split_flag
& (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
)) ==
3145 (EXT4_EXT_DATA_VALID1
| EXT4_EXT_DATA_VALID2
));
3147 ext_debug("ext4_split_extents_at: inode %lu, logical"
3148 "block %llu\n", inode
->i_ino
, (unsigned long long)split
);
3150 ext4_ext_show_leaf(inode
, path
);
3152 depth
= ext_depth(inode
);
3153 ex
= path
[depth
].p_ext
;
3154 ee_block
= le32_to_cpu(ex
->ee_block
);
3155 ee_len
= ext4_ext_get_actual_len(ex
);
3156 newblock
= split
- ee_block
+ ext4_ext_pblock(ex
);
3158 BUG_ON(split
< ee_block
|| split
>= (ee_block
+ ee_len
));
3159 BUG_ON(!ext4_ext_is_unwritten(ex
) &&
3160 split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3161 EXT4_EXT_MARK_UNWRIT1
|
3162 EXT4_EXT_MARK_UNWRIT2
));
3164 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3168 if (split
== ee_block
) {
3170 * case b: block @split is the block that the extent begins with
3171 * then we just change the state of the extent, and splitting
3174 if (split_flag
& EXT4_EXT_MARK_UNWRIT2
)
3175 ext4_ext_mark_unwritten(ex
);
3177 ext4_ext_mark_initialized(ex
);
3179 if (!(flags
& EXT4_GET_BLOCKS_PRE_IO
))
3180 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3182 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3187 memcpy(&orig_ex
, ex
, sizeof(orig_ex
));
3188 ex
->ee_len
= cpu_to_le16(split
- ee_block
);
3189 if (split_flag
& EXT4_EXT_MARK_UNWRIT1
)
3190 ext4_ext_mark_unwritten(ex
);
3193 * path may lead to new leaf, not to original leaf any more
3194 * after ext4_ext_insert_extent() returns,
3196 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
3198 goto fix_extent_len
;
3201 ex2
->ee_block
= cpu_to_le32(split
);
3202 ex2
->ee_len
= cpu_to_le16(ee_len
- (split
- ee_block
));
3203 ext4_ext_store_pblock(ex2
, newblock
);
3204 if (split_flag
& EXT4_EXT_MARK_UNWRIT2
)
3205 ext4_ext_mark_unwritten(ex2
);
3207 err
= ext4_ext_insert_extent(handle
, inode
, path
, &newex
, flags
);
3208 if (err
== -ENOSPC
&& (EXT4_EXT_MAY_ZEROOUT
& split_flag
)) {
3209 if (split_flag
& (EXT4_EXT_DATA_VALID1
|EXT4_EXT_DATA_VALID2
)) {
3210 if (split_flag
& EXT4_EXT_DATA_VALID1
) {
3211 err
= ext4_ext_zeroout(inode
, ex2
);
3212 zero_ex
.ee_block
= ex2
->ee_block
;
3213 zero_ex
.ee_len
= cpu_to_le16(
3214 ext4_ext_get_actual_len(ex2
));
3215 ext4_ext_store_pblock(&zero_ex
,
3216 ext4_ext_pblock(ex2
));
3218 err
= ext4_ext_zeroout(inode
, ex
);
3219 zero_ex
.ee_block
= ex
->ee_block
;
3220 zero_ex
.ee_len
= cpu_to_le16(
3221 ext4_ext_get_actual_len(ex
));
3222 ext4_ext_store_pblock(&zero_ex
,
3223 ext4_ext_pblock(ex
));
3226 err
= ext4_ext_zeroout(inode
, &orig_ex
);
3227 zero_ex
.ee_block
= orig_ex
.ee_block
;
3228 zero_ex
.ee_len
= cpu_to_le16(
3229 ext4_ext_get_actual_len(&orig_ex
));
3230 ext4_ext_store_pblock(&zero_ex
,
3231 ext4_ext_pblock(&orig_ex
));
3235 goto fix_extent_len
;
3236 /* update the extent length and mark as initialized */
3237 ex
->ee_len
= cpu_to_le16(ee_len
);
3238 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3239 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3241 goto fix_extent_len
;
3243 /* update extent status tree */
3244 err
= ext4_zeroout_es(inode
, &zero_ex
);
3248 goto fix_extent_len
;
3251 ext4_ext_show_leaf(inode
, path
);
3255 ex
->ee_len
= orig_ex
.ee_len
;
3256 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3261 * ext4_split_extents() splits an extent and mark extent which is covered
3262 * by @map as split_flags indicates
3264 * It may result in splitting the extent into multiple extents (up to three)
3265 * There are three possibilities:
3266 * a> There is no split required
3267 * b> Splits in two extents: Split is happening at either end of the extent
3268 * c> Splits in three extents: Somone is splitting in middle of the extent
3271 static int ext4_split_extent(handle_t
*handle
,
3272 struct inode
*inode
,
3273 struct ext4_ext_path
*path
,
3274 struct ext4_map_blocks
*map
,
3278 ext4_lblk_t ee_block
;
3279 struct ext4_extent
*ex
;
3280 unsigned int ee_len
, depth
;
3283 int split_flag1
, flags1
;
3284 int allocated
= map
->m_len
;
3286 depth
= ext_depth(inode
);
3287 ex
= path
[depth
].p_ext
;
3288 ee_block
= le32_to_cpu(ex
->ee_block
);
3289 ee_len
= ext4_ext_get_actual_len(ex
);
3290 unwritten
= ext4_ext_is_unwritten(ex
);
3292 if (map
->m_lblk
+ map
->m_len
< ee_block
+ ee_len
) {
3293 split_flag1
= split_flag
& EXT4_EXT_MAY_ZEROOUT
;
3294 flags1
= flags
| EXT4_GET_BLOCKS_PRE_IO
;
3296 split_flag1
|= EXT4_EXT_MARK_UNWRIT1
|
3297 EXT4_EXT_MARK_UNWRIT2
;
3298 if (split_flag
& EXT4_EXT_DATA_VALID2
)
3299 split_flag1
|= EXT4_EXT_DATA_VALID1
;
3300 err
= ext4_split_extent_at(handle
, inode
, path
,
3301 map
->m_lblk
+ map
->m_len
, split_flag1
, flags1
);
3305 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3308 * Update path is required because previous ext4_split_extent_at() may
3309 * result in split of original leaf or extent zeroout.
3311 ext4_ext_drop_refs(path
);
3312 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
, 0);
3314 return PTR_ERR(path
);
3315 depth
= ext_depth(inode
);
3316 ex
= path
[depth
].p_ext
;
3318 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
3319 (unsigned long) map
->m_lblk
);
3322 unwritten
= ext4_ext_is_unwritten(ex
);
3325 if (map
->m_lblk
>= ee_block
) {
3326 split_flag1
= split_flag
& EXT4_EXT_DATA_VALID2
;
3328 split_flag1
|= EXT4_EXT_MARK_UNWRIT1
;
3329 split_flag1
|= split_flag
& (EXT4_EXT_MAY_ZEROOUT
|
3330 EXT4_EXT_MARK_UNWRIT2
);
3332 err
= ext4_split_extent_at(handle
, inode
, path
,
3333 map
->m_lblk
, split_flag1
, flags
);
3338 ext4_ext_show_leaf(inode
, path
);
3340 return err
? err
: allocated
;
3344 * This function is called by ext4_ext_map_blocks() if someone tries to write
3345 * to an unwritten extent. It may result in splitting the unwritten
3346 * extent into multiple extents (up to three - one initialized and two
3348 * There are three possibilities:
3349 * a> There is no split required: Entire extent should be initialized
3350 * b> Splits in two extents: Write is happening at either end of the extent
3351 * c> Splits in three extents: Somone is writing in middle of the extent
3354 * - The extent pointed to by 'path' is unwritten.
3355 * - The extent pointed to by 'path' contains a superset
3356 * of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3358 * Post-conditions on success:
3359 * - the returned value is the number of blocks beyond map->l_lblk
3360 * that are allocated and initialized.
3361 * It is guaranteed to be >= map->m_len.
3363 static int ext4_ext_convert_to_initialized(handle_t
*handle
,
3364 struct inode
*inode
,
3365 struct ext4_map_blocks
*map
,
3366 struct ext4_ext_path
*path
,
3369 struct ext4_sb_info
*sbi
;
3370 struct ext4_extent_header
*eh
;
3371 struct ext4_map_blocks split_map
;
3372 struct ext4_extent zero_ex
;
3373 struct ext4_extent
*ex
, *abut_ex
;
3374 ext4_lblk_t ee_block
, eof_block
;
3375 unsigned int ee_len
, depth
, map_len
= map
->m_len
;
3376 int allocated
= 0, max_zeroout
= 0;
3380 ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3381 "block %llu, max_blocks %u\n", inode
->i_ino
,
3382 (unsigned long long)map
->m_lblk
, map_len
);
3384 sbi
= EXT4_SB(inode
->i_sb
);
3385 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3386 inode
->i_sb
->s_blocksize_bits
;
3387 if (eof_block
< map
->m_lblk
+ map_len
)
3388 eof_block
= map
->m_lblk
+ map_len
;
3390 depth
= ext_depth(inode
);
3391 eh
= path
[depth
].p_hdr
;
3392 ex
= path
[depth
].p_ext
;
3393 ee_block
= le32_to_cpu(ex
->ee_block
);
3394 ee_len
= ext4_ext_get_actual_len(ex
);
3397 trace_ext4_ext_convert_to_initialized_enter(inode
, map
, ex
);
3399 /* Pre-conditions */
3400 BUG_ON(!ext4_ext_is_unwritten(ex
));
3401 BUG_ON(!in_range(map
->m_lblk
, ee_block
, ee_len
));
3404 * Attempt to transfer newly initialized blocks from the currently
3405 * unwritten extent to its neighbor. This is much cheaper
3406 * than an insertion followed by a merge as those involve costly
3407 * memmove() calls. Transferring to the left is the common case in
3408 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3409 * followed by append writes.
3411 * Limitations of the current logic:
3412 * - L1: we do not deal with writes covering the whole extent.
3413 * This would require removing the extent if the transfer
3415 * - L2: we only attempt to merge with an extent stored in the
3416 * same extent tree node.
3418 if ((map
->m_lblk
== ee_block
) &&
3419 /* See if we can merge left */
3420 (map_len
< ee_len
) && /*L1*/
3421 (ex
> EXT_FIRST_EXTENT(eh
))) { /*L2*/
3422 ext4_lblk_t prev_lblk
;
3423 ext4_fsblk_t prev_pblk
, ee_pblk
;
3424 unsigned int prev_len
;
3427 prev_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3428 prev_len
= ext4_ext_get_actual_len(abut_ex
);
3429 prev_pblk
= ext4_ext_pblock(abut_ex
);
3430 ee_pblk
= ext4_ext_pblock(ex
);
3433 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3434 * upon those conditions:
3435 * - C1: abut_ex is initialized,
3436 * - C2: abut_ex is logically abutting ex,
3437 * - C3: abut_ex is physically abutting ex,
3438 * - C4: abut_ex can receive the additional blocks without
3439 * overflowing the (initialized) length limit.
3441 if ((!ext4_ext_is_unwritten(abut_ex
)) && /*C1*/
3442 ((prev_lblk
+ prev_len
) == ee_block
) && /*C2*/
3443 ((prev_pblk
+ prev_len
) == ee_pblk
) && /*C3*/
3444 (prev_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3445 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3449 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3452 /* Shift the start of ex by 'map_len' blocks */
3453 ex
->ee_block
= cpu_to_le32(ee_block
+ map_len
);
3454 ext4_ext_store_pblock(ex
, ee_pblk
+ map_len
);
3455 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3456 ext4_ext_mark_unwritten(ex
); /* Restore the flag */
3458 /* Extend abut_ex by 'map_len' blocks */
3459 abut_ex
->ee_len
= cpu_to_le16(prev_len
+ map_len
);
3461 /* Result: number of initialized blocks past m_lblk */
3462 allocated
= map_len
;
3464 } else if (((map
->m_lblk
+ map_len
) == (ee_block
+ ee_len
)) &&
3465 (map_len
< ee_len
) && /*L1*/
3466 ex
< EXT_LAST_EXTENT(eh
)) { /*L2*/
3467 /* See if we can merge right */
3468 ext4_lblk_t next_lblk
;
3469 ext4_fsblk_t next_pblk
, ee_pblk
;
3470 unsigned int next_len
;
3473 next_lblk
= le32_to_cpu(abut_ex
->ee_block
);
3474 next_len
= ext4_ext_get_actual_len(abut_ex
);
3475 next_pblk
= ext4_ext_pblock(abut_ex
);
3476 ee_pblk
= ext4_ext_pblock(ex
);
3479 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3480 * upon those conditions:
3481 * - C1: abut_ex is initialized,
3482 * - C2: abut_ex is logically abutting ex,
3483 * - C3: abut_ex is physically abutting ex,
3484 * - C4: abut_ex can receive the additional blocks without
3485 * overflowing the (initialized) length limit.
3487 if ((!ext4_ext_is_unwritten(abut_ex
)) && /*C1*/
3488 ((map
->m_lblk
+ map_len
) == next_lblk
) && /*C2*/
3489 ((ee_pblk
+ ee_len
) == next_pblk
) && /*C3*/
3490 (next_len
< (EXT_INIT_MAX_LEN
- map_len
))) { /*C4*/
3491 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3495 trace_ext4_ext_convert_to_initialized_fastpath(inode
,
3498 /* Shift the start of abut_ex by 'map_len' blocks */
3499 abut_ex
->ee_block
= cpu_to_le32(next_lblk
- map_len
);
3500 ext4_ext_store_pblock(abut_ex
, next_pblk
- map_len
);
3501 ex
->ee_len
= cpu_to_le16(ee_len
- map_len
);
3502 ext4_ext_mark_unwritten(ex
); /* Restore the flag */
3504 /* Extend abut_ex by 'map_len' blocks */
3505 abut_ex
->ee_len
= cpu_to_le16(next_len
+ map_len
);
3507 /* Result: number of initialized blocks past m_lblk */
3508 allocated
= map_len
;
3512 /* Mark the block containing both extents as dirty */
3513 ext4_ext_dirty(handle
, inode
, path
+ depth
);
3515 /* Update path to point to the right extent */
3516 path
[depth
].p_ext
= abut_ex
;
3519 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
3521 WARN_ON(map
->m_lblk
< ee_block
);
3523 * It is safe to convert extent to initialized via explicit
3524 * zeroout only if extent is fully inside i_size or new_size.
3526 split_flag
|= ee_block
+ ee_len
<= eof_block
? EXT4_EXT_MAY_ZEROOUT
: 0;
3528 if (EXT4_EXT_MAY_ZEROOUT
& split_flag
)
3529 max_zeroout
= sbi
->s_extent_max_zeroout_kb
>>
3530 (inode
->i_sb
->s_blocksize_bits
- 10);
3532 /* If extent is less than s_max_zeroout_kb, zeroout directly */
3533 if (max_zeroout
&& (ee_len
<= max_zeroout
)) {
3534 err
= ext4_ext_zeroout(inode
, ex
);
3537 zero_ex
.ee_block
= ex
->ee_block
;
3538 zero_ex
.ee_len
= cpu_to_le16(ext4_ext_get_actual_len(ex
));
3539 ext4_ext_store_pblock(&zero_ex
, ext4_ext_pblock(ex
));
3541 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3544 ext4_ext_mark_initialized(ex
);
3545 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3546 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3552 * 1. split the extent into three extents.
3553 * 2. split the extent into two extents, zeroout the first half.
3554 * 3. split the extent into two extents, zeroout the second half.
3555 * 4. split the extent into two extents with out zeroout.
3557 split_map
.m_lblk
= map
->m_lblk
;
3558 split_map
.m_len
= map
->m_len
;
3560 if (max_zeroout
&& (allocated
> map
->m_len
)) {
3561 if (allocated
<= max_zeroout
) {
3564 cpu_to_le32(map
->m_lblk
);
3565 zero_ex
.ee_len
= cpu_to_le16(allocated
);
3566 ext4_ext_store_pblock(&zero_ex
,
3567 ext4_ext_pblock(ex
) + map
->m_lblk
- ee_block
);
3568 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3571 split_map
.m_lblk
= map
->m_lblk
;
3572 split_map
.m_len
= allocated
;
3573 } else if (map
->m_lblk
- ee_block
+ map
->m_len
< max_zeroout
) {
3575 if (map
->m_lblk
!= ee_block
) {
3576 zero_ex
.ee_block
= ex
->ee_block
;
3577 zero_ex
.ee_len
= cpu_to_le16(map
->m_lblk
-
3579 ext4_ext_store_pblock(&zero_ex
,
3580 ext4_ext_pblock(ex
));
3581 err
= ext4_ext_zeroout(inode
, &zero_ex
);
3586 split_map
.m_lblk
= ee_block
;
3587 split_map
.m_len
= map
->m_lblk
- ee_block
+ map
->m_len
;
3588 allocated
= map
->m_len
;
3592 allocated
= ext4_split_extent(handle
, inode
, path
,
3593 &split_map
, split_flag
, flags
);
3598 /* If we have gotten a failure, don't zero out status tree */
3600 err
= ext4_zeroout_es(inode
, &zero_ex
);
3601 return err
? err
: allocated
;
3605 * This function is called by ext4_ext_map_blocks() from
3606 * ext4_get_blocks_dio_write() when DIO to write
3607 * to an unwritten extent.
3609 * Writing to an unwritten extent may result in splitting the unwritten
3610 * extent into multiple initialized/unwritten extents (up to three)
3611 * There are three possibilities:
3612 * a> There is no split required: Entire extent should be unwritten
3613 * b> Splits in two extents: Write is happening at either end of the extent
3614 * c> Splits in three extents: Somone is writing in middle of the extent
3616 * This works the same way in the case of initialized -> unwritten conversion.
3618 * One of more index blocks maybe needed if the extent tree grow after
3619 * the unwritten extent split. To prevent ENOSPC occur at the IO
3620 * complete, we need to split the unwritten extent before DIO submit
3621 * the IO. The unwritten extent called at this time will be split
3622 * into three unwritten extent(at most). After IO complete, the part
3623 * being filled will be convert to initialized by the end_io callback function
3624 * via ext4_convert_unwritten_extents().
3626 * Returns the size of unwritten extent to be written on success.
3628 static int ext4_split_convert_extents(handle_t
*handle
,
3629 struct inode
*inode
,
3630 struct ext4_map_blocks
*map
,
3631 struct ext4_ext_path
*path
,
3634 ext4_lblk_t eof_block
;
3635 ext4_lblk_t ee_block
;
3636 struct ext4_extent
*ex
;
3637 unsigned int ee_len
;
3638 int split_flag
= 0, depth
;
3640 ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3641 __func__
, inode
->i_ino
,
3642 (unsigned long long)map
->m_lblk
, map
->m_len
);
3644 eof_block
= (inode
->i_size
+ inode
->i_sb
->s_blocksize
- 1) >>
3645 inode
->i_sb
->s_blocksize_bits
;
3646 if (eof_block
< map
->m_lblk
+ map
->m_len
)
3647 eof_block
= map
->m_lblk
+ map
->m_len
;
3649 * It is safe to convert extent to initialized via explicit
3650 * zeroout only if extent is fully insde i_size or new_size.
3652 depth
= ext_depth(inode
);
3653 ex
= path
[depth
].p_ext
;
3654 ee_block
= le32_to_cpu(ex
->ee_block
);
3655 ee_len
= ext4_ext_get_actual_len(ex
);
3657 /* Convert to unwritten */
3658 if (flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
) {
3659 split_flag
|= EXT4_EXT_DATA_VALID1
;
3660 /* Convert to initialized */
3661 } else if (flags
& EXT4_GET_BLOCKS_CONVERT
) {
3662 split_flag
|= ee_block
+ ee_len
<= eof_block
?
3663 EXT4_EXT_MAY_ZEROOUT
: 0;
3664 split_flag
|= (EXT4_EXT_MARK_UNWRIT2
| EXT4_EXT_DATA_VALID2
);
3666 flags
|= EXT4_GET_BLOCKS_PRE_IO
;
3667 return ext4_split_extent(handle
, inode
, path
, map
, split_flag
, flags
);
3670 static int ext4_convert_initialized_extents(handle_t
*handle
,
3671 struct inode
*inode
,
3672 struct ext4_map_blocks
*map
,
3673 struct ext4_ext_path
*path
)
3675 struct ext4_extent
*ex
;
3676 ext4_lblk_t ee_block
;
3677 unsigned int ee_len
;
3681 depth
= ext_depth(inode
);
3682 ex
= path
[depth
].p_ext
;
3683 ee_block
= le32_to_cpu(ex
->ee_block
);
3684 ee_len
= ext4_ext_get_actual_len(ex
);
3686 ext_debug("%s: inode %lu, logical"
3687 "block %llu, max_blocks %u\n", __func__
, inode
->i_ino
,
3688 (unsigned long long)ee_block
, ee_len
);
3690 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3691 err
= ext4_split_convert_extents(handle
, inode
, map
, path
,
3692 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
);
3695 ext4_ext_drop_refs(path
);
3696 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
, 0);
3698 err
= PTR_ERR(path
);
3701 depth
= ext_depth(inode
);
3702 ex
= path
[depth
].p_ext
;
3704 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
3705 (unsigned long) map
->m_lblk
);
3711 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3714 /* first mark the extent as unwritten */
3715 ext4_ext_mark_unwritten(ex
);
3717 /* note: ext4_ext_correct_indexes() isn't needed here because
3718 * borders are not changed
3720 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3722 /* Mark modified extent as dirty */
3723 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3725 ext4_ext_show_leaf(inode
, path
);
3730 static int ext4_convert_unwritten_extents_endio(handle_t
*handle
,
3731 struct inode
*inode
,
3732 struct ext4_map_blocks
*map
,
3733 struct ext4_ext_path
*path
)
3735 struct ext4_extent
*ex
;
3736 ext4_lblk_t ee_block
;
3737 unsigned int ee_len
;
3741 depth
= ext_depth(inode
);
3742 ex
= path
[depth
].p_ext
;
3743 ee_block
= le32_to_cpu(ex
->ee_block
);
3744 ee_len
= ext4_ext_get_actual_len(ex
);
3746 ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3747 "block %llu, max_blocks %u\n", inode
->i_ino
,
3748 (unsigned long long)ee_block
, ee_len
);
3750 /* If extent is larger than requested it is a clear sign that we still
3751 * have some extent state machine issues left. So extent_split is still
3753 * TODO: Once all related issues will be fixed this situation should be
3756 if (ee_block
!= map
->m_lblk
|| ee_len
> map
->m_len
) {
3758 ext4_warning("Inode (%ld) finished: extent logical block %llu,"
3759 " len %u; IO logical block %llu, len %u\n",
3760 inode
->i_ino
, (unsigned long long)ee_block
, ee_len
,
3761 (unsigned long long)map
->m_lblk
, map
->m_len
);
3763 err
= ext4_split_convert_extents(handle
, inode
, map
, path
,
3764 EXT4_GET_BLOCKS_CONVERT
);
3767 ext4_ext_drop_refs(path
);
3768 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, path
, 0);
3770 err
= PTR_ERR(path
);
3773 depth
= ext_depth(inode
);
3774 ex
= path
[depth
].p_ext
;
3777 err
= ext4_ext_get_access(handle
, inode
, path
+ depth
);
3780 /* first mark the extent as initialized */
3781 ext4_ext_mark_initialized(ex
);
3783 /* note: ext4_ext_correct_indexes() isn't needed here because
3784 * borders are not changed
3786 ext4_ext_try_to_merge(handle
, inode
, path
, ex
);
3788 /* Mark modified extent as dirty */
3789 err
= ext4_ext_dirty(handle
, inode
, path
+ path
->p_depth
);
3791 ext4_ext_show_leaf(inode
, path
);
3795 static void unmap_underlying_metadata_blocks(struct block_device
*bdev
,
3796 sector_t block
, int count
)
3799 for (i
= 0; i
< count
; i
++)
3800 unmap_underlying_metadata(bdev
, block
+ i
);
3804 * Handle EOFBLOCKS_FL flag, clearing it if necessary
3806 static int check_eofblocks_fl(handle_t
*handle
, struct inode
*inode
,
3808 struct ext4_ext_path
*path
,
3812 struct ext4_extent_header
*eh
;
3813 struct ext4_extent
*last_ex
;
3815 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
))
3818 depth
= ext_depth(inode
);
3819 eh
= path
[depth
].p_hdr
;
3822 * We're going to remove EOFBLOCKS_FL entirely in future so we
3823 * do not care for this case anymore. Simply remove the flag
3824 * if there are no extents.
3826 if (unlikely(!eh
->eh_entries
))
3828 last_ex
= EXT_LAST_EXTENT(eh
);
3830 * We should clear the EOFBLOCKS_FL flag if we are writing the
3831 * last block in the last extent in the file. We test this by
3832 * first checking to see if the caller to
3833 * ext4_ext_get_blocks() was interested in the last block (or
3834 * a block beyond the last block) in the current extent. If
3835 * this turns out to be false, we can bail out from this
3836 * function immediately.
3838 if (lblk
+ len
< le32_to_cpu(last_ex
->ee_block
) +
3839 ext4_ext_get_actual_len(last_ex
))
3842 * If the caller does appear to be planning to write at or
3843 * beyond the end of the current extent, we then test to see
3844 * if the current extent is the last extent in the file, by
3845 * checking to make sure it was reached via the rightmost node
3846 * at each level of the tree.
3848 for (i
= depth
-1; i
>= 0; i
--)
3849 if (path
[i
].p_idx
!= EXT_LAST_INDEX(path
[i
].p_hdr
))
3852 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3853 return ext4_mark_inode_dirty(handle
, inode
);
3857 * ext4_find_delalloc_range: find delayed allocated block in the given range.
3859 * Return 1 if there is a delalloc block in the range, otherwise 0.
3861 int ext4_find_delalloc_range(struct inode
*inode
,
3862 ext4_lblk_t lblk_start
,
3863 ext4_lblk_t lblk_end
)
3865 struct extent_status es
;
3867 ext4_es_find_delayed_extent_range(inode
, lblk_start
, lblk_end
, &es
);
3869 return 0; /* there is no delay extent in this tree */
3870 else if (es
.es_lblk
<= lblk_start
&&
3871 lblk_start
< es
.es_lblk
+ es
.es_len
)
3873 else if (lblk_start
<= es
.es_lblk
&& es
.es_lblk
<= lblk_end
)
3879 int ext4_find_delalloc_cluster(struct inode
*inode
, ext4_lblk_t lblk
)
3881 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3882 ext4_lblk_t lblk_start
, lblk_end
;
3883 lblk_start
= EXT4_LBLK_CMASK(sbi
, lblk
);
3884 lblk_end
= lblk_start
+ sbi
->s_cluster_ratio
- 1;
3886 return ext4_find_delalloc_range(inode
, lblk_start
, lblk_end
);
3890 * Determines how many complete clusters (out of those specified by the 'map')
3891 * are under delalloc and were reserved quota for.
3892 * This function is called when we are writing out the blocks that were
3893 * originally written with their allocation delayed, but then the space was
3894 * allocated using fallocate() before the delayed allocation could be resolved.
3895 * The cases to look for are:
3896 * ('=' indicated delayed allocated blocks
3897 * '-' indicates non-delayed allocated blocks)
3898 * (a) partial clusters towards beginning and/or end outside of allocated range
3899 * are not delalloc'ed.
3901 * |----c---=|====c====|====c====|===-c----|
3902 * |++++++ allocated ++++++|
3903 * ==> 4 complete clusters in above example
3905 * (b) partial cluster (outside of allocated range) towards either end is
3906 * marked for delayed allocation. In this case, we will exclude that
3909 * |----====c========|========c========|
3910 * |++++++ allocated ++++++|
3911 * ==> 1 complete clusters in above example
3914 * |================c================|
3915 * |++++++ allocated ++++++|
3916 * ==> 0 complete clusters in above example
3918 * The ext4_da_update_reserve_space will be called only if we
3919 * determine here that there were some "entire" clusters that span
3920 * this 'allocated' range.
3921 * In the non-bigalloc case, this function will just end up returning num_blks
3922 * without ever calling ext4_find_delalloc_range.
3925 get_reserved_cluster_alloc(struct inode
*inode
, ext4_lblk_t lblk_start
,
3926 unsigned int num_blks
)
3928 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3929 ext4_lblk_t alloc_cluster_start
, alloc_cluster_end
;
3930 ext4_lblk_t lblk_from
, lblk_to
, c_offset
;
3931 unsigned int allocated_clusters
= 0;
3933 alloc_cluster_start
= EXT4_B2C(sbi
, lblk_start
);
3934 alloc_cluster_end
= EXT4_B2C(sbi
, lblk_start
+ num_blks
- 1);
3936 /* max possible clusters for this allocation */
3937 allocated_clusters
= alloc_cluster_end
- alloc_cluster_start
+ 1;
3939 trace_ext4_get_reserved_cluster_alloc(inode
, lblk_start
, num_blks
);
3941 /* Check towards left side */
3942 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
);
3944 lblk_from
= EXT4_LBLK_CMASK(sbi
, lblk_start
);
3945 lblk_to
= lblk_from
+ c_offset
- 1;
3947 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3948 allocated_clusters
--;
3951 /* Now check towards right. */
3952 c_offset
= EXT4_LBLK_COFF(sbi
, lblk_start
+ num_blks
);
3953 if (allocated_clusters
&& c_offset
) {
3954 lblk_from
= lblk_start
+ num_blks
;
3955 lblk_to
= lblk_from
+ (sbi
->s_cluster_ratio
- c_offset
) - 1;
3957 if (ext4_find_delalloc_range(inode
, lblk_from
, lblk_to
))
3958 allocated_clusters
--;
3961 return allocated_clusters
;
3965 ext4_ext_convert_initialized_extent(handle_t
*handle
, struct inode
*inode
,
3966 struct ext4_map_blocks
*map
,
3967 struct ext4_ext_path
*path
, int flags
,
3968 unsigned int allocated
, ext4_fsblk_t newblock
)
3974 * Make sure that the extent is no bigger than we support with
3977 if (map
->m_len
> EXT_UNWRITTEN_MAX_LEN
)
3978 map
->m_len
= EXT_UNWRITTEN_MAX_LEN
/ 2;
3980 ret
= ext4_convert_initialized_extents(handle
, inode
, map
,
3983 ext4_update_inode_fsync_trans(handle
, inode
, 1);
3984 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
3988 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
3989 if (allocated
> map
->m_len
)
3990 allocated
= map
->m_len
;
3991 map
->m_len
= allocated
;
3993 return err
? err
: allocated
;
3997 ext4_ext_handle_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
3998 struct ext4_map_blocks
*map
,
3999 struct ext4_ext_path
*path
, int flags
,
4000 unsigned int allocated
, ext4_fsblk_t newblock
)
4004 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
4006 ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4007 "block %llu, max_blocks %u, flags %x, allocated %u\n",
4008 inode
->i_ino
, (unsigned long long)map
->m_lblk
, map
->m_len
,
4010 ext4_ext_show_leaf(inode
, path
);
4013 * When writing into unwritten space, we should not fail to
4014 * allocate metadata blocks for the new extent block if needed.
4016 flags
|= EXT4_GET_BLOCKS_METADATA_NOFAIL
;
4018 trace_ext4_ext_handle_unwritten_extents(inode
, map
, flags
,
4019 allocated
, newblock
);
4021 /* get_block() before submit the IO, split the extent */
4022 if (flags
& EXT4_GET_BLOCKS_PRE_IO
) {
4023 ret
= ext4_split_convert_extents(handle
, inode
, map
,
4024 path
, flags
| EXT4_GET_BLOCKS_CONVERT
);
4028 * Flag the inode(non aio case) or end_io struct (aio case)
4029 * that this IO needs to conversion to written when IO is
4033 ext4_set_io_unwritten_flag(inode
, io
);
4035 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
4036 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4039 /* IO end_io complete, convert the filled extent to written */
4040 if (flags
& EXT4_GET_BLOCKS_CONVERT
) {
4041 ret
= ext4_convert_unwritten_extents_endio(handle
, inode
, map
,
4044 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4045 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4049 map
->m_flags
|= EXT4_MAP_MAPPED
;
4050 map
->m_pblk
= newblock
;
4051 if (allocated
> map
->m_len
)
4052 allocated
= map
->m_len
;
4053 map
->m_len
= allocated
;
4056 /* buffered IO case */
4058 * repeat fallocate creation request
4059 * we already have an unwritten extent
4061 if (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
) {
4062 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4066 /* buffered READ or buffered write_begin() lookup */
4067 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4069 * We have blocks reserved already. We
4070 * return allocated blocks so that delalloc
4071 * won't do block reservation for us. But
4072 * the buffer head will be unmapped so that
4073 * a read from the block returns 0s.
4075 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4079 /* buffered write, writepage time, convert*/
4080 ret
= ext4_ext_convert_to_initialized(handle
, inode
, map
, path
, flags
);
4082 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4089 map
->m_flags
|= EXT4_MAP_NEW
;
4091 * if we allocated more blocks than requested
4092 * we need to make sure we unmap the extra block
4093 * allocated. The actual needed block will get
4094 * unmapped later when we find the buffer_head marked
4097 if (allocated
> map
->m_len
) {
4098 unmap_underlying_metadata_blocks(inode
->i_sb
->s_bdev
,
4099 newblock
+ map
->m_len
,
4100 allocated
- map
->m_len
);
4101 allocated
= map
->m_len
;
4103 map
->m_len
= allocated
;
4106 * If we have done fallocate with the offset that is already
4107 * delayed allocated, we would have block reservation
4108 * and quota reservation done in the delayed write path.
4109 * But fallocate would have already updated quota and block
4110 * count for this offset. So cancel these reservation
4112 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4113 unsigned int reserved_clusters
;
4114 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4115 map
->m_lblk
, map
->m_len
);
4116 if (reserved_clusters
)
4117 ext4_da_update_reserve_space(inode
,
4123 map
->m_flags
|= EXT4_MAP_MAPPED
;
4124 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0) {
4125 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
, path
,
4131 if (allocated
> map
->m_len
)
4132 allocated
= map
->m_len
;
4133 ext4_ext_show_leaf(inode
, path
);
4134 map
->m_pblk
= newblock
;
4135 map
->m_len
= allocated
;
4137 return err
? err
: allocated
;
4141 * get_implied_cluster_alloc - check to see if the requested
4142 * allocation (in the map structure) overlaps with a cluster already
4143 * allocated in an extent.
4144 * @sb The filesystem superblock structure
4145 * @map The requested lblk->pblk mapping
4146 * @ex The extent structure which might contain an implied
4147 * cluster allocation
4149 * This function is called by ext4_ext_map_blocks() after we failed to
4150 * find blocks that were already in the inode's extent tree. Hence,
4151 * we know that the beginning of the requested region cannot overlap
4152 * the extent from the inode's extent tree. There are three cases we
4153 * want to catch. The first is this case:
4155 * |--- cluster # N--|
4156 * |--- extent ---| |---- requested region ---|
4159 * The second case that we need to test for is this one:
4161 * |--------- cluster # N ----------------|
4162 * |--- requested region --| |------- extent ----|
4163 * |=======================|
4165 * The third case is when the requested region lies between two extents
4166 * within the same cluster:
4167 * |------------- cluster # N-------------|
4168 * |----- ex -----| |---- ex_right ----|
4169 * |------ requested region ------|
4170 * |================|
4172 * In each of the above cases, we need to set the map->m_pblk and
4173 * map->m_len so it corresponds to the return the extent labelled as
4174 * "|====|" from cluster #N, since it is already in use for data in
4175 * cluster EXT4_B2C(sbi, map->m_lblk). We will then return 1 to
4176 * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4177 * as a new "allocated" block region. Otherwise, we will return 0 and
4178 * ext4_ext_map_blocks() will then allocate one or more new clusters
4179 * by calling ext4_mb_new_blocks().
4181 static int get_implied_cluster_alloc(struct super_block
*sb
,
4182 struct ext4_map_blocks
*map
,
4183 struct ext4_extent
*ex
,
4184 struct ext4_ext_path
*path
)
4186 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
4187 ext4_lblk_t c_offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4188 ext4_lblk_t ex_cluster_start
, ex_cluster_end
;
4189 ext4_lblk_t rr_cluster_start
;
4190 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4191 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4192 unsigned short ee_len
= ext4_ext_get_actual_len(ex
);
4194 /* The extent passed in that we are trying to match */
4195 ex_cluster_start
= EXT4_B2C(sbi
, ee_block
);
4196 ex_cluster_end
= EXT4_B2C(sbi
, ee_block
+ ee_len
- 1);
4198 /* The requested region passed into ext4_map_blocks() */
4199 rr_cluster_start
= EXT4_B2C(sbi
, map
->m_lblk
);
4201 if ((rr_cluster_start
== ex_cluster_end
) ||
4202 (rr_cluster_start
== ex_cluster_start
)) {
4203 if (rr_cluster_start
== ex_cluster_end
)
4204 ee_start
+= ee_len
- 1;
4205 map
->m_pblk
= EXT4_PBLK_CMASK(sbi
, ee_start
) + c_offset
;
4206 map
->m_len
= min(map
->m_len
,
4207 (unsigned) sbi
->s_cluster_ratio
- c_offset
);
4209 * Check for and handle this case:
4211 * |--------- cluster # N-------------|
4212 * |------- extent ----|
4213 * |--- requested region ---|
4217 if (map
->m_lblk
< ee_block
)
4218 map
->m_len
= min(map
->m_len
, ee_block
- map
->m_lblk
);
4221 * Check for the case where there is already another allocated
4222 * block to the right of 'ex' but before the end of the cluster.
4224 * |------------- cluster # N-------------|
4225 * |----- ex -----| |---- ex_right ----|
4226 * |------ requested region ------|
4227 * |================|
4229 if (map
->m_lblk
> ee_block
) {
4230 ext4_lblk_t next
= ext4_ext_next_allocated_block(path
);
4231 map
->m_len
= min(map
->m_len
, next
- map
->m_lblk
);
4234 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 1);
4238 trace_ext4_get_implied_cluster_alloc_exit(sb
, map
, 0);
4244 * Block allocation/map/preallocation routine for extents based files
4247 * Need to be called with
4248 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4249 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4251 * return > 0, number of of blocks already mapped/allocated
4252 * if create == 0 and these are pre-allocated blocks
4253 * buffer head is unmapped
4254 * otherwise blocks are mapped
4256 * return = 0, if plain look up failed (blocks have not been allocated)
4257 * buffer head is unmapped
4259 * return < 0, error case.
4261 int ext4_ext_map_blocks(handle_t
*handle
, struct inode
*inode
,
4262 struct ext4_map_blocks
*map
, int flags
)
4264 struct ext4_ext_path
*path
= NULL
;
4265 struct ext4_extent newex
, *ex
, *ex2
;
4266 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4267 ext4_fsblk_t newblock
= 0;
4268 int free_on_err
= 0, err
= 0, depth
, ret
;
4269 unsigned int allocated
= 0, offset
= 0;
4270 unsigned int allocated_clusters
= 0;
4271 struct ext4_allocation_request ar
;
4272 ext4_io_end_t
*io
= ext4_inode_aio(inode
);
4273 ext4_lblk_t cluster_offset
;
4274 int set_unwritten
= 0;
4276 ext_debug("blocks %u/%u requested for inode %lu\n",
4277 map
->m_lblk
, map
->m_len
, inode
->i_ino
);
4278 trace_ext4_ext_map_blocks_enter(inode
, map
->m_lblk
, map
->m_len
, flags
);
4280 /* find extent for this block */
4281 path
= ext4_ext_find_extent(inode
, map
->m_lblk
, NULL
, 0);
4283 err
= PTR_ERR(path
);
4288 depth
= ext_depth(inode
);
4291 * consistent leaf must not be empty;
4292 * this situation is possible, though, _during_ tree modification;
4293 * this is why assert can't be put in ext4_ext_find_extent()
4295 if (unlikely(path
[depth
].p_ext
== NULL
&& depth
!= 0)) {
4296 EXT4_ERROR_INODE(inode
, "bad extent address "
4297 "lblock: %lu, depth: %d pblock %lld",
4298 (unsigned long) map
->m_lblk
, depth
,
4299 path
[depth
].p_block
);
4304 ex
= path
[depth
].p_ext
;
4306 ext4_lblk_t ee_block
= le32_to_cpu(ex
->ee_block
);
4307 ext4_fsblk_t ee_start
= ext4_ext_pblock(ex
);
4308 unsigned short ee_len
;
4312 * unwritten extents are treated as holes, except that
4313 * we split out initialized portions during a write.
4315 ee_len
= ext4_ext_get_actual_len(ex
);
4317 trace_ext4_ext_show_extent(inode
, ee_block
, ee_start
, ee_len
);
4319 /* if found extent covers block, simply return it */
4320 if (in_range(map
->m_lblk
, ee_block
, ee_len
)) {
4321 newblock
= map
->m_lblk
- ee_block
+ ee_start
;
4322 /* number of remaining blocks in the extent */
4323 allocated
= ee_len
- (map
->m_lblk
- ee_block
);
4324 ext_debug("%u fit into %u:%d -> %llu\n", map
->m_lblk
,
4325 ee_block
, ee_len
, newblock
);
4328 * If the extent is initialized check whether the
4329 * caller wants to convert it to unwritten.
4331 if ((!ext4_ext_is_unwritten(ex
)) &&
4332 (flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
)) {
4333 allocated
= ext4_ext_convert_initialized_extent(
4334 handle
, inode
, map
, path
, flags
,
4335 allocated
, newblock
);
4337 } else if (!ext4_ext_is_unwritten(ex
))
4340 ret
= ext4_ext_handle_unwritten_extents(
4341 handle
, inode
, map
, path
, flags
,
4342 allocated
, newblock
);
4351 if ((sbi
->s_cluster_ratio
> 1) &&
4352 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
4353 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4356 * requested block isn't allocated yet;
4357 * we couldn't try to create block if create flag is zero
4359 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0) {
4361 * put just found gap into cache to speed up
4362 * subsequent requests
4364 if ((flags
& EXT4_GET_BLOCKS_NO_PUT_HOLE
) == 0)
4365 ext4_ext_put_gap_in_cache(inode
, path
, map
->m_lblk
);
4370 * Okay, we need to do block allocation.
4372 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
4373 newex
.ee_block
= cpu_to_le32(map
->m_lblk
);
4374 cluster_offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4377 * If we are doing bigalloc, check to see if the extent returned
4378 * by ext4_ext_find_extent() implies a cluster we can use.
4380 if (cluster_offset
&& ex
&&
4381 get_implied_cluster_alloc(inode
->i_sb
, map
, ex
, path
)) {
4382 ar
.len
= allocated
= map
->m_len
;
4383 newblock
= map
->m_pblk
;
4384 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4385 goto got_allocated_blocks
;
4388 /* find neighbour allocated blocks */
4389 ar
.lleft
= map
->m_lblk
;
4390 err
= ext4_ext_search_left(inode
, path
, &ar
.lleft
, &ar
.pleft
);
4393 ar
.lright
= map
->m_lblk
;
4395 err
= ext4_ext_search_right(inode
, path
, &ar
.lright
, &ar
.pright
, &ex2
);
4399 /* Check if the extent after searching to the right implies a
4400 * cluster we can use. */
4401 if ((sbi
->s_cluster_ratio
> 1) && ex2
&&
4402 get_implied_cluster_alloc(inode
->i_sb
, map
, ex2
, path
)) {
4403 ar
.len
= allocated
= map
->m_len
;
4404 newblock
= map
->m_pblk
;
4405 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
4406 goto got_allocated_blocks
;
4410 * See if request is beyond maximum number of blocks we can have in
4411 * a single extent. For an initialized extent this limit is
4412 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4413 * EXT_UNWRITTEN_MAX_LEN.
4415 if (map
->m_len
> EXT_INIT_MAX_LEN
&&
4416 !(flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
))
4417 map
->m_len
= EXT_INIT_MAX_LEN
;
4418 else if (map
->m_len
> EXT_UNWRITTEN_MAX_LEN
&&
4419 (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
))
4420 map
->m_len
= EXT_UNWRITTEN_MAX_LEN
;
4422 /* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4423 newex
.ee_len
= cpu_to_le16(map
->m_len
);
4424 err
= ext4_ext_check_overlap(sbi
, inode
, &newex
, path
);
4426 allocated
= ext4_ext_get_actual_len(&newex
);
4428 allocated
= map
->m_len
;
4430 /* allocate new block */
4432 ar
.goal
= ext4_ext_find_goal(inode
, path
, map
->m_lblk
);
4433 ar
.logical
= map
->m_lblk
;
4435 * We calculate the offset from the beginning of the cluster
4436 * for the logical block number, since when we allocate a
4437 * physical cluster, the physical block should start at the
4438 * same offset from the beginning of the cluster. This is
4439 * needed so that future calls to get_implied_cluster_alloc()
4442 offset
= EXT4_LBLK_COFF(sbi
, map
->m_lblk
);
4443 ar
.len
= EXT4_NUM_B2C(sbi
, offset
+allocated
);
4445 ar
.logical
-= offset
;
4446 if (S_ISREG(inode
->i_mode
))
4447 ar
.flags
= EXT4_MB_HINT_DATA
;
4449 /* disable in-core preallocation for non-regular files */
4451 if (flags
& EXT4_GET_BLOCKS_NO_NORMALIZE
)
4452 ar
.flags
|= EXT4_MB_HINT_NOPREALLOC
;
4453 newblock
= ext4_mb_new_blocks(handle
, &ar
, &err
);
4456 ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4457 ar
.goal
, newblock
, allocated
);
4459 allocated_clusters
= ar
.len
;
4460 ar
.len
= EXT4_C2B(sbi
, ar
.len
) - offset
;
4461 if (ar
.len
> allocated
)
4464 got_allocated_blocks
:
4465 /* try to insert new extent into found leaf and return */
4466 ext4_ext_store_pblock(&newex
, newblock
+ offset
);
4467 newex
.ee_len
= cpu_to_le16(ar
.len
);
4468 /* Mark unwritten */
4469 if (flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
){
4470 ext4_ext_mark_unwritten(&newex
);
4471 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
4473 * io_end structure was created for every IO write to an
4474 * unwritten extent. To avoid unnecessary conversion,
4475 * here we flag the IO that really needs the conversion.
4476 * For non asycn direct IO case, flag the inode state
4477 * that we need to perform conversion when IO is done.
4479 if (flags
& EXT4_GET_BLOCKS_PRE_IO
)
4484 if ((flags
& EXT4_GET_BLOCKS_KEEP_SIZE
) == 0)
4485 err
= check_eofblocks_fl(handle
, inode
, map
->m_lblk
,
4488 err
= ext4_ext_insert_extent(handle
, inode
, path
,
4491 if (!err
&& set_unwritten
) {
4493 ext4_set_io_unwritten_flag(inode
, io
);
4495 ext4_set_inode_state(inode
,
4496 EXT4_STATE_DIO_UNWRITTEN
);
4499 if (err
&& free_on_err
) {
4500 int fb_flags
= flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
?
4501 EXT4_FREE_BLOCKS_NO_QUOT_UPDATE
: 0;
4502 /* free data blocks we just allocated */
4503 /* not a good idea to call discard here directly,
4504 * but otherwise we'd need to call it every free() */
4505 ext4_discard_preallocations(inode
);
4506 ext4_free_blocks(handle
, inode
, NULL
, newblock
,
4507 EXT4_C2B(sbi
, allocated_clusters
), fb_flags
);
4511 /* previous routine could use block we allocated */
4512 newblock
= ext4_ext_pblock(&newex
);
4513 allocated
= ext4_ext_get_actual_len(&newex
);
4514 if (allocated
> map
->m_len
)
4515 allocated
= map
->m_len
;
4516 map
->m_flags
|= EXT4_MAP_NEW
;
4519 * Update reserved blocks/metadata blocks after successful
4520 * block allocation which had been deferred till now.
4522 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
4523 unsigned int reserved_clusters
;
4525 * Check how many clusters we had reserved this allocated range
4527 reserved_clusters
= get_reserved_cluster_alloc(inode
,
4528 map
->m_lblk
, allocated
);
4529 if (map
->m_flags
& EXT4_MAP_FROM_CLUSTER
) {
4530 if (reserved_clusters
) {
4532 * We have clusters reserved for this range.
4533 * But since we are not doing actual allocation
4534 * and are simply using blocks from previously
4535 * allocated cluster, we should release the
4536 * reservation and not claim quota.
4538 ext4_da_update_reserve_space(inode
,
4539 reserved_clusters
, 0);
4542 BUG_ON(allocated_clusters
< reserved_clusters
);
4543 if (reserved_clusters
< allocated_clusters
) {
4544 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4545 int reservation
= allocated_clusters
-
4548 * It seems we claimed few clusters outside of
4549 * the range of this allocation. We should give
4550 * it back to the reservation pool. This can
4551 * happen in the following case:
4553 * * Suppose s_cluster_ratio is 4 (i.e., each
4554 * cluster has 4 blocks. Thus, the clusters
4555 * are [0-3],[4-7],[8-11]...
4556 * * First comes delayed allocation write for
4557 * logical blocks 10 & 11. Since there were no
4558 * previous delayed allocated blocks in the
4559 * range [8-11], we would reserve 1 cluster
4561 * * Next comes write for logical blocks 3 to 8.
4562 * In this case, we will reserve 2 clusters
4563 * (for [0-3] and [4-7]; and not for [8-11] as
4564 * that range has a delayed allocated blocks.
4565 * Thus total reserved clusters now becomes 3.
4566 * * Now, during the delayed allocation writeout
4567 * time, we will first write blocks [3-8] and
4568 * allocate 3 clusters for writing these
4569 * blocks. Also, we would claim all these
4570 * three clusters above.
4571 * * Now when we come here to writeout the
4572 * blocks [10-11], we would expect to claim
4573 * the reservation of 1 cluster we had made
4574 * (and we would claim it since there are no
4575 * more delayed allocated blocks in the range
4576 * [8-11]. But our reserved cluster count had
4577 * already gone to 0.
4579 * Thus, at the step 4 above when we determine
4580 * that there are still some unwritten delayed
4581 * allocated blocks outside of our current
4582 * block range, we should increment the
4583 * reserved clusters count so that when the
4584 * remaining blocks finally gets written, we
4587 dquot_reserve_block(inode
,
4588 EXT4_C2B(sbi
, reservation
));
4589 spin_lock(&ei
->i_block_reservation_lock
);
4590 ei
->i_reserved_data_blocks
+= reservation
;
4591 spin_unlock(&ei
->i_block_reservation_lock
);
4594 * We will claim quota for all newly allocated blocks.
4595 * We're updating the reserved space *after* the
4596 * correction above so we do not accidentally free
4597 * all the metadata reservation because we might
4598 * actually need it later on.
4600 ext4_da_update_reserve_space(inode
, allocated_clusters
,
4606 * Cache the extent and update transaction to commit on fdatasync only
4607 * when it is _not_ an unwritten extent.
4609 if ((flags
& EXT4_GET_BLOCKS_UNWRIT_EXT
) == 0)
4610 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4612 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4614 if (allocated
> map
->m_len
)
4615 allocated
= map
->m_len
;
4616 ext4_ext_show_leaf(inode
, path
);
4617 map
->m_flags
|= EXT4_MAP_MAPPED
;
4618 map
->m_pblk
= newblock
;
4619 map
->m_len
= allocated
;
4622 ext4_ext_drop_refs(path
);
4626 trace_ext4_ext_map_blocks_exit(inode
, flags
, map
,
4627 err
? err
: allocated
);
4628 ext4_es_lru_add(inode
);
4629 return err
? err
: allocated
;
4632 void ext4_ext_truncate(handle_t
*handle
, struct inode
*inode
)
4634 struct super_block
*sb
= inode
->i_sb
;
4635 ext4_lblk_t last_block
;
4639 * TODO: optimization is possible here.
4640 * Probably we need not scan at all,
4641 * because page truncation is enough.
4644 /* we have to know where to truncate from in crash case */
4645 EXT4_I(inode
)->i_disksize
= inode
->i_size
;
4646 ext4_mark_inode_dirty(handle
, inode
);
4648 last_block
= (inode
->i_size
+ sb
->s_blocksize
- 1)
4649 >> EXT4_BLOCK_SIZE_BITS(sb
);
4651 err
= ext4_es_remove_extent(inode
, last_block
,
4652 EXT_MAX_BLOCKS
- last_block
);
4653 if (err
== -ENOMEM
) {
4655 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
4659 ext4_std_error(inode
->i_sb
, err
);
4662 err
= ext4_ext_remove_space(inode
, last_block
, EXT_MAX_BLOCKS
- 1);
4663 ext4_std_error(inode
->i_sb
, err
);
4666 static int ext4_alloc_file_blocks(struct file
*file
, ext4_lblk_t offset
,
4667 ext4_lblk_t len
, int flags
, int mode
)
4669 struct inode
*inode
= file_inode(file
);
4674 struct ext4_map_blocks map
;
4675 unsigned int credits
;
4677 map
.m_lblk
= offset
;
4679 * Don't normalize the request if it can fit in one extent so
4680 * that it doesn't get unnecessarily split into multiple
4683 if (len
<= EXT_UNWRITTEN_MAX_LEN
)
4684 flags
|= EXT4_GET_BLOCKS_NO_NORMALIZE
;
4687 * credits to insert 1 extent into extent tree
4689 credits
= ext4_chunk_trans_blocks(inode
, len
);
4692 while (ret
>= 0 && ret
< len
) {
4693 map
.m_lblk
= map
.m_lblk
+ ret
;
4694 map
.m_len
= len
= len
- ret
;
4695 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
4697 if (IS_ERR(handle
)) {
4698 ret
= PTR_ERR(handle
);
4701 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
4703 ext4_debug("inode #%lu: block %u: len %u: "
4704 "ext4_ext_map_blocks returned %d",
4705 inode
->i_ino
, map
.m_lblk
,
4707 ext4_mark_inode_dirty(handle
, inode
);
4708 ret2
= ext4_journal_stop(handle
);
4711 ret2
= ext4_journal_stop(handle
);
4715 if (ret
== -ENOSPC
&&
4716 ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
4721 return ret
> 0 ? ret2
: ret
;
4724 static long ext4_zero_range(struct file
*file
, loff_t offset
,
4725 loff_t len
, int mode
)
4727 struct inode
*inode
= file_inode(file
);
4728 handle_t
*handle
= NULL
;
4729 unsigned int max_blocks
;
4730 loff_t new_size
= 0;
4736 struct address_space
*mapping
= inode
->i_mapping
;
4737 unsigned int blkbits
= inode
->i_blkbits
;
4739 trace_ext4_zero_range(inode
, offset
, len
, mode
);
4741 if (!S_ISREG(inode
->i_mode
))
4744 /* Call ext4_force_commit to flush all data in case of data=journal. */
4745 if (ext4_should_journal_data(inode
)) {
4746 ret
= ext4_force_commit(inode
->i_sb
);
4752 * Write out all dirty pages to avoid race conditions
4753 * Then release them.
4755 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4756 ret
= filemap_write_and_wait_range(mapping
, offset
,
4763 * Round up offset. This is not fallocate, we neet to zero out
4764 * blocks, so convert interior block aligned part of the range to
4765 * unwritten and possibly manually zero out unaligned parts of the
4768 start
= round_up(offset
, 1 << blkbits
);
4769 end
= round_down((offset
+ len
), 1 << blkbits
);
4771 if (start
< offset
|| end
> offset
+ len
)
4773 partial
= (offset
+ len
) & ((1 << blkbits
) - 1);
4775 lblk
= start
>> blkbits
;
4776 max_blocks
= (end
>> blkbits
);
4777 if (max_blocks
< lblk
)
4782 flags
= EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT
|
4783 EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
;
4784 if (mode
& FALLOC_FL_KEEP_SIZE
)
4785 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4787 mutex_lock(&inode
->i_mutex
);
4790 * Indirect files do not support unwritten extnets
4792 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4797 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4798 offset
+ len
> i_size_read(inode
)) {
4799 new_size
= offset
+ len
;
4800 ret
= inode_newsize_ok(inode
, new_size
);
4804 * If we have a partial block after EOF we have to allocate
4811 if (max_blocks
> 0) {
4813 /* Now release the pages and zero block aligned part of pages*/
4814 truncate_pagecache_range(inode
, start
, end
- 1);
4816 /* Wait all existing dio workers, newcomers will block on i_mutex */
4817 ext4_inode_block_unlocked_dio(inode
);
4818 inode_dio_wait(inode
);
4821 * Remove entire range from the extent status tree.
4823 ret
= ext4_es_remove_extent(inode
, lblk
, max_blocks
);
4827 ret
= ext4_alloc_file_blocks(file
, lblk
, max_blocks
, flags
,
4833 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 4);
4834 if (IS_ERR(handle
)) {
4835 ret
= PTR_ERR(handle
);
4836 ext4_std_error(inode
->i_sb
, ret
);
4840 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
4843 if (new_size
> i_size_read(inode
))
4844 i_size_write(inode
, new_size
);
4845 if (new_size
> EXT4_I(inode
)->i_disksize
)
4846 ext4_update_i_disksize(inode
, new_size
);
4849 * Mark that we allocate beyond EOF so the subsequent truncate
4850 * can proceed even if the new size is the same as i_size.
4852 if ((offset
+ len
) > i_size_read(inode
))
4853 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4856 ext4_mark_inode_dirty(handle
, inode
);
4858 /* Zero out partial block at the edges of the range */
4859 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
, len
);
4861 if (file
->f_flags
& O_SYNC
)
4862 ext4_handle_sync(handle
);
4864 ext4_journal_stop(handle
);
4866 ext4_inode_resume_unlocked_dio(inode
);
4868 mutex_unlock(&inode
->i_mutex
);
4873 * preallocate space for a file. This implements ext4's fallocate file
4874 * operation, which gets called from sys_fallocate system call.
4875 * For block-mapped files, posix_fallocate should fall back to the method
4876 * of writing zeroes to the required new blocks (the same behavior which is
4877 * expected for file systems which do not support fallocate() system call).
4879 long ext4_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
4881 struct inode
*inode
= file_inode(file
);
4883 loff_t new_size
= 0;
4884 unsigned int max_blocks
;
4889 unsigned int blkbits
= inode
->i_blkbits
;
4891 /* Return error if mode is not supported */
4892 if (mode
& ~(FALLOC_FL_KEEP_SIZE
| FALLOC_FL_PUNCH_HOLE
|
4893 FALLOC_FL_COLLAPSE_RANGE
| FALLOC_FL_ZERO_RANGE
))
4896 if (mode
& FALLOC_FL_PUNCH_HOLE
)
4897 return ext4_punch_hole(inode
, offset
, len
);
4899 ret
= ext4_convert_inline_data(inode
);
4904 * currently supporting (pre)allocate mode for extent-based
4907 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4910 if (mode
& FALLOC_FL_COLLAPSE_RANGE
)
4911 return ext4_collapse_range(inode
, offset
, len
);
4913 if (mode
& FALLOC_FL_ZERO_RANGE
)
4914 return ext4_zero_range(file
, offset
, len
, mode
);
4916 trace_ext4_fallocate_enter(inode
, offset
, len
, mode
);
4917 lblk
= offset
>> blkbits
;
4919 * We can't just convert len to max_blocks because
4920 * If blocksize = 4096 offset = 3072 and len = 2048
4922 max_blocks
= (EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
)
4925 flags
= EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT
;
4926 if (mode
& FALLOC_FL_KEEP_SIZE
)
4927 flags
|= EXT4_GET_BLOCKS_KEEP_SIZE
;
4929 mutex_lock(&inode
->i_mutex
);
4931 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4932 offset
+ len
> i_size_read(inode
)) {
4933 new_size
= offset
+ len
;
4934 ret
= inode_newsize_ok(inode
, new_size
);
4939 ret
= ext4_alloc_file_blocks(file
, lblk
, max_blocks
, flags
, mode
);
4943 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
4947 tv
= inode
->i_ctime
= ext4_current_time(inode
);
4950 if (new_size
> i_size_read(inode
)) {
4951 i_size_write(inode
, new_size
);
4952 inode
->i_mtime
= tv
;
4954 if (new_size
> EXT4_I(inode
)->i_disksize
)
4955 ext4_update_i_disksize(inode
, new_size
);
4958 * Mark that we allocate beyond EOF so the subsequent truncate
4959 * can proceed even if the new size is the same as i_size.
4961 if ((offset
+ len
) > i_size_read(inode
))
4962 ext4_set_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4964 ext4_mark_inode_dirty(handle
, inode
);
4965 if (file
->f_flags
& O_SYNC
)
4966 ext4_handle_sync(handle
);
4968 ext4_journal_stop(handle
);
4970 mutex_unlock(&inode
->i_mutex
);
4971 trace_ext4_fallocate_exit(inode
, offset
, max_blocks
, ret
);
4976 * This function convert a range of blocks to written extents
4977 * The caller of this function will pass the start offset and the size.
4978 * all unwritten extents within this range will be converted to
4981 * This function is called from the direct IO end io call back
4982 * function, to convert the fallocated extents after IO is completed.
4983 * Returns 0 on success.
4985 int ext4_convert_unwritten_extents(handle_t
*handle
, struct inode
*inode
,
4986 loff_t offset
, ssize_t len
)
4988 unsigned int max_blocks
;
4991 struct ext4_map_blocks map
;
4992 unsigned int credits
, blkbits
= inode
->i_blkbits
;
4994 map
.m_lblk
= offset
>> blkbits
;
4996 * We can't just convert len to max_blocks because
4997 * If blocksize = 4096 offset = 3072 and len = 2048
4999 max_blocks
= ((EXT4_BLOCK_ALIGN(len
+ offset
, blkbits
) >> blkbits
) -
5002 * This is somewhat ugly but the idea is clear: When transaction is
5003 * reserved, everything goes into it. Otherwise we rather start several
5004 * smaller transactions for conversion of each extent separately.
5007 handle
= ext4_journal_start_reserved(handle
,
5008 EXT4_HT_EXT_CONVERT
);
5010 return PTR_ERR(handle
);
5014 * credits to insert 1 extent into extent tree
5016 credits
= ext4_chunk_trans_blocks(inode
, max_blocks
);
5018 while (ret
>= 0 && ret
< max_blocks
) {
5020 map
.m_len
= (max_blocks
-= ret
);
5022 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
5024 if (IS_ERR(handle
)) {
5025 ret
= PTR_ERR(handle
);
5029 ret
= ext4_map_blocks(handle
, inode
, &map
,
5030 EXT4_GET_BLOCKS_IO_CONVERT_EXT
);
5032 ext4_warning(inode
->i_sb
,
5033 "inode #%lu: block %u: len %u: "
5034 "ext4_ext_map_blocks returned %d",
5035 inode
->i_ino
, map
.m_lblk
,
5037 ext4_mark_inode_dirty(handle
, inode
);
5039 ret2
= ext4_journal_stop(handle
);
5040 if (ret
<= 0 || ret2
)
5044 ret2
= ext4_journal_stop(handle
);
5045 return ret
> 0 ? ret2
: ret
;
5049 * If newes is not existing extent (newes->ec_pblk equals zero) find
5050 * delayed extent at start of newes and update newes accordingly and
5051 * return start of the next delayed extent.
5053 * If newes is existing extent (newes->ec_pblk is not equal zero)
5054 * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5055 * extent found. Leave newes unmodified.
5057 static int ext4_find_delayed_extent(struct inode
*inode
,
5058 struct extent_status
*newes
)
5060 struct extent_status es
;
5061 ext4_lblk_t block
, next_del
;
5063 if (newes
->es_pblk
== 0) {
5064 ext4_es_find_delayed_extent_range(inode
, newes
->es_lblk
,
5065 newes
->es_lblk
+ newes
->es_len
- 1, &es
);
5068 * No extent in extent-tree contains block @newes->es_pblk,
5069 * then the block may stay in 1)a hole or 2)delayed-extent.
5075 if (es
.es_lblk
> newes
->es_lblk
) {
5077 newes
->es_len
= min(es
.es_lblk
- newes
->es_lblk
,
5082 newes
->es_len
= es
.es_lblk
+ es
.es_len
- newes
->es_lblk
;
5085 block
= newes
->es_lblk
+ newes
->es_len
;
5086 ext4_es_find_delayed_extent_range(inode
, block
, EXT_MAX_BLOCKS
, &es
);
5088 next_del
= EXT_MAX_BLOCKS
;
5090 next_del
= es
.es_lblk
;
5094 /* fiemap flags we can handle specified here */
5095 #define EXT4_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
5097 static int ext4_xattr_fiemap(struct inode
*inode
,
5098 struct fiemap_extent_info
*fieinfo
)
5102 __u32 flags
= FIEMAP_EXTENT_LAST
;
5103 int blockbits
= inode
->i_sb
->s_blocksize_bits
;
5107 if (ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
5108 struct ext4_iloc iloc
;
5109 int offset
; /* offset of xattr in inode */
5111 error
= ext4_get_inode_loc(inode
, &iloc
);
5114 physical
= (__u64
)iloc
.bh
->b_blocknr
<< blockbits
;
5115 offset
= EXT4_GOOD_OLD_INODE_SIZE
+
5116 EXT4_I(inode
)->i_extra_isize
;
5118 length
= EXT4_SB(inode
->i_sb
)->s_inode_size
- offset
;
5119 flags
|= FIEMAP_EXTENT_DATA_INLINE
;
5121 } else { /* external block */
5122 physical
= (__u64
)EXT4_I(inode
)->i_file_acl
<< blockbits
;
5123 length
= inode
->i_sb
->s_blocksize
;
5127 error
= fiemap_fill_next_extent(fieinfo
, 0, physical
,
5129 return (error
< 0 ? error
: 0);
5132 int ext4_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
5133 __u64 start
, __u64 len
)
5135 ext4_lblk_t start_blk
;
5138 if (ext4_has_inline_data(inode
)) {
5141 error
= ext4_inline_data_fiemap(inode
, fieinfo
, &has_inline
);
5147 if (fieinfo
->fi_flags
& FIEMAP_FLAG_CACHE
) {
5148 error
= ext4_ext_precache(inode
);
5153 /* fallback to generic here if not in extents fmt */
5154 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5155 return generic_block_fiemap(inode
, fieinfo
, start
, len
,
5158 if (fiemap_check_flags(fieinfo
, EXT4_FIEMAP_FLAGS
))
5161 if (fieinfo
->fi_flags
& FIEMAP_FLAG_XATTR
) {
5162 error
= ext4_xattr_fiemap(inode
, fieinfo
);
5164 ext4_lblk_t len_blks
;
5167 start_blk
= start
>> inode
->i_sb
->s_blocksize_bits
;
5168 last_blk
= (start
+ len
- 1) >> inode
->i_sb
->s_blocksize_bits
;
5169 if (last_blk
>= EXT_MAX_BLOCKS
)
5170 last_blk
= EXT_MAX_BLOCKS
-1;
5171 len_blks
= ((ext4_lblk_t
) last_blk
) - start_blk
+ 1;
5174 * Walk the extent tree gathering extent information
5175 * and pushing extents back to the user.
5177 error
= ext4_fill_fiemap_extents(inode
, start_blk
,
5180 ext4_es_lru_add(inode
);
5186 * Function to access the path buffer for marking it dirty.
5187 * It also checks if there are sufficient credits left in the journal handle
5191 ext4_access_path(handle_t
*handle
, struct inode
*inode
,
5192 struct ext4_ext_path
*path
)
5196 if (!ext4_handle_valid(handle
))
5200 * Check if need to extend journal credits
5201 * 3 for leaf, sb, and inode plus 2 (bmap and group
5202 * descriptor) for each block group; assume two block
5205 if (handle
->h_buffer_credits
< 7) {
5206 credits
= ext4_writepage_trans_blocks(inode
);
5207 err
= ext4_ext_truncate_extend_restart(handle
, inode
, credits
);
5208 /* EAGAIN is success */
5209 if (err
&& err
!= -EAGAIN
)
5213 err
= ext4_ext_get_access(handle
, inode
, path
);
5218 * ext4_ext_shift_path_extents:
5219 * Shift the extents of a path structure lying between path[depth].p_ext
5220 * and EXT_LAST_EXTENT(path[depth].p_hdr) downwards, by subtracting shift
5221 * from starting block for each extent.
5224 ext4_ext_shift_path_extents(struct ext4_ext_path
*path
, ext4_lblk_t shift
,
5225 struct inode
*inode
, handle_t
*handle
,
5229 struct ext4_extent
*ex_start
, *ex_last
;
5231 depth
= path
->p_depth
;
5233 while (depth
>= 0) {
5234 if (depth
== path
->p_depth
) {
5235 ex_start
= path
[depth
].p_ext
;
5239 ex_last
= EXT_LAST_EXTENT(path
[depth
].p_hdr
);
5243 err
= ext4_access_path(handle
, inode
, path
+ depth
);
5247 if (ex_start
== EXT_FIRST_EXTENT(path
[depth
].p_hdr
))
5250 *start
= le32_to_cpu(ex_last
->ee_block
) +
5251 ext4_ext_get_actual_len(ex_last
);
5253 while (ex_start
<= ex_last
) {
5254 le32_add_cpu(&ex_start
->ee_block
, -shift
);
5255 /* Try to merge to the left. */
5257 EXT_FIRST_EXTENT(path
[depth
].p_hdr
)) &&
5258 ext4_ext_try_to_merge_right(inode
,
5259 path
, ex_start
- 1))
5264 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
5268 if (--depth
< 0 || !update
)
5272 /* Update index too */
5273 err
= ext4_access_path(handle
, inode
, path
+ depth
);
5277 le32_add_cpu(&path
[depth
].p_idx
->ei_block
, -shift
);
5278 err
= ext4_ext_dirty(handle
, inode
, path
+ depth
);
5282 /* we are done if current index is not a starting index */
5283 if (path
[depth
].p_idx
!= EXT_FIRST_INDEX(path
[depth
].p_hdr
))
5294 * ext4_ext_shift_extents:
5295 * All the extents which lies in the range from start to the last allocated
5296 * block for the file are shifted downwards by shift blocks.
5297 * On success, 0 is returned, error otherwise.
5300 ext4_ext_shift_extents(struct inode
*inode
, handle_t
*handle
,
5301 ext4_lblk_t start
, ext4_lblk_t shift
)
5303 struct ext4_ext_path
*path
;
5305 struct ext4_extent
*extent
;
5306 ext4_lblk_t stop_block
, current_block
;
5307 ext4_lblk_t ex_start
, ex_end
;
5309 /* Let path point to the last extent */
5310 path
= ext4_ext_find_extent(inode
, EXT_MAX_BLOCKS
- 1, NULL
, 0);
5312 return PTR_ERR(path
);
5314 depth
= path
->p_depth
;
5315 extent
= path
[depth
].p_ext
;
5317 ext4_ext_drop_refs(path
);
5322 stop_block
= le32_to_cpu(extent
->ee_block
) +
5323 ext4_ext_get_actual_len(extent
);
5324 ext4_ext_drop_refs(path
);
5327 /* Nothing to shift, if hole is at the end of file */
5328 if (start
>= stop_block
)
5332 * Don't start shifting extents until we make sure the hole is big
5333 * enough to accomodate the shift.
5335 path
= ext4_ext_find_extent(inode
, start
- 1, NULL
, 0);
5337 return PTR_ERR(path
);
5338 depth
= path
->p_depth
;
5339 extent
= path
[depth
].p_ext
;
5341 ex_start
= le32_to_cpu(extent
->ee_block
);
5342 ex_end
= le32_to_cpu(extent
->ee_block
) +
5343 ext4_ext_get_actual_len(extent
);
5348 ext4_ext_drop_refs(path
);
5351 if ((start
== ex_start
&& shift
> ex_start
) ||
5352 (shift
> start
- ex_end
))
5355 /* Its safe to start updating extents */
5356 while (start
< stop_block
) {
5357 path
= ext4_ext_find_extent(inode
, start
, NULL
, 0);
5359 return PTR_ERR(path
);
5360 depth
= path
->p_depth
;
5361 extent
= path
[depth
].p_ext
;
5363 EXT4_ERROR_INODE(inode
, "unexpected hole at %lu",
5364 (unsigned long) start
);
5368 current_block
= le32_to_cpu(extent
->ee_block
);
5369 if (start
> current_block
) {
5370 /* Hole, move to the next extent */
5371 ret
= mext_next_extent(inode
, path
, &extent
);
5373 ext4_ext_drop_refs(path
);
5380 ret
= ext4_ext_shift_path_extents(path
, shift
, inode
,
5382 ext4_ext_drop_refs(path
);
5392 * ext4_collapse_range:
5393 * This implements the fallocate's collapse range functionality for ext4
5394 * Returns: 0 and non-zero on error.
5396 int ext4_collapse_range(struct inode
*inode
, loff_t offset
, loff_t len
)
5398 struct super_block
*sb
= inode
->i_sb
;
5399 ext4_lblk_t punch_start
, punch_stop
;
5401 unsigned int credits
;
5402 loff_t new_size
, ioffset
;
5405 /* Collapse range works only on fs block size aligned offsets. */
5406 if (offset
& (EXT4_BLOCK_SIZE(sb
) - 1) ||
5407 len
& (EXT4_BLOCK_SIZE(sb
) - 1))
5410 if (!S_ISREG(inode
->i_mode
))
5413 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1)
5416 trace_ext4_collapse_range(inode
, offset
, len
);
5418 punch_start
= offset
>> EXT4_BLOCK_SIZE_BITS(sb
);
5419 punch_stop
= (offset
+ len
) >> EXT4_BLOCK_SIZE_BITS(sb
);
5421 /* Call ext4_force_commit to flush all data in case of data=journal. */
5422 if (ext4_should_journal_data(inode
)) {
5423 ret
= ext4_force_commit(inode
->i_sb
);
5429 * Need to round down offset to be aligned with page size boundary
5430 * for page size > block size.
5432 ioffset
= round_down(offset
, PAGE_SIZE
);
5434 /* Write out all dirty pages */
5435 ret
= filemap_write_and_wait_range(inode
->i_mapping
, ioffset
,
5440 /* Take mutex lock */
5441 mutex_lock(&inode
->i_mutex
);
5444 * There is no need to overlap collapse range with EOF, in which case
5445 * it is effectively a truncate operation
5447 if (offset
+ len
>= i_size_read(inode
)) {
5452 /* Currently just for extent based files */
5453 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
5458 truncate_pagecache(inode
, ioffset
);
5460 /* Wait for existing dio to complete */
5461 ext4_inode_block_unlocked_dio(inode
);
5462 inode_dio_wait(inode
);
5464 credits
= ext4_writepage_trans_blocks(inode
);
5465 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
5466 if (IS_ERR(handle
)) {
5467 ret
= PTR_ERR(handle
);
5471 down_write(&EXT4_I(inode
)->i_data_sem
);
5472 ext4_discard_preallocations(inode
);
5474 ret
= ext4_es_remove_extent(inode
, punch_start
,
5475 EXT_MAX_BLOCKS
- punch_start
);
5477 up_write(&EXT4_I(inode
)->i_data_sem
);
5481 ret
= ext4_ext_remove_space(inode
, punch_start
, punch_stop
- 1);
5483 up_write(&EXT4_I(inode
)->i_data_sem
);
5486 ext4_discard_preallocations(inode
);
5488 ret
= ext4_ext_shift_extents(inode
, handle
, punch_stop
,
5489 punch_stop
- punch_start
);
5491 up_write(&EXT4_I(inode
)->i_data_sem
);
5495 new_size
= i_size_read(inode
) - len
;
5496 i_size_write(inode
, new_size
);
5497 EXT4_I(inode
)->i_disksize
= new_size
;
5499 up_write(&EXT4_I(inode
)->i_data_sem
);
5501 ext4_handle_sync(handle
);
5502 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
5503 ext4_mark_inode_dirty(handle
, inode
);
5506 ext4_journal_stop(handle
);
5508 ext4_inode_resume_unlocked_dio(inode
);
5510 mutex_unlock(&inode
->i_mutex
);