2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
11 * Implements Extendible Hashing as described in:
12 * "Extendible Hashing" by Fagin, et al in
13 * __ACM Trans. on Database Systems__, Sept 1979.
16 * Here's the layout of dirents which is essentially the same as that of ext2
17 * within a single block. The field de_name_len is the number of bytes
18 * actually required for the name (no null terminator). The field de_rec_len
19 * is the number of bytes allocated to the dirent. The offset of the next
20 * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
21 * deleted, the preceding dirent inherits its allocated space, ie
22 * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
23 * by adding de_rec_len to the current dirent, this essentially causes the
24 * deleted dirent to get jumped over when iterating through all the dirents.
26 * When deleting the first dirent in a block, there is no previous dirent so
27 * the field de_ino is set to zero to designate it as deleted. When allocating
28 * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
29 * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
30 * dirent is allocated. Otherwise it must go through all the 'used' dirents
31 * searching for one in which the amount of total space minus the amount of
32 * used space will provide enough space for the new dirent.
34 * There are two types of blocks in which dirents reside. In a stuffed dinode,
35 * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
36 * the block. In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
37 * beginning of the leaf block. The dirents reside in leaves when
39 * dip->i_di.di_flags & GFS2_DIF_EXHASH is true
41 * Otherwise, the dirents are "linear", within a single stuffed dinode block.
43 * When the dirents are in leaves, the actual contents of the directory file are
44 * used as an array of 64-bit block pointers pointing to the leaf blocks. The
45 * dirents are NOT in the directory file itself. There can be more than one
46 * block pointer in the array that points to the same leaf. In fact, when a
47 * directory is first converted from linear to exhash, all of the pointers
48 * point to the same leaf.
50 * When a leaf is completely full, the size of the hash table can be
51 * doubled unless it is already at the maximum size which is hard coded into
52 * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
53 * but never before the maximum hash table size has been reached.
56 #include <linux/sched.h>
57 #include <linux/slab.h>
58 #include <linux/spinlock.h>
59 #include <linux/buffer_head.h>
60 #include <linux/sort.h>
61 #include <linux/gfs2_ondisk.h>
62 #include <linux/crc32.h>
63 #include <linux/vmalloc.h>
64 #include <linux/lm_interface.h>
78 #define IS_LEAF 1 /* Hashed (leaf) directory */
79 #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */
81 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
82 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
84 typedef int (*leaf_call_t
) (struct gfs2_inode
*dip
, u32 index
, u32 len
,
85 u64 leaf_no
, void *data
);
86 typedef int (*gfs2_dscan_t
)(const struct gfs2_dirent
*dent
,
87 const struct qstr
*name
, void *opaque
);
90 int gfs2_dir_get_new_buffer(struct gfs2_inode
*ip
, u64 block
,
91 struct buffer_head
**bhp
)
93 struct buffer_head
*bh
;
95 bh
= gfs2_meta_new(ip
->i_gl
, block
);
96 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
97 gfs2_metatype_set(bh
, GFS2_METATYPE_JD
, GFS2_FORMAT_JD
);
98 gfs2_buffer_clear_tail(bh
, sizeof(struct gfs2_meta_header
));
103 static int gfs2_dir_get_existing_buffer(struct gfs2_inode
*ip
, u64 block
,
104 struct buffer_head
**bhp
)
106 struct buffer_head
*bh
;
109 error
= gfs2_meta_read(ip
->i_gl
, block
, DIO_WAIT
, &bh
);
112 if (gfs2_metatype_check(GFS2_SB(&ip
->i_inode
), bh
, GFS2_METATYPE_JD
)) {
120 static int gfs2_dir_write_stuffed(struct gfs2_inode
*ip
, const char *buf
,
121 unsigned int offset
, unsigned int size
)
123 struct buffer_head
*dibh
;
126 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
130 gfs2_trans_add_bh(ip
->i_gl
, dibh
, 1);
131 memcpy(dibh
->b_data
+ offset
+ sizeof(struct gfs2_dinode
), buf
, size
);
132 if (ip
->i_di
.di_size
< offset
+ size
)
133 ip
->i_di
.di_size
= offset
+ size
;
134 ip
->i_inode
.i_mtime
.tv_sec
= ip
->i_inode
.i_ctime
.tv_sec
= get_seconds();
135 gfs2_dinode_out(ip
, dibh
->b_data
);
145 * gfs2_dir_write_data - Write directory information to the inode
146 * @ip: The GFS2 inode
147 * @buf: The buffer containing information to be written
148 * @offset: The file offset to start writing at
149 * @size: The amount of data to write
151 * Returns: The number of bytes correctly written or error code
153 static int gfs2_dir_write_data(struct gfs2_inode
*ip
, const char *buf
,
154 u64 offset
, unsigned int size
)
156 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
157 struct buffer_head
*dibh
;
167 if (gfs2_is_stuffed(ip
) &&
168 offset
+ size
<= sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_dinode
))
169 return gfs2_dir_write_stuffed(ip
, buf
, (unsigned int)offset
,
172 if (gfs2_assert_warn(sdp
, gfs2_is_jdata(ip
)))
175 if (gfs2_is_stuffed(ip
)) {
176 error
= gfs2_unstuff_dinode(ip
, NULL
);
182 o
= do_div(lblock
, sdp
->sd_jbsize
) + sizeof(struct gfs2_meta_header
);
184 while (copied
< size
) {
186 struct buffer_head
*bh
;
189 amount
= size
- copied
;
190 if (amount
> sdp
->sd_sb
.sb_bsize
- o
)
191 amount
= sdp
->sd_sb
.sb_bsize
- o
;
195 error
= gfs2_extent_map(&ip
->i_inode
, lblock
, &new,
200 if (gfs2_assert_withdraw(sdp
, dblock
))
204 if (amount
== sdp
->sd_jbsize
|| new)
205 error
= gfs2_dir_get_new_buffer(ip
, dblock
, &bh
);
207 error
= gfs2_dir_get_existing_buffer(ip
, dblock
, &bh
);
212 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
213 memcpy(bh
->b_data
+ o
, buf
, amount
);
222 o
= sizeof(struct gfs2_meta_header
);
226 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
230 if (ip
->i_di
.di_size
< offset
+ copied
)
231 ip
->i_di
.di_size
= offset
+ copied
;
232 ip
->i_inode
.i_mtime
.tv_sec
= ip
->i_inode
.i_ctime
.tv_sec
= get_seconds();
234 gfs2_trans_add_bh(ip
->i_gl
, dibh
, 1);
235 gfs2_dinode_out(ip
, dibh
->b_data
);
245 static int gfs2_dir_read_stuffed(struct gfs2_inode
*ip
, char *buf
,
246 u64 offset
, unsigned int size
)
248 struct buffer_head
*dibh
;
251 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
253 offset
+= sizeof(struct gfs2_dinode
);
254 memcpy(buf
, dibh
->b_data
+ offset
, size
);
258 return (error
) ? error
: size
;
263 * gfs2_dir_read_data - Read a data from a directory inode
264 * @ip: The GFS2 Inode
265 * @buf: The buffer to place result into
266 * @offset: File offset to begin jdata_readng from
267 * @size: Amount of data to transfer
269 * Returns: The amount of data actually copied or the error
271 static int gfs2_dir_read_data(struct gfs2_inode
*ip
, char *buf
, u64 offset
,
272 unsigned int size
, unsigned ra
)
274 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
281 if (offset
>= ip
->i_di
.di_size
)
284 if (offset
+ size
> ip
->i_di
.di_size
)
285 size
= ip
->i_di
.di_size
- offset
;
290 if (gfs2_is_stuffed(ip
))
291 return gfs2_dir_read_stuffed(ip
, buf
, offset
, size
);
293 if (gfs2_assert_warn(sdp
, gfs2_is_jdata(ip
)))
297 o
= do_div(lblock
, sdp
->sd_jbsize
) + sizeof(struct gfs2_meta_header
);
299 while (copied
< size
) {
301 struct buffer_head
*bh
;
304 amount
= size
- copied
;
305 if (amount
> sdp
->sd_sb
.sb_bsize
- o
)
306 amount
= sdp
->sd_sb
.sb_bsize
- o
;
310 error
= gfs2_extent_map(&ip
->i_inode
, lblock
, &new,
312 if (error
|| !dblock
)
317 bh
= gfs2_meta_ra(ip
->i_gl
, dblock
, extlen
);
319 error
= gfs2_meta_read(ip
->i_gl
, dblock
, DIO_WAIT
, &bh
);
323 error
= gfs2_metatype_check(sdp
, bh
, GFS2_METATYPE_JD
);
330 memcpy(buf
, bh
->b_data
+ o
, amount
);
335 o
= sizeof(struct gfs2_meta_header
);
340 return (copied
) ? copied
: error
;
343 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent
*dent
)
345 return dent
->de_inum
.no_addr
== 0 || dent
->de_inum
.no_formal_ino
== 0;
348 static inline int __gfs2_dirent_find(const struct gfs2_dirent
*dent
,
349 const struct qstr
*name
, int ret
)
351 if (!gfs2_dirent_sentinel(dent
) &&
352 be32_to_cpu(dent
->de_hash
) == name
->hash
&&
353 be16_to_cpu(dent
->de_name_len
) == name
->len
&&
354 memcmp(dent
+1, name
->name
, name
->len
) == 0)
359 static int gfs2_dirent_find(const struct gfs2_dirent
*dent
,
360 const struct qstr
*name
,
363 return __gfs2_dirent_find(dent
, name
, 1);
366 static int gfs2_dirent_prev(const struct gfs2_dirent
*dent
,
367 const struct qstr
*name
,
370 return __gfs2_dirent_find(dent
, name
, 2);
374 * name->name holds ptr to start of block.
375 * name->len holds size of block.
377 static int gfs2_dirent_last(const struct gfs2_dirent
*dent
,
378 const struct qstr
*name
,
381 const char *start
= name
->name
;
382 const char *end
= (const char *)dent
+ be16_to_cpu(dent
->de_rec_len
);
383 if (name
->len
== (end
- start
))
388 static int gfs2_dirent_find_space(const struct gfs2_dirent
*dent
,
389 const struct qstr
*name
,
392 unsigned required
= GFS2_DIRENT_SIZE(name
->len
);
393 unsigned actual
= GFS2_DIRENT_SIZE(be16_to_cpu(dent
->de_name_len
));
394 unsigned totlen
= be16_to_cpu(dent
->de_rec_len
);
396 if (gfs2_dirent_sentinel(dent
))
397 actual
= GFS2_DIRENT_SIZE(0);
398 if (totlen
- actual
>= required
)
403 struct dirent_gather
{
404 const struct gfs2_dirent
**pdent
;
408 static int gfs2_dirent_gather(const struct gfs2_dirent
*dent
,
409 const struct qstr
*name
,
412 struct dirent_gather
*g
= opaque
;
413 if (!gfs2_dirent_sentinel(dent
)) {
414 g
->pdent
[g
->offset
++] = dent
;
420 * Other possible things to check:
421 * - Inode located within filesystem size (and on valid block)
422 * - Valid directory entry type
423 * Not sure how heavy-weight we want to make this... could also check
424 * hash is correct for example, but that would take a lot of extra time.
425 * For now the most important thing is to check that the various sizes
428 static int gfs2_check_dirent(struct gfs2_dirent
*dent
, unsigned int offset
,
429 unsigned int size
, unsigned int len
, int first
)
431 const char *msg
= "gfs2_dirent too small";
432 if (unlikely(size
< sizeof(struct gfs2_dirent
)))
434 msg
= "gfs2_dirent misaligned";
435 if (unlikely(offset
& 0x7))
437 msg
= "gfs2_dirent points beyond end of block";
438 if (unlikely(offset
+ size
> len
))
440 msg
= "zero inode number";
441 if (unlikely(!first
&& gfs2_dirent_sentinel(dent
)))
443 msg
= "name length is greater than space in dirent";
444 if (!gfs2_dirent_sentinel(dent
) &&
445 unlikely(sizeof(struct gfs2_dirent
)+be16_to_cpu(dent
->de_name_len
) >
450 printk(KERN_WARNING
"gfs2_check_dirent: %s (%s)\n", msg
,
451 first
? "first in block" : "not first in block");
455 static int gfs2_dirent_offset(const void *buf
)
457 const struct gfs2_meta_header
*h
= buf
;
462 switch(be32_to_cpu(h
->mh_type
)) {
463 case GFS2_METATYPE_LF
:
464 offset
= sizeof(struct gfs2_leaf
);
466 case GFS2_METATYPE_DI
:
467 offset
= sizeof(struct gfs2_dinode
);
474 printk(KERN_WARNING
"gfs2_scan_dirent: wrong block type %u\n",
475 be32_to_cpu(h
->mh_type
));
479 static struct gfs2_dirent
*gfs2_dirent_scan(struct inode
*inode
, void *buf
,
480 unsigned int len
, gfs2_dscan_t scan
,
481 const struct qstr
*name
,
484 struct gfs2_dirent
*dent
, *prev
;
489 ret
= gfs2_dirent_offset(buf
);
496 size
= be16_to_cpu(dent
->de_rec_len
);
497 if (gfs2_check_dirent(dent
, offset
, size
, len
, 1))
500 ret
= scan(dent
, name
, opaque
);
508 size
= be16_to_cpu(dent
->de_rec_len
);
509 if (gfs2_check_dirent(dent
, offset
, size
, len
, 0))
519 return prev
? prev
: dent
;
526 gfs2_consist_inode(GFS2_I(inode
));
527 return ERR_PTR(-EIO
);
532 * dirent_first - Return the first dirent
533 * @dip: the directory
535 * @dent: Pointer to list of dirents
537 * return first dirent whether bh points to leaf or stuffed dinode
539 * Returns: IS_LEAF, IS_DINODE, or -errno
542 static int dirent_first(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
543 struct gfs2_dirent
**dent
)
545 struct gfs2_meta_header
*h
= (struct gfs2_meta_header
*)bh
->b_data
;
547 if (be32_to_cpu(h
->mh_type
) == GFS2_METATYPE_LF
) {
548 if (gfs2_meta_check(GFS2_SB(&dip
->i_inode
), bh
))
550 *dent
= (struct gfs2_dirent
*)(bh
->b_data
+
551 sizeof(struct gfs2_leaf
));
554 if (gfs2_metatype_check(GFS2_SB(&dip
->i_inode
), bh
, GFS2_METATYPE_DI
))
556 *dent
= (struct gfs2_dirent
*)(bh
->b_data
+
557 sizeof(struct gfs2_dinode
));
562 static int dirent_check_reclen(struct gfs2_inode
*dip
,
563 const struct gfs2_dirent
*d
, const void *end_p
)
566 u16 rec_len
= be16_to_cpu(d
->de_rec_len
);
568 if (unlikely(rec_len
< sizeof(struct gfs2_dirent
)))
576 gfs2_consist_inode(dip
);
581 * dirent_next - Next dirent
582 * @dip: the directory
584 * @dent: Pointer to list of dirents
586 * Returns: 0 on success, error code otherwise
589 static int dirent_next(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
590 struct gfs2_dirent
**dent
)
592 struct gfs2_dirent
*cur
= *dent
, *tmp
;
593 char *bh_end
= bh
->b_data
+ bh
->b_size
;
596 ret
= dirent_check_reclen(dip
, cur
, bh_end
);
600 tmp
= (void *)cur
+ ret
;
601 ret
= dirent_check_reclen(dip
, tmp
, bh_end
);
605 /* Only the first dent could ever have de_inum.no_addr == 0 */
606 if (gfs2_dirent_sentinel(tmp
)) {
607 gfs2_consist_inode(dip
);
616 * dirent_del - Delete a dirent
617 * @dip: The GFS2 inode
619 * @prev: The previous dirent
620 * @cur: The current dirent
624 static void dirent_del(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
625 struct gfs2_dirent
*prev
, struct gfs2_dirent
*cur
)
627 u16 cur_rec_len
, prev_rec_len
;
629 if (gfs2_dirent_sentinel(cur
)) {
630 gfs2_consist_inode(dip
);
634 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
636 /* If there is no prev entry, this is the first entry in the block.
637 The de_rec_len is already as big as it needs to be. Just zero
638 out the inode number and return. */
641 cur
->de_inum
.no_addr
= 0;
642 cur
->de_inum
.no_formal_ino
= 0;
646 /* Combine this dentry with the previous one. */
648 prev_rec_len
= be16_to_cpu(prev
->de_rec_len
);
649 cur_rec_len
= be16_to_cpu(cur
->de_rec_len
);
651 if ((char *)prev
+ prev_rec_len
!= (char *)cur
)
652 gfs2_consist_inode(dip
);
653 if ((char *)cur
+ cur_rec_len
> bh
->b_data
+ bh
->b_size
)
654 gfs2_consist_inode(dip
);
656 prev_rec_len
+= cur_rec_len
;
657 prev
->de_rec_len
= cpu_to_be16(prev_rec_len
);
661 * Takes a dent from which to grab space as an argument. Returns the
662 * newly created dent.
664 static struct gfs2_dirent
*gfs2_init_dirent(struct inode
*inode
,
665 struct gfs2_dirent
*dent
,
666 const struct qstr
*name
,
667 struct buffer_head
*bh
)
669 struct gfs2_inode
*ip
= GFS2_I(inode
);
670 struct gfs2_dirent
*ndent
;
671 unsigned offset
= 0, totlen
;
673 if (!gfs2_dirent_sentinel(dent
))
674 offset
= GFS2_DIRENT_SIZE(be16_to_cpu(dent
->de_name_len
));
675 totlen
= be16_to_cpu(dent
->de_rec_len
);
676 BUG_ON(offset
+ name
->len
> totlen
);
677 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
678 ndent
= (struct gfs2_dirent
*)((char *)dent
+ offset
);
679 dent
->de_rec_len
= cpu_to_be16(offset
);
680 gfs2_qstr2dirent(name
, totlen
- offset
, ndent
);
684 static struct gfs2_dirent
*gfs2_dirent_alloc(struct inode
*inode
,
685 struct buffer_head
*bh
,
686 const struct qstr
*name
)
688 struct gfs2_dirent
*dent
;
689 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
,
690 gfs2_dirent_find_space
, name
, NULL
);
691 if (!dent
|| IS_ERR(dent
))
693 return gfs2_init_dirent(inode
, dent
, name
, bh
);
696 static int get_leaf(struct gfs2_inode
*dip
, u64 leaf_no
,
697 struct buffer_head
**bhp
)
701 error
= gfs2_meta_read(dip
->i_gl
, leaf_no
, DIO_WAIT
, bhp
);
702 if (!error
&& gfs2_metatype_check(GFS2_SB(&dip
->i_inode
), *bhp
, GFS2_METATYPE_LF
)) {
703 /* printk(KERN_INFO "block num=%llu\n", leaf_no); */
711 * get_leaf_nr - Get a leaf number associated with the index
712 * @dip: The GFS2 inode
716 * Returns: 0 on success, error code otherwise
719 static int get_leaf_nr(struct gfs2_inode
*dip
, u32 index
,
725 error
= gfs2_dir_read_data(dip
, (char *)&leaf_no
,
726 index
* sizeof(__be64
),
728 if (error
!= sizeof(u64
))
729 return (error
< 0) ? error
: -EIO
;
731 *leaf_out
= be64_to_cpu(leaf_no
);
736 static int get_first_leaf(struct gfs2_inode
*dip
, u32 index
,
737 struct buffer_head
**bh_out
)
742 error
= get_leaf_nr(dip
, index
, &leaf_no
);
744 error
= get_leaf(dip
, leaf_no
, bh_out
);
749 static struct gfs2_dirent
*gfs2_dirent_search(struct inode
*inode
,
750 const struct qstr
*name
,
752 struct buffer_head
**pbh
)
754 struct buffer_head
*bh
;
755 struct gfs2_dirent
*dent
;
756 struct gfs2_inode
*ip
= GFS2_I(inode
);
759 if (ip
->i_di
.di_flags
& GFS2_DIF_EXHASH
) {
760 struct gfs2_leaf
*leaf
;
761 unsigned hsize
= 1 << ip
->i_di
.di_depth
;
764 if (hsize
* sizeof(u64
) != ip
->i_di
.di_size
) {
765 gfs2_consist_inode(ip
);
766 return ERR_PTR(-EIO
);
769 index
= name
->hash
>> (32 - ip
->i_di
.di_depth
);
770 error
= get_first_leaf(ip
, index
, &bh
);
772 return ERR_PTR(error
);
774 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
,
778 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
779 ln
= be64_to_cpu(leaf
->lf_next
);
784 error
= get_leaf(ip
, ln
, &bh
);
787 return error
? ERR_PTR(error
) : NULL
;
791 error
= gfs2_meta_inode_buffer(ip
, &bh
);
793 return ERR_PTR(error
);
794 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
, scan
, name
, NULL
);
796 if (unlikely(dent
== NULL
|| IS_ERR(dent
))) {
804 static struct gfs2_leaf
*new_leaf(struct inode
*inode
, struct buffer_head
**pbh
, u16 depth
)
806 struct gfs2_inode
*ip
= GFS2_I(inode
);
807 u64 bn
= gfs2_alloc_meta(ip
);
808 struct buffer_head
*bh
= gfs2_meta_new(ip
->i_gl
, bn
);
809 struct gfs2_leaf
*leaf
;
810 struct gfs2_dirent
*dent
;
811 struct qstr name
= { .name
= "", .len
= 0, .hash
= 0 };
815 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
816 gfs2_metatype_set(bh
, GFS2_METATYPE_LF
, GFS2_FORMAT_LF
);
817 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
818 leaf
->lf_depth
= cpu_to_be16(depth
);
819 leaf
->lf_entries
= 0;
820 leaf
->lf_dirent_format
= cpu_to_be32(GFS2_FORMAT_DE
);
822 memset(leaf
->lf_reserved
, 0, sizeof(leaf
->lf_reserved
));
823 dent
= (struct gfs2_dirent
*)(leaf
+1);
824 gfs2_qstr2dirent(&name
, bh
->b_size
- sizeof(struct gfs2_leaf
), dent
);
830 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
831 * @dip: The GFS2 inode
833 * Returns: 0 on success, error code otherwise
836 static int dir_make_exhash(struct inode
*inode
)
838 struct gfs2_inode
*dip
= GFS2_I(inode
);
839 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
840 struct gfs2_dirent
*dent
;
842 struct buffer_head
*bh
, *dibh
;
843 struct gfs2_leaf
*leaf
;
850 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
854 /* Turn over a new leaf */
856 leaf
= new_leaf(inode
, &bh
, 0);
861 gfs2_assert(sdp
, dip
->i_di
.di_entries
< (1 << 16));
862 leaf
->lf_entries
= cpu_to_be16(dip
->i_di
.di_entries
);
866 gfs2_buffer_copy_tail(bh
, sizeof(struct gfs2_leaf
), dibh
,
867 sizeof(struct gfs2_dinode
));
869 /* Find last entry */
872 args
.len
= bh
->b_size
- sizeof(struct gfs2_dinode
) +
873 sizeof(struct gfs2_leaf
);
874 args
.name
= bh
->b_data
;
875 dent
= gfs2_dirent_scan(&dip
->i_inode
, bh
->b_data
, bh
->b_size
,
876 gfs2_dirent_last
, &args
, NULL
);
885 return PTR_ERR(dent
);
888 /* Adjust the last dirent's record length
889 (Remember that dent still points to the last entry.) */
891 dent
->de_rec_len
= cpu_to_be16(be16_to_cpu(dent
->de_rec_len
) +
892 sizeof(struct gfs2_dinode
) -
893 sizeof(struct gfs2_leaf
));
897 /* We're done with the new leaf block, now setup the new
900 gfs2_trans_add_bh(dip
->i_gl
, dibh
, 1);
901 gfs2_buffer_clear_tail(dibh
, sizeof(struct gfs2_dinode
));
903 lp
= (__be64
*)(dibh
->b_data
+ sizeof(struct gfs2_dinode
));
905 for (x
= sdp
->sd_hash_ptrs
; x
--; lp
++)
906 *lp
= cpu_to_be64(bn
);
908 dip
->i_di
.di_size
= sdp
->sd_sb
.sb_bsize
/ 2;
909 dip
->i_di
.di_blocks
++;
910 gfs2_set_inode_blocks(&dip
->i_inode
);
911 dip
->i_di
.di_flags
|= GFS2_DIF_EXHASH
;
913 for (x
= sdp
->sd_hash_ptrs
, y
= -1; x
; x
>>= 1, y
++) ;
914 dip
->i_di
.di_depth
= y
;
916 gfs2_dinode_out(dip
, dibh
->b_data
);
924 * dir_split_leaf - Split a leaf block into two
925 * @dip: The GFS2 inode
929 * Returns: 0 on success, error code on failure
932 static int dir_split_leaf(struct inode
*inode
, const struct qstr
*name
)
934 struct gfs2_inode
*dip
= GFS2_I(inode
);
935 struct buffer_head
*nbh
, *obh
, *dibh
;
936 struct gfs2_leaf
*nleaf
, *oleaf
;
937 struct gfs2_dirent
*dent
= NULL
, *prev
= NULL
, *next
= NULL
, *new;
938 u32 start
, len
, half_len
, divider
;
945 index
= name
->hash
>> (32 - dip
->i_di
.di_depth
);
946 error
= get_leaf_nr(dip
, index
, &leaf_no
);
950 /* Get the old leaf block */
951 error
= get_leaf(dip
, leaf_no
, &obh
);
955 oleaf
= (struct gfs2_leaf
*)obh
->b_data
;
956 if (dip
->i_di
.di_depth
== be16_to_cpu(oleaf
->lf_depth
)) {
958 return 1; /* can't split */
961 gfs2_trans_add_bh(dip
->i_gl
, obh
, 1);
963 nleaf
= new_leaf(inode
, &nbh
, be16_to_cpu(oleaf
->lf_depth
) + 1);
970 /* Compute the start and len of leaf pointers in the hash table. */
971 len
= 1 << (dip
->i_di
.di_depth
- be16_to_cpu(oleaf
->lf_depth
));
974 printk(KERN_WARNING
"di_depth %u lf_depth %u index %u\n", dip
->i_di
.di_depth
, be16_to_cpu(oleaf
->lf_depth
), index
);
975 gfs2_consist_inode(dip
);
980 start
= (index
& ~(len
- 1));
982 /* Change the pointers.
983 Don't bother distinguishing stuffed from non-stuffed.
984 This code is complicated enough already. */
985 lp
= kmalloc(half_len
* sizeof(__be64
), GFP_NOFS
| __GFP_NOFAIL
);
986 /* Change the pointers */
987 for (x
= 0; x
< half_len
; x
++)
988 lp
[x
] = cpu_to_be64(bn
);
990 error
= gfs2_dir_write_data(dip
, (char *)lp
, start
* sizeof(u64
),
991 half_len
* sizeof(u64
));
992 if (error
!= half_len
* sizeof(u64
)) {
1000 /* Compute the divider */
1001 divider
= (start
+ half_len
) << (32 - dip
->i_di
.di_depth
);
1003 /* Copy the entries */
1004 dirent_first(dip
, obh
, &dent
);
1008 if (dirent_next(dip
, obh
, &next
))
1011 if (!gfs2_dirent_sentinel(dent
) &&
1012 be32_to_cpu(dent
->de_hash
) < divider
) {
1014 str
.name
= (char*)(dent
+1);
1015 str
.len
= be16_to_cpu(dent
->de_name_len
);
1016 str
.hash
= be32_to_cpu(dent
->de_hash
);
1017 new = gfs2_dirent_alloc(inode
, nbh
, &str
);
1019 error
= PTR_ERR(new);
1023 new->de_inum
= dent
->de_inum
; /* No endian worries */
1024 new->de_type
= dent
->de_type
; /* No endian worries */
1025 nleaf
->lf_entries
= cpu_to_be16(be16_to_cpu(nleaf
->lf_entries
)+1);
1027 dirent_del(dip
, obh
, prev
, dent
);
1029 if (!oleaf
->lf_entries
)
1030 gfs2_consist_inode(dip
);
1031 oleaf
->lf_entries
= cpu_to_be16(be16_to_cpu(oleaf
->lf_entries
)-1);
1043 oleaf
->lf_depth
= nleaf
->lf_depth
;
1045 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1046 if (!gfs2_assert_withdraw(GFS2_SB(&dip
->i_inode
), !error
)) {
1047 dip
->i_di
.di_blocks
++;
1048 gfs2_set_inode_blocks(&dip
->i_inode
);
1049 gfs2_dinode_out(dip
, dibh
->b_data
);
1068 * dir_double_exhash - Double size of ExHash table
1069 * @dip: The GFS2 dinode
1071 * Returns: 0 on success, error code on failure
1074 static int dir_double_exhash(struct gfs2_inode
*dip
)
1076 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1077 struct buffer_head
*dibh
;
1085 hsize
= 1 << dip
->i_di
.di_depth
;
1086 if (hsize
* sizeof(u64
) != dip
->i_di
.di_size
) {
1087 gfs2_consist_inode(dip
);
1091 /* Allocate both the "from" and "to" buffers in one big chunk */
1093 buf
= kcalloc(3, sdp
->sd_hash_bsize
, GFP_KERNEL
| __GFP_NOFAIL
);
1095 for (block
= dip
->i_di
.di_size
>> sdp
->sd_hash_bsize_shift
; block
--;) {
1096 error
= gfs2_dir_read_data(dip
, (char *)buf
,
1097 block
* sdp
->sd_hash_bsize
,
1098 sdp
->sd_hash_bsize
, 1);
1099 if (error
!= sdp
->sd_hash_bsize
) {
1106 to
= (u64
*)((char *)buf
+ sdp
->sd_hash_bsize
);
1108 for (x
= sdp
->sd_hash_ptrs
; x
--; from
++) {
1109 *to
++ = *from
; /* No endianess worries */
1113 error
= gfs2_dir_write_data(dip
,
1114 (char *)buf
+ sdp
->sd_hash_bsize
,
1115 block
* sdp
->sd_sb
.sb_bsize
,
1116 sdp
->sd_sb
.sb_bsize
);
1117 if (error
!= sdp
->sd_sb
.sb_bsize
) {
1126 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1127 if (!gfs2_assert_withdraw(sdp
, !error
)) {
1128 dip
->i_di
.di_depth
++;
1129 gfs2_dinode_out(dip
, dibh
->b_data
);
1141 * compare_dents - compare directory entries by hash value
1145 * When comparing the hash entries of @a to @b:
1151 static int compare_dents(const void *a
, const void *b
)
1153 const struct gfs2_dirent
*dent_a
, *dent_b
;
1157 dent_a
= *(const struct gfs2_dirent
**)a
;
1158 hash_a
= be32_to_cpu(dent_a
->de_hash
);
1160 dent_b
= *(const struct gfs2_dirent
**)b
;
1161 hash_b
= be32_to_cpu(dent_b
->de_hash
);
1163 if (hash_a
> hash_b
)
1165 else if (hash_a
< hash_b
)
1168 unsigned int len_a
= be16_to_cpu(dent_a
->de_name_len
);
1169 unsigned int len_b
= be16_to_cpu(dent_b
->de_name_len
);
1173 else if (len_a
< len_b
)
1176 ret
= memcmp(dent_a
+ 1, dent_b
+ 1, len_a
);
1183 * do_filldir_main - read out directory entries
1184 * @dip: The GFS2 inode
1185 * @offset: The offset in the file to read from
1186 * @opaque: opaque data to pass to filldir
1187 * @filldir: The function to pass entries to
1188 * @darr: an array of struct gfs2_dirent pointers to read
1189 * @entries: the number of entries in darr
1190 * @copied: pointer to int that's non-zero if a entry has been copied out
1192 * Jump through some hoops to make sure that if there are hash collsions,
1193 * they are read out at the beginning of a buffer. We want to minimize
1194 * the possibility that they will fall into different readdir buffers or
1195 * that someone will want to seek to that location.
1197 * Returns: errno, >0 on exception from filldir
1200 static int do_filldir_main(struct gfs2_inode
*dip
, u64
*offset
,
1201 void *opaque
, gfs2_filldir_t filldir
,
1202 const struct gfs2_dirent
**darr
, u32 entries
,
1205 const struct gfs2_dirent
*dent
, *dent_next
;
1206 struct gfs2_inum_host inum
;
1212 sort(darr
, entries
, sizeof(struct gfs2_dirent
*), compare_dents
, NULL
);
1214 dent_next
= darr
[0];
1215 off_next
= be32_to_cpu(dent_next
->de_hash
);
1216 off_next
= gfs2_disk_hash2offset(off_next
);
1218 for (x
= 0, y
= 1; x
< entries
; x
++, y
++) {
1223 dent_next
= darr
[y
];
1224 off_next
= be32_to_cpu(dent_next
->de_hash
);
1225 off_next
= gfs2_disk_hash2offset(off_next
);
1231 if (off_next
== off
) {
1232 if (*copied
&& !run
)
1243 gfs2_inum_in(&inum
, (char *)&dent
->de_inum
);
1245 error
= filldir(opaque
, (const char *)(dent
+ 1),
1246 be16_to_cpu(dent
->de_name_len
),
1248 be16_to_cpu(dent
->de_type
));
1255 /* Increment the *offset by one, so the next time we come into the
1256 do_filldir fxn, we get the next entry instead of the last one in the
1264 static int gfs2_dir_read_leaf(struct inode
*inode
, u64
*offset
, void *opaque
,
1265 gfs2_filldir_t filldir
, int *copied
,
1266 unsigned *depth
, u64 leaf_no
)
1268 struct gfs2_inode
*ip
= GFS2_I(inode
);
1269 struct buffer_head
*bh
;
1270 struct gfs2_leaf
*lf
;
1271 unsigned entries
= 0;
1272 unsigned leaves
= 0;
1273 const struct gfs2_dirent
**darr
, *dent
;
1274 struct dirent_gather g
;
1275 struct buffer_head
**larr
;
1281 error
= get_leaf(ip
, lfn
, &bh
);
1284 lf
= (struct gfs2_leaf
*)bh
->b_data
;
1286 *depth
= be16_to_cpu(lf
->lf_depth
);
1287 entries
+= be16_to_cpu(lf
->lf_entries
);
1289 lfn
= be64_to_cpu(lf
->lf_next
);
1297 larr
= vmalloc((leaves
+ entries
) * sizeof(void *));
1300 darr
= (const struct gfs2_dirent
**)(larr
+ leaves
);
1306 error
= get_leaf(ip
, lfn
, &bh
);
1309 lf
= (struct gfs2_leaf
*)bh
->b_data
;
1310 lfn
= be64_to_cpu(lf
->lf_next
);
1311 if (lf
->lf_entries
) {
1312 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
,
1313 gfs2_dirent_gather
, NULL
, &g
);
1314 error
= PTR_ERR(dent
);
1325 error
= do_filldir_main(ip
, offset
, opaque
, filldir
, darr
,
1328 for(i
= 0; i
< leaf
; i
++)
1336 * dir_e_read - Reads the entries from a directory into a filldir buffer
1337 * @dip: dinode pointer
1338 * @offset: the hash of the last entry read shifted to the right once
1339 * @opaque: buffer for the filldir function to fill
1340 * @filldir: points to the filldir function to use
1345 static int dir_e_read(struct inode
*inode
, u64
*offset
, void *opaque
,
1346 gfs2_filldir_t filldir
)
1348 struct gfs2_inode
*dip
= GFS2_I(inode
);
1349 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1351 u32 ht_offset
, lp_offset
, ht_offset_cur
= -1;
1358 hsize
= 1 << dip
->i_di
.di_depth
;
1359 if (hsize
* sizeof(u64
) != dip
->i_di
.di_size
) {
1360 gfs2_consist_inode(dip
);
1364 hash
= gfs2_dir_offset2hash(*offset
);
1365 index
= hash
>> (32 - dip
->i_di
.di_depth
);
1367 lp
= kmalloc(sdp
->sd_hash_bsize
, GFP_KERNEL
);
1371 while (index
< hsize
) {
1372 lp_offset
= index
& (sdp
->sd_hash_ptrs
- 1);
1373 ht_offset
= index
- lp_offset
;
1375 if (ht_offset_cur
!= ht_offset
) {
1376 error
= gfs2_dir_read_data(dip
, (char *)lp
,
1377 ht_offset
* sizeof(__be64
),
1378 sdp
->sd_hash_bsize
, 1);
1379 if (error
!= sdp
->sd_hash_bsize
) {
1384 ht_offset_cur
= ht_offset
;
1387 error
= gfs2_dir_read_leaf(inode
, offset
, opaque
, filldir
,
1389 be64_to_cpu(lp
[lp_offset
]));
1393 len
= 1 << (dip
->i_di
.di_depth
- depth
);
1394 index
= (index
& ~(len
- 1)) + len
;
1404 int gfs2_dir_read(struct inode
*inode
, u64
*offset
, void *opaque
,
1405 gfs2_filldir_t filldir
)
1407 struct gfs2_inode
*dip
= GFS2_I(inode
);
1408 struct dirent_gather g
;
1409 const struct gfs2_dirent
**darr
, *dent
;
1410 struct buffer_head
*dibh
;
1414 if (!dip
->i_di
.di_entries
)
1417 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
)
1418 return dir_e_read(inode
, offset
, opaque
, filldir
);
1420 if (!gfs2_is_stuffed(dip
)) {
1421 gfs2_consist_inode(dip
);
1425 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1430 darr
= kmalloc(dip
->i_di
.di_entries
* sizeof(struct gfs2_dirent
*),
1435 dent
= gfs2_dirent_scan(inode
, dibh
->b_data
, dibh
->b_size
,
1436 gfs2_dirent_gather
, NULL
, &g
);
1438 error
= PTR_ERR(dent
);
1441 error
= do_filldir_main(dip
, offset
, opaque
, filldir
, darr
,
1442 dip
->i_di
.di_entries
, &copied
);
1456 * gfs2_dir_search - Search a directory
1457 * @dip: The GFS2 inode
1461 * This routine searches a directory for a file or another directory.
1462 * Assumes a glock is held on dip.
1467 int gfs2_dir_search(struct inode
*dir
, const struct qstr
*name
,
1468 struct gfs2_inum_host
*inum
, unsigned int *type
)
1470 struct buffer_head
*bh
;
1471 struct gfs2_dirent
*dent
;
1473 dent
= gfs2_dirent_search(dir
, name
, gfs2_dirent_find
, &bh
);
1476 return PTR_ERR(dent
);
1478 gfs2_inum_in(inum
, (char *)&dent
->de_inum
);
1480 *type
= be16_to_cpu(dent
->de_type
);
1487 static int dir_new_leaf(struct inode
*inode
, const struct qstr
*name
)
1489 struct buffer_head
*bh
, *obh
;
1490 struct gfs2_inode
*ip
= GFS2_I(inode
);
1491 struct gfs2_leaf
*leaf
, *oleaf
;
1496 index
= name
->hash
>> (32 - ip
->i_di
.di_depth
);
1497 error
= get_first_leaf(ip
, index
, &obh
);
1501 oleaf
= (struct gfs2_leaf
*)obh
->b_data
;
1502 bn
= be64_to_cpu(oleaf
->lf_next
);
1506 error
= get_leaf(ip
, bn
, &obh
);
1511 gfs2_trans_add_bh(ip
->i_gl
, obh
, 1);
1513 leaf
= new_leaf(inode
, &bh
, be16_to_cpu(oleaf
->lf_depth
));
1518 oleaf
->lf_next
= cpu_to_be64(bh
->b_blocknr
);
1522 error
= gfs2_meta_inode_buffer(ip
, &bh
);
1525 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
1526 ip
->i_di
.di_blocks
++;
1527 gfs2_set_inode_blocks(&ip
->i_inode
);
1528 gfs2_dinode_out(ip
, bh
->b_data
);
1534 * gfs2_dir_add - Add new filename into directory
1535 * @dip: The GFS2 inode
1536 * @filename: The new name
1537 * @inode: The inode number of the entry
1538 * @type: The type of the entry
1540 * Returns: 0 on success, error code on failure
1543 int gfs2_dir_add(struct inode
*inode
, const struct qstr
*name
,
1544 const struct gfs2_inum_host
*inum
, unsigned type
)
1546 struct gfs2_inode
*ip
= GFS2_I(inode
);
1547 struct buffer_head
*bh
;
1548 struct gfs2_dirent
*dent
;
1549 struct gfs2_leaf
*leaf
;
1553 dent
= gfs2_dirent_search(inode
, name
, gfs2_dirent_find_space
,
1557 return PTR_ERR(dent
);
1558 dent
= gfs2_init_dirent(inode
, dent
, name
, bh
);
1559 gfs2_inum_out(inum
, (char *)&dent
->de_inum
);
1560 dent
->de_type
= cpu_to_be16(type
);
1561 if (ip
->i_di
.di_flags
& GFS2_DIF_EXHASH
) {
1562 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1563 leaf
->lf_entries
= cpu_to_be16(be16_to_cpu(leaf
->lf_entries
) + 1);
1566 error
= gfs2_meta_inode_buffer(ip
, &bh
);
1569 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
1570 ip
->i_di
.di_entries
++;
1571 ip
->i_inode
.i_mtime
.tv_sec
= ip
->i_inode
.i_ctime
.tv_sec
= get_seconds();
1572 gfs2_dinode_out(ip
, bh
->b_data
);
1577 if (!(ip
->i_di
.di_flags
& GFS2_DIF_EXHASH
)) {
1578 error
= dir_make_exhash(inode
);
1583 error
= dir_split_leaf(inode
, name
);
1588 if (ip
->i_di
.di_depth
< GFS2_DIR_MAX_DEPTH
) {
1589 error
= dir_double_exhash(ip
);
1592 error
= dir_split_leaf(inode
, name
);
1598 error
= dir_new_leaf(inode
, name
);
1609 * gfs2_dir_del - Delete a directory entry
1610 * @dip: The GFS2 inode
1611 * @filename: The filename
1613 * Returns: 0 on success, error code on failure
1616 int gfs2_dir_del(struct gfs2_inode
*dip
, const struct qstr
*name
)
1618 struct gfs2_dirent
*dent
, *prev
= NULL
;
1619 struct buffer_head
*bh
;
1622 /* Returns _either_ the entry (if its first in block) or the
1623 previous entry otherwise */
1624 dent
= gfs2_dirent_search(&dip
->i_inode
, name
, gfs2_dirent_prev
, &bh
);
1626 gfs2_consist_inode(dip
);
1630 gfs2_consist_inode(dip
);
1631 return PTR_ERR(dent
);
1633 /* If not first in block, adjust pointers accordingly */
1634 if (gfs2_dirent_find(dent
, name
, NULL
) == 0) {
1636 dent
= (struct gfs2_dirent
*)((char *)dent
+ be16_to_cpu(prev
->de_rec_len
));
1639 dirent_del(dip
, bh
, prev
, dent
);
1640 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
) {
1641 struct gfs2_leaf
*leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1642 u16 entries
= be16_to_cpu(leaf
->lf_entries
);
1644 gfs2_consist_inode(dip
);
1645 leaf
->lf_entries
= cpu_to_be16(--entries
);
1649 error
= gfs2_meta_inode_buffer(dip
, &bh
);
1653 if (!dip
->i_di
.di_entries
)
1654 gfs2_consist_inode(dip
);
1655 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
1656 dip
->i_di
.di_entries
--;
1657 dip
->i_inode
.i_mtime
.tv_sec
= dip
->i_inode
.i_ctime
.tv_sec
= get_seconds();
1658 gfs2_dinode_out(dip
, bh
->b_data
);
1660 mark_inode_dirty(&dip
->i_inode
);
1666 * gfs2_dir_mvino - Change inode number of directory entry
1667 * @dip: The GFS2 inode
1671 * This routine changes the inode number of a directory entry. It's used
1672 * by rename to change ".." when a directory is moved.
1673 * Assumes a glock is held on dvp.
1678 int gfs2_dir_mvino(struct gfs2_inode
*dip
, const struct qstr
*filename
,
1679 struct gfs2_inum_host
*inum
, unsigned int new_type
)
1681 struct buffer_head
*bh
;
1682 struct gfs2_dirent
*dent
;
1685 dent
= gfs2_dirent_search(&dip
->i_inode
, filename
, gfs2_dirent_find
, &bh
);
1687 gfs2_consist_inode(dip
);
1691 return PTR_ERR(dent
);
1693 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
1694 gfs2_inum_out(inum
, (char *)&dent
->de_inum
);
1695 dent
->de_type
= cpu_to_be16(new_type
);
1697 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
) {
1699 error
= gfs2_meta_inode_buffer(dip
, &bh
);
1702 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
1705 dip
->i_inode
.i_mtime
.tv_sec
= dip
->i_inode
.i_ctime
.tv_sec
= get_seconds();
1706 gfs2_dinode_out(dip
, bh
->b_data
);
1712 * foreach_leaf - call a function for each leaf in a directory
1713 * @dip: the directory
1714 * @lc: the function to call for each each
1715 * @data: private data to pass to it
1720 static int foreach_leaf(struct gfs2_inode
*dip
, leaf_call_t lc
, void *data
)
1722 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1723 struct buffer_head
*bh
;
1724 struct gfs2_leaf
*leaf
;
1726 u32 ht_offset
, lp_offset
, ht_offset_cur
= -1;
1732 hsize
= 1 << dip
->i_di
.di_depth
;
1733 if (hsize
* sizeof(u64
) != dip
->i_di
.di_size
) {
1734 gfs2_consist_inode(dip
);
1738 lp
= kmalloc(sdp
->sd_hash_bsize
, GFP_KERNEL
);
1742 while (index
< hsize
) {
1743 lp_offset
= index
& (sdp
->sd_hash_ptrs
- 1);
1744 ht_offset
= index
- lp_offset
;
1746 if (ht_offset_cur
!= ht_offset
) {
1747 error
= gfs2_dir_read_data(dip
, (char *)lp
,
1748 ht_offset
* sizeof(__be64
),
1749 sdp
->sd_hash_bsize
, 1);
1750 if (error
!= sdp
->sd_hash_bsize
) {
1755 ht_offset_cur
= ht_offset
;
1758 leaf_no
= be64_to_cpu(lp
[lp_offset
]);
1760 error
= get_leaf(dip
, leaf_no
, &bh
);
1763 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1764 len
= 1 << (dip
->i_di
.di_depth
- be16_to_cpu(leaf
->lf_depth
));
1767 error
= lc(dip
, index
, len
, leaf_no
, data
);
1771 index
= (index
& ~(len
- 1)) + len
;
1776 if (index
!= hsize
) {
1777 gfs2_consist_inode(dip
);
1788 * leaf_dealloc - Deallocate a directory leaf
1789 * @dip: the directory
1790 * @index: the hash table offset in the directory
1791 * @len: the number of pointers to this leaf
1792 * @leaf_no: the leaf number
1798 static int leaf_dealloc(struct gfs2_inode
*dip
, u32 index
, u32 len
,
1799 u64 leaf_no
, void *data
)
1801 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1802 struct gfs2_leaf
*tmp_leaf
;
1803 struct gfs2_rgrp_list rlist
;
1804 struct buffer_head
*bh
, *dibh
;
1806 unsigned int rg_blocks
= 0, l_blocks
= 0;
1808 unsigned int x
, size
= len
* sizeof(u64
);
1811 memset(&rlist
, 0, sizeof(struct gfs2_rgrp_list
));
1813 ht
= kzalloc(size
, GFP_KERNEL
);
1817 gfs2_alloc_get(dip
);
1819 error
= gfs2_quota_hold(dip
, NO_QUOTA_CHANGE
, NO_QUOTA_CHANGE
);
1823 error
= gfs2_rindex_hold(sdp
, &dip
->i_alloc
.al_ri_gh
);
1827 /* Count the number of leaves */
1829 for (blk
= leaf_no
; blk
; blk
= nblk
) {
1830 error
= get_leaf(dip
, blk
, &bh
);
1833 tmp_leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1834 nblk
= be64_to_cpu(tmp_leaf
->lf_next
);
1837 gfs2_rlist_add(sdp
, &rlist
, blk
);
1841 gfs2_rlist_alloc(&rlist
, LM_ST_EXCLUSIVE
, 0);
1843 for (x
= 0; x
< rlist
.rl_rgrps
; x
++) {
1844 struct gfs2_rgrpd
*rgd
;
1845 rgd
= rlist
.rl_ghs
[x
].gh_gl
->gl_object
;
1846 rg_blocks
+= rgd
->rd_ri
.ri_length
;
1849 error
= gfs2_glock_nq_m(rlist
.rl_rgrps
, rlist
.rl_ghs
);
1853 error
= gfs2_trans_begin(sdp
,
1854 rg_blocks
+ (DIV_ROUND_UP(size
, sdp
->sd_jbsize
) + 1) +
1855 RES_DINODE
+ RES_STATFS
+ RES_QUOTA
, l_blocks
);
1857 goto out_rg_gunlock
;
1859 for (blk
= leaf_no
; blk
; blk
= nblk
) {
1860 error
= get_leaf(dip
, blk
, &bh
);
1863 tmp_leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1864 nblk
= be64_to_cpu(tmp_leaf
->lf_next
);
1867 gfs2_free_meta(dip
, blk
, 1);
1869 if (!dip
->i_di
.di_blocks
)
1870 gfs2_consist_inode(dip
);
1871 dip
->i_di
.di_blocks
--;
1872 gfs2_set_inode_blocks(&dip
->i_inode
);
1875 error
= gfs2_dir_write_data(dip
, ht
, index
* sizeof(u64
), size
);
1876 if (error
!= size
) {
1882 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1886 gfs2_trans_add_bh(dip
->i_gl
, dibh
, 1);
1887 gfs2_dinode_out(dip
, dibh
->b_data
);
1891 gfs2_trans_end(sdp
);
1893 gfs2_glock_dq_m(rlist
.rl_rgrps
, rlist
.rl_ghs
);
1895 gfs2_rlist_free(&rlist
);
1896 gfs2_glock_dq_uninit(&dip
->i_alloc
.al_ri_gh
);
1898 gfs2_quota_unhold(dip
);
1900 gfs2_alloc_put(dip
);
1906 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
1907 * @dip: the directory
1909 * Dealloc all on-disk directory leaves to FREEMETA state
1910 * Change on-disk inode type to "regular file"
1915 int gfs2_dir_exhash_dealloc(struct gfs2_inode
*dip
)
1917 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1918 struct buffer_head
*bh
;
1921 /* Dealloc on-disk leaves to FREEMETA state */
1922 error
= foreach_leaf(dip
, leaf_dealloc
, NULL
);
1926 /* Make this a regular file in case we crash.
1927 (We don't want to free these blocks a second time.) */
1929 error
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
1933 error
= gfs2_meta_inode_buffer(dip
, &bh
);
1935 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
1936 ((struct gfs2_dinode
*)bh
->b_data
)->di_mode
=
1937 cpu_to_be32(S_IFREG
);
1941 gfs2_trans_end(sdp
);
1947 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
1948 * @ip: the file being written to
1949 * @filname: the filename that's going to be added
1951 * Returns: 1 if alloc required, 0 if not, -ve on error
1954 int gfs2_diradd_alloc_required(struct inode
*inode
, const struct qstr
*name
)
1956 struct gfs2_dirent
*dent
;
1957 struct buffer_head
*bh
;
1959 dent
= gfs2_dirent_search(inode
, name
, gfs2_dirent_find_space
, &bh
);
1964 return PTR_ERR(dent
);