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_di
.di_mtime
= ip
->i_di
.di_ctime
= get_seconds();
135 gfs2_dinode_out(&ip
->i_di
, 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_di
.di_mtime
= ip
->i_di
.di_ctime
= get_seconds();
234 gfs2_trans_add_bh(ip
->i_gl
, dibh
, 1);
235 gfs2_dinode_out(&ip
->i_di
, 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_find(const struct gfs2_dirent
*dent
,
344 const struct qstr
*name
, int ret
)
346 if (dent
->de_inum
.no_addr
!= 0 &&
347 be32_to_cpu(dent
->de_hash
) == name
->hash
&&
348 be16_to_cpu(dent
->de_name_len
) == name
->len
&&
349 memcmp(dent
+1, name
->name
, name
->len
) == 0)
354 static int gfs2_dirent_find(const struct gfs2_dirent
*dent
,
355 const struct qstr
*name
,
358 return __gfs2_dirent_find(dent
, name
, 1);
361 static int gfs2_dirent_prev(const struct gfs2_dirent
*dent
,
362 const struct qstr
*name
,
365 return __gfs2_dirent_find(dent
, name
, 2);
369 * name->name holds ptr to start of block.
370 * name->len holds size of block.
372 static int gfs2_dirent_last(const struct gfs2_dirent
*dent
,
373 const struct qstr
*name
,
376 const char *start
= name
->name
;
377 const char *end
= (const char *)dent
+ be16_to_cpu(dent
->de_rec_len
);
378 if (name
->len
== (end
- start
))
383 static int gfs2_dirent_find_space(const struct gfs2_dirent
*dent
,
384 const struct qstr
*name
,
387 unsigned required
= GFS2_DIRENT_SIZE(name
->len
);
388 unsigned actual
= GFS2_DIRENT_SIZE(be16_to_cpu(dent
->de_name_len
));
389 unsigned totlen
= be16_to_cpu(dent
->de_rec_len
);
391 if (!dent
->de_inum
.no_addr
)
392 actual
= GFS2_DIRENT_SIZE(0);
393 if (totlen
- actual
>= required
)
398 struct dirent_gather
{
399 const struct gfs2_dirent
**pdent
;
403 static int gfs2_dirent_gather(const struct gfs2_dirent
*dent
,
404 const struct qstr
*name
,
407 struct dirent_gather
*g
= opaque
;
408 if (dent
->de_inum
.no_addr
) {
409 g
->pdent
[g
->offset
++] = dent
;
415 * Other possible things to check:
416 * - Inode located within filesystem size (and on valid block)
417 * - Valid directory entry type
418 * Not sure how heavy-weight we want to make this... could also check
419 * hash is correct for example, but that would take a lot of extra time.
420 * For now the most important thing is to check that the various sizes
423 static int gfs2_check_dirent(struct gfs2_dirent
*dent
, unsigned int offset
,
424 unsigned int size
, unsigned int len
, int first
)
426 const char *msg
= "gfs2_dirent too small";
427 if (unlikely(size
< sizeof(struct gfs2_dirent
)))
429 msg
= "gfs2_dirent misaligned";
430 if (unlikely(offset
& 0x7))
432 msg
= "gfs2_dirent points beyond end of block";
433 if (unlikely(offset
+ size
> len
))
435 msg
= "zero inode number";
436 if (unlikely(!first
&& !dent
->de_inum
.no_addr
))
438 msg
= "name length is greater than space in dirent";
439 if (dent
->de_inum
.no_addr
&&
440 unlikely(sizeof(struct gfs2_dirent
)+be16_to_cpu(dent
->de_name_len
) >
445 printk(KERN_WARNING
"gfs2_check_dirent: %s (%s)\n", msg
,
446 first
? "first in block" : "not first in block");
450 static int gfs2_dirent_offset(const void *buf
)
452 const struct gfs2_meta_header
*h
= buf
;
457 switch(be32_to_cpu(h
->mh_type
)) {
458 case GFS2_METATYPE_LF
:
459 offset
= sizeof(struct gfs2_leaf
);
461 case GFS2_METATYPE_DI
:
462 offset
= sizeof(struct gfs2_dinode
);
469 printk(KERN_WARNING
"gfs2_scan_dirent: wrong block type %u\n",
470 be32_to_cpu(h
->mh_type
));
474 static struct gfs2_dirent
*gfs2_dirent_scan(struct inode
*inode
, void *buf
,
475 unsigned int len
, gfs2_dscan_t scan
,
476 const struct qstr
*name
,
479 struct gfs2_dirent
*dent
, *prev
;
484 ret
= gfs2_dirent_offset(buf
);
491 size
= be16_to_cpu(dent
->de_rec_len
);
492 if (gfs2_check_dirent(dent
, offset
, size
, len
, 1))
495 ret
= scan(dent
, name
, opaque
);
503 size
= be16_to_cpu(dent
->de_rec_len
);
504 if (gfs2_check_dirent(dent
, offset
, size
, len
, 0))
514 return prev
? prev
: dent
;
521 gfs2_consist_inode(GFS2_I(inode
));
522 return ERR_PTR(-EIO
);
527 * dirent_first - Return the first dirent
528 * @dip: the directory
530 * @dent: Pointer to list of dirents
532 * return first dirent whether bh points to leaf or stuffed dinode
534 * Returns: IS_LEAF, IS_DINODE, or -errno
537 static int dirent_first(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
538 struct gfs2_dirent
**dent
)
540 struct gfs2_meta_header
*h
= (struct gfs2_meta_header
*)bh
->b_data
;
542 if (be32_to_cpu(h
->mh_type
) == GFS2_METATYPE_LF
) {
543 if (gfs2_meta_check(GFS2_SB(&dip
->i_inode
), bh
))
545 *dent
= (struct gfs2_dirent
*)(bh
->b_data
+
546 sizeof(struct gfs2_leaf
));
549 if (gfs2_metatype_check(GFS2_SB(&dip
->i_inode
), bh
, GFS2_METATYPE_DI
))
551 *dent
= (struct gfs2_dirent
*)(bh
->b_data
+
552 sizeof(struct gfs2_dinode
));
557 static int dirent_check_reclen(struct gfs2_inode
*dip
,
558 const struct gfs2_dirent
*d
, const void *end_p
)
561 u16 rec_len
= be16_to_cpu(d
->de_rec_len
);
563 if (unlikely(rec_len
< sizeof(struct gfs2_dirent
)))
571 gfs2_consist_inode(dip
);
576 * dirent_next - Next dirent
577 * @dip: the directory
579 * @dent: Pointer to list of dirents
581 * Returns: 0 on success, error code otherwise
584 static int dirent_next(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
585 struct gfs2_dirent
**dent
)
587 struct gfs2_dirent
*cur
= *dent
, *tmp
;
588 char *bh_end
= bh
->b_data
+ bh
->b_size
;
591 ret
= dirent_check_reclen(dip
, cur
, bh_end
);
595 tmp
= (void *)cur
+ ret
;
596 ret
= dirent_check_reclen(dip
, tmp
, bh_end
);
600 /* Only the first dent could ever have de_inum.no_addr == 0 */
601 if (!tmp
->de_inum
.no_addr
) {
602 gfs2_consist_inode(dip
);
611 * dirent_del - Delete a dirent
612 * @dip: The GFS2 inode
614 * @prev: The previous dirent
615 * @cur: The current dirent
619 static void dirent_del(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
620 struct gfs2_dirent
*prev
, struct gfs2_dirent
*cur
)
622 u16 cur_rec_len
, prev_rec_len
;
624 if (!cur
->de_inum
.no_addr
) {
625 gfs2_consist_inode(dip
);
629 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
631 /* If there is no prev entry, this is the first entry in the block.
632 The de_rec_len is already as big as it needs to be. Just zero
633 out the inode number and return. */
636 cur
->de_inum
.no_addr
= 0; /* No endianess worries */
640 /* Combine this dentry with the previous one. */
642 prev_rec_len
= be16_to_cpu(prev
->de_rec_len
);
643 cur_rec_len
= be16_to_cpu(cur
->de_rec_len
);
645 if ((char *)prev
+ prev_rec_len
!= (char *)cur
)
646 gfs2_consist_inode(dip
);
647 if ((char *)cur
+ cur_rec_len
> bh
->b_data
+ bh
->b_size
)
648 gfs2_consist_inode(dip
);
650 prev_rec_len
+= cur_rec_len
;
651 prev
->de_rec_len
= cpu_to_be16(prev_rec_len
);
655 * Takes a dent from which to grab space as an argument. Returns the
656 * newly created dent.
658 static struct gfs2_dirent
*gfs2_init_dirent(struct inode
*inode
,
659 struct gfs2_dirent
*dent
,
660 const struct qstr
*name
,
661 struct buffer_head
*bh
)
663 struct gfs2_inode
*ip
= GFS2_I(inode
);
664 struct gfs2_dirent
*ndent
;
665 unsigned offset
= 0, totlen
;
667 if (dent
->de_inum
.no_addr
)
668 offset
= GFS2_DIRENT_SIZE(be16_to_cpu(dent
->de_name_len
));
669 totlen
= be16_to_cpu(dent
->de_rec_len
);
670 BUG_ON(offset
+ name
->len
> totlen
);
671 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
672 ndent
= (struct gfs2_dirent
*)((char *)dent
+ offset
);
673 dent
->de_rec_len
= cpu_to_be16(offset
);
674 gfs2_qstr2dirent(name
, totlen
- offset
, ndent
);
678 static struct gfs2_dirent
*gfs2_dirent_alloc(struct inode
*inode
,
679 struct buffer_head
*bh
,
680 const struct qstr
*name
)
682 struct gfs2_dirent
*dent
;
683 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
,
684 gfs2_dirent_find_space
, name
, NULL
);
685 if (!dent
|| IS_ERR(dent
))
687 return gfs2_init_dirent(inode
, dent
, name
, bh
);
690 static int get_leaf(struct gfs2_inode
*dip
, u64 leaf_no
,
691 struct buffer_head
**bhp
)
695 error
= gfs2_meta_read(dip
->i_gl
, leaf_no
, DIO_WAIT
, bhp
);
696 if (!error
&& gfs2_metatype_check(GFS2_SB(&dip
->i_inode
), *bhp
, GFS2_METATYPE_LF
)) {
697 /* printk(KERN_INFO "block num=%llu\n", leaf_no); */
705 * get_leaf_nr - Get a leaf number associated with the index
706 * @dip: The GFS2 inode
710 * Returns: 0 on success, error code otherwise
713 static int get_leaf_nr(struct gfs2_inode
*dip
, u32 index
,
719 error
= gfs2_dir_read_data(dip
, (char *)&leaf_no
,
722 if (error
!= sizeof(u64
))
723 return (error
< 0) ? error
: -EIO
;
725 *leaf_out
= be64_to_cpu(leaf_no
);
730 static int get_first_leaf(struct gfs2_inode
*dip
, u32 index
,
731 struct buffer_head
**bh_out
)
736 error
= get_leaf_nr(dip
, index
, &leaf_no
);
738 error
= get_leaf(dip
, leaf_no
, bh_out
);
743 static struct gfs2_dirent
*gfs2_dirent_search(struct inode
*inode
,
744 const struct qstr
*name
,
746 struct buffer_head
**pbh
)
748 struct buffer_head
*bh
;
749 struct gfs2_dirent
*dent
;
750 struct gfs2_inode
*ip
= GFS2_I(inode
);
753 if (ip
->i_di
.di_flags
& GFS2_DIF_EXHASH
) {
754 struct gfs2_leaf
*leaf
;
755 unsigned hsize
= 1 << ip
->i_di
.di_depth
;
758 if (hsize
* sizeof(u64
) != ip
->i_di
.di_size
) {
759 gfs2_consist_inode(ip
);
760 return ERR_PTR(-EIO
);
763 index
= name
->hash
>> (32 - ip
->i_di
.di_depth
);
764 error
= get_first_leaf(ip
, index
, &bh
);
766 return ERR_PTR(error
);
768 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
,
772 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
773 ln
= be64_to_cpu(leaf
->lf_next
);
778 error
= get_leaf(ip
, ln
, &bh
);
781 return error
? ERR_PTR(error
) : NULL
;
785 error
= gfs2_meta_inode_buffer(ip
, &bh
);
787 return ERR_PTR(error
);
788 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
, scan
, name
, NULL
);
790 if (unlikely(dent
== NULL
|| IS_ERR(dent
))) {
798 static struct gfs2_leaf
*new_leaf(struct inode
*inode
, struct buffer_head
**pbh
, u16 depth
)
800 struct gfs2_inode
*ip
= GFS2_I(inode
);
801 u64 bn
= gfs2_alloc_meta(ip
);
802 struct buffer_head
*bh
= gfs2_meta_new(ip
->i_gl
, bn
);
803 struct gfs2_leaf
*leaf
;
804 struct gfs2_dirent
*dent
;
805 struct qstr name
= { .name
= "", .len
= 0, .hash
= 0 };
809 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
810 gfs2_metatype_set(bh
, GFS2_METATYPE_LF
, GFS2_FORMAT_LF
);
811 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
812 leaf
->lf_depth
= cpu_to_be16(depth
);
813 leaf
->lf_entries
= 0;
814 leaf
->lf_dirent_format
= cpu_to_be32(GFS2_FORMAT_DE
);
816 memset(leaf
->lf_reserved
, 0, sizeof(leaf
->lf_reserved
));
817 dent
= (struct gfs2_dirent
*)(leaf
+1);
818 gfs2_qstr2dirent(&name
, bh
->b_size
- sizeof(struct gfs2_leaf
), dent
);
824 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
825 * @dip: The GFS2 inode
827 * Returns: 0 on success, error code otherwise
830 static int dir_make_exhash(struct inode
*inode
)
832 struct gfs2_inode
*dip
= GFS2_I(inode
);
833 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
834 struct gfs2_dirent
*dent
;
836 struct buffer_head
*bh
, *dibh
;
837 struct gfs2_leaf
*leaf
;
843 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
847 /* Turn over a new leaf */
849 leaf
= new_leaf(inode
, &bh
, 0);
854 gfs2_assert(sdp
, dip
->i_di
.di_entries
< (1 << 16));
855 leaf
->lf_entries
= cpu_to_be16(dip
->i_di
.di_entries
);
859 gfs2_buffer_copy_tail(bh
, sizeof(struct gfs2_leaf
), dibh
,
860 sizeof(struct gfs2_dinode
));
862 /* Find last entry */
865 args
.len
= bh
->b_size
- sizeof(struct gfs2_dinode
) +
866 sizeof(struct gfs2_leaf
);
867 args
.name
= bh
->b_data
;
868 dent
= gfs2_dirent_scan(&dip
->i_inode
, bh
->b_data
, bh
->b_size
,
869 gfs2_dirent_last
, &args
, NULL
);
878 return PTR_ERR(dent
);
881 /* Adjust the last dirent's record length
882 (Remember that dent still points to the last entry.) */
884 dent
->de_rec_len
= cpu_to_be16(be16_to_cpu(dent
->de_rec_len
) +
885 sizeof(struct gfs2_dinode
) -
886 sizeof(struct gfs2_leaf
));
890 /* We're done with the new leaf block, now setup the new
893 gfs2_trans_add_bh(dip
->i_gl
, dibh
, 1);
894 gfs2_buffer_clear_tail(dibh
, sizeof(struct gfs2_dinode
));
896 lp
= (u64
*)(dibh
->b_data
+ sizeof(struct gfs2_dinode
));
898 for (x
= sdp
->sd_hash_ptrs
; x
--; lp
++)
899 *lp
= cpu_to_be64(bn
);
901 dip
->i_di
.di_size
= sdp
->sd_sb
.sb_bsize
/ 2;
902 dip
->i_di
.di_blocks
++;
903 dip
->i_di
.di_flags
|= GFS2_DIF_EXHASH
;
904 dip
->i_di
.di_payload_format
= 0;
906 for (x
= sdp
->sd_hash_ptrs
, y
= -1; x
; x
>>= 1, y
++) ;
907 dip
->i_di
.di_depth
= y
;
909 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
917 * dir_split_leaf - Split a leaf block into two
918 * @dip: The GFS2 inode
922 * Returns: 0 on success, error code on failure
925 static int dir_split_leaf(struct inode
*inode
, const struct qstr
*name
)
927 struct gfs2_inode
*dip
= GFS2_I(inode
);
928 struct buffer_head
*nbh
, *obh
, *dibh
;
929 struct gfs2_leaf
*nleaf
, *oleaf
;
930 struct gfs2_dirent
*dent
= NULL
, *prev
= NULL
, *next
= NULL
, *new;
931 u32 start
, len
, half_len
, divider
;
932 u64 bn
, *lp
, leaf_no
;
937 index
= name
->hash
>> (32 - dip
->i_di
.di_depth
);
938 error
= get_leaf_nr(dip
, index
, &leaf_no
);
942 /* Get the old leaf block */
943 error
= get_leaf(dip
, leaf_no
, &obh
);
947 oleaf
= (struct gfs2_leaf
*)obh
->b_data
;
948 if (dip
->i_di
.di_depth
== be16_to_cpu(oleaf
->lf_depth
)) {
950 return 1; /* can't split */
953 gfs2_trans_add_bh(dip
->i_gl
, obh
, 1);
955 nleaf
= new_leaf(inode
, &nbh
, be16_to_cpu(oleaf
->lf_depth
) + 1);
962 /* Compute the start and len of leaf pointers in the hash table. */
963 len
= 1 << (dip
->i_di
.di_depth
- be16_to_cpu(oleaf
->lf_depth
));
966 printk(KERN_WARNING
"di_depth %u lf_depth %u index %u\n", dip
->i_di
.di_depth
, be16_to_cpu(oleaf
->lf_depth
), index
);
967 gfs2_consist_inode(dip
);
972 start
= (index
& ~(len
- 1));
974 /* Change the pointers.
975 Don't bother distinguishing stuffed from non-stuffed.
976 This code is complicated enough already. */
977 lp
= kmalloc(half_len
* sizeof(u64
), GFP_NOFS
| __GFP_NOFAIL
);
978 /* Change the pointers */
979 for (x
= 0; x
< half_len
; x
++)
980 lp
[x
] = cpu_to_be64(bn
);
982 error
= gfs2_dir_write_data(dip
, (char *)lp
, start
* sizeof(u64
),
983 half_len
* sizeof(u64
));
984 if (error
!= half_len
* sizeof(u64
)) {
992 /* Compute the divider */
993 divider
= (start
+ half_len
) << (32 - dip
->i_di
.di_depth
);
995 /* Copy the entries */
996 dirent_first(dip
, obh
, &dent
);
1000 if (dirent_next(dip
, obh
, &next
))
1003 if (dent
->de_inum
.no_addr
&&
1004 be32_to_cpu(dent
->de_hash
) < divider
) {
1006 str
.name
= (char*)(dent
+1);
1007 str
.len
= be16_to_cpu(dent
->de_name_len
);
1008 str
.hash
= be32_to_cpu(dent
->de_hash
);
1009 new = gfs2_dirent_alloc(inode
, nbh
, &str
);
1011 error
= PTR_ERR(new);
1015 new->de_inum
= dent
->de_inum
; /* No endian worries */
1016 new->de_type
= dent
->de_type
; /* No endian worries */
1017 nleaf
->lf_entries
= cpu_to_be16(be16_to_cpu(nleaf
->lf_entries
)+1);
1019 dirent_del(dip
, obh
, prev
, dent
);
1021 if (!oleaf
->lf_entries
)
1022 gfs2_consist_inode(dip
);
1023 oleaf
->lf_entries
= cpu_to_be16(be16_to_cpu(oleaf
->lf_entries
)-1);
1035 oleaf
->lf_depth
= nleaf
->lf_depth
;
1037 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1038 if (!gfs2_assert_withdraw(GFS2_SB(&dip
->i_inode
), !error
)) {
1039 dip
->i_di
.di_blocks
++;
1040 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
1059 * dir_double_exhash - Double size of ExHash table
1060 * @dip: The GFS2 dinode
1062 * Returns: 0 on success, error code on failure
1065 static int dir_double_exhash(struct gfs2_inode
*dip
)
1067 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1068 struct buffer_head
*dibh
;
1076 hsize
= 1 << dip
->i_di
.di_depth
;
1077 if (hsize
* sizeof(u64
) != dip
->i_di
.di_size
) {
1078 gfs2_consist_inode(dip
);
1082 /* Allocate both the "from" and "to" buffers in one big chunk */
1084 buf
= kcalloc(3, sdp
->sd_hash_bsize
, GFP_KERNEL
| __GFP_NOFAIL
);
1086 for (block
= dip
->i_di
.di_size
>> sdp
->sd_hash_bsize_shift
; block
--;) {
1087 error
= gfs2_dir_read_data(dip
, (char *)buf
,
1088 block
* sdp
->sd_hash_bsize
,
1089 sdp
->sd_hash_bsize
, 1);
1090 if (error
!= sdp
->sd_hash_bsize
) {
1097 to
= (u64
*)((char *)buf
+ sdp
->sd_hash_bsize
);
1099 for (x
= sdp
->sd_hash_ptrs
; x
--; from
++) {
1100 *to
++ = *from
; /* No endianess worries */
1104 error
= gfs2_dir_write_data(dip
,
1105 (char *)buf
+ sdp
->sd_hash_bsize
,
1106 block
* sdp
->sd_sb
.sb_bsize
,
1107 sdp
->sd_sb
.sb_bsize
);
1108 if (error
!= sdp
->sd_sb
.sb_bsize
) {
1117 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1118 if (!gfs2_assert_withdraw(sdp
, !error
)) {
1119 dip
->i_di
.di_depth
++;
1120 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
1132 * compare_dents - compare directory entries by hash value
1136 * When comparing the hash entries of @a to @b:
1142 static int compare_dents(const void *a
, const void *b
)
1144 const struct gfs2_dirent
*dent_a
, *dent_b
;
1148 dent_a
= *(const struct gfs2_dirent
**)a
;
1149 hash_a
= be32_to_cpu(dent_a
->de_hash
);
1151 dent_b
= *(const struct gfs2_dirent
**)b
;
1152 hash_b
= be32_to_cpu(dent_b
->de_hash
);
1154 if (hash_a
> hash_b
)
1156 else if (hash_a
< hash_b
)
1159 unsigned int len_a
= be16_to_cpu(dent_a
->de_name_len
);
1160 unsigned int len_b
= be16_to_cpu(dent_b
->de_name_len
);
1164 else if (len_a
< len_b
)
1167 ret
= memcmp(dent_a
+ 1, dent_b
+ 1, len_a
);
1174 * do_filldir_main - read out directory entries
1175 * @dip: The GFS2 inode
1176 * @offset: The offset in the file to read from
1177 * @opaque: opaque data to pass to filldir
1178 * @filldir: The function to pass entries to
1179 * @darr: an array of struct gfs2_dirent pointers to read
1180 * @entries: the number of entries in darr
1181 * @copied: pointer to int that's non-zero if a entry has been copied out
1183 * Jump through some hoops to make sure that if there are hash collsions,
1184 * they are read out at the beginning of a buffer. We want to minimize
1185 * the possibility that they will fall into different readdir buffers or
1186 * that someone will want to seek to that location.
1188 * Returns: errno, >0 on exception from filldir
1191 static int do_filldir_main(struct gfs2_inode
*dip
, u64
*offset
,
1192 void *opaque
, gfs2_filldir_t filldir
,
1193 const struct gfs2_dirent
**darr
, u32 entries
,
1196 const struct gfs2_dirent
*dent
, *dent_next
;
1197 struct gfs2_inum inum
;
1203 sort(darr
, entries
, sizeof(struct gfs2_dirent
*), compare_dents
, NULL
);
1205 dent_next
= darr
[0];
1206 off_next
= be32_to_cpu(dent_next
->de_hash
);
1207 off_next
= gfs2_disk_hash2offset(off_next
);
1209 for (x
= 0, y
= 1; x
< entries
; x
++, y
++) {
1214 dent_next
= darr
[y
];
1215 off_next
= be32_to_cpu(dent_next
->de_hash
);
1216 off_next
= gfs2_disk_hash2offset(off_next
);
1222 if (off_next
== off
) {
1223 if (*copied
&& !run
)
1234 gfs2_inum_in(&inum
, (char *)&dent
->de_inum
);
1236 error
= filldir(opaque
, (const char *)(dent
+ 1),
1237 be16_to_cpu(dent
->de_name_len
),
1239 be16_to_cpu(dent
->de_type
));
1246 /* Increment the *offset by one, so the next time we come into the
1247 do_filldir fxn, we get the next entry instead of the last one in the
1255 static int gfs2_dir_read_leaf(struct inode
*inode
, u64
*offset
, void *opaque
,
1256 gfs2_filldir_t filldir
, int *copied
,
1257 unsigned *depth
, u64 leaf_no
)
1259 struct gfs2_inode
*ip
= GFS2_I(inode
);
1260 struct buffer_head
*bh
;
1261 struct gfs2_leaf
*lf
;
1262 unsigned entries
= 0;
1263 unsigned leaves
= 0;
1264 const struct gfs2_dirent
**darr
, *dent
;
1265 struct dirent_gather g
;
1266 struct buffer_head
**larr
;
1272 error
= get_leaf(ip
, lfn
, &bh
);
1275 lf
= (struct gfs2_leaf
*)bh
->b_data
;
1277 *depth
= be16_to_cpu(lf
->lf_depth
);
1278 entries
+= be16_to_cpu(lf
->lf_entries
);
1280 lfn
= be64_to_cpu(lf
->lf_next
);
1288 larr
= vmalloc((leaves
+ entries
) * sizeof(void *));
1291 darr
= (const struct gfs2_dirent
**)(larr
+ leaves
);
1297 error
= get_leaf(ip
, lfn
, &bh
);
1300 lf
= (struct gfs2_leaf
*)bh
->b_data
;
1301 lfn
= be64_to_cpu(lf
->lf_next
);
1302 if (lf
->lf_entries
) {
1303 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
,
1304 gfs2_dirent_gather
, NULL
, &g
);
1305 error
= PTR_ERR(dent
);
1316 error
= do_filldir_main(ip
, offset
, opaque
, filldir
, darr
,
1319 for(i
= 0; i
< leaf
; i
++)
1327 * dir_e_read - Reads the entries from a directory into a filldir buffer
1328 * @dip: dinode pointer
1329 * @offset: the hash of the last entry read shifted to the right once
1330 * @opaque: buffer for the filldir function to fill
1331 * @filldir: points to the filldir function to use
1336 static int dir_e_read(struct inode
*inode
, u64
*offset
, void *opaque
,
1337 gfs2_filldir_t filldir
)
1339 struct gfs2_inode
*dip
= GFS2_I(inode
);
1340 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1342 u32 ht_offset
, lp_offset
, ht_offset_cur
= -1;
1349 hsize
= 1 << dip
->i_di
.di_depth
;
1350 if (hsize
* sizeof(u64
) != dip
->i_di
.di_size
) {
1351 gfs2_consist_inode(dip
);
1355 hash
= gfs2_dir_offset2hash(*offset
);
1356 index
= hash
>> (32 - dip
->i_di
.di_depth
);
1358 lp
= kmalloc(sdp
->sd_hash_bsize
, GFP_KERNEL
);
1362 while (index
< hsize
) {
1363 lp_offset
= index
& (sdp
->sd_hash_ptrs
- 1);
1364 ht_offset
= index
- lp_offset
;
1366 if (ht_offset_cur
!= ht_offset
) {
1367 error
= gfs2_dir_read_data(dip
, (char *)lp
,
1368 ht_offset
* sizeof(u64
),
1369 sdp
->sd_hash_bsize
, 1);
1370 if (error
!= sdp
->sd_hash_bsize
) {
1375 ht_offset_cur
= ht_offset
;
1378 error
= gfs2_dir_read_leaf(inode
, offset
, opaque
, filldir
,
1380 be64_to_cpu(lp
[lp_offset
]));
1384 len
= 1 << (dip
->i_di
.di_depth
- depth
);
1385 index
= (index
& ~(len
- 1)) + len
;
1395 int gfs2_dir_read(struct inode
*inode
, u64
*offset
, void *opaque
,
1396 gfs2_filldir_t filldir
)
1398 struct gfs2_inode
*dip
= GFS2_I(inode
);
1399 struct dirent_gather g
;
1400 const struct gfs2_dirent
**darr
, *dent
;
1401 struct buffer_head
*dibh
;
1405 if (!dip
->i_di
.di_entries
)
1408 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
)
1409 return dir_e_read(inode
, offset
, opaque
, filldir
);
1411 if (!gfs2_is_stuffed(dip
)) {
1412 gfs2_consist_inode(dip
);
1416 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1421 darr
= kmalloc(dip
->i_di
.di_entries
* sizeof(struct gfs2_dirent
*),
1426 dent
= gfs2_dirent_scan(inode
, dibh
->b_data
, dibh
->b_size
,
1427 gfs2_dirent_gather
, NULL
, &g
);
1429 error
= PTR_ERR(dent
);
1432 error
= do_filldir_main(dip
, offset
, opaque
, filldir
, darr
,
1433 dip
->i_di
.di_entries
, &copied
);
1447 * gfs2_dir_search - Search a directory
1448 * @dip: The GFS2 inode
1452 * This routine searches a directory for a file or another directory.
1453 * Assumes a glock is held on dip.
1458 int gfs2_dir_search(struct inode
*dir
, const struct qstr
*name
,
1459 struct gfs2_inum
*inum
, unsigned int *type
)
1461 struct buffer_head
*bh
;
1462 struct gfs2_dirent
*dent
;
1464 dent
= gfs2_dirent_search(dir
, name
, gfs2_dirent_find
, &bh
);
1467 return PTR_ERR(dent
);
1469 gfs2_inum_in(inum
, (char *)&dent
->de_inum
);
1471 *type
= be16_to_cpu(dent
->de_type
);
1478 static int dir_new_leaf(struct inode
*inode
, const struct qstr
*name
)
1480 struct buffer_head
*bh
, *obh
;
1481 struct gfs2_inode
*ip
= GFS2_I(inode
);
1482 struct gfs2_leaf
*leaf
, *oleaf
;
1487 index
= name
->hash
>> (32 - ip
->i_di
.di_depth
);
1488 error
= get_first_leaf(ip
, index
, &obh
);
1492 oleaf
= (struct gfs2_leaf
*)obh
->b_data
;
1493 bn
= be64_to_cpu(oleaf
->lf_next
);
1497 error
= get_leaf(ip
, bn
, &obh
);
1502 gfs2_trans_add_bh(ip
->i_gl
, obh
, 1);
1504 leaf
= new_leaf(inode
, &bh
, be16_to_cpu(oleaf
->lf_depth
));
1509 oleaf
->lf_next
= cpu_to_be64(bh
->b_blocknr
);
1513 error
= gfs2_meta_inode_buffer(ip
, &bh
);
1516 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
1517 ip
->i_di
.di_blocks
++;
1518 gfs2_dinode_out(&ip
->i_di
, bh
->b_data
);
1524 * gfs2_dir_add - Add new filename into directory
1525 * @dip: The GFS2 inode
1526 * @filename: The new name
1527 * @inode: The inode number of the entry
1528 * @type: The type of the entry
1530 * Returns: 0 on success, error code on failure
1533 int gfs2_dir_add(struct inode
*inode
, const struct qstr
*name
,
1534 const struct gfs2_inum
*inum
, unsigned type
)
1536 struct gfs2_inode
*ip
= GFS2_I(inode
);
1537 struct buffer_head
*bh
;
1538 struct gfs2_dirent
*dent
;
1539 struct gfs2_leaf
*leaf
;
1543 dent
= gfs2_dirent_search(inode
, name
, gfs2_dirent_find_space
,
1547 return PTR_ERR(dent
);
1548 dent
= gfs2_init_dirent(inode
, dent
, name
, bh
);
1549 gfs2_inum_out(inum
, (char *)&dent
->de_inum
);
1550 dent
->de_type
= cpu_to_be16(type
);
1551 if (ip
->i_di
.di_flags
& GFS2_DIF_EXHASH
) {
1552 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1553 leaf
->lf_entries
= cpu_to_be16(be16_to_cpu(leaf
->lf_entries
) + 1);
1556 error
= gfs2_meta_inode_buffer(ip
, &bh
);
1559 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
1560 ip
->i_di
.di_entries
++;
1561 ip
->i_di
.di_mtime
= ip
->i_di
.di_ctime
= get_seconds();
1562 gfs2_dinode_out(&ip
->i_di
, bh
->b_data
);
1567 if (!(ip
->i_di
.di_flags
& GFS2_DIF_EXHASH
)) {
1568 error
= dir_make_exhash(inode
);
1573 error
= dir_split_leaf(inode
, name
);
1578 if (ip
->i_di
.di_depth
< GFS2_DIR_MAX_DEPTH
) {
1579 error
= dir_double_exhash(ip
);
1582 error
= dir_split_leaf(inode
, name
);
1588 error
= dir_new_leaf(inode
, name
);
1599 * gfs2_dir_del - Delete a directory entry
1600 * @dip: The GFS2 inode
1601 * @filename: The filename
1603 * Returns: 0 on success, error code on failure
1606 int gfs2_dir_del(struct gfs2_inode
*dip
, const struct qstr
*name
)
1608 struct gfs2_dirent
*dent
, *prev
= NULL
;
1609 struct buffer_head
*bh
;
1612 /* Returns _either_ the entry (if its first in block) or the
1613 previous entry otherwise */
1614 dent
= gfs2_dirent_search(&dip
->i_inode
, name
, gfs2_dirent_prev
, &bh
);
1616 gfs2_consist_inode(dip
);
1620 gfs2_consist_inode(dip
);
1621 return PTR_ERR(dent
);
1623 /* If not first in block, adjust pointers accordingly */
1624 if (gfs2_dirent_find(dent
, name
, NULL
) == 0) {
1626 dent
= (struct gfs2_dirent
*)((char *)dent
+ be16_to_cpu(prev
->de_rec_len
));
1629 dirent_del(dip
, bh
, prev
, dent
);
1630 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
) {
1631 struct gfs2_leaf
*leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1632 u16 entries
= be16_to_cpu(leaf
->lf_entries
);
1634 gfs2_consist_inode(dip
);
1635 leaf
->lf_entries
= cpu_to_be16(--entries
);
1639 error
= gfs2_meta_inode_buffer(dip
, &bh
);
1643 if (!dip
->i_di
.di_entries
)
1644 gfs2_consist_inode(dip
);
1645 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
1646 dip
->i_di
.di_entries
--;
1647 dip
->i_di
.di_mtime
= dip
->i_di
.di_ctime
= get_seconds();
1648 gfs2_dinode_out(&dip
->i_di
, bh
->b_data
);
1650 mark_inode_dirty(&dip
->i_inode
);
1656 * gfs2_dir_mvino - Change inode number of directory entry
1657 * @dip: The GFS2 inode
1661 * This routine changes the inode number of a directory entry. It's used
1662 * by rename to change ".." when a directory is moved.
1663 * Assumes a glock is held on dvp.
1668 int gfs2_dir_mvino(struct gfs2_inode
*dip
, const struct qstr
*filename
,
1669 struct gfs2_inum
*inum
, unsigned int new_type
)
1671 struct buffer_head
*bh
;
1672 struct gfs2_dirent
*dent
;
1675 dent
= gfs2_dirent_search(&dip
->i_inode
, filename
, gfs2_dirent_find
, &bh
);
1677 gfs2_consist_inode(dip
);
1681 return PTR_ERR(dent
);
1683 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
1684 gfs2_inum_out(inum
, (char *)&dent
->de_inum
);
1685 dent
->de_type
= cpu_to_be16(new_type
);
1687 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
) {
1689 error
= gfs2_meta_inode_buffer(dip
, &bh
);
1692 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
1695 dip
->i_di
.di_mtime
= dip
->i_di
.di_ctime
= get_seconds();
1696 gfs2_dinode_out(&dip
->i_di
, bh
->b_data
);
1702 * foreach_leaf - call a function for each leaf in a directory
1703 * @dip: the directory
1704 * @lc: the function to call for each each
1705 * @data: private data to pass to it
1710 static int foreach_leaf(struct gfs2_inode
*dip
, leaf_call_t lc
, void *data
)
1712 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1713 struct buffer_head
*bh
;
1714 struct gfs2_leaf
*leaf
;
1716 u32 ht_offset
, lp_offset
, ht_offset_cur
= -1;
1722 hsize
= 1 << dip
->i_di
.di_depth
;
1723 if (hsize
* sizeof(u64
) != dip
->i_di
.di_size
) {
1724 gfs2_consist_inode(dip
);
1728 lp
= kmalloc(sdp
->sd_hash_bsize
, GFP_KERNEL
);
1732 while (index
< hsize
) {
1733 lp_offset
= index
& (sdp
->sd_hash_ptrs
- 1);
1734 ht_offset
= index
- lp_offset
;
1736 if (ht_offset_cur
!= ht_offset
) {
1737 error
= gfs2_dir_read_data(dip
, (char *)lp
,
1738 ht_offset
* sizeof(u64
),
1739 sdp
->sd_hash_bsize
, 1);
1740 if (error
!= sdp
->sd_hash_bsize
) {
1745 ht_offset_cur
= ht_offset
;
1748 leaf_no
= be64_to_cpu(lp
[lp_offset
]);
1750 error
= get_leaf(dip
, leaf_no
, &bh
);
1753 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1754 len
= 1 << (dip
->i_di
.di_depth
- be16_to_cpu(leaf
->lf_depth
));
1757 error
= lc(dip
, index
, len
, leaf_no
, data
);
1761 index
= (index
& ~(len
- 1)) + len
;
1766 if (index
!= hsize
) {
1767 gfs2_consist_inode(dip
);
1778 * leaf_dealloc - Deallocate a directory leaf
1779 * @dip: the directory
1780 * @index: the hash table offset in the directory
1781 * @len: the number of pointers to this leaf
1782 * @leaf_no: the leaf number
1788 static int leaf_dealloc(struct gfs2_inode
*dip
, u32 index
, u32 len
,
1789 u64 leaf_no
, void *data
)
1791 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1792 struct gfs2_leaf
*tmp_leaf
;
1793 struct gfs2_rgrp_list rlist
;
1794 struct buffer_head
*bh
, *dibh
;
1796 unsigned int rg_blocks
= 0, l_blocks
= 0;
1798 unsigned int x
, size
= len
* sizeof(u64
);
1801 memset(&rlist
, 0, sizeof(struct gfs2_rgrp_list
));
1803 ht
= kzalloc(size
, GFP_KERNEL
);
1807 gfs2_alloc_get(dip
);
1809 error
= gfs2_quota_hold(dip
, NO_QUOTA_CHANGE
, NO_QUOTA_CHANGE
);
1813 error
= gfs2_rindex_hold(sdp
, &dip
->i_alloc
.al_ri_gh
);
1817 /* Count the number of leaves */
1819 for (blk
= leaf_no
; blk
; blk
= nblk
) {
1820 error
= get_leaf(dip
, blk
, &bh
);
1823 tmp_leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1824 nblk
= be64_to_cpu(tmp_leaf
->lf_next
);
1827 gfs2_rlist_add(sdp
, &rlist
, blk
);
1831 gfs2_rlist_alloc(&rlist
, LM_ST_EXCLUSIVE
, 0);
1833 for (x
= 0; x
< rlist
.rl_rgrps
; x
++) {
1834 struct gfs2_rgrpd
*rgd
;
1835 rgd
= rlist
.rl_ghs
[x
].gh_gl
->gl_object
;
1836 rg_blocks
+= rgd
->rd_ri
.ri_length
;
1839 error
= gfs2_glock_nq_m(rlist
.rl_rgrps
, rlist
.rl_ghs
);
1843 error
= gfs2_trans_begin(sdp
,
1844 rg_blocks
+ (DIV_ROUND_UP(size
, sdp
->sd_jbsize
) + 1) +
1845 RES_DINODE
+ RES_STATFS
+ RES_QUOTA
, l_blocks
);
1847 goto out_rg_gunlock
;
1849 for (blk
= leaf_no
; blk
; blk
= nblk
) {
1850 error
= get_leaf(dip
, blk
, &bh
);
1853 tmp_leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1854 nblk
= be64_to_cpu(tmp_leaf
->lf_next
);
1857 gfs2_free_meta(dip
, blk
, 1);
1859 if (!dip
->i_di
.di_blocks
)
1860 gfs2_consist_inode(dip
);
1861 dip
->i_di
.di_blocks
--;
1864 error
= gfs2_dir_write_data(dip
, ht
, index
* sizeof(u64
), size
);
1865 if (error
!= size
) {
1871 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1875 gfs2_trans_add_bh(dip
->i_gl
, dibh
, 1);
1876 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
1880 gfs2_trans_end(sdp
);
1882 gfs2_glock_dq_m(rlist
.rl_rgrps
, rlist
.rl_ghs
);
1884 gfs2_rlist_free(&rlist
);
1885 gfs2_glock_dq_uninit(&dip
->i_alloc
.al_ri_gh
);
1887 gfs2_quota_unhold(dip
);
1889 gfs2_alloc_put(dip
);
1895 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
1896 * @dip: the directory
1898 * Dealloc all on-disk directory leaves to FREEMETA state
1899 * Change on-disk inode type to "regular file"
1904 int gfs2_dir_exhash_dealloc(struct gfs2_inode
*dip
)
1906 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1907 struct buffer_head
*bh
;
1910 /* Dealloc on-disk leaves to FREEMETA state */
1911 error
= foreach_leaf(dip
, leaf_dealloc
, NULL
);
1915 /* Make this a regular file in case we crash.
1916 (We don't want to free these blocks a second time.) */
1918 error
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
1922 error
= gfs2_meta_inode_buffer(dip
, &bh
);
1924 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
1925 ((struct gfs2_dinode
*)bh
->b_data
)->di_mode
=
1926 cpu_to_be32(S_IFREG
);
1930 gfs2_trans_end(sdp
);
1936 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
1937 * @ip: the file being written to
1938 * @filname: the filename that's going to be added
1940 * Returns: 1 if alloc required, 0 if not, -ve on error
1943 int gfs2_diradd_alloc_required(struct inode
*inode
, const struct qstr
*name
)
1945 struct gfs2_dirent
*dent
;
1946 struct buffer_head
*bh
;
1948 dent
= gfs2_dirent_search(inode
, name
, gfs2_dirent_find_space
, &bh
);
1953 return PTR_ERR(dent
);