]>
git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/gfs2/dir.c
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/slab.h>
57 #include <linux/spinlock.h>
58 #include <linux/buffer_head.h>
59 #include <linux/sort.h>
60 #include <linux/gfs2_ondisk.h>
61 #include <linux/crc32.h>
62 #include <linux/vmalloc.h>
63 #include <linux/lm_interface.h>
77 #define IS_LEAF 1 /* Hashed (leaf) directory */
78 #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */
80 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
81 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
83 typedef int (*leaf_call_t
) (struct gfs2_inode
*dip
, u32 index
, u32 len
,
84 u64 leaf_no
, void *data
);
85 typedef int (*gfs2_dscan_t
)(const struct gfs2_dirent
*dent
,
86 const struct qstr
*name
, void *opaque
);
89 int gfs2_dir_get_new_buffer(struct gfs2_inode
*ip
, u64 block
,
90 struct buffer_head
**bhp
)
92 struct buffer_head
*bh
;
94 bh
= gfs2_meta_new(ip
->i_gl
, block
);
95 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
96 gfs2_metatype_set(bh
, GFS2_METATYPE_JD
, GFS2_FORMAT_JD
);
97 gfs2_buffer_clear_tail(bh
, sizeof(struct gfs2_meta_header
));
102 static int gfs2_dir_get_existing_buffer(struct gfs2_inode
*ip
, u64 block
,
103 struct buffer_head
**bhp
)
105 struct buffer_head
*bh
;
108 error
= gfs2_meta_read(ip
->i_gl
, block
, DIO_WAIT
, &bh
);
111 if (gfs2_metatype_check(GFS2_SB(&ip
->i_inode
), bh
, GFS2_METATYPE_JD
)) {
119 static int gfs2_dir_write_stuffed(struct gfs2_inode
*ip
, const char *buf
,
120 unsigned int offset
, unsigned int size
)
122 struct buffer_head
*dibh
;
125 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
129 gfs2_trans_add_bh(ip
->i_gl
, dibh
, 1);
130 memcpy(dibh
->b_data
+ offset
+ sizeof(struct gfs2_dinode
), buf
, size
);
131 if (ip
->i_di
.di_size
< offset
+ size
)
132 ip
->i_di
.di_size
= offset
+ size
;
133 ip
->i_inode
.i_mtime
= ip
->i_inode
.i_ctime
= CURRENT_TIME
;
134 gfs2_dinode_out(ip
, dibh
->b_data
);
144 * gfs2_dir_write_data - Write directory information to the inode
145 * @ip: The GFS2 inode
146 * @buf: The buffer containing information to be written
147 * @offset: The file offset to start writing at
148 * @size: The amount of data to write
150 * Returns: The number of bytes correctly written or error code
152 static int gfs2_dir_write_data(struct gfs2_inode
*ip
, const char *buf
,
153 u64 offset
, unsigned int size
)
155 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
156 struct buffer_head
*dibh
;
166 if (gfs2_is_stuffed(ip
) &&
167 offset
+ size
<= sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_dinode
))
168 return gfs2_dir_write_stuffed(ip
, buf
, (unsigned int)offset
,
171 if (gfs2_assert_warn(sdp
, gfs2_is_jdata(ip
)))
174 if (gfs2_is_stuffed(ip
)) {
175 error
= gfs2_unstuff_dinode(ip
, NULL
);
181 o
= do_div(lblock
, sdp
->sd_jbsize
) + sizeof(struct gfs2_meta_header
);
183 while (copied
< size
) {
185 struct buffer_head
*bh
;
188 amount
= size
- copied
;
189 if (amount
> sdp
->sd_sb
.sb_bsize
- o
)
190 amount
= sdp
->sd_sb
.sb_bsize
- o
;
194 error
= gfs2_extent_map(&ip
->i_inode
, lblock
, &new,
199 if (gfs2_assert_withdraw(sdp
, dblock
))
203 if (amount
== sdp
->sd_jbsize
|| new)
204 error
= gfs2_dir_get_new_buffer(ip
, dblock
, &bh
);
206 error
= gfs2_dir_get_existing_buffer(ip
, dblock
, &bh
);
211 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
212 memcpy(bh
->b_data
+ o
, buf
, amount
);
221 o
= sizeof(struct gfs2_meta_header
);
225 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
229 if (ip
->i_di
.di_size
< offset
+ copied
)
230 ip
->i_di
.di_size
= offset
+ copied
;
231 ip
->i_inode
.i_mtime
= ip
->i_inode
.i_ctime
= CURRENT_TIME
;
233 gfs2_trans_add_bh(ip
->i_gl
, dibh
, 1);
234 gfs2_dinode_out(ip
, dibh
->b_data
);
244 static int gfs2_dir_read_stuffed(struct gfs2_inode
*ip
, char *buf
,
245 u64 offset
, unsigned int size
)
247 struct buffer_head
*dibh
;
250 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
252 offset
+= sizeof(struct gfs2_dinode
);
253 memcpy(buf
, dibh
->b_data
+ offset
, size
);
257 return (error
) ? error
: size
;
262 * gfs2_dir_read_data - Read a data from a directory inode
263 * @ip: The GFS2 Inode
264 * @buf: The buffer to place result into
265 * @offset: File offset to begin jdata_readng from
266 * @size: Amount of data to transfer
268 * Returns: The amount of data actually copied or the error
270 static int gfs2_dir_read_data(struct gfs2_inode
*ip
, char *buf
, u64 offset
,
271 unsigned int size
, unsigned ra
)
273 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
280 if (offset
>= ip
->i_di
.di_size
)
283 if (offset
+ size
> ip
->i_di
.di_size
)
284 size
= ip
->i_di
.di_size
- offset
;
289 if (gfs2_is_stuffed(ip
))
290 return gfs2_dir_read_stuffed(ip
, buf
, offset
, size
);
292 if (gfs2_assert_warn(sdp
, gfs2_is_jdata(ip
)))
296 o
= do_div(lblock
, sdp
->sd_jbsize
) + sizeof(struct gfs2_meta_header
);
298 while (copied
< size
) {
300 struct buffer_head
*bh
;
303 amount
= size
- copied
;
304 if (amount
> sdp
->sd_sb
.sb_bsize
- o
)
305 amount
= sdp
->sd_sb
.sb_bsize
- o
;
309 error
= gfs2_extent_map(&ip
->i_inode
, lblock
, &new,
311 if (error
|| !dblock
)
316 bh
= gfs2_meta_ra(ip
->i_gl
, dblock
, extlen
);
318 error
= gfs2_meta_read(ip
->i_gl
, dblock
, DIO_WAIT
, &bh
);
322 error
= gfs2_metatype_check(sdp
, bh
, GFS2_METATYPE_JD
);
329 memcpy(buf
, bh
->b_data
+ o
, amount
);
334 o
= sizeof(struct gfs2_meta_header
);
339 return (copied
) ? copied
: error
;
342 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent
*dent
)
344 return dent
->de_inum
.no_addr
== 0 || dent
->de_inum
.no_formal_ino
== 0;
347 static inline int __gfs2_dirent_find(const struct gfs2_dirent
*dent
,
348 const struct qstr
*name
, int ret
)
350 if (!gfs2_dirent_sentinel(dent
) &&
351 be32_to_cpu(dent
->de_hash
) == name
->hash
&&
352 be16_to_cpu(dent
->de_name_len
) == name
->len
&&
353 memcmp(dent
+1, name
->name
, name
->len
) == 0)
358 static int gfs2_dirent_find(const struct gfs2_dirent
*dent
,
359 const struct qstr
*name
,
362 return __gfs2_dirent_find(dent
, name
, 1);
365 static int gfs2_dirent_prev(const struct gfs2_dirent
*dent
,
366 const struct qstr
*name
,
369 return __gfs2_dirent_find(dent
, name
, 2);
373 * name->name holds ptr to start of block.
374 * name->len holds size of block.
376 static int gfs2_dirent_last(const struct gfs2_dirent
*dent
,
377 const struct qstr
*name
,
380 const char *start
= name
->name
;
381 const char *end
= (const char *)dent
+ be16_to_cpu(dent
->de_rec_len
);
382 if (name
->len
== (end
- start
))
387 static int gfs2_dirent_find_space(const struct gfs2_dirent
*dent
,
388 const struct qstr
*name
,
391 unsigned required
= GFS2_DIRENT_SIZE(name
->len
);
392 unsigned actual
= GFS2_DIRENT_SIZE(be16_to_cpu(dent
->de_name_len
));
393 unsigned totlen
= be16_to_cpu(dent
->de_rec_len
);
395 if (gfs2_dirent_sentinel(dent
))
396 actual
= GFS2_DIRENT_SIZE(0);
397 if (totlen
- actual
>= required
)
402 struct dirent_gather
{
403 const struct gfs2_dirent
**pdent
;
407 static int gfs2_dirent_gather(const struct gfs2_dirent
*dent
,
408 const struct qstr
*name
,
411 struct dirent_gather
*g
= opaque
;
412 if (!gfs2_dirent_sentinel(dent
)) {
413 g
->pdent
[g
->offset
++] = dent
;
419 * Other possible things to check:
420 * - Inode located within filesystem size (and on valid block)
421 * - Valid directory entry type
422 * Not sure how heavy-weight we want to make this... could also check
423 * hash is correct for example, but that would take a lot of extra time.
424 * For now the most important thing is to check that the various sizes
427 static int gfs2_check_dirent(struct gfs2_dirent
*dent
, unsigned int offset
,
428 unsigned int size
, unsigned int len
, int first
)
430 const char *msg
= "gfs2_dirent too small";
431 if (unlikely(size
< sizeof(struct gfs2_dirent
)))
433 msg
= "gfs2_dirent misaligned";
434 if (unlikely(offset
& 0x7))
436 msg
= "gfs2_dirent points beyond end of block";
437 if (unlikely(offset
+ size
> len
))
439 msg
= "zero inode number";
440 if (unlikely(!first
&& gfs2_dirent_sentinel(dent
)))
442 msg
= "name length is greater than space in dirent";
443 if (!gfs2_dirent_sentinel(dent
) &&
444 unlikely(sizeof(struct gfs2_dirent
)+be16_to_cpu(dent
->de_name_len
) >
449 printk(KERN_WARNING
"gfs2_check_dirent: %s (%s)\n", msg
,
450 first
? "first in block" : "not first in block");
454 static int gfs2_dirent_offset(const void *buf
)
456 const struct gfs2_meta_header
*h
= buf
;
461 switch(be32_to_cpu(h
->mh_type
)) {
462 case GFS2_METATYPE_LF
:
463 offset
= sizeof(struct gfs2_leaf
);
465 case GFS2_METATYPE_DI
:
466 offset
= sizeof(struct gfs2_dinode
);
473 printk(KERN_WARNING
"gfs2_scan_dirent: wrong block type %u\n",
474 be32_to_cpu(h
->mh_type
));
478 static struct gfs2_dirent
*gfs2_dirent_scan(struct inode
*inode
, void *buf
,
479 unsigned int len
, gfs2_dscan_t scan
,
480 const struct qstr
*name
,
483 struct gfs2_dirent
*dent
, *prev
;
488 ret
= gfs2_dirent_offset(buf
);
495 size
= be16_to_cpu(dent
->de_rec_len
);
496 if (gfs2_check_dirent(dent
, offset
, size
, len
, 1))
499 ret
= scan(dent
, name
, opaque
);
507 size
= be16_to_cpu(dent
->de_rec_len
);
508 if (gfs2_check_dirent(dent
, offset
, size
, len
, 0))
518 return prev
? prev
: dent
;
525 gfs2_consist_inode(GFS2_I(inode
));
526 return ERR_PTR(-EIO
);
531 * dirent_first - Return the first dirent
532 * @dip: the directory
534 * @dent: Pointer to list of dirents
536 * return first dirent whether bh points to leaf or stuffed dinode
538 * Returns: IS_LEAF, IS_DINODE, or -errno
541 static int dirent_first(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
542 struct gfs2_dirent
**dent
)
544 struct gfs2_meta_header
*h
= (struct gfs2_meta_header
*)bh
->b_data
;
546 if (be32_to_cpu(h
->mh_type
) == GFS2_METATYPE_LF
) {
547 if (gfs2_meta_check(GFS2_SB(&dip
->i_inode
), bh
))
549 *dent
= (struct gfs2_dirent
*)(bh
->b_data
+
550 sizeof(struct gfs2_leaf
));
553 if (gfs2_metatype_check(GFS2_SB(&dip
->i_inode
), bh
, GFS2_METATYPE_DI
))
555 *dent
= (struct gfs2_dirent
*)(bh
->b_data
+
556 sizeof(struct gfs2_dinode
));
561 static int dirent_check_reclen(struct gfs2_inode
*dip
,
562 const struct gfs2_dirent
*d
, const void *end_p
)
565 u16 rec_len
= be16_to_cpu(d
->de_rec_len
);
567 if (unlikely(rec_len
< sizeof(struct gfs2_dirent
)))
575 gfs2_consist_inode(dip
);
580 * dirent_next - Next dirent
581 * @dip: the directory
583 * @dent: Pointer to list of dirents
585 * Returns: 0 on success, error code otherwise
588 static int dirent_next(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
589 struct gfs2_dirent
**dent
)
591 struct gfs2_dirent
*cur
= *dent
, *tmp
;
592 char *bh_end
= bh
->b_data
+ bh
->b_size
;
595 ret
= dirent_check_reclen(dip
, cur
, bh_end
);
599 tmp
= (void *)cur
+ ret
;
600 ret
= dirent_check_reclen(dip
, tmp
, bh_end
);
604 /* Only the first dent could ever have de_inum.no_addr == 0 */
605 if (gfs2_dirent_sentinel(tmp
)) {
606 gfs2_consist_inode(dip
);
615 * dirent_del - Delete a dirent
616 * @dip: The GFS2 inode
618 * @prev: The previous dirent
619 * @cur: The current dirent
623 static void dirent_del(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
624 struct gfs2_dirent
*prev
, struct gfs2_dirent
*cur
)
626 u16 cur_rec_len
, prev_rec_len
;
628 if (gfs2_dirent_sentinel(cur
)) {
629 gfs2_consist_inode(dip
);
633 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
635 /* If there is no prev entry, this is the first entry in the block.
636 The de_rec_len is already as big as it needs to be. Just zero
637 out the inode number and return. */
640 cur
->de_inum
.no_addr
= 0;
641 cur
->de_inum
.no_formal_ino
= 0;
645 /* Combine this dentry with the previous one. */
647 prev_rec_len
= be16_to_cpu(prev
->de_rec_len
);
648 cur_rec_len
= be16_to_cpu(cur
->de_rec_len
);
650 if ((char *)prev
+ prev_rec_len
!= (char *)cur
)
651 gfs2_consist_inode(dip
);
652 if ((char *)cur
+ cur_rec_len
> bh
->b_data
+ bh
->b_size
)
653 gfs2_consist_inode(dip
);
655 prev_rec_len
+= cur_rec_len
;
656 prev
->de_rec_len
= cpu_to_be16(prev_rec_len
);
660 * Takes a dent from which to grab space as an argument. Returns the
661 * newly created dent.
663 static struct gfs2_dirent
*gfs2_init_dirent(struct inode
*inode
,
664 struct gfs2_dirent
*dent
,
665 const struct qstr
*name
,
666 struct buffer_head
*bh
)
668 struct gfs2_inode
*ip
= GFS2_I(inode
);
669 struct gfs2_dirent
*ndent
;
670 unsigned offset
= 0, totlen
;
672 if (!gfs2_dirent_sentinel(dent
))
673 offset
= GFS2_DIRENT_SIZE(be16_to_cpu(dent
->de_name_len
));
674 totlen
= be16_to_cpu(dent
->de_rec_len
);
675 BUG_ON(offset
+ name
->len
> totlen
);
676 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
677 ndent
= (struct gfs2_dirent
*)((char *)dent
+ offset
);
678 dent
->de_rec_len
= cpu_to_be16(offset
);
679 gfs2_qstr2dirent(name
, totlen
- offset
, ndent
);
683 static struct gfs2_dirent
*gfs2_dirent_alloc(struct inode
*inode
,
684 struct buffer_head
*bh
,
685 const struct qstr
*name
)
687 struct gfs2_dirent
*dent
;
688 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
,
689 gfs2_dirent_find_space
, name
, NULL
);
690 if (!dent
|| IS_ERR(dent
))
692 return gfs2_init_dirent(inode
, dent
, name
, bh
);
695 static int get_leaf(struct gfs2_inode
*dip
, u64 leaf_no
,
696 struct buffer_head
**bhp
)
700 error
= gfs2_meta_read(dip
->i_gl
, leaf_no
, DIO_WAIT
, bhp
);
701 if (!error
&& gfs2_metatype_check(GFS2_SB(&dip
->i_inode
), *bhp
, GFS2_METATYPE_LF
)) {
702 /* printk(KERN_INFO "block num=%llu\n", leaf_no); */
710 * get_leaf_nr - Get a leaf number associated with the index
711 * @dip: The GFS2 inode
715 * Returns: 0 on success, error code otherwise
718 static int get_leaf_nr(struct gfs2_inode
*dip
, u32 index
,
724 error
= gfs2_dir_read_data(dip
, (char *)&leaf_no
,
725 index
* sizeof(__be64
),
727 if (error
!= sizeof(u64
))
728 return (error
< 0) ? error
: -EIO
;
730 *leaf_out
= be64_to_cpu(leaf_no
);
735 static int get_first_leaf(struct gfs2_inode
*dip
, u32 index
,
736 struct buffer_head
**bh_out
)
741 error
= get_leaf_nr(dip
, index
, &leaf_no
);
743 error
= get_leaf(dip
, leaf_no
, bh_out
);
748 static struct gfs2_dirent
*gfs2_dirent_search(struct inode
*inode
,
749 const struct qstr
*name
,
751 struct buffer_head
**pbh
)
753 struct buffer_head
*bh
;
754 struct gfs2_dirent
*dent
;
755 struct gfs2_inode
*ip
= GFS2_I(inode
);
758 if (ip
->i_di
.di_flags
& GFS2_DIF_EXHASH
) {
759 struct gfs2_leaf
*leaf
;
760 unsigned hsize
= 1 << ip
->i_di
.di_depth
;
763 if (hsize
* sizeof(u64
) != ip
->i_di
.di_size
) {
764 gfs2_consist_inode(ip
);
765 return ERR_PTR(-EIO
);
768 index
= name
->hash
>> (32 - ip
->i_di
.di_depth
);
769 error
= get_first_leaf(ip
, index
, &bh
);
771 return ERR_PTR(error
);
773 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
,
777 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
778 ln
= be64_to_cpu(leaf
->lf_next
);
783 error
= get_leaf(ip
, ln
, &bh
);
786 return error
? ERR_PTR(error
) : NULL
;
790 error
= gfs2_meta_inode_buffer(ip
, &bh
);
792 return ERR_PTR(error
);
793 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
, scan
, name
, NULL
);
795 if (unlikely(dent
== NULL
|| IS_ERR(dent
))) {
803 static struct gfs2_leaf
*new_leaf(struct inode
*inode
, struct buffer_head
**pbh
, u16 depth
)
805 struct gfs2_inode
*ip
= GFS2_I(inode
);
806 u64 bn
= gfs2_alloc_meta(ip
);
807 struct buffer_head
*bh
= gfs2_meta_new(ip
->i_gl
, bn
);
808 struct gfs2_leaf
*leaf
;
809 struct gfs2_dirent
*dent
;
810 struct qstr name
= { .name
= "", .len
= 0, .hash
= 0 };
814 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
815 gfs2_metatype_set(bh
, GFS2_METATYPE_LF
, GFS2_FORMAT_LF
);
816 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
817 leaf
->lf_depth
= cpu_to_be16(depth
);
818 leaf
->lf_entries
= 0;
819 leaf
->lf_dirent_format
= cpu_to_be32(GFS2_FORMAT_DE
);
821 memset(leaf
->lf_reserved
, 0, sizeof(leaf
->lf_reserved
));
822 dent
= (struct gfs2_dirent
*)(leaf
+1);
823 gfs2_qstr2dirent(&name
, bh
->b_size
- sizeof(struct gfs2_leaf
), dent
);
829 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
830 * @dip: The GFS2 inode
832 * Returns: 0 on success, error code otherwise
835 static int dir_make_exhash(struct inode
*inode
)
837 struct gfs2_inode
*dip
= GFS2_I(inode
);
838 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
839 struct gfs2_dirent
*dent
;
841 struct buffer_head
*bh
, *dibh
;
842 struct gfs2_leaf
*leaf
;
849 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
853 /* Turn over a new leaf */
855 leaf
= new_leaf(inode
, &bh
, 0);
860 gfs2_assert(sdp
, dip
->i_di
.di_entries
< (1 << 16));
861 leaf
->lf_entries
= cpu_to_be16(dip
->i_di
.di_entries
);
865 gfs2_buffer_copy_tail(bh
, sizeof(struct gfs2_leaf
), dibh
,
866 sizeof(struct gfs2_dinode
));
868 /* Find last entry */
871 args
.len
= bh
->b_size
- sizeof(struct gfs2_dinode
) +
872 sizeof(struct gfs2_leaf
);
873 args
.name
= bh
->b_data
;
874 dent
= gfs2_dirent_scan(&dip
->i_inode
, bh
->b_data
, bh
->b_size
,
875 gfs2_dirent_last
, &args
, NULL
);
884 return PTR_ERR(dent
);
887 /* Adjust the last dirent's record length
888 (Remember that dent still points to the last entry.) */
890 dent
->de_rec_len
= cpu_to_be16(be16_to_cpu(dent
->de_rec_len
) +
891 sizeof(struct gfs2_dinode
) -
892 sizeof(struct gfs2_leaf
));
896 /* We're done with the new leaf block, now setup the new
899 gfs2_trans_add_bh(dip
->i_gl
, dibh
, 1);
900 gfs2_buffer_clear_tail(dibh
, sizeof(struct gfs2_dinode
));
902 lp
= (__be64
*)(dibh
->b_data
+ sizeof(struct gfs2_dinode
));
904 for (x
= sdp
->sd_hash_ptrs
; x
--; lp
++)
905 *lp
= cpu_to_be64(bn
);
907 dip
->i_di
.di_size
= sdp
->sd_sb
.sb_bsize
/ 2;
908 dip
->i_di
.di_blocks
++;
909 gfs2_set_inode_blocks(&dip
->i_inode
);
910 dip
->i_di
.di_flags
|= GFS2_DIF_EXHASH
;
912 for (x
= sdp
->sd_hash_ptrs
, y
= -1; x
; x
>>= 1, y
++) ;
913 dip
->i_di
.di_depth
= y
;
915 gfs2_dinode_out(dip
, dibh
->b_data
);
923 * dir_split_leaf - Split a leaf block into two
924 * @dip: The GFS2 inode
928 * Returns: 0 on success, error code on failure
931 static int dir_split_leaf(struct inode
*inode
, const struct qstr
*name
)
933 struct gfs2_inode
*dip
= GFS2_I(inode
);
934 struct buffer_head
*nbh
, *obh
, *dibh
;
935 struct gfs2_leaf
*nleaf
, *oleaf
;
936 struct gfs2_dirent
*dent
= NULL
, *prev
= NULL
, *next
= NULL
, *new;
937 u32 start
, len
, half_len
, divider
;
944 index
= name
->hash
>> (32 - dip
->i_di
.di_depth
);
945 error
= get_leaf_nr(dip
, index
, &leaf_no
);
949 /* Get the old leaf block */
950 error
= get_leaf(dip
, leaf_no
, &obh
);
954 oleaf
= (struct gfs2_leaf
*)obh
->b_data
;
955 if (dip
->i_di
.di_depth
== be16_to_cpu(oleaf
->lf_depth
)) {
957 return 1; /* can't split */
960 gfs2_trans_add_bh(dip
->i_gl
, obh
, 1);
962 nleaf
= new_leaf(inode
, &nbh
, be16_to_cpu(oleaf
->lf_depth
) + 1);
969 /* Compute the start and len of leaf pointers in the hash table. */
970 len
= 1 << (dip
->i_di
.di_depth
- be16_to_cpu(oleaf
->lf_depth
));
973 printk(KERN_WARNING
"di_depth %u lf_depth %u index %u\n", dip
->i_di
.di_depth
, be16_to_cpu(oleaf
->lf_depth
), index
);
974 gfs2_consist_inode(dip
);
979 start
= (index
& ~(len
- 1));
981 /* Change the pointers.
982 Don't bother distinguishing stuffed from non-stuffed.
983 This code is complicated enough already. */
984 lp
= kmalloc(half_len
* sizeof(__be64
), GFP_NOFS
| __GFP_NOFAIL
);
985 /* Change the pointers */
986 for (x
= 0; x
< half_len
; x
++)
987 lp
[x
] = cpu_to_be64(bn
);
989 error
= gfs2_dir_write_data(dip
, (char *)lp
, start
* sizeof(u64
),
990 half_len
* sizeof(u64
));
991 if (error
!= half_len
* sizeof(u64
)) {
999 /* Compute the divider */
1000 divider
= (start
+ half_len
) << (32 - dip
->i_di
.di_depth
);
1002 /* Copy the entries */
1003 dirent_first(dip
, obh
, &dent
);
1007 if (dirent_next(dip
, obh
, &next
))
1010 if (!gfs2_dirent_sentinel(dent
) &&
1011 be32_to_cpu(dent
->de_hash
) < divider
) {
1013 str
.name
= (char*)(dent
+1);
1014 str
.len
= be16_to_cpu(dent
->de_name_len
);
1015 str
.hash
= be32_to_cpu(dent
->de_hash
);
1016 new = gfs2_dirent_alloc(inode
, nbh
, &str
);
1018 error
= PTR_ERR(new);
1022 new->de_inum
= dent
->de_inum
; /* No endian worries */
1023 new->de_type
= dent
->de_type
; /* No endian worries */
1024 nleaf
->lf_entries
= cpu_to_be16(be16_to_cpu(nleaf
->lf_entries
)+1);
1026 dirent_del(dip
, obh
, prev
, dent
);
1028 if (!oleaf
->lf_entries
)
1029 gfs2_consist_inode(dip
);
1030 oleaf
->lf_entries
= cpu_to_be16(be16_to_cpu(oleaf
->lf_entries
)-1);
1042 oleaf
->lf_depth
= nleaf
->lf_depth
;
1044 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1045 if (!gfs2_assert_withdraw(GFS2_SB(&dip
->i_inode
), !error
)) {
1046 dip
->i_di
.di_blocks
++;
1047 gfs2_set_inode_blocks(&dip
->i_inode
);
1048 gfs2_dinode_out(dip
, dibh
->b_data
);
1067 * dir_double_exhash - Double size of ExHash table
1068 * @dip: The GFS2 dinode
1070 * Returns: 0 on success, error code on failure
1073 static int dir_double_exhash(struct gfs2_inode
*dip
)
1075 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1076 struct buffer_head
*dibh
;
1084 hsize
= 1 << dip
->i_di
.di_depth
;
1085 if (hsize
* sizeof(u64
) != dip
->i_di
.di_size
) {
1086 gfs2_consist_inode(dip
);
1090 /* Allocate both the "from" and "to" buffers in one big chunk */
1092 buf
= kcalloc(3, sdp
->sd_hash_bsize
, GFP_KERNEL
| __GFP_NOFAIL
);
1094 for (block
= dip
->i_di
.di_size
>> sdp
->sd_hash_bsize_shift
; block
--;) {
1095 error
= gfs2_dir_read_data(dip
, (char *)buf
,
1096 block
* sdp
->sd_hash_bsize
,
1097 sdp
->sd_hash_bsize
, 1);
1098 if (error
!= sdp
->sd_hash_bsize
) {
1105 to
= (u64
*)((char *)buf
+ sdp
->sd_hash_bsize
);
1107 for (x
= sdp
->sd_hash_ptrs
; x
--; from
++) {
1108 *to
++ = *from
; /* No endianess worries */
1112 error
= gfs2_dir_write_data(dip
,
1113 (char *)buf
+ sdp
->sd_hash_bsize
,
1114 block
* sdp
->sd_sb
.sb_bsize
,
1115 sdp
->sd_sb
.sb_bsize
);
1116 if (error
!= sdp
->sd_sb
.sb_bsize
) {
1125 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1126 if (!gfs2_assert_withdraw(sdp
, !error
)) {
1127 dip
->i_di
.di_depth
++;
1128 gfs2_dinode_out(dip
, dibh
->b_data
);
1140 * compare_dents - compare directory entries by hash value
1144 * When comparing the hash entries of @a to @b:
1150 static int compare_dents(const void *a
, const void *b
)
1152 const struct gfs2_dirent
*dent_a
, *dent_b
;
1156 dent_a
= *(const struct gfs2_dirent
**)a
;
1157 hash_a
= be32_to_cpu(dent_a
->de_hash
);
1159 dent_b
= *(const struct gfs2_dirent
**)b
;
1160 hash_b
= be32_to_cpu(dent_b
->de_hash
);
1162 if (hash_a
> hash_b
)
1164 else if (hash_a
< hash_b
)
1167 unsigned int len_a
= be16_to_cpu(dent_a
->de_name_len
);
1168 unsigned int len_b
= be16_to_cpu(dent_b
->de_name_len
);
1172 else if (len_a
< len_b
)
1175 ret
= memcmp(dent_a
+ 1, dent_b
+ 1, len_a
);
1182 * do_filldir_main - read out directory entries
1183 * @dip: The GFS2 inode
1184 * @offset: The offset in the file to read from
1185 * @opaque: opaque data to pass to filldir
1186 * @filldir: The function to pass entries to
1187 * @darr: an array of struct gfs2_dirent pointers to read
1188 * @entries: the number of entries in darr
1189 * @copied: pointer to int that's non-zero if a entry has been copied out
1191 * Jump through some hoops to make sure that if there are hash collsions,
1192 * they are read out at the beginning of a buffer. We want to minimize
1193 * the possibility that they will fall into different readdir buffers or
1194 * that someone will want to seek to that location.
1196 * Returns: errno, >0 on exception from filldir
1199 static int do_filldir_main(struct gfs2_inode
*dip
, u64
*offset
,
1200 void *opaque
, filldir_t filldir
,
1201 const struct gfs2_dirent
**darr
, u32 entries
,
1204 const struct gfs2_dirent
*dent
, *dent_next
;
1210 sort(darr
, entries
, sizeof(struct gfs2_dirent
*), compare_dents
, NULL
);
1212 dent_next
= darr
[0];
1213 off_next
= be32_to_cpu(dent_next
->de_hash
);
1214 off_next
= gfs2_disk_hash2offset(off_next
);
1216 for (x
= 0, y
= 1; x
< entries
; x
++, y
++) {
1221 dent_next
= darr
[y
];
1222 off_next
= be32_to_cpu(dent_next
->de_hash
);
1223 off_next
= gfs2_disk_hash2offset(off_next
);
1229 if (off_next
== off
) {
1230 if (*copied
&& !run
)
1241 error
= filldir(opaque
, (const char *)(dent
+ 1),
1242 be16_to_cpu(dent
->de_name_len
),
1243 off
, be64_to_cpu(dent
->de_inum
.no_addr
),
1244 be16_to_cpu(dent
->de_type
));
1251 /* Increment the *offset by one, so the next time we come into the
1252 do_filldir fxn, we get the next entry instead of the last one in the
1260 static int gfs2_dir_read_leaf(struct inode
*inode
, u64
*offset
, void *opaque
,
1261 filldir_t filldir
, int *copied
, unsigned *depth
,
1264 struct gfs2_inode
*ip
= GFS2_I(inode
);
1265 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1266 struct buffer_head
*bh
;
1267 struct gfs2_leaf
*lf
;
1268 unsigned entries
= 0, entries2
= 0;
1269 unsigned leaves
= 0;
1270 const struct gfs2_dirent
**darr
, *dent
;
1271 struct dirent_gather g
;
1272 struct buffer_head
**larr
;
1278 error
= get_leaf(ip
, lfn
, &bh
);
1281 lf
= (struct gfs2_leaf
*)bh
->b_data
;
1283 *depth
= be16_to_cpu(lf
->lf_depth
);
1284 entries
+= be16_to_cpu(lf
->lf_entries
);
1286 lfn
= be64_to_cpu(lf
->lf_next
);
1295 * The extra 99 entries are not normally used, but are a buffer
1296 * zone in case the number of entries in the leaf is corrupt.
1297 * 99 is the maximum number of entries that can fit in a single
1300 larr
= vmalloc((leaves
+ entries
+ 99) * sizeof(void *));
1303 darr
= (const struct gfs2_dirent
**)(larr
+ leaves
);
1309 error
= get_leaf(ip
, lfn
, &bh
);
1312 lf
= (struct gfs2_leaf
*)bh
->b_data
;
1313 lfn
= be64_to_cpu(lf
->lf_next
);
1314 if (lf
->lf_entries
) {
1315 entries2
+= be16_to_cpu(lf
->lf_entries
);
1316 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
,
1317 gfs2_dirent_gather
, NULL
, &g
);
1318 error
= PTR_ERR(dent
);
1321 if (entries2
!= g
.offset
) {
1322 fs_warn(sdp
, "Number of entries corrupt in dir "
1323 "leaf %llu, entries2 (%u) != "
1325 (unsigned long long)bh
->b_blocknr
,
1326 entries2
, g
.offset
);
1338 BUG_ON(entries2
!= entries
);
1339 error
= do_filldir_main(ip
, offset
, opaque
, filldir
, darr
,
1342 for(i
= 0; i
< leaf
; i
++)
1350 * dir_e_read - Reads the entries from a directory into a filldir buffer
1351 * @dip: dinode pointer
1352 * @offset: the hash of the last entry read shifted to the right once
1353 * @opaque: buffer for the filldir function to fill
1354 * @filldir: points to the filldir function to use
1359 static int dir_e_read(struct inode
*inode
, u64
*offset
, void *opaque
,
1362 struct gfs2_inode
*dip
= GFS2_I(inode
);
1363 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1365 u32 ht_offset
, lp_offset
, ht_offset_cur
= -1;
1372 hsize
= 1 << dip
->i_di
.di_depth
;
1373 if (hsize
* sizeof(u64
) != dip
->i_di
.di_size
) {
1374 gfs2_consist_inode(dip
);
1378 hash
= gfs2_dir_offset2hash(*offset
);
1379 index
= hash
>> (32 - dip
->i_di
.di_depth
);
1381 lp
= kmalloc(sdp
->sd_hash_bsize
, GFP_KERNEL
);
1385 while (index
< hsize
) {
1386 lp_offset
= index
& (sdp
->sd_hash_ptrs
- 1);
1387 ht_offset
= index
- lp_offset
;
1389 if (ht_offset_cur
!= ht_offset
) {
1390 error
= gfs2_dir_read_data(dip
, (char *)lp
,
1391 ht_offset
* sizeof(__be64
),
1392 sdp
->sd_hash_bsize
, 1);
1393 if (error
!= sdp
->sd_hash_bsize
) {
1398 ht_offset_cur
= ht_offset
;
1401 error
= gfs2_dir_read_leaf(inode
, offset
, opaque
, filldir
,
1403 be64_to_cpu(lp
[lp_offset
]));
1407 len
= 1 << (dip
->i_di
.di_depth
- depth
);
1408 index
= (index
& ~(len
- 1)) + len
;
1418 int gfs2_dir_read(struct inode
*inode
, u64
*offset
, void *opaque
,
1421 struct gfs2_inode
*dip
= GFS2_I(inode
);
1422 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1423 struct dirent_gather g
;
1424 const struct gfs2_dirent
**darr
, *dent
;
1425 struct buffer_head
*dibh
;
1429 if (!dip
->i_di
.di_entries
)
1432 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
)
1433 return dir_e_read(inode
, offset
, opaque
, filldir
);
1435 if (!gfs2_is_stuffed(dip
)) {
1436 gfs2_consist_inode(dip
);
1440 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1445 /* 96 is max number of dirents which can be stuffed into an inode */
1446 darr
= kmalloc(96 * sizeof(struct gfs2_dirent
*), GFP_KERNEL
);
1450 dent
= gfs2_dirent_scan(inode
, dibh
->b_data
, dibh
->b_size
,
1451 gfs2_dirent_gather
, NULL
, &g
);
1453 error
= PTR_ERR(dent
);
1456 if (dip
->i_di
.di_entries
!= g
.offset
) {
1457 fs_warn(sdp
, "Number of entries corrupt in dir %llu, "
1458 "ip->i_di.di_entries (%u) != g.offset (%u)\n",
1459 (unsigned long long)dip
->i_no_addr
,
1460 dip
->i_di
.di_entries
,
1465 error
= do_filldir_main(dip
, offset
, opaque
, filldir
, darr
,
1466 dip
->i_di
.di_entries
, &copied
);
1480 * gfs2_dir_search - Search a directory
1481 * @dip: The GFS2 inode
1485 * This routine searches a directory for a file or another directory.
1486 * Assumes a glock is held on dip.
1491 struct inode
*gfs2_dir_search(struct inode
*dir
, const struct qstr
*name
)
1493 struct buffer_head
*bh
;
1494 struct gfs2_dirent
*dent
;
1495 struct inode
*inode
;
1497 dent
= gfs2_dirent_search(dir
, name
, gfs2_dirent_find
, &bh
);
1500 return ERR_PTR(PTR_ERR(dent
));
1501 inode
= gfs2_inode_lookup(dir
->i_sb
,
1502 be64_to_cpu(dent
->de_inum
.no_addr
),
1503 be16_to_cpu(dent
->de_type
));
1507 return ERR_PTR(-ENOENT
);
1510 int gfs2_dir_check(struct inode
*dir
, const struct qstr
*name
,
1511 const struct gfs2_inode
*ip
)
1513 struct buffer_head
*bh
;
1514 struct gfs2_dirent
*dent
;
1517 dent
= gfs2_dirent_search(dir
, name
, gfs2_dirent_find
, &bh
);
1520 return PTR_ERR(dent
);
1522 if (be64_to_cpu(dent
->de_inum
.no_addr
) != ip
->i_no_addr
)
1524 if (be64_to_cpu(dent
->de_inum
.no_formal_ino
) !=
1525 ip
->i_no_formal_ino
)
1527 if (unlikely(IF2DT(ip
->i_inode
.i_mode
) !=
1528 be16_to_cpu(dent
->de_type
))) {
1529 gfs2_consist_inode(GFS2_I(dir
));
1541 static int dir_new_leaf(struct inode
*inode
, const struct qstr
*name
)
1543 struct buffer_head
*bh
, *obh
;
1544 struct gfs2_inode
*ip
= GFS2_I(inode
);
1545 struct gfs2_leaf
*leaf
, *oleaf
;
1550 index
= name
->hash
>> (32 - ip
->i_di
.di_depth
);
1551 error
= get_first_leaf(ip
, index
, &obh
);
1555 oleaf
= (struct gfs2_leaf
*)obh
->b_data
;
1556 bn
= be64_to_cpu(oleaf
->lf_next
);
1560 error
= get_leaf(ip
, bn
, &obh
);
1565 gfs2_trans_add_bh(ip
->i_gl
, obh
, 1);
1567 leaf
= new_leaf(inode
, &bh
, be16_to_cpu(oleaf
->lf_depth
));
1572 oleaf
->lf_next
= cpu_to_be64(bh
->b_blocknr
);
1576 error
= gfs2_meta_inode_buffer(ip
, &bh
);
1579 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
1580 ip
->i_di
.di_blocks
++;
1581 gfs2_set_inode_blocks(&ip
->i_inode
);
1582 gfs2_dinode_out(ip
, bh
->b_data
);
1588 * gfs2_dir_add - Add new filename into directory
1589 * @dip: The GFS2 inode
1590 * @filename: The new name
1591 * @inode: The inode number of the entry
1592 * @type: The type of the entry
1594 * Returns: 0 on success, error code on failure
1597 int gfs2_dir_add(struct inode
*inode
, const struct qstr
*name
,
1598 const struct gfs2_inode
*nip
, unsigned type
)
1600 struct gfs2_inode
*ip
= GFS2_I(inode
);
1601 struct buffer_head
*bh
;
1602 struct gfs2_dirent
*dent
;
1603 struct gfs2_leaf
*leaf
;
1607 dent
= gfs2_dirent_search(inode
, name
, gfs2_dirent_find_space
,
1611 return PTR_ERR(dent
);
1612 dent
= gfs2_init_dirent(inode
, dent
, name
, bh
);
1613 gfs2_inum_out(nip
, dent
);
1614 dent
->de_type
= cpu_to_be16(type
);
1615 if (ip
->i_di
.di_flags
& GFS2_DIF_EXHASH
) {
1616 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1617 leaf
->lf_entries
= cpu_to_be16(be16_to_cpu(leaf
->lf_entries
) + 1);
1620 error
= gfs2_meta_inode_buffer(ip
, &bh
);
1623 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
1624 ip
->i_di
.di_entries
++;
1625 ip
->i_inode
.i_mtime
= ip
->i_inode
.i_ctime
= CURRENT_TIME
;
1626 gfs2_dinode_out(ip
, bh
->b_data
);
1631 if (!(ip
->i_di
.di_flags
& GFS2_DIF_EXHASH
)) {
1632 error
= dir_make_exhash(inode
);
1637 error
= dir_split_leaf(inode
, name
);
1642 if (ip
->i_di
.di_depth
< GFS2_DIR_MAX_DEPTH
) {
1643 error
= dir_double_exhash(ip
);
1646 error
= dir_split_leaf(inode
, name
);
1652 error
= dir_new_leaf(inode
, name
);
1663 * gfs2_dir_del - Delete a directory entry
1664 * @dip: The GFS2 inode
1665 * @filename: The filename
1667 * Returns: 0 on success, error code on failure
1670 int gfs2_dir_del(struct gfs2_inode
*dip
, const struct qstr
*name
)
1672 struct gfs2_dirent
*dent
, *prev
= NULL
;
1673 struct buffer_head
*bh
;
1676 /* Returns _either_ the entry (if its first in block) or the
1677 previous entry otherwise */
1678 dent
= gfs2_dirent_search(&dip
->i_inode
, name
, gfs2_dirent_prev
, &bh
);
1680 gfs2_consist_inode(dip
);
1684 gfs2_consist_inode(dip
);
1685 return PTR_ERR(dent
);
1687 /* If not first in block, adjust pointers accordingly */
1688 if (gfs2_dirent_find(dent
, name
, NULL
) == 0) {
1690 dent
= (struct gfs2_dirent
*)((char *)dent
+ be16_to_cpu(prev
->de_rec_len
));
1693 dirent_del(dip
, bh
, prev
, dent
);
1694 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
) {
1695 struct gfs2_leaf
*leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1696 u16 entries
= be16_to_cpu(leaf
->lf_entries
);
1698 gfs2_consist_inode(dip
);
1699 leaf
->lf_entries
= cpu_to_be16(--entries
);
1703 error
= gfs2_meta_inode_buffer(dip
, &bh
);
1707 if (!dip
->i_di
.di_entries
)
1708 gfs2_consist_inode(dip
);
1709 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
1710 dip
->i_di
.di_entries
--;
1711 dip
->i_inode
.i_mtime
= dip
->i_inode
.i_ctime
= CURRENT_TIME
;
1712 gfs2_dinode_out(dip
, bh
->b_data
);
1714 mark_inode_dirty(&dip
->i_inode
);
1720 * gfs2_dir_mvino - Change inode number of directory entry
1721 * @dip: The GFS2 inode
1725 * This routine changes the inode number of a directory entry. It's used
1726 * by rename to change ".." when a directory is moved.
1727 * Assumes a glock is held on dvp.
1732 int gfs2_dir_mvino(struct gfs2_inode
*dip
, const struct qstr
*filename
,
1733 const struct gfs2_inode
*nip
, unsigned int new_type
)
1735 struct buffer_head
*bh
;
1736 struct gfs2_dirent
*dent
;
1739 dent
= gfs2_dirent_search(&dip
->i_inode
, filename
, gfs2_dirent_find
, &bh
);
1741 gfs2_consist_inode(dip
);
1745 return PTR_ERR(dent
);
1747 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
1748 gfs2_inum_out(nip
, dent
);
1749 dent
->de_type
= cpu_to_be16(new_type
);
1751 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
) {
1753 error
= gfs2_meta_inode_buffer(dip
, &bh
);
1756 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
1759 dip
->i_inode
.i_mtime
= dip
->i_inode
.i_ctime
= CURRENT_TIME
;
1760 gfs2_dinode_out(dip
, bh
->b_data
);
1766 * foreach_leaf - call a function for each leaf in a directory
1767 * @dip: the directory
1768 * @lc: the function to call for each each
1769 * @data: private data to pass to it
1774 static int foreach_leaf(struct gfs2_inode
*dip
, leaf_call_t lc
, void *data
)
1776 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1777 struct buffer_head
*bh
;
1778 struct gfs2_leaf
*leaf
;
1780 u32 ht_offset
, lp_offset
, ht_offset_cur
= -1;
1786 hsize
= 1 << dip
->i_di
.di_depth
;
1787 if (hsize
* sizeof(u64
) != dip
->i_di
.di_size
) {
1788 gfs2_consist_inode(dip
);
1792 lp
= kmalloc(sdp
->sd_hash_bsize
, GFP_KERNEL
);
1796 while (index
< hsize
) {
1797 lp_offset
= index
& (sdp
->sd_hash_ptrs
- 1);
1798 ht_offset
= index
- lp_offset
;
1800 if (ht_offset_cur
!= ht_offset
) {
1801 error
= gfs2_dir_read_data(dip
, (char *)lp
,
1802 ht_offset
* sizeof(__be64
),
1803 sdp
->sd_hash_bsize
, 1);
1804 if (error
!= sdp
->sd_hash_bsize
) {
1809 ht_offset_cur
= ht_offset
;
1812 leaf_no
= be64_to_cpu(lp
[lp_offset
]);
1814 error
= get_leaf(dip
, leaf_no
, &bh
);
1817 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1818 len
= 1 << (dip
->i_di
.di_depth
- be16_to_cpu(leaf
->lf_depth
));
1821 error
= lc(dip
, index
, len
, leaf_no
, data
);
1825 index
= (index
& ~(len
- 1)) + len
;
1830 if (index
!= hsize
) {
1831 gfs2_consist_inode(dip
);
1842 * leaf_dealloc - Deallocate a directory leaf
1843 * @dip: the directory
1844 * @index: the hash table offset in the directory
1845 * @len: the number of pointers to this leaf
1846 * @leaf_no: the leaf number
1852 static int leaf_dealloc(struct gfs2_inode
*dip
, u32 index
, u32 len
,
1853 u64 leaf_no
, void *data
)
1855 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1856 struct gfs2_leaf
*tmp_leaf
;
1857 struct gfs2_rgrp_list rlist
;
1858 struct buffer_head
*bh
, *dibh
;
1860 unsigned int rg_blocks
= 0, l_blocks
= 0;
1862 unsigned int x
, size
= len
* sizeof(u64
);
1865 memset(&rlist
, 0, sizeof(struct gfs2_rgrp_list
));
1867 ht
= kzalloc(size
, GFP_KERNEL
);
1871 gfs2_alloc_get(dip
);
1873 error
= gfs2_quota_hold(dip
, NO_QUOTA_CHANGE
, NO_QUOTA_CHANGE
);
1877 error
= gfs2_rindex_hold(sdp
, &dip
->i_alloc
.al_ri_gh
);
1881 /* Count the number of leaves */
1883 for (blk
= leaf_no
; blk
; blk
= nblk
) {
1884 error
= get_leaf(dip
, blk
, &bh
);
1887 tmp_leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1888 nblk
= be64_to_cpu(tmp_leaf
->lf_next
);
1891 gfs2_rlist_add(sdp
, &rlist
, blk
);
1895 gfs2_rlist_alloc(&rlist
, LM_ST_EXCLUSIVE
, 0);
1897 for (x
= 0; x
< rlist
.rl_rgrps
; x
++) {
1898 struct gfs2_rgrpd
*rgd
;
1899 rgd
= rlist
.rl_ghs
[x
].gh_gl
->gl_object
;
1900 rg_blocks
+= rgd
->rd_length
;
1903 error
= gfs2_glock_nq_m(rlist
.rl_rgrps
, rlist
.rl_ghs
);
1907 error
= gfs2_trans_begin(sdp
,
1908 rg_blocks
+ (DIV_ROUND_UP(size
, sdp
->sd_jbsize
) + 1) +
1909 RES_DINODE
+ RES_STATFS
+ RES_QUOTA
, l_blocks
);
1911 goto out_rg_gunlock
;
1913 for (blk
= leaf_no
; blk
; blk
= nblk
) {
1914 error
= get_leaf(dip
, blk
, &bh
);
1917 tmp_leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1918 nblk
= be64_to_cpu(tmp_leaf
->lf_next
);
1921 gfs2_free_meta(dip
, blk
, 1);
1923 if (!dip
->i_di
.di_blocks
)
1924 gfs2_consist_inode(dip
);
1925 dip
->i_di
.di_blocks
--;
1926 gfs2_set_inode_blocks(&dip
->i_inode
);
1929 error
= gfs2_dir_write_data(dip
, ht
, index
* sizeof(u64
), size
);
1930 if (error
!= size
) {
1936 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1940 gfs2_trans_add_bh(dip
->i_gl
, dibh
, 1);
1941 gfs2_dinode_out(dip
, dibh
->b_data
);
1945 gfs2_trans_end(sdp
);
1947 gfs2_glock_dq_m(rlist
.rl_rgrps
, rlist
.rl_ghs
);
1949 gfs2_rlist_free(&rlist
);
1950 gfs2_glock_dq_uninit(&dip
->i_alloc
.al_ri_gh
);
1952 gfs2_quota_unhold(dip
);
1954 gfs2_alloc_put(dip
);
1960 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
1961 * @dip: the directory
1963 * Dealloc all on-disk directory leaves to FREEMETA state
1964 * Change on-disk inode type to "regular file"
1969 int gfs2_dir_exhash_dealloc(struct gfs2_inode
*dip
)
1971 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1972 struct buffer_head
*bh
;
1975 /* Dealloc on-disk leaves to FREEMETA state */
1976 error
= foreach_leaf(dip
, leaf_dealloc
, NULL
);
1980 /* Make this a regular file in case we crash.
1981 (We don't want to free these blocks a second time.) */
1983 error
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
1987 error
= gfs2_meta_inode_buffer(dip
, &bh
);
1989 gfs2_trans_add_bh(dip
->i_gl
, bh
, 1);
1990 ((struct gfs2_dinode
*)bh
->b_data
)->di_mode
=
1991 cpu_to_be32(S_IFREG
);
1995 gfs2_trans_end(sdp
);
2001 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2002 * @ip: the file being written to
2003 * @filname: the filename that's going to be added
2005 * Returns: 1 if alloc required, 0 if not, -ve on error
2008 int gfs2_diradd_alloc_required(struct inode
*inode
, const struct qstr
*name
)
2010 struct gfs2_dirent
*dent
;
2011 struct buffer_head
*bh
;
2013 dent
= gfs2_dirent_search(inode
, name
, gfs2_dirent_find_space
, &bh
);
2018 return PTR_ERR(dent
);