2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 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 v.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 block
46 * pointer in the array that points to the same leaf. In fact, when a directory
47 * is first converted from linear to exhash, all of the pointers point to the
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/completion.h>
60 #include <linux/buffer_head.h>
61 #include <linux/sort.h>
62 #include <asm/semaphore.h>
74 #define IS_LEAF 1 /* Hashed (leaf) directory */
75 #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */
78 #define gfs2_disk_hash2offset(h) (((uint64_t)(h)) >> 1)
79 #define gfs2_dir_offset2hash(p) ((uint32_t)(((uint64_t)(p)) << 1))
81 #define gfs2_disk_hash2offset(h) (((uint64_t)(h)))
82 #define gfs2_dir_offset2hash(p) ((uint32_t)(((uint64_t)(p))))
85 typedef int (*leaf_call_t
) (struct gfs2_inode
*dip
,
86 uint32_t index
, uint32_t len
, uint64_t leaf_no
,
90 * int gfs2_filecmp - Compare two filenames
91 * @file1: The first filename
92 * @file2: The second filename
93 * @len_of_file2: The length of the second file
95 * This routine compares two filenames and returns 1 if they are equal.
97 * Returns: 1 if the files are the same, otherwise 0.
100 int gfs2_filecmp(struct qstr
*file1
, char *file2
, int len_of_file2
)
102 if (file1
->len
!= len_of_file2
)
104 if (memcmp(file1
->name
, file2
, file1
->len
))
110 * dirent_first - Return the first dirent
111 * @dip: the directory
113 * @dent: Pointer to list of dirents
115 * return first dirent whether bh points to leaf or stuffed dinode
117 * Returns: IS_LEAF, IS_DINODE, or -errno
120 static int dirent_first(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
121 struct gfs2_dirent
**dent
)
123 struct gfs2_meta_header
*h
= (struct gfs2_meta_header
*)bh
->b_data
;
125 if (be16_to_cpu(h
->mh_type
) == GFS2_METATYPE_LF
) {
126 if (gfs2_meta_check(dip
->i_sbd
, bh
))
128 *dent
= (struct gfs2_dirent
*)(bh
->b_data
+
129 sizeof(struct gfs2_leaf
));
132 if (gfs2_metatype_check(dip
->i_sbd
, bh
, GFS2_METATYPE_DI
))
134 *dent
= (struct gfs2_dirent
*)(bh
->b_data
+
135 sizeof(struct gfs2_dinode
));
141 * dirent_next - Next dirent
142 * @dip: the directory
144 * @dent: Pointer to list of dirents
146 * Returns: 0 on success, error code otherwise
149 static int dirent_next(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
150 struct gfs2_dirent
**dent
)
152 struct gfs2_dirent
*tmp
, *cur
;
154 uint32_t cur_rec_len
;
157 bh_end
= bh
->b_data
+ bh
->b_size
;
158 cur_rec_len
= be32_to_cpu(cur
->de_rec_len
);
160 if ((char *)cur
+ cur_rec_len
>= bh_end
) {
161 if ((char *)cur
+ cur_rec_len
> bh_end
) {
162 gfs2_consist_inode(dip
);
168 tmp
= (struct gfs2_dirent
*)((char *)cur
+ cur_rec_len
);
170 if ((char *)tmp
+ be32_to_cpu(tmp
->de_rec_len
) > bh_end
) {
171 gfs2_consist_inode(dip
);
174 /* Only the first dent could ever have de_inum.no_addr == 0 */
175 if (!tmp
->de_inum
.no_addr
) {
176 gfs2_consist_inode(dip
);
186 * dirent_del - Delete a dirent
187 * @dip: The GFS2 inode
189 * @prev: The previous dirent
190 * @cur: The current dirent
194 static void dirent_del(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
195 struct gfs2_dirent
*prev
, struct gfs2_dirent
*cur
)
197 uint32_t cur_rec_len
, prev_rec_len
;
199 if (!cur
->de_inum
.no_addr
) {
200 gfs2_consist_inode(dip
);
204 gfs2_trans_add_bh(dip
->i_gl
, bh
);
206 /* If there is no prev entry, this is the first entry in the block.
207 The de_rec_len is already as big as it needs to be. Just zero
208 out the inode number and return. */
211 cur
->de_inum
.no_addr
= 0; /* No endianess worries */
215 /* Combine this dentry with the previous one. */
217 prev_rec_len
= be32_to_cpu(prev
->de_rec_len
);
218 cur_rec_len
= be32_to_cpu(cur
->de_rec_len
);
220 if ((char *)prev
+ prev_rec_len
!= (char *)cur
)
221 gfs2_consist_inode(dip
);
222 if ((char *)cur
+ cur_rec_len
> bh
->b_data
+ bh
->b_size
)
223 gfs2_consist_inode(dip
);
225 prev_rec_len
+= cur_rec_len
;
226 prev
->de_rec_len
= cpu_to_be32(prev_rec_len
);
230 * gfs2_dirent_alloc - Allocate a directory entry
231 * @dip: The GFS2 inode
233 * @name_len: The length of the name
234 * @dent_out: Pointer to list of dirents
236 * Returns: 0 on success, error code otherwise
239 int gfs2_dirent_alloc(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
240 int name_len
, struct gfs2_dirent
**dent_out
)
242 struct gfs2_dirent
*dent
, *new;
243 unsigned int rec_len
= GFS2_DIRENT_SIZE(name_len
);
244 unsigned int entries
= 0, offset
= 0;
247 type
= dirent_first(dip
, bh
, &dent
);
251 if (type
== IS_LEAF
) {
252 struct gfs2_leaf
*leaf
= (struct gfs2_leaf
*)bh
->b_data
;
253 entries
= be16_to_cpu(leaf
->lf_entries
);
254 offset
= sizeof(struct gfs2_leaf
);
256 struct gfs2_dinode
*dinode
= (struct gfs2_dinode
*)bh
->b_data
;
257 entries
= be32_to_cpu(dinode
->di_entries
);
258 offset
= sizeof(struct gfs2_dinode
);
262 if (dent
->de_inum
.no_addr
) {
263 gfs2_consist_inode(dip
);
267 gfs2_trans_add_bh(dip
->i_gl
, bh
);
269 dent
->de_rec_len
= bh
->b_size
- offset
;
270 dent
->de_rec_len
= cpu_to_be32(dent
->de_rec_len
);
271 dent
->de_name_len
= name_len
;
278 uint32_t cur_rec_len
, cur_name_len
;
280 cur_rec_len
= be32_to_cpu(dent
->de_rec_len
);
281 cur_name_len
= dent
->de_name_len
;
283 if ((!dent
->de_inum
.no_addr
&& cur_rec_len
>= rec_len
) ||
284 (cur_rec_len
>= GFS2_DIRENT_SIZE(cur_name_len
) + rec_len
)) {
285 gfs2_trans_add_bh(dip
->i_gl
, bh
);
287 if (dent
->de_inum
.no_addr
) {
288 new = (struct gfs2_dirent
*)((char *)dent
+
289 GFS2_DIRENT_SIZE(cur_name_len
));
290 memset(new, 0, sizeof(struct gfs2_dirent
));
292 new->de_rec_len
= cur_rec_len
- GFS2_DIRENT_SIZE(cur_name_len
);
293 new->de_rec_len
= cpu_to_be32(new->de_rec_len
);
294 new->de_name_len
= name_len
;
296 dent
->de_rec_len
= cur_rec_len
- be32_to_cpu(new->de_rec_len
);
297 dent
->de_rec_len
= cpu_to_be32(dent
->de_rec_len
);
303 dent
->de_name_len
= name_len
;
308 } while (dirent_next(dip
, bh
, &dent
) == 0);
314 * dirent_fits - See if we can fit a entry in this buffer
315 * @dip: The GFS2 inode
317 * @name_len: The length of the name
319 * Returns: 1 if it can fit, 0 otherwise
322 static int dirent_fits(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
325 struct gfs2_dirent
*dent
;
326 unsigned int rec_len
= GFS2_DIRENT_SIZE(name_len
);
327 unsigned int entries
= 0;
330 type
= dirent_first(dip
, bh
, &dent
);
334 if (type
== IS_LEAF
) {
335 struct gfs2_leaf
*leaf
= (struct gfs2_leaf
*)bh
->b_data
;
336 entries
= be16_to_cpu(leaf
->lf_entries
);
338 struct gfs2_dinode
*dinode
= (struct gfs2_dinode
*)bh
->b_data
;
339 entries
= be32_to_cpu(dinode
->di_entries
);
346 uint32_t cur_rec_len
, cur_name_len
;
348 cur_rec_len
= be32_to_cpu(dent
->de_rec_len
);
349 cur_name_len
= dent
->de_name_len
;
351 if ((!dent
->de_inum
.no_addr
&& cur_rec_len
>= rec_len
) ||
352 (cur_rec_len
>= GFS2_DIRENT_SIZE(cur_name_len
) + rec_len
))
354 } while (dirent_next(dip
, bh
, &dent
) == 0);
359 static int leaf_search(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
360 struct qstr
*filename
, struct gfs2_dirent
**dent_out
,
361 struct gfs2_dirent
**dent_prev
)
364 struct gfs2_dirent
*dent
, *prev
= NULL
;
365 unsigned int entries
= 0;
368 type
= dirent_first(dip
, bh
, &dent
);
372 if (type
== IS_LEAF
) {
373 struct gfs2_leaf
*leaf
= (struct gfs2_leaf
*)bh
->b_data
;
374 entries
= be16_to_cpu(leaf
->lf_entries
);
375 } else if (type
== IS_DINODE
) {
376 struct gfs2_dinode
*dinode
= (struct gfs2_dinode
*)bh
->b_data
;
377 entries
= be32_to_cpu(dinode
->di_entries
);
380 hash
= gfs2_disk_hash(filename
->name
, filename
->len
);
383 if (!dent
->de_inum
.no_addr
) {
388 if (be32_to_cpu(dent
->de_hash
) == hash
&&
389 gfs2_filecmp(filename
, (char *)(dent
+ 1),
390 dent
->de_name_len
)) {
399 } while (dirent_next(dip
, bh
, &dent
) == 0);
404 static int get_leaf(struct gfs2_inode
*dip
, uint64_t leaf_no
,
405 struct buffer_head
**bhp
)
409 error
= gfs2_meta_read(dip
->i_gl
, leaf_no
, DIO_START
| DIO_WAIT
, bhp
);
410 if (!error
&& gfs2_metatype_check(dip
->i_sbd
, *bhp
, GFS2_METATYPE_LF
))
417 * get_leaf_nr - Get a leaf number associated with the index
418 * @dip: The GFS2 inode
422 * Returns: 0 on success, error code otherwise
425 static int get_leaf_nr(struct gfs2_inode
*dip
, uint32_t index
,
431 error
= gfs2_jdata_read_mem(dip
, (char *)&leaf_no
,
432 index
* sizeof(uint64_t),
434 if (error
!= sizeof(uint64_t))
435 return (error
< 0) ? error
: -EIO
;
437 *leaf_out
= be64_to_cpu(leaf_no
);
442 static int get_first_leaf(struct gfs2_inode
*dip
, uint32_t index
,
443 struct buffer_head
**bh_out
)
448 error
= get_leaf_nr(dip
, index
, &leaf_no
);
450 error
= get_leaf(dip
, leaf_no
, bh_out
);
455 static int get_next_leaf(struct gfs2_inode
*dip
, struct buffer_head
*bh_in
,
456 struct buffer_head
**bh_out
)
458 struct gfs2_leaf
*leaf
;
461 leaf
= (struct gfs2_leaf
*)bh_in
->b_data
;
466 error
= get_leaf(dip
, be64_to_cpu(leaf
->lf_next
), bh_out
);
471 static int linked_leaf_search(struct gfs2_inode
*dip
, struct qstr
*filename
,
472 struct gfs2_dirent
**dent_out
,
473 struct gfs2_dirent
**dent_prev
,
474 struct buffer_head
**bh_out
)
476 struct buffer_head
*bh
= NULL
, *bh_next
;
477 uint32_t hsize
, index
;
481 hsize
= 1 << dip
->i_di
.di_depth
;
482 if (hsize
* sizeof(uint64_t) != dip
->i_di
.di_size
) {
483 gfs2_consist_inode(dip
);
487 /* Figure out the address of the leaf node. */
489 hash
= gfs2_disk_hash(filename
->name
, filename
->len
);
490 index
= hash
>> (32 - dip
->i_di
.di_depth
);
492 error
= get_first_leaf(dip
, index
, &bh_next
);
503 error
= leaf_search(dip
, bh
, filename
, dent_out
, dent_prev
);
517 error
= get_next_leaf(dip
, bh
, &bh_next
);
527 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
528 * @dip: The GFS2 inode
530 * Returns: 0 on success, error code otherwise
533 static int dir_make_exhash(struct gfs2_inode
*dip
)
535 struct gfs2_sbd
*sdp
= dip
->i_sbd
;
536 struct gfs2_dirent
*dent
;
537 struct buffer_head
*bh
, *dibh
;
538 struct gfs2_leaf
*leaf
;
544 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
548 /* Allocate a new block for the first leaf node */
550 bn
= gfs2_alloc_meta(dip
);
552 /* Turn over a new leaf */
554 bh
= gfs2_meta_new(dip
->i_gl
, bn
);
555 gfs2_trans_add_bh(dip
->i_gl
, bh
);
556 gfs2_metatype_set(bh
, GFS2_METATYPE_LF
, GFS2_FORMAT_LF
);
557 gfs2_buffer_clear_tail(bh
, sizeof(struct gfs2_meta_header
));
559 /* Fill in the leaf structure */
561 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
563 gfs2_assert(sdp
, dip
->i_di
.di_entries
< (1 << 16));
565 leaf
->lf_dirent_format
= cpu_to_be32(GFS2_FORMAT_DE
);
566 leaf
->lf_entries
= cpu_to_be16(dip
->i_di
.di_entries
);
570 gfs2_buffer_copy_tail(bh
, sizeof(struct gfs2_leaf
), dibh
,
571 sizeof(struct gfs2_dinode
));
573 /* Find last entry */
576 dirent_first(dip
, bh
, &dent
);
579 if (!dent
->de_inum
.no_addr
)
581 if (++x
== dip
->i_di
.di_entries
)
584 while (dirent_next(dip
, bh
, &dent
) == 0);
586 /* Adjust the last dirent's record length
587 (Remember that dent still points to the last entry.) */
589 dent
->de_rec_len
= be32_to_cpu(dent
->de_rec_len
) +
590 sizeof(struct gfs2_dinode
) -
591 sizeof(struct gfs2_leaf
);
592 dent
->de_rec_len
= cpu_to_be32(dent
->de_rec_len
);
596 /* We're done with the new leaf block, now setup the new
599 gfs2_trans_add_bh(dip
->i_gl
, dibh
);
600 gfs2_buffer_clear_tail(dibh
, sizeof(struct gfs2_dinode
));
602 lp
= (uint64_t *)(dibh
->b_data
+ sizeof(struct gfs2_dinode
));
604 for (x
= sdp
->sd_hash_ptrs
; x
--; lp
++)
605 *lp
= cpu_to_be64(bn
);
607 dip
->i_di
.di_size
= sdp
->sd_sb
.sb_bsize
/ 2;
608 dip
->i_di
.di_blocks
++;
609 dip
->i_di
.di_flags
|= GFS2_DIF_EXHASH
;
610 dip
->i_di
.di_payload_format
= 0;
612 for (x
= sdp
->sd_hash_ptrs
, y
= -1; x
; x
>>= 1, y
++) ;
613 dip
->i_di
.di_depth
= y
;
615 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
623 * dir_split_leaf - Split a leaf block into two
624 * @dip: The GFS2 inode
628 * Returns: 0 on success, error code on failure
631 static int dir_split_leaf(struct gfs2_inode
*dip
, uint32_t index
,
634 struct buffer_head
*nbh
, *obh
, *dibh
;
635 struct gfs2_leaf
*nleaf
, *oleaf
;
636 struct gfs2_dirent
*dent
, *prev
= NULL
, *next
= NULL
, *new;
637 uint32_t start
, len
, half_len
, divider
;
643 /* Allocate the new leaf block */
645 bn
= gfs2_alloc_meta(dip
);
647 /* Get the new leaf block */
649 nbh
= gfs2_meta_new(dip
->i_gl
, bn
);
650 gfs2_trans_add_bh(dip
->i_gl
, nbh
);
651 gfs2_metatype_set(nbh
, GFS2_METATYPE_LF
, GFS2_FORMAT_LF
);
652 gfs2_buffer_clear_tail(nbh
, sizeof(struct gfs2_meta_header
));
654 nleaf
= (struct gfs2_leaf
*)nbh
->b_data
;
656 nleaf
->lf_dirent_format
= cpu_to_be32(GFS2_FORMAT_DE
);
658 /* Get the old leaf block */
660 error
= get_leaf(dip
, leaf_no
, &obh
);
664 gfs2_trans_add_bh(dip
->i_gl
, obh
);
666 oleaf
= (struct gfs2_leaf
*)obh
->b_data
;
668 /* Compute the start and len of leaf pointers in the hash table. */
670 len
= 1 << (dip
->i_di
.di_depth
- be16_to_cpu(oleaf
->lf_depth
));
673 gfs2_consist_inode(dip
);
678 start
= (index
& ~(len
- 1));
680 /* Change the pointers.
681 Don't bother distinguishing stuffed from non-stuffed.
682 This code is complicated enough already. */
684 lp
= kcalloc(half_len
, sizeof(uint64_t), GFP_KERNEL
| __GFP_NOFAIL
);
686 error
= gfs2_jdata_read_mem(dip
, (char *)lp
, start
* sizeof(uint64_t),
687 half_len
* sizeof(uint64_t));
688 if (error
!= half_len
* sizeof(uint64_t)) {
694 /* Change the pointers */
696 for (x
= 0; x
< half_len
; x
++)
697 lp
[x
] = cpu_to_be64(bn
);
699 error
= gfs2_jdata_write_mem(dip
, (char *)lp
, start
* sizeof(uint64_t),
700 half_len
* sizeof(uint64_t));
701 if (error
!= half_len
* sizeof(uint64_t)) {
709 /* Compute the divider */
711 divider
= (start
+ half_len
) << (32 - dip
->i_di
.di_depth
);
713 /* Copy the entries */
715 dirent_first(dip
, obh
, &dent
);
719 if (dirent_next(dip
, obh
, &next
))
722 if (dent
->de_inum
.no_addr
&&
723 be32_to_cpu(dent
->de_hash
) < divider
) {
724 name_len
= dent
->de_name_len
;
726 gfs2_dirent_alloc(dip
, nbh
, name_len
, &new);
728 new->de_inum
= dent
->de_inum
; /* No endian worries */
729 new->de_hash
= dent
->de_hash
; /* No endian worries */
730 new->de_type
= dent
->de_type
; /* No endian worries */
731 memcpy((char *)(new + 1), (char *)(dent
+ 1),
734 nleaf
->lf_entries
= be16_to_cpu(nleaf
->lf_entries
)+1;
735 nleaf
->lf_entries
= cpu_to_be16(nleaf
->lf_entries
);
737 dirent_del(dip
, obh
, prev
, dent
);
739 if (!oleaf
->lf_entries
)
740 gfs2_consist_inode(dip
);
741 oleaf
->lf_entries
= be16_to_cpu(oleaf
->lf_entries
)-1;
742 oleaf
->lf_entries
= cpu_to_be16(oleaf
->lf_entries
);
755 /* If none of the entries got moved into the new leaf,
756 artificially fill in the first entry. */
759 gfs2_dirent_alloc(dip
, nbh
, 0, &new);
760 new->de_inum
.no_addr
= 0;
763 oleaf
->lf_depth
= be16_to_cpu(oleaf
->lf_depth
) + 1;
764 oleaf
->lf_depth
= cpu_to_be16(oleaf
->lf_depth
);
765 nleaf
->lf_depth
= oleaf
->lf_depth
;
767 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
768 if (!gfs2_assert_withdraw(dip
->i_sbd
, !error
)) {
769 dip
->i_di
.di_blocks
++;
770 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
791 * dir_double_exhash - Double size of ExHash table
792 * @dip: The GFS2 dinode
794 * Returns: 0 on success, error code on failure
797 static int dir_double_exhash(struct gfs2_inode
*dip
)
799 struct gfs2_sbd
*sdp
= dip
->i_sbd
;
800 struct buffer_head
*dibh
;
808 hsize
= 1 << dip
->i_di
.di_depth
;
809 if (hsize
* sizeof(uint64_t) != dip
->i_di
.di_size
) {
810 gfs2_consist_inode(dip
);
814 /* Allocate both the "from" and "to" buffers in one big chunk */
816 buf
= kcalloc(3, sdp
->sd_hash_bsize
, GFP_KERNEL
| __GFP_NOFAIL
);
818 for (block
= dip
->i_di
.di_size
>> sdp
->sd_hash_bsize_shift
; block
--;) {
819 error
= gfs2_jdata_read_mem(dip
, (char *)buf
,
820 block
* sdp
->sd_hash_bsize
,
822 if (error
!= sdp
->sd_hash_bsize
) {
829 to
= (uint64_t *)((char *)buf
+ sdp
->sd_hash_bsize
);
831 for (x
= sdp
->sd_hash_ptrs
; x
--; from
++) {
832 *to
++ = *from
; /* No endianess worries */
836 error
= gfs2_jdata_write_mem(dip
,
837 (char *)buf
+ sdp
->sd_hash_bsize
,
838 block
* sdp
->sd_sb
.sb_bsize
,
839 sdp
->sd_sb
.sb_bsize
);
840 if (error
!= sdp
->sd_sb
.sb_bsize
) {
849 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
850 if (!gfs2_assert_withdraw(sdp
, !error
)) {
851 dip
->i_di
.di_depth
++;
852 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
865 * compare_dents - compare directory entries by hash value
869 * When comparing the hash entries of @a to @b:
875 static int compare_dents(const void *a
, const void *b
)
877 struct gfs2_dirent
*dent_a
, *dent_b
;
878 uint32_t hash_a
, hash_b
;
881 dent_a
= *(struct gfs2_dirent
**)a
;
882 hash_a
= dent_a
->de_hash
;
883 hash_a
= be32_to_cpu(hash_a
);
885 dent_b
= *(struct gfs2_dirent
**)b
;
886 hash_b
= dent_b
->de_hash
;
887 hash_b
= be32_to_cpu(hash_b
);
891 else if (hash_a
< hash_b
)
894 unsigned int len_a
= dent_a
->de_name_len
;
895 unsigned int len_b
= dent_b
->de_name_len
;
899 else if (len_a
< len_b
)
902 ret
= memcmp((char *)(dent_a
+ 1),
903 (char *)(dent_b
+ 1),
911 * do_filldir_main - read out directory entries
912 * @dip: The GFS2 inode
913 * @offset: The offset in the file to read from
914 * @opaque: opaque data to pass to filldir
915 * @filldir: The function to pass entries to
916 * @darr: an array of struct gfs2_dirent pointers to read
917 * @entries: the number of entries in darr
918 * @copied: pointer to int that's non-zero if a entry has been copied out
920 * Jump through some hoops to make sure that if there are hash collsions,
921 * they are read out at the beginning of a buffer. We want to minimize
922 * the possibility that they will fall into different readdir buffers or
923 * that someone will want to seek to that location.
925 * Returns: errno, >0 on exception from filldir
928 static int do_filldir_main(struct gfs2_inode
*dip
, uint64_t *offset
,
929 void *opaque
, gfs2_filldir_t filldir
,
930 struct gfs2_dirent
**darr
, uint32_t entries
,
933 struct gfs2_dirent
*dent
, *dent_next
;
934 struct gfs2_inum inum
;
935 uint64_t off
, off_next
;
940 sort(darr
, entries
, sizeof(struct gfs2_dirent
*), compare_dents
, NULL
);
943 off_next
= be32_to_cpu(dent_next
->de_hash
);
944 off_next
= gfs2_disk_hash2offset(off_next
);
946 for (x
= 0, y
= 1; x
< entries
; x
++, y
++) {
952 off_next
= be32_to_cpu(dent_next
->de_hash
);
953 off_next
= gfs2_disk_hash2offset(off_next
);
959 if (off_next
== off
) {
971 gfs2_inum_in(&inum
, (char *)&dent
->de_inum
);
973 error
= filldir(opaque
, (char *)(dent
+ 1),
983 /* Increment the *offset by one, so the next time we come into the
984 do_filldir fxn, we get the next entry instead of the last one in the
993 * do_filldir_single - Read directory entries out of a single block
994 * @dip: The GFS2 inode
995 * @offset: The offset in the file to read from
996 * @opaque: opaque data to pass to filldir
997 * @filldir: The function to pass entries to
999 * @entries: the number of entries in the block
1000 * @copied: pointer to int that's non-zero if a entry has been copied out
1002 * Returns: errno, >0 on exception from filldir
1005 static int do_filldir_single(struct gfs2_inode
*dip
, uint64_t *offset
,
1006 void *opaque
, gfs2_filldir_t filldir
,
1007 struct buffer_head
*bh
, uint32_t entries
,
1010 struct gfs2_dirent
**darr
;
1011 struct gfs2_dirent
*de
;
1018 darr
= kcalloc(entries
, sizeof(struct gfs2_dirent
*), GFP_KERNEL
);
1022 dirent_first(dip
, bh
, &de
);
1024 if (!de
->de_inum
.no_addr
)
1027 gfs2_consist_inode(dip
);
1033 while (dirent_next(dip
, bh
, &de
) == 0);
1036 gfs2_consist_inode(dip
);
1041 error
= do_filldir_main(dip
, offset
, opaque
, filldir
, darr
,
1051 * do_filldir_multi - Read directory entries out of a linked leaf list
1052 * @dip: The GFS2 inode
1053 * @offset: The offset in the file to read from
1054 * @opaque: opaque data to pass to filldir
1055 * @filldir: The function to pass entries to
1056 * @bh: the first leaf in the list
1057 * @copied: pointer to int that's non-zero if a entry has been copied out
1059 * Returns: errno, >0 on exception from filldir
1062 static int do_filldir_multi(struct gfs2_inode
*dip
, uint64_t *offset
,
1063 void *opaque
, gfs2_filldir_t filldir
,
1064 struct buffer_head
*bh
, int *copied
)
1066 struct buffer_head
**larr
= NULL
;
1067 struct gfs2_dirent
**darr
;
1068 struct gfs2_leaf
*leaf
;
1069 struct buffer_head
*tmp_bh
;
1070 struct gfs2_dirent
*de
;
1071 unsigned int entries
, e
= 0;
1072 unsigned int leaves
= 0, l
= 0;
1077 /* Count leaves and entries */
1079 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1080 entries
= be16_to_cpu(leaf
->lf_entries
);
1084 ln
= be64_to_cpu(ln
);
1086 error
= get_leaf(dip
, ln
, &tmp_bh
);
1090 leaf
= (struct gfs2_leaf
*)tmp_bh
->b_data
;
1091 if (leaf
->lf_entries
) {
1092 entries
+= be16_to_cpu(leaf
->lf_entries
);
1104 larr
= kcalloc(leaves
, sizeof(struct buffer_head
*),GFP_KERNEL
);
1109 darr
= kcalloc(entries
, sizeof(struct gfs2_dirent
*), GFP_KERNEL
);
1115 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1116 if (leaf
->lf_entries
) {
1117 dirent_first(dip
, bh
, &de
);
1119 if (!de
->de_inum
.no_addr
)
1122 gfs2_consist_inode(dip
);
1128 while (dirent_next(dip
, bh
, &de
) == 0);
1133 ln
= be64_to_cpu(ln
);
1135 error
= get_leaf(dip
, ln
, &tmp_bh
);
1139 leaf
= (struct gfs2_leaf
*)tmp_bh
->b_data
;
1140 if (leaf
->lf_entries
) {
1141 dirent_first(dip
, tmp_bh
, &de
);
1143 if (!de
->de_inum
.no_addr
)
1146 gfs2_consist_inode(dip
);
1152 while (dirent_next(dip
, tmp_bh
, &de
) == 0);
1163 if (gfs2_assert_withdraw(dip
->i_sbd
, l
== leaves
)) {
1168 gfs2_consist_inode(dip
);
1173 error
= do_filldir_main(dip
, offset
, opaque
, filldir
, darr
,
1178 for (x
= 0; x
< l
; x
++)
1186 * dir_e_search - Search exhash (leaf) dir for inode matching name
1187 * @dip: The GFS2 inode
1188 * @filename: Filename string
1189 * @inode: If non-NULL, function fills with formal inode # and block address
1190 * @type: If non-NULL, function fills with DT_... dinode type
1195 static int dir_e_search(struct gfs2_inode
*dip
, struct qstr
*filename
,
1196 struct gfs2_inum
*inum
, unsigned int *type
)
1198 struct buffer_head
*bh
;
1199 struct gfs2_dirent
*dent
;
1202 error
= linked_leaf_search(dip
, filename
, &dent
, NULL
, &bh
);
1207 gfs2_inum_in(inum
, (char *)&dent
->de_inum
);
1209 *type
= dent
->de_type
;
1216 static int dir_e_add(struct gfs2_inode
*dip
, struct qstr
*filename
,
1217 struct gfs2_inum
*inum
, unsigned int type
)
1219 struct buffer_head
*bh
, *nbh
, *dibh
;
1220 struct gfs2_leaf
*leaf
, *nleaf
;
1221 struct gfs2_dirent
*dent
;
1222 uint32_t hsize
, index
;
1224 uint64_t leaf_no
, bn
;
1228 hsize
= 1 << dip
->i_di
.di_depth
;
1229 if (hsize
* sizeof(uint64_t) != dip
->i_di
.di_size
) {
1230 gfs2_consist_inode(dip
);
1234 /* Figure out the address of the leaf node. */
1236 hash
= gfs2_disk_hash(filename
->name
, filename
->len
);
1237 index
= hash
>> (32 - dip
->i_di
.di_depth
);
1239 error
= get_leaf_nr(dip
, index
, &leaf_no
);
1243 /* Add entry to the leaf */
1246 error
= get_leaf(dip
, leaf_no
, &bh
);
1250 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1252 if (gfs2_dirent_alloc(dip
, bh
, filename
->len
, &dent
)) {
1254 if (be16_to_cpu(leaf
->lf_depth
) < dip
->i_di
.di_depth
) {
1255 /* Can we split the leaf? */
1259 error
= dir_split_leaf(dip
, index
, leaf_no
);
1265 } else if (dip
->i_di
.di_depth
< GFS2_DIR_MAX_DEPTH
) {
1266 /* Can we double the hash table? */
1270 error
= dir_double_exhash(dip
);
1276 } else if (leaf
->lf_next
) {
1277 /* Can we try the next leaf in the list? */
1278 leaf_no
= be64_to_cpu(leaf
->lf_next
);
1283 /* Create a new leaf and add it to the list. */
1285 bn
= gfs2_alloc_meta(dip
);
1287 nbh
= gfs2_meta_new(dip
->i_gl
, bn
);
1288 gfs2_trans_add_bh(dip
->i_gl
, nbh
);
1289 gfs2_metatype_set(nbh
,
1292 gfs2_buffer_clear_tail(nbh
,
1293 sizeof(struct gfs2_meta_header
));
1295 gfs2_trans_add_bh(dip
->i_gl
, bh
);
1296 leaf
->lf_next
= cpu_to_be64(bn
);
1298 nleaf
= (struct gfs2_leaf
*)nbh
->b_data
;
1299 nleaf
->lf_depth
= leaf
->lf_depth
;
1300 nleaf
->lf_dirent_format
= cpu_to_be32(GFS2_FORMAT_DE
);
1302 gfs2_dirent_alloc(dip
, nbh
, filename
->len
,
1305 dip
->i_di
.di_blocks
++;
1314 /* If the gfs2_dirent_alloc() succeeded, it pinned the "bh" */
1316 gfs2_inum_out(inum
, (char *)&dent
->de_inum
);
1317 dent
->de_hash
= cpu_to_be32(hash
);
1318 dent
->de_type
= type
;
1319 memcpy((char *)(dent
+ 1), filename
->name
, filename
->len
);
1321 leaf
->lf_entries
= be16_to_cpu(leaf
->lf_entries
) + 1;
1322 leaf
->lf_entries
= cpu_to_be16(leaf
->lf_entries
);
1326 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1330 dip
->i_di
.di_entries
++;
1331 dip
->i_di
.di_mtime
= dip
->i_di
.di_ctime
= get_seconds();
1333 gfs2_trans_add_bh(dip
->i_gl
, dibh
);
1334 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
1343 static int dir_e_del(struct gfs2_inode
*dip
, struct qstr
*filename
)
1345 struct buffer_head
*bh
, *dibh
;
1346 struct gfs2_dirent
*dent
, *prev
;
1347 struct gfs2_leaf
*leaf
;
1348 unsigned int entries
;
1351 error
= linked_leaf_search(dip
, filename
, &dent
, &prev
, &bh
);
1352 if (error
== -ENOENT
) {
1353 gfs2_consist_inode(dip
);
1359 dirent_del(dip
, bh
, prev
, dent
); /* Pins bh */
1361 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1362 entries
= be16_to_cpu(leaf
->lf_entries
);
1364 gfs2_consist_inode(dip
);
1366 leaf
->lf_entries
= cpu_to_be16(entries
);
1370 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1374 if (!dip
->i_di
.di_entries
)
1375 gfs2_consist_inode(dip
);
1376 dip
->i_di
.di_entries
--;
1377 dip
->i_di
.di_mtime
= dip
->i_di
.di_ctime
= get_seconds();
1379 gfs2_trans_add_bh(dip
->i_gl
, dibh
);
1380 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
1387 * dir_e_read - Reads the entries from a directory into a filldir buffer
1388 * @dip: dinode pointer
1389 * @offset: the hash of the last entry read shifted to the right once
1390 * @opaque: buffer for the filldir function to fill
1391 * @filldir: points to the filldir function to use
1396 static int dir_e_read(struct gfs2_inode
*dip
, uint64_t *offset
, void *opaque
,
1397 gfs2_filldir_t filldir
)
1399 struct gfs2_sbd
*sdp
= dip
->i_sbd
;
1400 struct buffer_head
*bh
;
1401 struct gfs2_leaf leaf
;
1402 uint32_t hsize
, len
;
1403 uint32_t ht_offset
, lp_offset
, ht_offset_cur
= -1;
1404 uint32_t hash
, index
;
1409 hsize
= 1 << dip
->i_di
.di_depth
;
1410 if (hsize
* sizeof(uint64_t) != dip
->i_di
.di_size
) {
1411 gfs2_consist_inode(dip
);
1415 hash
= gfs2_dir_offset2hash(*offset
);
1416 index
= hash
>> (32 - dip
->i_di
.di_depth
);
1418 lp
= kmalloc(sdp
->sd_hash_bsize
, GFP_KERNEL
);
1422 while (index
< hsize
) {
1423 lp_offset
= index
& (sdp
->sd_hash_ptrs
- 1);
1424 ht_offset
= index
- lp_offset
;
1426 if (ht_offset_cur
!= ht_offset
) {
1427 error
= gfs2_jdata_read_mem(dip
, (char *)lp
,
1428 ht_offset
* sizeof(uint64_t),
1429 sdp
->sd_hash_bsize
);
1430 if (error
!= sdp
->sd_hash_bsize
) {
1435 ht_offset_cur
= ht_offset
;
1438 error
= get_leaf(dip
, be64_to_cpu(lp
[lp_offset
]), &bh
);
1442 gfs2_leaf_in(&leaf
, bh
->b_data
);
1445 error
= do_filldir_multi(dip
, offset
, opaque
, filldir
,
1448 error
= do_filldir_single(dip
, offset
, opaque
, filldir
,
1449 bh
, leaf
.lf_entries
, &copied
);
1459 len
= 1 << (dip
->i_di
.di_depth
- leaf
.lf_depth
);
1460 index
= (index
& ~(len
- 1)) + len
;
1469 static int dir_e_mvino(struct gfs2_inode
*dip
, struct qstr
*filename
,
1470 struct gfs2_inum
*inum
, unsigned int new_type
)
1472 struct buffer_head
*bh
, *dibh
;
1473 struct gfs2_dirent
*dent
;
1476 error
= linked_leaf_search(dip
, filename
, &dent
, NULL
, &bh
);
1477 if (error
== -ENOENT
) {
1478 gfs2_consist_inode(dip
);
1484 gfs2_trans_add_bh(dip
->i_gl
, bh
);
1486 gfs2_inum_out(inum
, (char *)&dent
->de_inum
);
1487 dent
->de_type
= new_type
;
1491 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1495 dip
->i_di
.di_mtime
= dip
->i_di
.di_ctime
= get_seconds();
1497 gfs2_trans_add_bh(dip
->i_gl
, dibh
);
1498 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
1505 * dir_l_search - Search linear (stuffed dinode) dir for inode matching name
1506 * @dip: The GFS2 inode
1507 * @filename: Filename string
1508 * @inode: If non-NULL, function fills with formal inode # and block address
1509 * @type: If non-NULL, function fills with DT_... dinode type
1514 static int dir_l_search(struct gfs2_inode
*dip
, struct qstr
*filename
,
1515 struct gfs2_inum
*inum
, unsigned int *type
)
1517 struct buffer_head
*dibh
;
1518 struct gfs2_dirent
*dent
;
1521 if (!gfs2_is_stuffed(dip
)) {
1522 gfs2_consist_inode(dip
);
1526 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1530 error
= leaf_search(dip
, dibh
, filename
, &dent
, NULL
);
1533 gfs2_inum_in(inum
, (char *)&dent
->de_inum
);
1535 *type
= dent
->de_type
;
1543 static int dir_l_add(struct gfs2_inode
*dip
, struct qstr
*filename
,
1544 struct gfs2_inum
*inum
, unsigned int type
)
1546 struct buffer_head
*dibh
;
1547 struct gfs2_dirent
*dent
;
1550 if (!gfs2_is_stuffed(dip
)) {
1551 gfs2_consist_inode(dip
);
1555 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1559 if (gfs2_dirent_alloc(dip
, dibh
, filename
->len
, &dent
)) {
1562 error
= dir_make_exhash(dip
);
1564 error
= dir_e_add(dip
, filename
, inum
, type
);
1569 /* gfs2_dirent_alloc() pins */
1571 gfs2_inum_out(inum
, (char *)&dent
->de_inum
);
1572 dent
->de_hash
= gfs2_disk_hash(filename
->name
, filename
->len
);
1573 dent
->de_hash
= cpu_to_be32(dent
->de_hash
);
1574 dent
->de_type
= type
;
1575 memcpy((char *)(dent
+ 1), filename
->name
, filename
->len
);
1577 dip
->i_di
.di_entries
++;
1578 dip
->i_di
.di_mtime
= dip
->i_di
.di_ctime
= get_seconds();
1580 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
1586 static int dir_l_del(struct gfs2_inode
*dip
, struct qstr
*filename
)
1588 struct buffer_head
*dibh
;
1589 struct gfs2_dirent
*dent
, *prev
;
1592 if (!gfs2_is_stuffed(dip
)) {
1593 gfs2_consist_inode(dip
);
1597 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1601 error
= leaf_search(dip
, dibh
, filename
, &dent
, &prev
);
1602 if (error
== -ENOENT
) {
1603 gfs2_consist_inode(dip
);
1610 dirent_del(dip
, dibh
, prev
, dent
);
1612 /* dirent_del() pins */
1614 if (!dip
->i_di
.di_entries
)
1615 gfs2_consist_inode(dip
);
1616 dip
->i_di
.di_entries
--;
1618 dip
->i_di
.di_mtime
= dip
->i_di
.di_ctime
= get_seconds();
1620 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
1628 static int dir_l_read(struct gfs2_inode
*dip
, uint64_t *offset
, void *opaque
,
1629 gfs2_filldir_t filldir
)
1631 struct buffer_head
*dibh
;
1635 if (!gfs2_is_stuffed(dip
)) {
1636 gfs2_consist_inode(dip
);
1640 if (!dip
->i_di
.di_entries
)
1643 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1647 error
= do_filldir_single(dip
, offset
,
1649 dibh
, dip
->i_di
.di_entries
,
1659 static int dir_l_mvino(struct gfs2_inode
*dip
, struct qstr
*filename
,
1660 struct gfs2_inum
*inum
, unsigned int new_type
)
1662 struct buffer_head
*dibh
;
1663 struct gfs2_dirent
*dent
;
1666 if (!gfs2_is_stuffed(dip
)) {
1667 gfs2_consist_inode(dip
);
1671 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1675 error
= leaf_search(dip
, dibh
, filename
, &dent
, NULL
);
1676 if (error
== -ENOENT
) {
1677 gfs2_consist_inode(dip
);
1684 gfs2_trans_add_bh(dip
->i_gl
, dibh
);
1686 gfs2_inum_out(inum
, (char *)&dent
->de_inum
);
1687 dent
->de_type
= new_type
;
1689 dip
->i_di
.di_mtime
= dip
->i_di
.di_ctime
= get_seconds();
1691 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
1700 * gfs2_dir_search - Search a directory
1701 * @dip: The GFS2 inode
1705 * This routine searches a directory for a file or another directory.
1706 * Assumes a glock is held on dip.
1711 int gfs2_dir_search(struct gfs2_inode
*dip
, struct qstr
*filename
,
1712 struct gfs2_inum
*inum
, unsigned int *type
)
1716 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
)
1717 error
= dir_e_search(dip
, filename
, inum
, type
);
1719 error
= dir_l_search(dip
, filename
, inum
, type
);
1725 * gfs2_dir_add - Add new filename into directory
1726 * @dip: The GFS2 inode
1727 * @filename: The new name
1728 * @inode: The inode number of the entry
1729 * @type: The type of the entry
1731 * Returns: 0 on success, error code on failure
1734 int gfs2_dir_add(struct gfs2_inode
*dip
, struct qstr
*filename
,
1735 struct gfs2_inum
*inum
, unsigned int type
)
1739 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
)
1740 error
= dir_e_add(dip
, filename
, inum
, type
);
1742 error
= dir_l_add(dip
, filename
, inum
, type
);
1748 * gfs2_dir_del - Delete a directory entry
1749 * @dip: The GFS2 inode
1750 * @filename: The filename
1752 * Returns: 0 on success, error code on failure
1755 int gfs2_dir_del(struct gfs2_inode
*dip
, struct qstr
*filename
)
1759 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
)
1760 error
= dir_e_del(dip
, filename
);
1762 error
= dir_l_del(dip
, filename
);
1767 int gfs2_dir_read(struct gfs2_inode
*dip
, uint64_t *offset
, void *opaque
,
1768 gfs2_filldir_t filldir
)
1772 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
)
1773 error
= dir_e_read(dip
, offset
, opaque
, filldir
);
1775 error
= dir_l_read(dip
, offset
, opaque
, filldir
);
1781 * gfs2_dir_mvino - Change inode number of directory entry
1782 * @dip: The GFS2 inode
1786 * This routine changes the inode number of a directory entry. It's used
1787 * by rename to change ".." when a directory is moved.
1788 * Assumes a glock is held on dvp.
1793 int gfs2_dir_mvino(struct gfs2_inode
*dip
, struct qstr
*filename
,
1794 struct gfs2_inum
*inum
, unsigned int new_type
)
1798 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
)
1799 error
= dir_e_mvino(dip
, filename
, inum
, new_type
);
1801 error
= dir_l_mvino(dip
, filename
, inum
, new_type
);
1807 * foreach_leaf - call a function for each leaf in a directory
1808 * @dip: the directory
1809 * @lc: the function to call for each each
1810 * @data: private data to pass to it
1815 static int foreach_leaf(struct gfs2_inode
*dip
, leaf_call_t lc
, void *data
)
1817 struct gfs2_sbd
*sdp
= dip
->i_sbd
;
1818 struct buffer_head
*bh
;
1819 struct gfs2_leaf leaf
;
1820 uint32_t hsize
, len
;
1821 uint32_t ht_offset
, lp_offset
, ht_offset_cur
= -1;
1827 hsize
= 1 << dip
->i_di
.di_depth
;
1828 if (hsize
* sizeof(uint64_t) != dip
->i_di
.di_size
) {
1829 gfs2_consist_inode(dip
);
1833 lp
= kmalloc(sdp
->sd_hash_bsize
, GFP_KERNEL
);
1837 while (index
< hsize
) {
1838 lp_offset
= index
& (sdp
->sd_hash_ptrs
- 1);
1839 ht_offset
= index
- lp_offset
;
1841 if (ht_offset_cur
!= ht_offset
) {
1842 error
= gfs2_jdata_read_mem(dip
, (char *)lp
,
1843 ht_offset
* sizeof(uint64_t),
1844 sdp
->sd_hash_bsize
);
1845 if (error
!= sdp
->sd_hash_bsize
) {
1850 ht_offset_cur
= ht_offset
;
1853 leaf_no
= be64_to_cpu(lp
[lp_offset
]);
1855 error
= get_leaf(dip
, leaf_no
, &bh
);
1858 gfs2_leaf_in(&leaf
, bh
->b_data
);
1861 len
= 1 << (dip
->i_di
.di_depth
- leaf
.lf_depth
);
1863 error
= lc(dip
, index
, len
, leaf_no
, data
);
1867 index
= (index
& ~(len
- 1)) + len
;
1872 if (index
!= hsize
) {
1873 gfs2_consist_inode(dip
);
1884 * leaf_dealloc - Deallocate a directory leaf
1885 * @dip: the directory
1886 * @index: the hash table offset in the directory
1887 * @len: the number of pointers to this leaf
1888 * @leaf_no: the leaf number
1894 static int leaf_dealloc(struct gfs2_inode
*dip
, uint32_t index
, uint32_t len
,
1895 uint64_t leaf_no
, void *data
)
1897 struct gfs2_sbd
*sdp
= dip
->i_sbd
;
1898 struct gfs2_leaf tmp_leaf
;
1899 struct gfs2_rgrp_list rlist
;
1900 struct buffer_head
*bh
, *dibh
;
1902 unsigned int rg_blocks
= 0, l_blocks
= 0;
1904 unsigned int x
, size
= len
* sizeof(uint64_t);
1907 memset(&rlist
, 0, sizeof(struct gfs2_rgrp_list
));
1909 ht
= kzalloc(size
, GFP_KERNEL
);
1913 gfs2_alloc_get(dip
);
1915 error
= gfs2_quota_hold(dip
, NO_QUOTA_CHANGE
, NO_QUOTA_CHANGE
);
1919 error
= gfs2_rindex_hold(sdp
, &dip
->i_alloc
.al_ri_gh
);
1923 /* Count the number of leaves */
1925 for (blk
= leaf_no
; blk
; blk
= tmp_leaf
.lf_next
) {
1926 error
= get_leaf(dip
, blk
, &bh
);
1929 gfs2_leaf_in(&tmp_leaf
, (bh
)->b_data
);
1932 gfs2_rlist_add(sdp
, &rlist
, blk
);
1936 gfs2_rlist_alloc(&rlist
, LM_ST_EXCLUSIVE
, 0);
1938 for (x
= 0; x
< rlist
.rl_rgrps
; x
++) {
1939 struct gfs2_rgrpd
*rgd
;
1940 rgd
= get_gl2rgd(rlist
.rl_ghs
[x
].gh_gl
);
1941 rg_blocks
+= rgd
->rd_ri
.ri_length
;
1944 error
= gfs2_glock_nq_m(rlist
.rl_rgrps
, rlist
.rl_ghs
);
1948 error
= gfs2_trans_begin(sdp
,
1949 rg_blocks
+ (DIV_RU(size
, sdp
->sd_jbsize
) + 1) +
1950 RES_DINODE
+ RES_STATFS
+ RES_QUOTA
, l_blocks
);
1952 goto out_rg_gunlock
;
1954 for (blk
= leaf_no
; blk
; blk
= tmp_leaf
.lf_next
) {
1955 error
= get_leaf(dip
, blk
, &bh
);
1958 gfs2_leaf_in(&tmp_leaf
, bh
->b_data
);
1961 gfs2_free_meta(dip
, blk
, 1);
1963 if (!dip
->i_di
.di_blocks
)
1964 gfs2_consist_inode(dip
);
1965 dip
->i_di
.di_blocks
--;
1968 error
= gfs2_jdata_write_mem(dip
, ht
, index
* sizeof(uint64_t), size
);
1969 if (error
!= size
) {
1975 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1979 gfs2_trans_add_bh(dip
->i_gl
, dibh
);
1980 gfs2_dinode_out(&dip
->i_di
, dibh
->b_data
);
1984 gfs2_trans_end(sdp
);
1987 gfs2_glock_dq_m(rlist
.rl_rgrps
, rlist
.rl_ghs
);
1990 gfs2_rlist_free(&rlist
);
1991 gfs2_glock_dq_uninit(&dip
->i_alloc
.al_ri_gh
);
1994 gfs2_quota_unhold(dip
);
1997 gfs2_alloc_put(dip
);
2004 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
2005 * @dip: the directory
2007 * Dealloc all on-disk directory leaves to FREEMETA state
2008 * Change on-disk inode type to "regular file"
2013 int gfs2_dir_exhash_dealloc(struct gfs2_inode
*dip
)
2015 struct gfs2_sbd
*sdp
= dip
->i_sbd
;
2016 struct buffer_head
*bh
;
2019 /* Dealloc on-disk leaves to FREEMETA state */
2020 error
= foreach_leaf(dip
, leaf_dealloc
, NULL
);
2024 /* Make this a regular file in case we crash.
2025 (We don't want to free these blocks a second time.) */
2027 error
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
2031 error
= gfs2_meta_inode_buffer(dip
, &bh
);
2033 gfs2_trans_add_bh(dip
->i_gl
, bh
);
2034 ((struct gfs2_dinode
*)bh
->b_data
)->di_mode
= cpu_to_be32(S_IFREG
);
2038 gfs2_trans_end(sdp
);
2044 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2045 * @ip: the file being written to
2046 * @filname: the filename that's going to be added
2047 * @alloc_required: set to 1 if an alloc is required, 0 otherwise
2052 int gfs2_diradd_alloc_required(struct gfs2_inode
*dip
, struct qstr
*filename
,
2053 int *alloc_required
)
2055 struct buffer_head
*bh
= NULL
, *bh_next
;
2056 uint32_t hsize
, hash
, index
;
2059 *alloc_required
= 0;
2061 if (dip
->i_di
.di_flags
& GFS2_DIF_EXHASH
) {
2062 hsize
= 1 << dip
->i_di
.di_depth
;
2063 if (hsize
* sizeof(uint64_t) != dip
->i_di
.di_size
) {
2064 gfs2_consist_inode(dip
);
2068 hash
= gfs2_disk_hash(filename
->name
, filename
->len
);
2069 index
= hash
>> (32 - dip
->i_di
.di_depth
);
2071 error
= get_first_leaf(dip
, index
, &bh_next
);
2080 if (dirent_fits(dip
, bh
, filename
->len
))
2083 error
= get_next_leaf(dip
, bh
, &bh_next
);
2084 if (error
== -ENOENT
) {
2085 *alloc_required
= 1;
2094 error
= gfs2_meta_inode_buffer(dip
, &bh
);
2098 if (!dirent_fits(dip
, bh
, filename
->len
))
2099 *alloc_required
= 1;