1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
8 * Implements Extendible Hashing as described in:
9 * "Extendible Hashing" by Fagin, et al in
10 * __ACM Trans. on Database Systems__, Sept 1979.
13 * Here's the layout of dirents which is essentially the same as that of ext2
14 * within a single block. The field de_name_len is the number of bytes
15 * actually required for the name (no null terminator). The field de_rec_len
16 * is the number of bytes allocated to the dirent. The offset of the next
17 * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
18 * deleted, the preceding dirent inherits its allocated space, ie
19 * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
20 * by adding de_rec_len to the current dirent, this essentially causes the
21 * deleted dirent to get jumped over when iterating through all the dirents.
23 * When deleting the first dirent in a block, there is no previous dirent so
24 * the field de_ino is set to zero to designate it as deleted. When allocating
25 * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
26 * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
27 * dirent is allocated. Otherwise it must go through all the 'used' dirents
28 * searching for one in which the amount of total space minus the amount of
29 * used space will provide enough space for the new dirent.
31 * There are two types of blocks in which dirents reside. In a stuffed dinode,
32 * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
33 * the block. In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
34 * beginning of the leaf block. The dirents reside in leaves when
36 * dip->i_diskflags & GFS2_DIF_EXHASH is true
38 * Otherwise, the dirents are "linear", within a single stuffed dinode block.
40 * When the dirents are in leaves, the actual contents of the directory file are
41 * used as an array of 64-bit block pointers pointing to the leaf blocks. The
42 * dirents are NOT in the directory file itself. There can be more than one
43 * block pointer in the array that points to the same leaf. In fact, when a
44 * directory is first converted from linear to exhash, all of the pointers
45 * point to the same leaf.
47 * When a leaf is completely full, the size of the hash table can be
48 * doubled unless it is already at the maximum size which is hard coded into
49 * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
50 * but never before the maximum hash table size has been reached.
53 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
55 #include <linux/slab.h>
56 #include <linux/spinlock.h>
57 #include <linux/buffer_head.h>
58 #include <linux/sort.h>
59 #include <linux/gfs2_ondisk.h>
60 #include <linux/crc32.h>
61 #include <linux/vmalloc.h>
62 #include <linux/bio.h>
76 #define MAX_RA_BLOCKS 32 /* max read-ahead blocks */
78 #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
79 #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
80 #define GFS2_HASH_INDEX_MASK 0xffffc000
81 #define GFS2_USE_HASH_FLAG 0x2000
83 struct qstr gfs2_qdot __read_mostly
;
84 struct qstr gfs2_qdotdot __read_mostly
;
86 typedef int (*gfs2_dscan_t
)(const struct gfs2_dirent
*dent
,
87 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_meta(ip
->i_gl
, bh
);
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
, 0, &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_meta(ip
->i_gl
, dibh
);
130 memcpy(dibh
->b_data
+ offset
+ sizeof(struct gfs2_dinode
), buf
, size
);
131 if (ip
->i_inode
.i_size
< offset
+ size
)
132 i_size_write(&ip
->i_inode
, offset
+ size
);
133 ip
->i_inode
.i_mtime
= ip
->i_inode
.i_ctime
= current_time(&ip
->i_inode
);
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
;
167 if (gfs2_is_stuffed(ip
) && offset
+ size
<= gfs2_max_stuffed_size(ip
))
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
);
181 o
= do_div(lblock
, sdp
->sd_jbsize
) + sizeof(struct gfs2_meta_header
);
183 while (copied
< size
) {
185 struct buffer_head
*bh
;
187 amount
= size
- copied
;
188 if (amount
> sdp
->sd_sb
.sb_bsize
- o
)
189 amount
= sdp
->sd_sb
.sb_bsize
- o
;
193 error
= gfs2_alloc_extent(&ip
->i_inode
, lblock
, &dblock
,
198 if (gfs2_assert_withdraw(sdp
, dblock
))
202 if (amount
== sdp
->sd_jbsize
|| new)
203 error
= gfs2_dir_get_new_buffer(ip
, dblock
, &bh
);
205 error
= gfs2_dir_get_existing_buffer(ip
, dblock
, &bh
);
210 gfs2_trans_add_meta(ip
->i_gl
, bh
);
211 memcpy(bh
->b_data
+ o
, buf
, amount
);
220 o
= sizeof(struct gfs2_meta_header
);
224 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
228 if (ip
->i_inode
.i_size
< offset
+ copied
)
229 i_size_write(&ip
->i_inode
, offset
+ copied
);
230 ip
->i_inode
.i_mtime
= ip
->i_inode
.i_ctime
= current_time(&ip
->i_inode
);
232 gfs2_trans_add_meta(ip
->i_gl
, dibh
);
233 gfs2_dinode_out(ip
, dibh
->b_data
);
243 static int gfs2_dir_read_stuffed(struct gfs2_inode
*ip
, __be64
*buf
,
246 struct buffer_head
*dibh
;
249 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
251 memcpy(buf
, dibh
->b_data
+ sizeof(struct gfs2_dinode
), size
);
255 return (error
) ? error
: size
;
260 * gfs2_dir_read_data - Read a data from a directory inode
261 * @ip: The GFS2 Inode
262 * @buf: The buffer to place result into
263 * @size: Amount of data to transfer
265 * Returns: The amount of data actually copied or the error
267 static int gfs2_dir_read_data(struct gfs2_inode
*ip
, __be64
*buf
,
270 struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
277 if (gfs2_is_stuffed(ip
))
278 return gfs2_dir_read_stuffed(ip
, buf
, size
);
280 if (gfs2_assert_warn(sdp
, gfs2_is_jdata(ip
)))
284 o
= do_div(lblock
, sdp
->sd_jbsize
) + sizeof(struct gfs2_meta_header
);
286 while (copied
< size
) {
288 struct buffer_head
*bh
;
290 amount
= size
- copied
;
291 if (amount
> sdp
->sd_sb
.sb_bsize
- o
)
292 amount
= sdp
->sd_sb
.sb_bsize
- o
;
296 error
= gfs2_get_extent(&ip
->i_inode
, lblock
,
298 if (error
|| !dblock
)
301 bh
= gfs2_meta_ra(ip
->i_gl
, dblock
, extlen
);
303 error
= gfs2_meta_read(ip
->i_gl
, dblock
, DIO_WAIT
, 0, &bh
);
307 error
= gfs2_metatype_check(sdp
, bh
, GFS2_METATYPE_JD
);
314 memcpy(buf
, bh
->b_data
+ o
, amount
);
316 buf
+= (amount
/sizeof(__be64
));
319 o
= sizeof(struct gfs2_meta_header
);
324 return (copied
) ? copied
: error
;
328 * gfs2_dir_get_hash_table - Get pointer to the dir hash table
329 * @ip: The inode in question
331 * Returns: The hash table or an error
334 static __be64
*gfs2_dir_get_hash_table(struct gfs2_inode
*ip
)
336 struct inode
*inode
= &ip
->i_inode
;
341 BUG_ON(!(ip
->i_diskflags
& GFS2_DIF_EXHASH
));
343 hc
= ip
->i_hash_cache
;
347 hsize
= BIT(ip
->i_depth
);
348 hsize
*= sizeof(__be64
);
349 if (hsize
!= i_size_read(&ip
->i_inode
)) {
350 gfs2_consist_inode(ip
);
351 return ERR_PTR(-EIO
);
354 hc
= kmalloc(hsize
, GFP_NOFS
| __GFP_NOWARN
);
356 hc
= __vmalloc(hsize
, GFP_NOFS
);
359 return ERR_PTR(-ENOMEM
);
361 ret
= gfs2_dir_read_data(ip
, hc
, hsize
);
367 spin_lock(&inode
->i_lock
);
368 if (likely(!ip
->i_hash_cache
)) {
369 ip
->i_hash_cache
= hc
;
372 spin_unlock(&inode
->i_lock
);
375 return ip
->i_hash_cache
;
379 * gfs2_dir_hash_inval - Invalidate dir hash
380 * @ip: The directory inode
382 * Must be called with an exclusive glock, or during glock invalidation.
384 void gfs2_dir_hash_inval(struct gfs2_inode
*ip
)
388 spin_lock(&ip
->i_inode
.i_lock
);
389 hc
= ip
->i_hash_cache
;
390 ip
->i_hash_cache
= NULL
;
391 spin_unlock(&ip
->i_inode
.i_lock
);
396 static inline int gfs2_dirent_sentinel(const struct gfs2_dirent
*dent
)
398 return dent
->de_inum
.no_addr
== 0 || dent
->de_inum
.no_formal_ino
== 0;
401 static inline int __gfs2_dirent_find(const struct gfs2_dirent
*dent
,
402 const struct qstr
*name
, int ret
)
404 if (!gfs2_dirent_sentinel(dent
) &&
405 be32_to_cpu(dent
->de_hash
) == name
->hash
&&
406 be16_to_cpu(dent
->de_name_len
) == name
->len
&&
407 memcmp(dent
+1, name
->name
, name
->len
) == 0)
412 static int gfs2_dirent_find(const struct gfs2_dirent
*dent
,
413 const struct qstr
*name
,
416 return __gfs2_dirent_find(dent
, name
, 1);
419 static int gfs2_dirent_prev(const struct gfs2_dirent
*dent
,
420 const struct qstr
*name
,
423 return __gfs2_dirent_find(dent
, name
, 2);
427 * name->name holds ptr to start of block.
428 * name->len holds size of block.
430 static int gfs2_dirent_last(const struct gfs2_dirent
*dent
,
431 const struct qstr
*name
,
434 const char *start
= name
->name
;
435 const char *end
= (const char *)dent
+ be16_to_cpu(dent
->de_rec_len
);
436 if (name
->len
== (end
- start
))
441 /* Look for the dirent that contains the offset specified in data. Once we
442 * find that dirent, there must be space available there for the new dirent */
443 static int gfs2_dirent_find_offset(const struct gfs2_dirent
*dent
,
444 const struct qstr
*name
,
447 unsigned required
= GFS2_DIRENT_SIZE(name
->len
);
448 unsigned actual
= GFS2_DIRENT_SIZE(be16_to_cpu(dent
->de_name_len
));
449 unsigned totlen
= be16_to_cpu(dent
->de_rec_len
);
451 if (ptr
< (void *)dent
|| ptr
>= (void *)dent
+ totlen
)
453 if (gfs2_dirent_sentinel(dent
))
455 if (ptr
< (void *)dent
+ actual
)
457 if ((void *)dent
+ totlen
>= ptr
+ required
)
462 static int gfs2_dirent_find_space(const struct gfs2_dirent
*dent
,
463 const struct qstr
*name
,
466 unsigned required
= GFS2_DIRENT_SIZE(name
->len
);
467 unsigned actual
= GFS2_DIRENT_SIZE(be16_to_cpu(dent
->de_name_len
));
468 unsigned totlen
= be16_to_cpu(dent
->de_rec_len
);
470 if (gfs2_dirent_sentinel(dent
))
472 if (totlen
- actual
>= required
)
477 struct dirent_gather
{
478 const struct gfs2_dirent
**pdent
;
482 static int gfs2_dirent_gather(const struct gfs2_dirent
*dent
,
483 const struct qstr
*name
,
486 struct dirent_gather
*g
= opaque
;
487 if (!gfs2_dirent_sentinel(dent
)) {
488 g
->pdent
[g
->offset
++] = dent
;
494 * Other possible things to check:
495 * - Inode located within filesystem size (and on valid block)
496 * - Valid directory entry type
497 * Not sure how heavy-weight we want to make this... could also check
498 * hash is correct for example, but that would take a lot of extra time.
499 * For now the most important thing is to check that the various sizes
502 static int gfs2_check_dirent(struct gfs2_sbd
*sdp
,
503 struct gfs2_dirent
*dent
, unsigned int offset
,
504 unsigned int size
, unsigned int len
, int first
)
506 const char *msg
= "gfs2_dirent too small";
507 if (unlikely(size
< sizeof(struct gfs2_dirent
)))
509 msg
= "gfs2_dirent misaligned";
510 if (unlikely(offset
& 0x7))
512 msg
= "gfs2_dirent points beyond end of block";
513 if (unlikely(offset
+ size
> len
))
515 msg
= "zero inode number";
516 if (unlikely(!first
&& gfs2_dirent_sentinel(dent
)))
518 msg
= "name length is greater than space in dirent";
519 if (!gfs2_dirent_sentinel(dent
) &&
520 unlikely(sizeof(struct gfs2_dirent
)+be16_to_cpu(dent
->de_name_len
) >
525 fs_warn(sdp
, "%s: %s (%s)\n",
526 __func__
, msg
, first
? "first in block" : "not first in block");
530 static int gfs2_dirent_offset(struct gfs2_sbd
*sdp
, const void *buf
)
532 const struct gfs2_meta_header
*h
= buf
;
537 switch(be32_to_cpu(h
->mh_type
)) {
538 case GFS2_METATYPE_LF
:
539 offset
= sizeof(struct gfs2_leaf
);
541 case GFS2_METATYPE_DI
:
542 offset
= sizeof(struct gfs2_dinode
);
549 fs_warn(sdp
, "%s: wrong block type %u\n", __func__
,
550 be32_to_cpu(h
->mh_type
));
554 static struct gfs2_dirent
*gfs2_dirent_scan(struct inode
*inode
, void *buf
,
555 unsigned int len
, gfs2_dscan_t scan
,
556 const struct qstr
*name
,
559 struct gfs2_dirent
*dent
, *prev
;
564 ret
= gfs2_dirent_offset(GFS2_SB(inode
), buf
);
571 size
= be16_to_cpu(dent
->de_rec_len
);
572 if (gfs2_check_dirent(GFS2_SB(inode
), dent
, offset
, size
, len
, 1))
575 ret
= scan(dent
, name
, opaque
);
583 size
= be16_to_cpu(dent
->de_rec_len
);
584 if (gfs2_check_dirent(GFS2_SB(inode
), dent
, offset
, size
,
595 return prev
? prev
: dent
;
602 gfs2_consist_inode(GFS2_I(inode
));
603 return ERR_PTR(-EIO
);
606 static int dirent_check_reclen(struct gfs2_inode
*dip
,
607 const struct gfs2_dirent
*d
, const void *end_p
)
610 u16 rec_len
= be16_to_cpu(d
->de_rec_len
);
612 if (unlikely(rec_len
< sizeof(struct gfs2_dirent
)))
620 gfs2_consist_inode(dip
);
625 * dirent_next - Next dirent
626 * @dip: the directory
628 * @dent: Pointer to list of dirents
630 * Returns: 0 on success, error code otherwise
633 static int dirent_next(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
634 struct gfs2_dirent
**dent
)
636 struct gfs2_dirent
*cur
= *dent
, *tmp
;
637 char *bh_end
= bh
->b_data
+ bh
->b_size
;
640 ret
= dirent_check_reclen(dip
, cur
, bh_end
);
644 tmp
= (void *)cur
+ ret
;
645 ret
= dirent_check_reclen(dip
, tmp
, bh_end
);
649 /* Only the first dent could ever have de_inum.no_addr == 0 */
650 if (gfs2_dirent_sentinel(tmp
)) {
651 gfs2_consist_inode(dip
);
660 * dirent_del - Delete a dirent
661 * @dip: The GFS2 inode
663 * @prev: The previous dirent
664 * @cur: The current dirent
668 static void dirent_del(struct gfs2_inode
*dip
, struct buffer_head
*bh
,
669 struct gfs2_dirent
*prev
, struct gfs2_dirent
*cur
)
671 u16 cur_rec_len
, prev_rec_len
;
673 if (gfs2_dirent_sentinel(cur
)) {
674 gfs2_consist_inode(dip
);
678 gfs2_trans_add_meta(dip
->i_gl
, bh
);
680 /* If there is no prev entry, this is the first entry in the block.
681 The de_rec_len is already as big as it needs to be. Just zero
682 out the inode number and return. */
685 cur
->de_inum
.no_addr
= 0;
686 cur
->de_inum
.no_formal_ino
= 0;
690 /* Combine this dentry with the previous one. */
692 prev_rec_len
= be16_to_cpu(prev
->de_rec_len
);
693 cur_rec_len
= be16_to_cpu(cur
->de_rec_len
);
695 if ((char *)prev
+ prev_rec_len
!= (char *)cur
)
696 gfs2_consist_inode(dip
);
697 if ((char *)cur
+ cur_rec_len
> bh
->b_data
+ bh
->b_size
)
698 gfs2_consist_inode(dip
);
700 prev_rec_len
+= cur_rec_len
;
701 prev
->de_rec_len
= cpu_to_be16(prev_rec_len
);
705 static struct gfs2_dirent
*do_init_dirent(struct inode
*inode
,
706 struct gfs2_dirent
*dent
,
707 const struct qstr
*name
,
708 struct buffer_head
*bh
,
711 struct gfs2_inode
*ip
= GFS2_I(inode
);
712 struct gfs2_dirent
*ndent
;
715 totlen
= be16_to_cpu(dent
->de_rec_len
);
716 BUG_ON(offset
+ name
->len
> totlen
);
717 gfs2_trans_add_meta(ip
->i_gl
, bh
);
718 ndent
= (struct gfs2_dirent
*)((char *)dent
+ offset
);
719 dent
->de_rec_len
= cpu_to_be16(offset
);
720 gfs2_qstr2dirent(name
, totlen
- offset
, ndent
);
726 * Takes a dent from which to grab space as an argument. Returns the
727 * newly created dent.
729 static struct gfs2_dirent
*gfs2_init_dirent(struct inode
*inode
,
730 struct gfs2_dirent
*dent
,
731 const struct qstr
*name
,
732 struct buffer_head
*bh
)
736 if (!gfs2_dirent_sentinel(dent
))
737 offset
= GFS2_DIRENT_SIZE(be16_to_cpu(dent
->de_name_len
));
738 return do_init_dirent(inode
, dent
, name
, bh
, offset
);
741 static struct gfs2_dirent
*gfs2_dirent_split_alloc(struct inode
*inode
,
742 struct buffer_head
*bh
,
743 const struct qstr
*name
,
746 struct gfs2_dirent
*dent
;
747 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
,
748 gfs2_dirent_find_offset
, name
, ptr
);
749 if (IS_ERR_OR_NULL(dent
))
751 return do_init_dirent(inode
, dent
, name
, bh
,
752 (unsigned)(ptr
- (void *)dent
));
755 static int get_leaf(struct gfs2_inode
*dip
, u64 leaf_no
,
756 struct buffer_head
**bhp
)
760 error
= gfs2_meta_read(dip
->i_gl
, leaf_no
, DIO_WAIT
, 0, bhp
);
761 if (!error
&& gfs2_metatype_check(GFS2_SB(&dip
->i_inode
), *bhp
, GFS2_METATYPE_LF
)) {
762 /* pr_info("block num=%llu\n", leaf_no); */
770 * get_leaf_nr - Get a leaf number associated with the index
771 * @dip: The GFS2 inode
772 * @index: hash table index of the targeted leaf
773 * @leaf_out: Resulting leaf block number
775 * Returns: 0 on success, error code otherwise
778 static int get_leaf_nr(struct gfs2_inode
*dip
, u32 index
, u64
*leaf_out
)
783 hash
= gfs2_dir_get_hash_table(dip
);
784 error
= PTR_ERR_OR_ZERO(hash
);
787 *leaf_out
= be64_to_cpu(*(hash
+ index
));
792 static int get_first_leaf(struct gfs2_inode
*dip
, u32 index
,
793 struct buffer_head
**bh_out
)
798 error
= get_leaf_nr(dip
, index
, &leaf_no
);
800 error
= get_leaf(dip
, leaf_no
, bh_out
);
805 static struct gfs2_dirent
*gfs2_dirent_search(struct inode
*inode
,
806 const struct qstr
*name
,
808 struct buffer_head
**pbh
)
810 struct buffer_head
*bh
;
811 struct gfs2_dirent
*dent
;
812 struct gfs2_inode
*ip
= GFS2_I(inode
);
815 if (ip
->i_diskflags
& GFS2_DIF_EXHASH
) {
816 struct gfs2_leaf
*leaf
;
817 unsigned int hsize
= BIT(ip
->i_depth
);
820 if (hsize
* sizeof(u64
) != i_size_read(inode
)) {
821 gfs2_consist_inode(ip
);
822 return ERR_PTR(-EIO
);
825 index
= name
->hash
>> (32 - ip
->i_depth
);
826 error
= get_first_leaf(ip
, index
, &bh
);
828 return ERR_PTR(error
);
830 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
,
834 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
835 ln
= be64_to_cpu(leaf
->lf_next
);
840 error
= get_leaf(ip
, ln
, &bh
);
843 return error
? ERR_PTR(error
) : NULL
;
847 error
= gfs2_meta_inode_buffer(ip
, &bh
);
849 return ERR_PTR(error
);
850 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
, scan
, name
, NULL
);
852 if (IS_ERR_OR_NULL(dent
)) {
860 static struct gfs2_leaf
*new_leaf(struct inode
*inode
, struct buffer_head
**pbh
, u16 depth
)
862 struct gfs2_inode
*ip
= GFS2_I(inode
);
866 struct buffer_head
*bh
;
867 struct gfs2_leaf
*leaf
;
868 struct gfs2_dirent
*dent
;
869 struct timespec64 tv
= current_time(inode
);
871 error
= gfs2_alloc_blocks(ip
, &bn
, &n
, 0, NULL
);
874 bh
= gfs2_meta_new(ip
->i_gl
, bn
);
878 gfs2_trans_remove_revoke(GFS2_SB(inode
), bn
, 1);
879 gfs2_trans_add_meta(ip
->i_gl
, bh
);
880 gfs2_metatype_set(bh
, GFS2_METATYPE_LF
, GFS2_FORMAT_LF
);
881 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
882 leaf
->lf_depth
= cpu_to_be16(depth
);
883 leaf
->lf_entries
= 0;
884 leaf
->lf_dirent_format
= cpu_to_be32(GFS2_FORMAT_DE
);
886 leaf
->lf_inode
= cpu_to_be64(ip
->i_no_addr
);
887 leaf
->lf_dist
= cpu_to_be32(1);
888 leaf
->lf_nsec
= cpu_to_be32(tv
.tv_nsec
);
889 leaf
->lf_sec
= cpu_to_be64(tv
.tv_sec
);
890 memset(leaf
->lf_reserved2
, 0, sizeof(leaf
->lf_reserved2
));
891 dent
= (struct gfs2_dirent
*)(leaf
+1);
892 gfs2_qstr2dirent(&empty_name
, bh
->b_size
- sizeof(struct gfs2_leaf
), dent
);
898 * dir_make_exhash - Convert a stuffed directory into an ExHash directory
899 * @inode: The directory inode to be converted to exhash
901 * Returns: 0 on success, error code otherwise
904 static int dir_make_exhash(struct inode
*inode
)
906 struct gfs2_inode
*dip
= GFS2_I(inode
);
907 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
908 struct gfs2_dirent
*dent
;
910 struct buffer_head
*bh
, *dibh
;
911 struct gfs2_leaf
*leaf
;
918 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
922 /* Turn over a new leaf */
924 leaf
= new_leaf(inode
, &bh
, 0);
929 gfs2_assert(sdp
, dip
->i_entries
< BIT(16));
930 leaf
->lf_entries
= cpu_to_be16(dip
->i_entries
);
934 gfs2_buffer_copy_tail(bh
, sizeof(struct gfs2_leaf
), dibh
,
935 sizeof(struct gfs2_dinode
));
937 /* Find last entry */
940 args
.len
= bh
->b_size
- sizeof(struct gfs2_dinode
) +
941 sizeof(struct gfs2_leaf
);
942 args
.name
= bh
->b_data
;
943 dent
= gfs2_dirent_scan(&dip
->i_inode
, bh
->b_data
, bh
->b_size
,
944 gfs2_dirent_last
, &args
, NULL
);
953 return PTR_ERR(dent
);
956 /* Adjust the last dirent's record length
957 (Remember that dent still points to the last entry.) */
959 dent
->de_rec_len
= cpu_to_be16(be16_to_cpu(dent
->de_rec_len
) +
960 sizeof(struct gfs2_dinode
) -
961 sizeof(struct gfs2_leaf
));
965 /* We're done with the new leaf block, now setup the new
968 gfs2_trans_add_meta(dip
->i_gl
, dibh
);
969 gfs2_buffer_clear_tail(dibh
, sizeof(struct gfs2_dinode
));
971 lp
= (__be64
*)(dibh
->b_data
+ sizeof(struct gfs2_dinode
));
973 for (x
= sdp
->sd_hash_ptrs
; x
--; lp
++)
974 *lp
= cpu_to_be64(bn
);
976 i_size_write(inode
, sdp
->sd_sb
.sb_bsize
/ 2);
977 gfs2_add_inode_blocks(&dip
->i_inode
, 1);
978 dip
->i_diskflags
|= GFS2_DIF_EXHASH
;
980 for (x
= sdp
->sd_hash_ptrs
, y
= -1; x
; x
>>= 1, y
++) ;
983 gfs2_dinode_out(dip
, dibh
->b_data
);
991 * dir_split_leaf - Split a leaf block into two
992 * @inode: The directory inode to be split
993 * @name: name of the dirent we're trying to insert
995 * Returns: 0 on success, error code on failure
998 static int dir_split_leaf(struct inode
*inode
, const struct qstr
*name
)
1000 struct gfs2_inode
*dip
= GFS2_I(inode
);
1001 struct buffer_head
*nbh
, *obh
, *dibh
;
1002 struct gfs2_leaf
*nleaf
, *oleaf
;
1003 struct gfs2_dirent
*dent
= NULL
, *prev
= NULL
, *next
= NULL
, *new;
1004 u32 start
, len
, half_len
, divider
;
1011 index
= name
->hash
>> (32 - dip
->i_depth
);
1012 error
= get_leaf_nr(dip
, index
, &leaf_no
);
1016 /* Get the old leaf block */
1017 error
= get_leaf(dip
, leaf_no
, &obh
);
1021 oleaf
= (struct gfs2_leaf
*)obh
->b_data
;
1022 if (dip
->i_depth
== be16_to_cpu(oleaf
->lf_depth
)) {
1024 return 1; /* can't split */
1027 gfs2_trans_add_meta(dip
->i_gl
, obh
);
1029 nleaf
= new_leaf(inode
, &nbh
, be16_to_cpu(oleaf
->lf_depth
) + 1);
1034 bn
= nbh
->b_blocknr
;
1036 /* Compute the start and len of leaf pointers in the hash table. */
1037 len
= BIT(dip
->i_depth
- be16_to_cpu(oleaf
->lf_depth
));
1038 half_len
= len
>> 1;
1040 fs_warn(GFS2_SB(inode
), "i_depth %u lf_depth %u index %u\n",
1041 dip
->i_depth
, be16_to_cpu(oleaf
->lf_depth
), index
);
1042 gfs2_consist_inode(dip
);
1047 start
= (index
& ~(len
- 1));
1049 /* Change the pointers.
1050 Don't bother distinguishing stuffed from non-stuffed.
1051 This code is complicated enough already. */
1052 lp
= kmalloc_array(half_len
, sizeof(__be64
), GFP_NOFS
);
1058 /* Change the pointers */
1059 for (x
= 0; x
< half_len
; x
++)
1060 lp
[x
] = cpu_to_be64(bn
);
1062 gfs2_dir_hash_inval(dip
);
1064 error
= gfs2_dir_write_data(dip
, (char *)lp
, start
* sizeof(u64
),
1065 half_len
* sizeof(u64
));
1066 if (error
!= half_len
* sizeof(u64
)) {
1074 /* Compute the divider */
1075 divider
= (start
+ half_len
) << (32 - dip
->i_depth
);
1077 /* Copy the entries */
1078 dent
= (struct gfs2_dirent
*)(obh
->b_data
+ sizeof(struct gfs2_leaf
));
1082 if (dirent_next(dip
, obh
, &next
))
1085 if (!gfs2_dirent_sentinel(dent
) &&
1086 be32_to_cpu(dent
->de_hash
) < divider
) {
1088 void *ptr
= ((char *)dent
- obh
->b_data
) + nbh
->b_data
;
1089 str
.name
= (char*)(dent
+1);
1090 str
.len
= be16_to_cpu(dent
->de_name_len
);
1091 str
.hash
= be32_to_cpu(dent
->de_hash
);
1092 new = gfs2_dirent_split_alloc(inode
, nbh
, &str
, ptr
);
1094 error
= PTR_ERR(new);
1098 new->de_inum
= dent
->de_inum
; /* No endian worries */
1099 new->de_type
= dent
->de_type
; /* No endian worries */
1100 be16_add_cpu(&nleaf
->lf_entries
, 1);
1102 dirent_del(dip
, obh
, prev
, dent
);
1104 if (!oleaf
->lf_entries
)
1105 gfs2_consist_inode(dip
);
1106 be16_add_cpu(&oleaf
->lf_entries
, -1);
1116 oleaf
->lf_depth
= nleaf
->lf_depth
;
1118 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1119 if (!gfs2_assert_withdraw(GFS2_SB(&dip
->i_inode
), !error
)) {
1120 gfs2_trans_add_meta(dip
->i_gl
, dibh
);
1121 gfs2_add_inode_blocks(&dip
->i_inode
, 1);
1122 gfs2_dinode_out(dip
, dibh
->b_data
);
1141 * dir_double_exhash - Double size of ExHash table
1142 * @dip: The GFS2 dinode
1144 * Returns: 0 on success, error code on failure
1147 static int dir_double_exhash(struct gfs2_inode
*dip
)
1149 struct buffer_head
*dibh
;
1157 hsize
= BIT(dip
->i_depth
);
1158 hsize_bytes
= hsize
* sizeof(__be64
);
1160 hc
= gfs2_dir_get_hash_table(dip
);
1164 hc2
= kmalloc_array(hsize_bytes
, 2, GFP_NOFS
| __GFP_NOWARN
);
1166 hc2
= __vmalloc(hsize_bytes
* 2, GFP_NOFS
);
1172 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1176 for (x
= 0; x
< hsize
; x
++) {
1182 error
= gfs2_dir_write_data(dip
, (char *)hc2
, 0, hsize_bytes
* 2);
1183 if (error
!= (hsize_bytes
* 2))
1186 gfs2_dir_hash_inval(dip
);
1187 dip
->i_hash_cache
= hc2
;
1189 gfs2_dinode_out(dip
, dibh
->b_data
);
1194 /* Replace original hash table & size */
1195 gfs2_dir_write_data(dip
, (char *)hc
, 0, hsize_bytes
);
1196 i_size_write(&dip
->i_inode
, hsize_bytes
);
1197 gfs2_dinode_out(dip
, dibh
->b_data
);
1205 * compare_dents - compare directory entries by hash value
1209 * When comparing the hash entries of @a to @b:
1215 static int compare_dents(const void *a
, const void *b
)
1217 const struct gfs2_dirent
*dent_a
, *dent_b
;
1221 dent_a
= *(const struct gfs2_dirent
**)a
;
1222 hash_a
= dent_a
->de_cookie
;
1224 dent_b
= *(const struct gfs2_dirent
**)b
;
1225 hash_b
= dent_b
->de_cookie
;
1227 if (hash_a
> hash_b
)
1229 else if (hash_a
< hash_b
)
1232 unsigned int len_a
= be16_to_cpu(dent_a
->de_name_len
);
1233 unsigned int len_b
= be16_to_cpu(dent_b
->de_name_len
);
1237 else if (len_a
< len_b
)
1240 ret
= memcmp(dent_a
+ 1, dent_b
+ 1, len_a
);
1247 * do_filldir_main - read out directory entries
1248 * @dip: The GFS2 inode
1249 * @ctx: what to feed the entries to
1250 * @darr: an array of struct gfs2_dirent pointers to read
1251 * @entries: the number of entries in darr
1252 * @sort_start: index of the directory array to start our sort
1253 * @copied: pointer to int that's non-zero if a entry has been copied out
1255 * Jump through some hoops to make sure that if there are hash collsions,
1256 * they are read out at the beginning of a buffer. We want to minimize
1257 * the possibility that they will fall into different readdir buffers or
1258 * that someone will want to seek to that location.
1260 * Returns: errno, >0 if the actor tells you to stop
1263 static int do_filldir_main(struct gfs2_inode
*dip
, struct dir_context
*ctx
,
1264 struct gfs2_dirent
**darr
, u32 entries
,
1265 u32 sort_start
, int *copied
)
1267 const struct gfs2_dirent
*dent
, *dent_next
;
1272 if (sort_start
< entries
)
1273 sort(&darr
[sort_start
], entries
- sort_start
,
1274 sizeof(struct gfs2_dirent
*), compare_dents
, NULL
);
1276 dent_next
= darr
[0];
1277 off_next
= dent_next
->de_cookie
;
1279 for (x
= 0, y
= 1; x
< entries
; x
++, y
++) {
1284 dent_next
= darr
[y
];
1285 off_next
= dent_next
->de_cookie
;
1291 if (off_next
== off
) {
1292 if (*copied
&& !run
)
1303 if (!dir_emit(ctx
, (const char *)(dent
+ 1),
1304 be16_to_cpu(dent
->de_name_len
),
1305 be64_to_cpu(dent
->de_inum
.no_addr
),
1306 be16_to_cpu(dent
->de_type
)))
1312 /* Increment the ctx->pos by one, so the next time we come into the
1313 do_filldir fxn, we get the next entry instead of the last one in the
1321 static void *gfs2_alloc_sort_buffer(unsigned size
)
1325 if (size
< KMALLOC_MAX_SIZE
)
1326 ptr
= kmalloc(size
, GFP_NOFS
| __GFP_NOWARN
);
1328 ptr
= __vmalloc(size
, GFP_NOFS
);
1333 static int gfs2_set_cookies(struct gfs2_sbd
*sdp
, struct buffer_head
*bh
,
1334 unsigned leaf_nr
, struct gfs2_dirent
**darr
,
1340 for (i
= 0; i
< entries
; i
++) {
1343 darr
[i
]->de_cookie
= be32_to_cpu(darr
[i
]->de_hash
);
1344 darr
[i
]->de_cookie
= gfs2_disk_hash2offset(darr
[i
]->de_cookie
);
1346 if (!sdp
->sd_args
.ar_loccookie
)
1348 offset
= (char *)(darr
[i
]) -
1349 (bh
->b_data
+ gfs2_dirent_offset(sdp
, bh
->b_data
));
1350 offset
/= GFS2_MIN_DIRENT_SIZE
;
1351 offset
+= leaf_nr
* sdp
->sd_max_dents_per_leaf
;
1352 if (offset
>= GFS2_USE_HASH_FLAG
||
1353 leaf_nr
>= GFS2_USE_HASH_FLAG
) {
1354 darr
[i
]->de_cookie
|= GFS2_USE_HASH_FLAG
;
1359 darr
[i
]->de_cookie
&= GFS2_HASH_INDEX_MASK
;
1360 darr
[i
]->de_cookie
|= offset
;
1366 static int gfs2_dir_read_leaf(struct inode
*inode
, struct dir_context
*ctx
,
1367 int *copied
, unsigned *depth
,
1370 struct gfs2_inode
*ip
= GFS2_I(inode
);
1371 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1372 struct buffer_head
*bh
;
1373 struct gfs2_leaf
*lf
;
1374 unsigned entries
= 0, entries2
= 0;
1375 unsigned leaves
= 0, leaf
= 0, offset
, sort_offset
;
1376 struct gfs2_dirent
**darr
, *dent
;
1377 struct dirent_gather g
;
1378 struct buffer_head
**larr
;
1379 int error
, i
, need_sort
= 0, sort_id
;
1383 error
= get_leaf(ip
, lfn
, &bh
);
1386 lf
= (struct gfs2_leaf
*)bh
->b_data
;
1388 *depth
= be16_to_cpu(lf
->lf_depth
);
1389 entries
+= be16_to_cpu(lf
->lf_entries
);
1391 lfn
= be64_to_cpu(lf
->lf_next
);
1395 if (*depth
< GFS2_DIR_MAX_DEPTH
|| !sdp
->sd_args
.ar_loccookie
) {
1405 * The extra 99 entries are not normally used, but are a buffer
1406 * zone in case the number of entries in the leaf is corrupt.
1407 * 99 is the maximum number of entries that can fit in a single
1410 larr
= gfs2_alloc_sort_buffer((leaves
+ entries
+ 99) * sizeof(void *));
1413 darr
= (struct gfs2_dirent
**)(larr
+ leaves
);
1414 g
.pdent
= (const struct gfs2_dirent
**)darr
;
1419 error
= get_leaf(ip
, lfn
, &bh
);
1422 lf
= (struct gfs2_leaf
*)bh
->b_data
;
1423 lfn
= be64_to_cpu(lf
->lf_next
);
1424 if (lf
->lf_entries
) {
1426 entries2
+= be16_to_cpu(lf
->lf_entries
);
1427 dent
= gfs2_dirent_scan(inode
, bh
->b_data
, bh
->b_size
,
1428 gfs2_dirent_gather
, NULL
, &g
);
1429 error
= PTR_ERR(dent
);
1432 if (entries2
!= g
.offset
) {
1433 fs_warn(sdp
, "Number of entries corrupt in dir "
1434 "leaf %llu, entries2 (%u) != "
1436 (unsigned long long)bh
->b_blocknr
,
1437 entries2
, g
.offset
);
1438 gfs2_consist_inode(ip
);
1443 sort_id
= gfs2_set_cookies(sdp
, bh
, leaf
, &darr
[offset
],
1444 be16_to_cpu(lf
->lf_entries
));
1445 if (!need_sort
&& sort_id
>= 0) {
1447 sort_offset
= offset
+ sort_id
;
1451 larr
[leaf
++] = NULL
;
1456 BUG_ON(entries2
!= entries
);
1457 error
= do_filldir_main(ip
, ctx
, darr
, entries
, need_sort
?
1458 sort_offset
: entries
, copied
);
1460 for(i
= 0; i
< leaf
; i
++)
1468 * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks.
1469 * @inode: the directory inode
1470 * @hsize: hash table size
1471 * @index: index into the hash table
1472 * @f_ra: read-ahead parameters
1474 * Note: we can't calculate each index like dir_e_read can because we don't
1475 * have the leaf, and therefore we don't have the depth, and therefore we
1476 * don't have the length. So we have to just read enough ahead to make up
1477 * for the loss of information.
1479 static void gfs2_dir_readahead(struct inode
*inode
, unsigned hsize
, u32 index
,
1480 struct file_ra_state
*f_ra
)
1482 struct gfs2_inode
*ip
= GFS2_I(inode
);
1483 struct gfs2_glock
*gl
= ip
->i_gl
;
1484 struct buffer_head
*bh
;
1485 u64 blocknr
= 0, last
;
1488 /* First check if we've already read-ahead for the whole range. */
1489 if (index
+ MAX_RA_BLOCKS
< f_ra
->start
)
1492 f_ra
->start
= max((pgoff_t
)index
, f_ra
->start
);
1493 for (count
= 0; count
< MAX_RA_BLOCKS
; count
++) {
1494 if (f_ra
->start
>= hsize
) /* if exceeded the hash table */
1498 blocknr
= be64_to_cpu(ip
->i_hash_cache
[f_ra
->start
]);
1500 if (blocknr
== last
)
1503 bh
= gfs2_getbuf(gl
, blocknr
, 1);
1504 if (trylock_buffer(bh
)) {
1505 if (buffer_uptodate(bh
)) {
1510 bh
->b_end_io
= end_buffer_read_sync
;
1511 submit_bh(REQ_OP_READ
,
1512 REQ_RAHEAD
| REQ_META
| REQ_PRIO
,
1521 * dir_e_read - Reads the entries from a directory into a filldir buffer
1522 * @inode: the directory inode
1523 * @ctx: actor to feed the entries to
1524 * @f_ra: read-ahead parameters
1529 static int dir_e_read(struct inode
*inode
, struct dir_context
*ctx
,
1530 struct file_ra_state
*f_ra
)
1532 struct gfs2_inode
*dip
= GFS2_I(inode
);
1540 hsize
= BIT(dip
->i_depth
);
1541 hash
= gfs2_dir_offset2hash(ctx
->pos
);
1542 index
= hash
>> (32 - dip
->i_depth
);
1544 if (dip
->i_hash_cache
== NULL
)
1546 lp
= gfs2_dir_get_hash_table(dip
);
1550 gfs2_dir_readahead(inode
, hsize
, index
, f_ra
);
1552 while (index
< hsize
) {
1553 error
= gfs2_dir_read_leaf(inode
, ctx
,
1555 be64_to_cpu(lp
[index
]));
1559 len
= BIT(dip
->i_depth
- depth
);
1560 index
= (index
& ~(len
- 1)) + len
;
1568 int gfs2_dir_read(struct inode
*inode
, struct dir_context
*ctx
,
1569 struct file_ra_state
*f_ra
)
1571 struct gfs2_inode
*dip
= GFS2_I(inode
);
1572 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1573 struct dirent_gather g
;
1574 struct gfs2_dirent
**darr
, *dent
;
1575 struct buffer_head
*dibh
;
1579 if (!dip
->i_entries
)
1582 if (dip
->i_diskflags
& GFS2_DIF_EXHASH
)
1583 return dir_e_read(inode
, ctx
, f_ra
);
1585 if (!gfs2_is_stuffed(dip
)) {
1586 gfs2_consist_inode(dip
);
1590 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
1595 /* 96 is max number of dirents which can be stuffed into an inode */
1596 darr
= kmalloc_array(96, sizeof(struct gfs2_dirent
*), GFP_NOFS
);
1598 g
.pdent
= (const struct gfs2_dirent
**)darr
;
1600 dent
= gfs2_dirent_scan(inode
, dibh
->b_data
, dibh
->b_size
,
1601 gfs2_dirent_gather
, NULL
, &g
);
1603 error
= PTR_ERR(dent
);
1606 if (dip
->i_entries
!= g
.offset
) {
1607 fs_warn(sdp
, "Number of entries corrupt in dir %llu, "
1608 "ip->i_entries (%u) != g.offset (%u)\n",
1609 (unsigned long long)dip
->i_no_addr
,
1612 gfs2_consist_inode(dip
);
1616 gfs2_set_cookies(sdp
, dibh
, 0, darr
, dip
->i_entries
);
1617 error
= do_filldir_main(dip
, ctx
, darr
,
1618 dip
->i_entries
, 0, &copied
);
1632 * gfs2_dir_search - Search a directory
1633 * @dir: The GFS2 directory inode
1634 * @name: The name we are looking up
1635 * @fail_on_exist: Fail if the name exists rather than looking it up
1637 * This routine searches a directory for a file or another directory.
1638 * Assumes a glock is held on dip.
1643 struct inode
*gfs2_dir_search(struct inode
*dir
, const struct qstr
*name
,
1646 struct buffer_head
*bh
;
1647 struct gfs2_dirent
*dent
;
1648 u64 addr
, formal_ino
;
1651 dent
= gfs2_dirent_search(dir
, name
, gfs2_dirent_find
, &bh
);
1653 struct inode
*inode
;
1657 return ERR_CAST(dent
);
1658 dtype
= be16_to_cpu(dent
->de_type
);
1659 rahead
= be16_to_cpu(dent
->de_rahead
);
1660 addr
= be64_to_cpu(dent
->de_inum
.no_addr
);
1661 formal_ino
= be64_to_cpu(dent
->de_inum
.no_formal_ino
);
1664 return ERR_PTR(-EEXIST
);
1665 inode
= gfs2_inode_lookup(dir
->i_sb
, dtype
, addr
, formal_ino
,
1666 GFS2_BLKST_FREE
/* ignore */);
1668 GFS2_I(inode
)->i_rahead
= rahead
;
1671 return ERR_PTR(-ENOENT
);
1674 int gfs2_dir_check(struct inode
*dir
, const struct qstr
*name
,
1675 const struct gfs2_inode
*ip
)
1677 struct buffer_head
*bh
;
1678 struct gfs2_dirent
*dent
;
1681 dent
= gfs2_dirent_search(dir
, name
, gfs2_dirent_find
, &bh
);
1684 return PTR_ERR(dent
);
1686 if (be64_to_cpu(dent
->de_inum
.no_addr
) != ip
->i_no_addr
)
1688 if (be64_to_cpu(dent
->de_inum
.no_formal_ino
) !=
1689 ip
->i_no_formal_ino
)
1691 if (unlikely(IF2DT(ip
->i_inode
.i_mode
) !=
1692 be16_to_cpu(dent
->de_type
))) {
1693 gfs2_consist_inode(GFS2_I(dir
));
1706 * dir_new_leaf - Add a new leaf onto hash chain
1707 * @inode: The directory
1708 * @name: The name we are adding
1710 * This adds a new dir leaf onto an existing leaf when there is not
1711 * enough space to add a new dir entry. This is a last resort after
1712 * we've expanded the hash table to max size and also split existing
1713 * leaf blocks, so it will only occur for very large directories.
1715 * The dist parameter is set to 1 for leaf blocks directly attached
1716 * to the hash table, 2 for one layer of indirection, 3 for two layers
1717 * etc. We are thus able to tell the difference between an old leaf
1718 * with dist set to zero (i.e. "don't know") and a new one where we
1719 * set this information for debug/fsck purposes.
1721 * Returns: 0 on success, or -ve on error
1724 static int dir_new_leaf(struct inode
*inode
, const struct qstr
*name
)
1726 struct buffer_head
*bh
, *obh
;
1727 struct gfs2_inode
*ip
= GFS2_I(inode
);
1728 struct gfs2_leaf
*leaf
, *oleaf
;
1734 index
= name
->hash
>> (32 - ip
->i_depth
);
1735 error
= get_first_leaf(ip
, index
, &obh
);
1740 oleaf
= (struct gfs2_leaf
*)obh
->b_data
;
1741 bn
= be64_to_cpu(oleaf
->lf_next
);
1745 error
= get_leaf(ip
, bn
, &obh
);
1750 gfs2_trans_add_meta(ip
->i_gl
, obh
);
1752 leaf
= new_leaf(inode
, &bh
, be16_to_cpu(oleaf
->lf_depth
));
1757 leaf
->lf_dist
= cpu_to_be32(dist
);
1758 oleaf
->lf_next
= cpu_to_be64(bh
->b_blocknr
);
1762 error
= gfs2_meta_inode_buffer(ip
, &bh
);
1765 gfs2_trans_add_meta(ip
->i_gl
, bh
);
1766 gfs2_add_inode_blocks(&ip
->i_inode
, 1);
1767 gfs2_dinode_out(ip
, bh
->b_data
);
1772 static u16
gfs2_inode_ra_len(const struct gfs2_inode
*ip
)
1774 u64 where
= ip
->i_no_addr
+ 1;
1775 if (ip
->i_eattr
== where
)
1781 * gfs2_dir_add - Add new filename into directory
1782 * @inode: The directory inode
1783 * @name: The new name
1784 * @nip: The GFS2 inode to be linked in to the directory
1785 * @da: The directory addition info
1787 * If the call to gfs2_diradd_alloc_required resulted in there being
1788 * no need to allocate any new directory blocks, then it will contain
1789 * a pointer to the directory entry and the bh in which it resides. We
1790 * can use that without having to repeat the search. If there was no
1791 * free space, then we must now create more space.
1793 * Returns: 0 on success, error code on failure
1796 int gfs2_dir_add(struct inode
*inode
, const struct qstr
*name
,
1797 const struct gfs2_inode
*nip
, struct gfs2_diradd
*da
)
1799 struct gfs2_inode
*ip
= GFS2_I(inode
);
1800 struct buffer_head
*bh
= da
->bh
;
1801 struct gfs2_dirent
*dent
= da
->dent
;
1802 struct timespec64 tv
;
1803 struct gfs2_leaf
*leaf
;
1807 if (da
->bh
== NULL
) {
1808 dent
= gfs2_dirent_search(inode
, name
,
1809 gfs2_dirent_find_space
, &bh
);
1813 return PTR_ERR(dent
);
1814 dent
= gfs2_init_dirent(inode
, dent
, name
, bh
);
1815 gfs2_inum_out(nip
, dent
);
1816 dent
->de_type
= cpu_to_be16(IF2DT(nip
->i_inode
.i_mode
));
1817 dent
->de_rahead
= cpu_to_be16(gfs2_inode_ra_len(nip
));
1818 tv
= current_time(&ip
->i_inode
);
1819 if (ip
->i_diskflags
& GFS2_DIF_EXHASH
) {
1820 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1821 be16_add_cpu(&leaf
->lf_entries
, 1);
1822 leaf
->lf_nsec
= cpu_to_be32(tv
.tv_nsec
);
1823 leaf
->lf_sec
= cpu_to_be64(tv
.tv_sec
);
1829 ip
->i_inode
.i_mtime
= ip
->i_inode
.i_ctime
= tv
;
1830 if (S_ISDIR(nip
->i_inode
.i_mode
))
1831 inc_nlink(&ip
->i_inode
);
1832 mark_inode_dirty(inode
);
1836 if (!(ip
->i_diskflags
& GFS2_DIF_EXHASH
)) {
1837 error
= dir_make_exhash(inode
);
1842 error
= dir_split_leaf(inode
, name
);
1847 if (ip
->i_depth
< GFS2_DIR_MAX_DEPTH
) {
1848 error
= dir_double_exhash(ip
);
1851 error
= dir_split_leaf(inode
, name
);
1857 error
= dir_new_leaf(inode
, name
);
1868 * gfs2_dir_del - Delete a directory entry
1869 * @dip: The GFS2 inode
1870 * @dentry: The directory entry we want to delete
1872 * Returns: 0 on success, error code on failure
1875 int gfs2_dir_del(struct gfs2_inode
*dip
, const struct dentry
*dentry
)
1877 const struct qstr
*name
= &dentry
->d_name
;
1878 struct gfs2_dirent
*dent
, *prev
= NULL
;
1879 struct buffer_head
*bh
;
1880 struct timespec64 tv
= current_time(&dip
->i_inode
);
1882 /* Returns _either_ the entry (if its first in block) or the
1883 previous entry otherwise */
1884 dent
= gfs2_dirent_search(&dip
->i_inode
, name
, gfs2_dirent_prev
, &bh
);
1886 gfs2_consist_inode(dip
);
1890 gfs2_consist_inode(dip
);
1891 return PTR_ERR(dent
);
1893 /* If not first in block, adjust pointers accordingly */
1894 if (gfs2_dirent_find(dent
, name
, NULL
) == 0) {
1896 dent
= (struct gfs2_dirent
*)((char *)dent
+ be16_to_cpu(prev
->de_rec_len
));
1899 dirent_del(dip
, bh
, prev
, dent
);
1900 if (dip
->i_diskflags
& GFS2_DIF_EXHASH
) {
1901 struct gfs2_leaf
*leaf
= (struct gfs2_leaf
*)bh
->b_data
;
1902 u16 entries
= be16_to_cpu(leaf
->lf_entries
);
1904 gfs2_consist_inode(dip
);
1905 leaf
->lf_entries
= cpu_to_be16(--entries
);
1906 leaf
->lf_nsec
= cpu_to_be32(tv
.tv_nsec
);
1907 leaf
->lf_sec
= cpu_to_be64(tv
.tv_sec
);
1911 if (!dip
->i_entries
)
1912 gfs2_consist_inode(dip
);
1914 dip
->i_inode
.i_mtime
= dip
->i_inode
.i_ctime
= tv
;
1915 if (d_is_dir(dentry
))
1916 drop_nlink(&dip
->i_inode
);
1917 mark_inode_dirty(&dip
->i_inode
);
1923 * gfs2_dir_mvino - Change inode number of directory entry
1924 * @dip: The GFS2 directory inode
1925 * @filename: the filename to be moved
1926 * @nip: the new GFS2 inode
1927 * @new_type: the de_type of the new dirent
1929 * This routine changes the inode number of a directory entry. It's used
1930 * by rename to change ".." when a directory is moved.
1931 * Assumes a glock is held on dvp.
1936 int gfs2_dir_mvino(struct gfs2_inode
*dip
, const struct qstr
*filename
,
1937 const struct gfs2_inode
*nip
, unsigned int new_type
)
1939 struct buffer_head
*bh
;
1940 struct gfs2_dirent
*dent
;
1942 dent
= gfs2_dirent_search(&dip
->i_inode
, filename
, gfs2_dirent_find
, &bh
);
1944 gfs2_consist_inode(dip
);
1948 return PTR_ERR(dent
);
1950 gfs2_trans_add_meta(dip
->i_gl
, bh
);
1951 gfs2_inum_out(nip
, dent
);
1952 dent
->de_type
= cpu_to_be16(new_type
);
1955 dip
->i_inode
.i_mtime
= dip
->i_inode
.i_ctime
= current_time(&dip
->i_inode
);
1956 mark_inode_dirty_sync(&dip
->i_inode
);
1961 * leaf_dealloc - Deallocate a directory leaf
1962 * @dip: the directory
1963 * @index: the hash table offset in the directory
1964 * @len: the number of pointers to this leaf
1965 * @leaf_no: the leaf number
1966 * @leaf_bh: buffer_head for the starting leaf
1967 * @last_dealloc: 1 if this is the final dealloc for the leaf, else 0
1972 static int leaf_dealloc(struct gfs2_inode
*dip
, u32 index
, u32 len
,
1973 u64 leaf_no
, struct buffer_head
*leaf_bh
,
1976 struct gfs2_sbd
*sdp
= GFS2_SB(&dip
->i_inode
);
1977 struct gfs2_leaf
*tmp_leaf
;
1978 struct gfs2_rgrp_list rlist
;
1979 struct buffer_head
*bh
, *dibh
;
1981 unsigned int rg_blocks
= 0, l_blocks
= 0;
1983 unsigned int x
, size
= len
* sizeof(u64
);
1986 error
= gfs2_rindex_update(sdp
);
1990 memset(&rlist
, 0, sizeof(struct gfs2_rgrp_list
));
1992 ht
= kzalloc(size
, GFP_NOFS
| __GFP_NOWARN
);
1994 ht
= __vmalloc(size
, GFP_NOFS
| __GFP_NOWARN
| __GFP_ZERO
);
1998 error
= gfs2_quota_hold(dip
, NO_UID_QUOTA_CHANGE
, NO_GID_QUOTA_CHANGE
);
2002 /* Count the number of leaves */
2005 for (blk
= leaf_no
; blk
; blk
= nblk
) {
2006 if (blk
!= leaf_no
) {
2007 error
= get_leaf(dip
, blk
, &bh
);
2011 tmp_leaf
= (struct gfs2_leaf
*)bh
->b_data
;
2012 nblk
= be64_to_cpu(tmp_leaf
->lf_next
);
2016 gfs2_rlist_add(dip
, &rlist
, blk
);
2020 gfs2_rlist_alloc(&rlist
);
2022 for (x
= 0; x
< rlist
.rl_rgrps
; x
++) {
2023 struct gfs2_rgrpd
*rgd
= gfs2_glock2rgrp(rlist
.rl_ghs
[x
].gh_gl
);
2025 rg_blocks
+= rgd
->rd_length
;
2028 error
= gfs2_glock_nq_m(rlist
.rl_rgrps
, rlist
.rl_ghs
);
2032 error
= gfs2_trans_begin(sdp
,
2033 rg_blocks
+ (DIV_ROUND_UP(size
, sdp
->sd_jbsize
) + 1) +
2034 RES_DINODE
+ RES_STATFS
+ RES_QUOTA
, RES_DINODE
+
2037 goto out_rg_gunlock
;
2041 for (blk
= leaf_no
; blk
; blk
= nblk
) {
2042 struct gfs2_rgrpd
*rgd
;
2044 if (blk
!= leaf_no
) {
2045 error
= get_leaf(dip
, blk
, &bh
);
2049 tmp_leaf
= (struct gfs2_leaf
*)bh
->b_data
;
2050 nblk
= be64_to_cpu(tmp_leaf
->lf_next
);
2054 rgd
= gfs2_blk2rgrpd(sdp
, blk
, true);
2055 gfs2_free_meta(dip
, rgd
, blk
, 1);
2056 gfs2_add_inode_blocks(&dip
->i_inode
, -1);
2059 error
= gfs2_dir_write_data(dip
, ht
, index
* sizeof(u64
), size
);
2060 if (error
!= size
) {
2066 error
= gfs2_meta_inode_buffer(dip
, &dibh
);
2070 gfs2_trans_add_meta(dip
->i_gl
, dibh
);
2071 /* On the last dealloc, make this a regular file in case we crash.
2072 (We don't want to free these blocks a second time.) */
2074 dip
->i_inode
.i_mode
= S_IFREG
;
2075 gfs2_dinode_out(dip
, dibh
->b_data
);
2079 gfs2_trans_end(sdp
);
2081 gfs2_glock_dq_m(rlist
.rl_rgrps
, rlist
.rl_ghs
);
2083 gfs2_rlist_free(&rlist
);
2084 gfs2_quota_unhold(dip
);
2091 * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
2092 * @dip: the directory
2094 * Dealloc all on-disk directory leaves to FREEMETA state
2095 * Change on-disk inode type to "regular file"
2100 int gfs2_dir_exhash_dealloc(struct gfs2_inode
*dip
)
2102 struct buffer_head
*bh
;
2103 struct gfs2_leaf
*leaf
;
2105 u32 index
= 0, next_index
;
2108 int error
= 0, last
;
2110 hsize
= BIT(dip
->i_depth
);
2112 lp
= gfs2_dir_get_hash_table(dip
);
2116 while (index
< hsize
) {
2117 leaf_no
= be64_to_cpu(lp
[index
]);
2119 error
= get_leaf(dip
, leaf_no
, &bh
);
2122 leaf
= (struct gfs2_leaf
*)bh
->b_data
;
2123 len
= BIT(dip
->i_depth
- be16_to_cpu(leaf
->lf_depth
));
2125 next_index
= (index
& ~(len
- 1)) + len
;
2126 last
= ((next_index
>= hsize
) ? 1 : 0);
2127 error
= leaf_dealloc(dip
, index
, len
, leaf_no
, bh
,
2137 if (index
!= hsize
) {
2138 gfs2_consist_inode(dip
);
2148 * gfs2_diradd_alloc_required - find if adding entry will require an allocation
2149 * @inode: the directory inode being written to
2150 * @name: the filename that's going to be added
2151 * @da: The structure to return dir alloc info
2153 * Returns: 0 if ok, -ve on error
2156 int gfs2_diradd_alloc_required(struct inode
*inode
, const struct qstr
*name
,
2157 struct gfs2_diradd
*da
)
2159 struct gfs2_inode
*ip
= GFS2_I(inode
);
2160 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
2161 const unsigned int extra
= sizeof(struct gfs2_dinode
) - sizeof(struct gfs2_leaf
);
2162 struct gfs2_dirent
*dent
;
2163 struct buffer_head
*bh
;
2169 dent
= gfs2_dirent_search(inode
, name
, gfs2_dirent_find_space
, &bh
);
2171 da
->nr_blocks
= sdp
->sd_max_dirres
;
2172 if (!(ip
->i_diskflags
& GFS2_DIF_EXHASH
) &&
2173 (GFS2_DIRENT_SIZE(name
->len
) < extra
))
2178 return PTR_ERR(dent
);