xref: /freebsd/sys/ufs/ufs/dirhash.h (revision 7bd6fde3)
1 /*-
2  * Copyright (c) 2001 Ian Dowse.  All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  *
13  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23  * SUCH DAMAGE.
24  *
25  * $FreeBSD$
26  */
27 
28 #ifndef _UFS_UFS_DIRHASH_H_
29 #define _UFS_UFS_DIRHASH_H_
30 
31 /*
32  * For fast operations on large directories, we maintain a hash
33  * that maps the file name to the offset of the directory entry within
34  * the directory file.
35  *
36  * The hashing uses a dumb spillover to the next free slot on
37  * collisions, so we must keep the utilisation low to avoid
38  * long linear searches. Deleted entries that are not the last
39  * in a chain must be marked DIRHASH_DEL.
40  *
41  * We also maintain information about free space in each block
42  * to speed up creations.
43  */
44 #define DIRHASH_EMPTY	(-1)	/* entry unused */
45 #define DIRHASH_DEL	(-2)	/* deleted entry; may be part of chain */
46 
47 #define DIRALIGN	4
48 #define DH_NFSTATS	(DIRECTSIZ(MAXNAMLEN + 1) / DIRALIGN)
49 				 /* max DIRALIGN words in a directory entry */
50 
51 /*
52  * Dirhash uses a score mechanism to achieve a hybrid between a
53  * least-recently-used and a least-often-used algorithm for entry
54  * recycling. The score is incremented when a directory is used, and
55  * decremented when the directory is a candidate for recycling. When
56  * the score reaches zero, the hash is recycled. Hashes are linked
57  * together on a TAILQ list, and hashes with higher scores filter
58  * towards the tail (most recently used) end of the list.
59  *
60  * New hash entries are given an inital score of DH_SCOREINIT and are
61  * placed at the most-recently-used end of the list. This helps a lot
62  * in the worst-case case scenario where every directory access is
63  * to a directory that is not hashed (i.e. the working set of hash
64  * candidates is much larger than the configured memry limit). In this
65  * case it limits the number of hash builds to 1/DH_SCOREINIT of the
66  * number of accesses.
67  */
68 #define DH_SCOREINIT	8	/* initial dh_score when dirhash built */
69 #define DH_SCOREMAX	64	/* max dh_score value */
70 
71 /*
72  * The main hash table has 2 levels. It is an array of pointers to
73  * blocks of DH_NBLKOFF offsets.
74  */
75 #define DH_BLKOFFSHIFT	8
76 #define DH_NBLKOFF	(1 << DH_BLKOFFSHIFT)
77 #define DH_BLKOFFMASK	(DH_NBLKOFF - 1)
78 
79 #define DH_ENTRY(dh, slot) \
80     ((dh)->dh_hash[(slot) >> DH_BLKOFFSHIFT][(slot) & DH_BLKOFFMASK])
81 
82 struct dirhash {
83 	struct mtx dh_mtx;	/* protects all fields except dh_list */
84 
85 	doff_t	**dh_hash;	/* the hash array (2-level) */
86 	int	dh_narrays;	/* number of entries in dh_hash */
87 	int	dh_hlen;	/* total slots in the 2-level hash array */
88 	int	dh_hused;	/* entries in use */
89 
90 	/* Free space statistics. XXX assumes DIRBLKSIZ is 512. */
91 	u_int8_t *dh_blkfree;	/* free DIRALIGN words in each dir block */
92 	int	dh_nblk;	/* size of dh_blkfree array */
93 	int	dh_dirblks;	/* number of DIRBLKSIZ blocks in dir */
94 	int	dh_firstfree[DH_NFSTATS + 1]; /* first blk with N words free */
95 
96 	int	dh_seqopt;	/* sequential access optimisation enabled */
97 	doff_t	dh_seqoff;	/* sequential access optimisation offset */
98 
99 	int	dh_score;	/* access count for this dirhash */
100 
101 	int	dh_onlist;	/* true if on the ufsdirhash_list chain */
102 
103 	/* Protected by ufsdirhash_mtx. */
104 	TAILQ_ENTRY(dirhash) dh_list;	/* chain of all dirhashes */
105 };
106 
107 
108 /*
109  * Dirhash functions.
110  */
111 void	ufsdirhash_init(void);
112 void	ufsdirhash_uninit(void);
113 int	ufsdirhash_build(struct inode *);
114 doff_t	ufsdirhash_findfree(struct inode *, int, int *);
115 doff_t	ufsdirhash_enduseful(struct inode *);
116 int	ufsdirhash_lookup(struct inode *, char *, int, doff_t *, struct buf **,
117 	    doff_t *);
118 void	ufsdirhash_newblk(struct inode *, doff_t);
119 void	ufsdirhash_add(struct inode *, struct direct *, doff_t);
120 void	ufsdirhash_remove(struct inode *, struct direct *, doff_t);
121 void	ufsdirhash_move(struct inode *, struct direct *, doff_t, doff_t);
122 void	ufsdirhash_dirtrunc(struct inode *, doff_t);
123 void	ufsdirhash_free(struct inode *);
124 
125 void	ufsdirhash_checkblock(struct inode *, char *, doff_t);
126 
127 #endif /* !_UFS_UFS_DIRHASH_H_ */
128