hash.h revision 1.16 1 1.16 christos /* $NetBSD: hash.h,v 1.16 2015/11/18 18:22:42 christos Exp $ */
2 1.5 cgd
3 1.1 cgd /*-
4 1.4 cgd * Copyright (c) 1990, 1993, 1994
5 1.2 cgd * The Regents of the University of California. All rights reserved.
6 1.1 cgd *
7 1.1 cgd * This code is derived from software contributed to Berkeley by
8 1.1 cgd * Margo Seltzer.
9 1.1 cgd *
10 1.1 cgd * Redistribution and use in source and binary forms, with or without
11 1.1 cgd * modification, are permitted provided that the following conditions
12 1.1 cgd * are met:
13 1.1 cgd * 1. Redistributions of source code must retain the above copyright
14 1.1 cgd * notice, this list of conditions and the following disclaimer.
15 1.1 cgd * 2. Redistributions in binary form must reproduce the above copyright
16 1.1 cgd * notice, this list of conditions and the following disclaimer in the
17 1.1 cgd * documentation and/or other materials provided with the distribution.
18 1.12 agc * 3. Neither the name of the University nor the names of its contributors
19 1.1 cgd * may be used to endorse or promote products derived from this software
20 1.1 cgd * without specific prior written permission.
21 1.1 cgd *
22 1.1 cgd * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 1.1 cgd * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 1.1 cgd * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 1.1 cgd * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 1.1 cgd * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 1.1 cgd * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 1.1 cgd * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 1.1 cgd * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 1.1 cgd * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 1.1 cgd * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 1.1 cgd * SUCH DAMAGE.
33 1.1 cgd *
34 1.4 cgd * @(#)hash.h 8.3 (Berkeley) 5/31/94
35 1.1 cgd */
36 1.10 tv
37 1.13 lukem #if HAVE_NBTOOL_CONFIG_H
38 1.13 lukem #include "nbtool_config.h"
39 1.10 tv #endif
40 1.1 cgd
41 1.1 cgd /* Operations */
42 1.2 cgd typedef enum {
43 1.2 cgd HASH_GET, HASH_PUT, HASH_PUTNEW, HASH_DELETE, HASH_FIRST, HASH_NEXT
44 1.2 cgd } ACTION;
45 1.1 cgd
46 1.1 cgd /* Buffer Management structures */
47 1.1 cgd typedef struct _bufhead BUFHEAD;
48 1.1 cgd
49 1.1 cgd struct _bufhead {
50 1.4 cgd BUFHEAD *prev; /* LRU links */
51 1.4 cgd BUFHEAD *next; /* LRU links */
52 1.4 cgd BUFHEAD *ovfl; /* Overflow page buffer header */
53 1.15 joerg uint32_t addr; /* Address of this page */
54 1.4 cgd char *page; /* Actual page data */
55 1.4 cgd char flags;
56 1.1 cgd #define BUF_MOD 0x0001
57 1.1 cgd #define BUF_DISK 0x0002
58 1.1 cgd #define BUF_BUCKET 0x0004
59 1.1 cgd #define BUF_PIN 0x0008
60 1.1 cgd };
61 1.1 cgd
62 1.2 cgd #define IS_BUCKET(X) ((X) & BUF_BUCKET)
63 1.1 cgd
64 1.2 cgd typedef BUFHEAD **SEGMENT;
65 1.1 cgd
66 1.1 cgd /* Hash Table Information */
67 1.4 cgd typedef struct hashhdr { /* Disk resident portion */
68 1.11 itojun int32_t magic; /* Magic NO for hash tables */
69 1.11 itojun int32_t version; /* Version ID */
70 1.15 joerg uint32_t lorder; /* Byte Order */
71 1.11 itojun int32_t bsize; /* Bucket/Page Size */
72 1.11 itojun int32_t bshift; /* Bucket shift */
73 1.11 itojun int32_t dsize; /* Directory Size */
74 1.11 itojun int32_t ssize; /* Segment Size */
75 1.11 itojun int32_t sshift; /* Segment shift */
76 1.11 itojun int32_t ovfl_point; /* Where overflow pages are being
77 1.4 cgd * allocated */
78 1.11 itojun int32_t last_freed; /* Last overflow page freed */
79 1.11 itojun int32_t max_bucket; /* ID of Maximum bucket in use */
80 1.11 itojun int32_t high_mask; /* Mask to modulo into entire table */
81 1.11 itojun int32_t low_mask; /* Mask to modulo into lower half of
82 1.4 cgd * table */
83 1.11 itojun int32_t ffactor; /* Fill factor */
84 1.11 itojun int32_t nkeys; /* Number of keys in hash table */
85 1.11 itojun int32_t hdrpages; /* Size of table header */
86 1.11 itojun int32_t h_charkey; /* value of hash(CHARKEY) */
87 1.4 cgd #define NCACHED 32 /* number of bit maps and spare
88 1.4 cgd * points */
89 1.11 itojun int32_t spares[NCACHED];/* spare pages for overflow */
90 1.15 joerg uint16_t bitmaps[NCACHED]; /* address of overflow page
91 1.4 cgd * bitmaps */
92 1.1 cgd } HASHHDR;
93 1.1 cgd
94 1.4 cgd typedef struct htab { /* Memory resident data structure */
95 1.4 cgd HASHHDR hdr; /* Header */
96 1.4 cgd int nsegs; /* Number of allocated segments */
97 1.4 cgd int exsegs; /* Number of extra allocated
98 1.4 cgd * segments */
99 1.15 joerg uint32_t (*hash)(const void *, size_t); /* Hash function */
100 1.4 cgd int flags; /* Flag values */
101 1.4 cgd int fp; /* File pointer */
102 1.4 cgd char *tmp_buf; /* Temporary Buffer for BIG data */
103 1.4 cgd char *tmp_key; /* Temporary Buffer for BIG keys */
104 1.4 cgd BUFHEAD *cpage; /* Current page */
105 1.4 cgd int cbucket; /* Current bucket */
106 1.4 cgd int cndx; /* Index of next item on cpage */
107 1.7 jtc int err; /* Error Number -- for DBM
108 1.9 wiz * compatibility */
109 1.4 cgd int new_file; /* Indicates if fd is backing store
110 1.4 cgd * or no */
111 1.4 cgd int save_file; /* Indicates whether we need to flush
112 1.4 cgd * file at
113 1.4 cgd * exit */
114 1.15 joerg uint32_t *mapp[NCACHED]; /* Pointers to page maps */
115 1.4 cgd int nmaps; /* Initial number of bitmaps */
116 1.4 cgd int nbufs; /* Number of buffers left to
117 1.4 cgd * allocate */
118 1.4 cgd BUFHEAD bufhead; /* Header of buffer lru list */
119 1.4 cgd SEGMENT *dir; /* Hash Bucket directory */
120 1.1 cgd } HTAB;
121 1.1 cgd
122 1.1 cgd /*
123 1.1 cgd * Constants
124 1.1 cgd */
125 1.2 cgd #define MAX_BSIZE 65536 /* 2^16 */
126 1.16 christos /*
127 1.16 christos * Make it fit in uint16_t; a better way would be to store size - 1, but
128 1.16 christos * then we'd need to bump the version.
129 1.16 christos */
130 1.16 christos #define HASH_BSIZE(hp) ((hp)->BSIZE == MAX_BSIZE ? MAX_BSIZE - 1 : (hp)->BSIZE)
131 1.1 cgd #define MIN_BUFFERS 6
132 1.1 cgd #define MINHDRSIZE 512
133 1.2 cgd #define DEF_BUFSIZE 65536 /* 64 K */
134 1.2 cgd #define DEF_BUCKET_SIZE 4096
135 1.2 cgd #define DEF_BUCKET_SHIFT 12 /* log2(BUCKET) */
136 1.1 cgd #define DEF_SEGSIZE 256
137 1.2 cgd #define DEF_SEGSIZE_SHIFT 8 /* log2(SEGSIZE) */
138 1.1 cgd #define DEF_DIRSIZE 256
139 1.2 cgd #define DEF_FFACTOR 65536
140 1.2 cgd #define MIN_FFACTOR 4
141 1.2 cgd #define SPLTMAX 8
142 1.2 cgd #define CHARKEY "%$sniglet^&"
143 1.1 cgd #define NUMKEY 1038583
144 1.1 cgd #define BYTE_SHIFT 3
145 1.1 cgd #define INT_TO_BYTE 2
146 1.1 cgd #define INT_BYTE_SHIFT 5
147 1.15 joerg #define ALL_SET ((uint32_t)0xFFFFFFFF)
148 1.1 cgd #define ALL_CLEAR 0
149 1.1 cgd
150 1.8 christos #define PTROF(X) ((BUFHEAD *)(void *)((u_long)(X)&~0x3))
151 1.15 joerg #define ISMOD(X) ((uint32_t)(u_long)(X)&0x1)
152 1.8 christos #define DOMOD(X) ((X) = (char *)(void *)((u_long)(X)|0x1))
153 1.15 joerg #define ISDISK(X) ((uint32_t)(u_long)(X)&0x2)
154 1.8 christos #define DODISK(X) ((X) = (char *)(void *)((u_long)(X)|0x2))
155 1.1 cgd
156 1.2 cgd #define BITS_PER_MAP 32
157 1.1 cgd
158 1.1 cgd /* Given the address of the beginning of a big map, clear/set the nth bit */
159 1.2 cgd #define CLRBIT(A, N) ((A)[(N)/BITS_PER_MAP] &= ~(1<<((N)%BITS_PER_MAP)))
160 1.2 cgd #define SETBIT(A, N) ((A)[(N)/BITS_PER_MAP] |= (1<<((N)%BITS_PER_MAP)))
161 1.2 cgd #define ISSET(A, N) ((A)[(N)/BITS_PER_MAP] & (1<<((N)%BITS_PER_MAP)))
162 1.1 cgd
163 1.1 cgd /* Overflow management */
164 1.1 cgd /*
165 1.2 cgd * Overflow page numbers are allocated per split point. At each doubling of
166 1.2 cgd * the table, we can allocate extra pages. So, an overflow page number has
167 1.2 cgd * the top 5 bits indicate which split point and the lower 11 bits indicate
168 1.2 cgd * which page at that split point is indicated (pages within split points are
169 1.2 cgd * numberered starting with 1).
170 1.2 cgd */
171 1.1 cgd
172 1.1 cgd #define SPLITSHIFT 11
173 1.1 cgd #define SPLITMASK 0x7FF
174 1.15 joerg #define SPLITNUM(N) (((uint32_t)(N)) >> SPLITSHIFT)
175 1.2 cgd #define OPAGENUM(N) ((N) & SPLITMASK)
176 1.15 joerg #define OADDR_OF(S,O) ((uint32_t)((uint32_t)(S) << SPLITSHIFT) + (O))
177 1.1 cgd
178 1.1 cgd #define BUCKET_TO_PAGE(B) \
179 1.8 christos (B) + hashp->HDRPAGES + \
180 1.15 joerg ((B) ? hashp->SPARES[__log2((uint32_t)((B)+1))-1] : 0)
181 1.1 cgd #define OADDR_TO_PAGE(B) \
182 1.2 cgd BUCKET_TO_PAGE ( (1 << SPLITNUM((B))) -1 ) + OPAGENUM((B));
183 1.1 cgd
184 1.1 cgd /*
185 1.2 cgd * page.h contains a detailed description of the page format.
186 1.2 cgd *
187 1.2 cgd * Normally, keys and data are accessed from offset tables in the top of
188 1.2 cgd * each page which point to the beginning of the key and data. There are
189 1.2 cgd * four flag values which may be stored in these offset tables which indicate
190 1.2 cgd * the following:
191 1.2 cgd *
192 1.2 cgd *
193 1.2 cgd * OVFLPAGE Rather than a key data pair, this pair contains
194 1.2 cgd * the address of an overflow page. The format of
195 1.2 cgd * the pair is:
196 1.2 cgd * OVERFLOW_PAGE_NUMBER OVFLPAGE
197 1.2 cgd *
198 1.2 cgd * PARTIAL_KEY This must be the first key/data pair on a page
199 1.2 cgd * and implies that page contains only a partial key.
200 1.2 cgd * That is, the key is too big to fit on a single page
201 1.2 cgd * so it starts on this page and continues on the next.
202 1.2 cgd * The format of the page is:
203 1.2 cgd * KEY_OFF PARTIAL_KEY OVFL_PAGENO OVFLPAGE
204 1.2 cgd *
205 1.2 cgd * KEY_OFF -- offset of the beginning of the key
206 1.2 cgd * PARTIAL_KEY -- 1
207 1.2 cgd * OVFL_PAGENO - page number of the next overflow page
208 1.2 cgd * OVFLPAGE -- 0
209 1.2 cgd *
210 1.2 cgd * FULL_KEY This must be the first key/data pair on the page. It
211 1.2 cgd * is used in two cases.
212 1.2 cgd *
213 1.2 cgd * Case 1:
214 1.2 cgd * There is a complete key on the page but no data
215 1.2 cgd * (because it wouldn't fit). The next page contains
216 1.2 cgd * the data.
217 1.2 cgd *
218 1.2 cgd * Page format it:
219 1.2 cgd * KEY_OFF FULL_KEY OVFL_PAGENO OVFL_PAGE
220 1.2 cgd *
221 1.2 cgd * KEY_OFF -- offset of the beginning of the key
222 1.2 cgd * FULL_KEY -- 2
223 1.2 cgd * OVFL_PAGENO - page number of the next overflow page
224 1.2 cgd * OVFLPAGE -- 0
225 1.2 cgd *
226 1.2 cgd * Case 2:
227 1.2 cgd * This page contains no key, but part of a large
228 1.2 cgd * data field, which is continued on the next page.
229 1.2 cgd *
230 1.2 cgd * Page format it:
231 1.2 cgd * DATA_OFF FULL_KEY OVFL_PAGENO OVFL_PAGE
232 1.2 cgd *
233 1.2 cgd * KEY_OFF -- offset of the beginning of the data on
234 1.2 cgd * this page
235 1.2 cgd * FULL_KEY -- 2
236 1.2 cgd * OVFL_PAGENO - page number of the next overflow page
237 1.2 cgd * OVFLPAGE -- 0
238 1.2 cgd *
239 1.2 cgd * FULL_KEY_DATA
240 1.2 cgd * This must be the first key/data pair on the page.
241 1.2 cgd * There are two cases:
242 1.2 cgd *
243 1.2 cgd * Case 1:
244 1.2 cgd * This page contains a key and the beginning of the
245 1.2 cgd * data field, but the data field is continued on the
246 1.2 cgd * next page.
247 1.2 cgd *
248 1.2 cgd * Page format is:
249 1.2 cgd * KEY_OFF FULL_KEY_DATA OVFL_PAGENO DATA_OFF
250 1.2 cgd *
251 1.2 cgd * KEY_OFF -- offset of the beginning of the key
252 1.2 cgd * FULL_KEY_DATA -- 3
253 1.2 cgd * OVFL_PAGENO - page number of the next overflow page
254 1.2 cgd * DATA_OFF -- offset of the beginning of the data
255 1.2 cgd *
256 1.2 cgd * Case 2:
257 1.2 cgd * This page contains the last page of a big data pair.
258 1.2 cgd * There is no key, only the tail end of the data
259 1.2 cgd * on this page.
260 1.2 cgd *
261 1.2 cgd * Page format is:
262 1.2 cgd * DATA_OFF FULL_KEY_DATA <OVFL_PAGENO> <OVFLPAGE>
263 1.2 cgd *
264 1.2 cgd * DATA_OFF -- offset of the beginning of the data on
265 1.2 cgd * this page
266 1.2 cgd * FULL_KEY_DATA -- 3
267 1.2 cgd * OVFL_PAGENO - page number of the next overflow page
268 1.2 cgd * OVFLPAGE -- 0
269 1.2 cgd *
270 1.2 cgd * OVFL_PAGENO and OVFLPAGE are optional (they are
271 1.2 cgd * not present if there is no next page).
272 1.2 cgd */
273 1.1 cgd
274 1.1 cgd #define OVFLPAGE 0
275 1.1 cgd #define PARTIAL_KEY 1
276 1.1 cgd #define FULL_KEY 2
277 1.1 cgd #define FULL_KEY_DATA 3
278 1.1 cgd #define REAL_KEY 4
279 1.2 cgd
280 1.1 cgd /* Short hands for accessing structure */
281 1.2 cgd #define BSIZE hdr.bsize
282 1.2 cgd #define BSHIFT hdr.bshift
283 1.2 cgd #define DSIZE hdr.dsize
284 1.2 cgd #define SGSIZE hdr.ssize
285 1.2 cgd #define SSHIFT hdr.sshift
286 1.2 cgd #define LORDER hdr.lorder
287 1.2 cgd #define OVFL_POINT hdr.ovfl_point
288 1.2 cgd #define LAST_FREED hdr.last_freed
289 1.1 cgd #define MAX_BUCKET hdr.max_bucket
290 1.1 cgd #define FFACTOR hdr.ffactor
291 1.1 cgd #define HIGH_MASK hdr.high_mask
292 1.1 cgd #define LOW_MASK hdr.low_mask
293 1.1 cgd #define NKEYS hdr.nkeys
294 1.1 cgd #define HDRPAGES hdr.hdrpages
295 1.1 cgd #define SPARES hdr.spares
296 1.1 cgd #define BITMAPS hdr.bitmaps
297 1.1 cgd #define VERSION hdr.version
298 1.1 cgd #define MAGIC hdr.magic
299 1.1 cgd #define NEXT_FREE hdr.next_free
300 1.1 cgd #define H_CHARKEY hdr.h_charkey
301