1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1990, 1993, 1994
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Margo Seltzer.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 */
34
35 /*
36 * PACKAGE: hash
37 *
38 * DESCRIPTION:
39 * Contains buffer management
40 *
41 * ROUTINES:
42 * External
43 * __buf_init
44 * __get_buf
45 * __buf_free
46 * __reclaim_buf
47 * Internal
48 * newbuf
49 */
50
51 #include <sys/param.h>
52
53 #include <stddef.h>
54 #include <stdio.h>
55 #include <stdlib.h>
56 #include <string.h>
57
58 #ifdef DEBUG
59 #include <assert.h>
60 #endif
61
62 #include <db.h>
63 #include "hash.h"
64 #include "page.h"
65 #include "extern.h"
66
67 static BUFHEAD *newbuf(HTAB *, u_int32_t, BUFHEAD *);
68
69 /* Unlink B from its place in the lru */
70 #define BUF_REMOVE(B) { \
71 (B)->prev->next = (B)->next; \
72 (B)->next->prev = (B)->prev; \
73 }
74
75 /* Insert B after P */
76 #define BUF_INSERT(B, P) { \
77 (B)->next = (P)->next; \
78 (B)->prev = (P); \
79 (P)->next = (B); \
80 (B)->next->prev = (B); \
81 }
82
83 #define MRU hashp->bufhead.next
84 #define LRU hashp->bufhead.prev
85
86 #define MRU_INSERT(B) BUF_INSERT((B), &hashp->bufhead)
87 #define LRU_INSERT(B) BUF_INSERT((B), LRU)
88
89 /*
90 * We are looking for a buffer with address "addr". If prev_bp is NULL, then
91 * address is a bucket index. If prev_bp is not NULL, then it points to the
92 * page previous to an overflow page that we are trying to find.
93 *
94 * CAVEAT: The buffer header accessed via prev_bp's ovfl field may no longer
95 * be valid. Therefore, you must always verify that its address matches the
96 * address you are seeking.
97 */
98 BUFHEAD *
__get_buf(HTAB * hashp,u_int32_t addr,BUFHEAD * prev_bp,int newpage)99 __get_buf(HTAB *hashp, u_int32_t addr,
100 BUFHEAD *prev_bp, /* If prev_bp set, indicates a new overflow page. */
101 int newpage)
102 {
103 BUFHEAD *bp;
104 u_int32_t is_disk_mask;
105 int is_disk, segment_ndx;
106 SEGMENT segp;
107
108 is_disk = 0;
109 is_disk_mask = 0;
110 if (prev_bp) {
111 bp = prev_bp->ovfl;
112 if (!bp || (bp->addr != addr))
113 bp = NULL;
114 if (!newpage)
115 is_disk = BUF_DISK;
116 } else {
117 /* Grab buffer out of directory */
118 segment_ndx = addr & (hashp->SGSIZE - 1);
119
120 /* valid segment ensured by __call_hash() */
121 segp = hashp->dir[addr >> hashp->SSHIFT];
122 #ifdef DEBUG
123 assert(segp != NULL);
124 #endif
125 bp = PTROF(segp[segment_ndx]);
126 is_disk_mask = ISDISK(segp[segment_ndx]);
127 is_disk = is_disk_mask || !hashp->new_file;
128 }
129
130 if (!bp) {
131 bp = newbuf(hashp, addr, prev_bp);
132 if (!bp ||
133 __get_page(hashp, bp->page, addr, !prev_bp, is_disk, 0))
134 return (NULL);
135 if (!prev_bp)
136 segp[segment_ndx] =
137 (BUFHEAD *)((intptr_t)bp | is_disk_mask);
138 } else {
139 BUF_REMOVE(bp);
140 MRU_INSERT(bp);
141 }
142 return (bp);
143 }
144
145 /*
146 * We need a buffer for this page. Either allocate one, or evict a resident
147 * one (if we have as many buffers as we're allowed) and put this one in.
148 *
149 * If newbuf finds an error (returning NULL), it also sets errno.
150 */
151 static BUFHEAD *
newbuf(HTAB * hashp,u_int32_t addr,BUFHEAD * prev_bp)152 newbuf(HTAB *hashp, u_int32_t addr, BUFHEAD *prev_bp)
153 {
154 BUFHEAD *bp; /* The buffer we're going to use */
155 BUFHEAD *xbp; /* Temp pointer */
156 BUFHEAD *next_xbp;
157 SEGMENT segp;
158 int segment_ndx;
159 u_int16_t oaddr, *shortp;
160
161 oaddr = 0;
162 bp = LRU;
163
164 /* It is bad to overwrite the page under the cursor. */
165 if (bp == hashp->cpage) {
166 BUF_REMOVE(bp);
167 MRU_INSERT(bp);
168 bp = LRU;
169 }
170
171 /* If prev_bp is part of bp overflow, create a new buffer. */
172 if (hashp->nbufs == 0 && prev_bp && bp->ovfl) {
173 BUFHEAD *ovfl;
174
175 for (ovfl = bp->ovfl; ovfl ; ovfl = ovfl->ovfl) {
176 if (ovfl == prev_bp) {
177 hashp->nbufs++;
178 break;
179 }
180 }
181 }
182
183 /*
184 * If LRU buffer is pinned, the buffer pool is too small. We need to
185 * allocate more buffers.
186 */
187 if (hashp->nbufs || (bp->flags & BUF_PIN) || bp == hashp->cpage) {
188 /* Allocate a new one */
189 if ((bp = (BUFHEAD *)calloc(1, sizeof(BUFHEAD))) == NULL)
190 return (NULL);
191 if ((bp->page = (char *)calloc(1, hashp->BSIZE)) == NULL) {
192 free(bp);
193 return (NULL);
194 }
195 if (hashp->nbufs)
196 hashp->nbufs--;
197 } else {
198 /* Kick someone out */
199 BUF_REMOVE(bp);
200 /*
201 * If this is an overflow page with addr 0, it's already been
202 * flushed back in an overflow chain and initialized.
203 */
204 if ((bp->addr != 0) || (bp->flags & BUF_BUCKET)) {
205 /*
206 * Set oaddr before __put_page so that you get it
207 * before bytes are swapped.
208 */
209 shortp = (u_int16_t *)bp->page;
210 if (shortp[0])
211 oaddr = shortp[shortp[0] - 1];
212 if ((bp->flags & BUF_MOD) && __put_page(hashp, bp->page,
213 bp->addr, (int)IS_BUCKET(bp->flags), 0))
214 return (NULL);
215 /*
216 * Update the pointer to this page (i.e. invalidate it).
217 *
218 * If this is a new file (i.e. we created it at open
219 * time), make sure that we mark pages which have been
220 * written to disk so we retrieve them from disk later,
221 * rather than allocating new pages.
222 */
223 if (IS_BUCKET(bp->flags)) {
224 segment_ndx = bp->addr & (hashp->SGSIZE - 1);
225 segp = hashp->dir[bp->addr >> hashp->SSHIFT];
226 #ifdef DEBUG
227 assert(segp != NULL);
228 #endif
229
230 if (hashp->new_file &&
231 ((bp->flags & BUF_MOD) ||
232 ISDISK(segp[segment_ndx])))
233 segp[segment_ndx] = (BUFHEAD *)BUF_DISK;
234 else
235 segp[segment_ndx] = NULL;
236 }
237 /*
238 * Since overflow pages can only be access by means of
239 * their bucket, free overflow pages associated with
240 * this bucket.
241 */
242 for (xbp = bp; xbp->ovfl;) {
243 next_xbp = xbp->ovfl;
244 xbp->ovfl = NULL;
245 xbp = next_xbp;
246
247 /* Check that ovfl pointer is up date. */
248 if (IS_BUCKET(xbp->flags) ||
249 (oaddr != xbp->addr))
250 break;
251
252 shortp = (u_int16_t *)xbp->page;
253 if (shortp[0])
254 /* set before __put_page */
255 oaddr = shortp[shortp[0] - 1];
256 if ((xbp->flags & BUF_MOD) && __put_page(hashp,
257 xbp->page, xbp->addr, 0, 0))
258 return (NULL);
259 xbp->addr = 0;
260 xbp->flags = 0;
261 BUF_REMOVE(xbp);
262 LRU_INSERT(xbp);
263 }
264 }
265 }
266
267 /* Now assign this buffer */
268 bp->addr = addr;
269 #ifdef DEBUG1
270 (void)fprintf(stderr, "NEWBUF1: %d->ovfl was %d is now %d\n",
271 bp->addr, (bp->ovfl ? bp->ovfl->addr : 0), 0);
272 #endif
273 bp->ovfl = NULL;
274 if (prev_bp) {
275 /*
276 * If prev_bp is set, this is an overflow page, hook it in to
277 * the buffer overflow links.
278 */
279 #ifdef DEBUG1
280 (void)fprintf(stderr, "NEWBUF2: %d->ovfl was %d is now %d\n",
281 prev_bp->addr, (prev_bp->ovfl ? prev_bp->ovfl->addr : 0),
282 (bp ? bp->addr : 0));
283 #endif
284 prev_bp->ovfl = bp;
285 bp->flags = 0;
286 } else
287 bp->flags = BUF_BUCKET;
288 MRU_INSERT(bp);
289 return (bp);
290 }
291
292 void
__buf_init(HTAB * hashp,int nbytes)293 __buf_init(HTAB *hashp, int nbytes)
294 {
295 BUFHEAD *bfp;
296 int npages;
297
298 bfp = &(hashp->bufhead);
299 npages = (nbytes + hashp->BSIZE - 1) >> hashp->BSHIFT;
300 npages = MAX(npages, MIN_BUFFERS);
301
302 hashp->nbufs = npages;
303 bfp->next = bfp;
304 bfp->prev = bfp;
305 /*
306 * This space is calloc'd so these are already null.
307 *
308 * bfp->ovfl = NULL;
309 * bfp->flags = 0;
310 * bfp->page = NULL;
311 * bfp->addr = 0;
312 */
313 }
314
315 int
__buf_free(HTAB * hashp,int do_free,int to_disk)316 __buf_free(HTAB *hashp, int do_free, int to_disk)
317 {
318 BUFHEAD *bp;
319
320 /* Need to make sure that buffer manager has been initialized */
321 if (!LRU)
322 return (0);
323 for (bp = LRU; bp != &hashp->bufhead;) {
324 /* Check that the buffer is valid */
325 if (bp->addr || IS_BUCKET(bp->flags)) {
326 if (to_disk && (bp->flags & BUF_MOD) &&
327 __put_page(hashp, bp->page,
328 bp->addr, IS_BUCKET(bp->flags), 0))
329 return (-1);
330 }
331 /* Check if we are freeing stuff */
332 if (do_free) {
333 if (bp->page) {
334 (void)memset(bp->page, 0, hashp->BSIZE);
335 free(bp->page);
336 }
337 BUF_REMOVE(bp);
338 free(bp);
339 bp = LRU;
340 } else
341 bp = bp->prev;
342 }
343 return (0);
344 }
345
346 void
__reclaim_buf(HTAB * hashp,BUFHEAD * bp)347 __reclaim_buf(HTAB *hashp, BUFHEAD *bp)
348 {
349 bp->ovfl = NULL;
350 bp->addr = 0;
351 bp->flags = 0;
352 BUF_REMOVE(bp);
353 LRU_INSERT(bp);
354 }
355