1 /*-
2  * Copyright (c) 1990, 1993, 1994
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Margo Seltzer.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #if defined(LIBC_SCCS) && !defined(lint)
38 static char sccsid[] = "@(#)hash_buf.c	8.5 (Berkeley) 7/15/94";
39 #endif /* LIBC_SCCS and not lint */
40 
41 /*
42  * PACKAGE: hash
43  *
44  * DESCRIPTION:
45  *	Contains buffer management
46  *
47  * ROUTINES:
48  * External
49  *	__buf_init
50  *	__get_buf
51  *	__buf_free
52  *	__reclaim_buf
53  * Internal
54  *	newbuf
55  */
56 
57 #include <sys/param.h>
58 
59 #include <errno.h>
60 #include <stddef.h>
61 #include <stdio.h>
62 #include <stdlib.h>
63 
64 #ifdef DEBUG
65 #include <assert.h>
66 #endif
67 
68 #include <db.h>
69 #include "hash.h"
70 #include "page.h"
71 #include "extern.h"
72 
73 static BUFHEAD *newbuf __P((HTAB *, u_int32_t, BUFHEAD *));
74 
75 /* Unlink B from its place in the lru */
76 #define BUF_REMOVE(B) { \
77 	(B)->prev->next = (B)->next; \
78 	(B)->next->prev = (B)->prev; \
79 }
80 
81 /* Insert B after P */
82 #define BUF_INSERT(B, P) { \
83 	(B)->next = (P)->next; \
84 	(B)->prev = (P); \
85 	(P)->next = (B); \
86 	(B)->next->prev = (B); \
87 }
88 
89 #define	MRU	hashp->bufhead.next
90 #define	LRU	hashp->bufhead.prev
91 
92 #define MRU_INSERT(B)	BUF_INSERT((B), &hashp->bufhead)
93 #define LRU_INSERT(B)	BUF_INSERT((B), LRU)
94 
95 /*
96  * We are looking for a buffer with address "addr".  If prev_bp is NULL, then
97  * address is a bucket index.  If prev_bp is not NULL, then it points to the
98  * page previous to an overflow page that we are trying to find.
99  *
100  * CAVEAT:  The buffer header accessed via prev_bp's ovfl field may no longer
101  * be valid.  Therefore, you must always verify that its address matches the
102  * address you are seeking.
103  */
104 extern BUFHEAD *
105 __get_buf(hashp, addr, prev_bp, newpage)
106 	HTAB *hashp;
107 	u_int32_t addr;
108 	BUFHEAD *prev_bp;
109 	int newpage;	/* If prev_bp set, indicates a new overflow page. */
110 {
111 	register BUFHEAD *bp;
112 	register u_int32_t is_disk_mask;
113 	register int is_disk, segment_ndx;
114 	SEGMENT segp;
115 
116 	is_disk = 0;
117 	is_disk_mask = 0;
118 	if (prev_bp) {
119 		bp = prev_bp->ovfl;
120 		if (!bp || (bp->addr != addr))
121 			bp = NULL;
122 		if (!newpage)
123 			is_disk = BUF_DISK;
124 	} else {
125 		/* Grab buffer out of directory */
126 		segment_ndx = addr & (hashp->SGSIZE - 1);
127 
128 		/* valid segment ensured by __call_hash() */
129 		segp = hashp->dir[addr >> hashp->SSHIFT];
130 #ifdef DEBUG
131 		assert(segp != NULL);
132 #endif
133 		bp = PTROF(segp[segment_ndx]);
134 		is_disk_mask = ISDISK(segp[segment_ndx]);
135 		is_disk = is_disk_mask || !hashp->new_file;
136 	}
137 
138 	if (!bp) {
139 		bp = newbuf(hashp, addr, prev_bp);
140 		if (!bp ||
141 		    __get_page(hashp, bp->page, addr, !prev_bp, is_disk, 0))
142 			return (NULL);
143 		if (!prev_bp)
144 			segp[segment_ndx] =
145 			    (BUFHEAD *)((ptrdiff_t)bp | is_disk_mask);
146 	} else {
147 		BUF_REMOVE(bp);
148 		MRU_INSERT(bp);
149 	}
150 	return (bp);
151 }
152 
153 /*
154  * We need a buffer for this page. Either allocate one, or evict a resident
155  * one (if we have as many buffers as we're allowed) and put this one in.
156  *
157  * If newbuf finds an error (returning NULL), it also sets errno.
158  */
159 static BUFHEAD *
newbuf(hashp,addr,prev_bp)160 newbuf(hashp, addr, prev_bp)
161 	HTAB *hashp;
162 	u_int32_t addr;
163 	BUFHEAD *prev_bp;
164 {
165 	register BUFHEAD *bp;		/* The buffer we're going to use */
166 	register BUFHEAD *xbp;		/* Temp pointer */
167 	register BUFHEAD *next_xbp;
168 	SEGMENT segp;
169 	int segment_ndx;
170 	u_int16_t oaddr, *shortp;
171 
172 	oaddr = 0;
173 	bp = LRU;
174 	/*
175 	 * If LRU buffer is pinned, the buffer pool is too small. We need to
176 	 * allocate more buffers.
177 	 */
178 	if (hashp->nbufs || (bp->flags & BUF_PIN)) {
179 		/* Allocate a new one */
180 		if ((bp = (BUFHEAD *)malloc(sizeof(BUFHEAD))) == NULL)
181 			return (NULL);
182 #ifdef PURIFY
183 		memset(bp, 0xff, sizeof(BUFHEAD));
184 #endif
185 		if ((bp->page = (char *)malloc(hashp->BSIZE)) == NULL) {
186 			free(bp);
187 			return (NULL);
188 		}
189 #ifdef PURIFY
190 		memset(bp->page, 0xff, hashp->BSIZE);
191 #endif
192 		if (hashp->nbufs)
193 			hashp->nbufs--;
194 	} else {
195 		/* Kick someone out */
196 		BUF_REMOVE(bp);
197 		/*
198 		 * If this is an overflow page with addr 0, it's already been
199 		 * flushed back in an overflow chain and initialized.
200 		 */
201 		if ((bp->addr != 0) || (bp->flags & BUF_BUCKET)) {
202 			/*
203 			 * Set oaddr before __put_page so that you get it
204 			 * before bytes are swapped.
205 			 */
206 			shortp = (u_int16_t *)bp->page;
207 			if (shortp[0])
208 				oaddr = shortp[shortp[0] - 1];
209 			if ((bp->flags & BUF_MOD) && __put_page(hashp, bp->page,
210 			    bp->addr, (int)IS_BUCKET(bp->flags), 0))
211 				return (NULL);
212 			/*
213 			 * Update the pointer to this page (i.e. invalidate it).
214 			 *
215 			 * If this is a new file (i.e. we created it at open
216 			 * time), make sure that we mark pages which have been
217 			 * written to disk so we retrieve them from disk later,
218 			 * rather than allocating new pages.
219 			 */
220 			if (IS_BUCKET(bp->flags)) {
221 				segment_ndx = bp->addr & (hashp->SGSIZE - 1);
222 				segp = hashp->dir[bp->addr >> hashp->SSHIFT];
223 #ifdef DEBUG
224 				assert(segp != NULL);
225 #endif
226 
227 				if (hashp->new_file &&
228 				    ((bp->flags & BUF_MOD) ||
229 				    ISDISK(segp[segment_ndx])))
230 					segp[segment_ndx] = (BUFHEAD *)BUF_DISK;
231 				else
232 					segp[segment_ndx] = NULL;
233 			}
234 			/*
235 			 * Since overflow pages can only be access by means of
236 			 * their bucket, free overflow pages associated with
237 			 * this bucket.
238 			 */
239 			for (xbp = bp; xbp->ovfl;) {
240 				next_xbp = xbp->ovfl;
241 				xbp->ovfl = 0;
242 				xbp = next_xbp;
243 
244 				/* Check that ovfl pointer is up date. */
245 				if (IS_BUCKET(xbp->flags) ||
246 				    (oaddr != xbp->addr))
247 					break;
248 
249 				shortp = (u_int16_t *)xbp->page;
250 				if (shortp[0])
251 					/* set before __put_page */
252 					oaddr = shortp[shortp[0] - 1];
253 				if ((xbp->flags & BUF_MOD) && __put_page(hashp,
254 				    xbp->page, xbp->addr, 0, 0))
255 					return (NULL);
256 				xbp->addr = 0;
257 				xbp->flags = 0;
258 				BUF_REMOVE(xbp);
259 				LRU_INSERT(xbp);
260 			}
261 		}
262 	}
263 
264 	/* Now assign this buffer */
265 	bp->addr = addr;
266 #ifdef DEBUG1
267 	(void)fprintf(stderr, "NEWBUF1: %d->ovfl was %d is now %d\n",
268 	    bp->addr, (bp->ovfl ? bp->ovfl->addr : 0), 0);
269 #endif
270 	bp->ovfl = NULL;
271 	if (prev_bp) {
272 		/*
273 		 * If prev_bp is set, this is an overflow page, hook it in to
274 		 * the buffer overflow links.
275 		 */
276 #ifdef DEBUG1
277 		(void)fprintf(stderr, "NEWBUF2: %d->ovfl was %d is now %d\n",
278 		    prev_bp->addr, (prev_bp->ovfl ? bp->ovfl->addr : 0),
279 		    (bp ? bp->addr : 0));
280 #endif
281 		prev_bp->ovfl = bp;
282 		bp->flags = 0;
283 	} else
284 		bp->flags = BUF_BUCKET;
285 	MRU_INSERT(bp);
286 	return (bp);
287 }
288 
289 extern void
290 __buf_init(hashp, nbytes)
291 	HTAB *hashp;
292 	int nbytes;
293 {
294 	BUFHEAD *bfp;
295 	int npages;
296 
297 	bfp = &(hashp->bufhead);
298 	npages = (nbytes + hashp->BSIZE - 1) >> hashp->BSHIFT;
299 	npages = MAX(npages, MIN_BUFFERS);
300 
301 	hashp->nbufs = npages;
302 	bfp->next = bfp;
303 	bfp->prev = bfp;
304 	/*
305 	 * This space is calloc'd so these are already null.
306 	 *
307 	 * bfp->ovfl = NULL;
308 	 * bfp->flags = 0;
309 	 * bfp->page = NULL;
310 	 * bfp->addr = 0;
311 	 */
312 }
313 
314 extern int
315 __buf_free(hashp, do_free, to_disk)
316 	HTAB *hashp;
317 	int do_free, to_disk;
318 {
319 	BUFHEAD *bp;
320 
321 	/* Need to make sure that buffer manager has been initialized */
322 	if (!LRU)
323 		return (0);
324 	for (bp = LRU; bp != &hashp->bufhead;) {
325 		/* Check that the buffer is valid */
326 		if (bp->addr || IS_BUCKET(bp->flags)) {
327 			if (to_disk && (bp->flags & BUF_MOD) &&
328 			    __put_page(hashp, bp->page,
329 			    bp->addr, IS_BUCKET(bp->flags), 0))
330 				return (-1);
331 		}
332 		/* Check if we are freeing stuff */
333 		if (do_free) {
334 			if (bp->page)
335 				free(bp->page);
336 			BUF_REMOVE(bp);
337 			free(bp);
338 			bp = LRU;
339 		} else
340 			bp = bp->prev;
341 	}
342 	return (0);
343 }
344 
345 extern void
346 __reclaim_buf(hashp, bp)
347 	HTAB *hashp;
348 	BUFHEAD *bp;
349 {
350 	bp->ovfl = 0;
351 	bp->addr = 0;
352 	bp->flags = 0;
353 	BUF_REMOVE(bp);
354 	LRU_INSERT(bp);
355 }
356