bufcache.c revision 1.1 1 /* $NetBSD: bufcache.c,v 1.1 2003/03/28 08:09:52 perseant Exp $ */
2 /*-
3 * Copyright (c) 2003 The NetBSD Foundation, Inc.
4 * All rights reserved.
5 *
6 * This code is derived from software contributed to The NetBSD Foundation
7 * by Konrad E. Schroder <perseant (at) hhhh.org>.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the NetBSD
20 * Foundation, Inc. and its contributors.
21 * 4. Neither the name of The NetBSD Foundation nor the names of its
22 * contributors may be used to endorse or promote products derived
23 * from this software without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
26 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
28 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35 * POSSIBILITY OF SUCH DAMAGE.
36 */
37
38 #include <sys/types.h>
39 #include <sys/param.h>
40 #include <sys/time.h>
41 #include <sys/buf.h>
42 #include <sys/queue.h>
43 #include <sys/mount.h>
44
45 #include <assert.h>
46 #include <err.h>
47 #include <stdio.h>
48 #include <stdlib.h>
49 #include <string.h>
50 #include <unistd.h>
51
52 #include "bufcache.h"
53 #include "vnode.h"
54
55 /*
56 * Definitions for the buffer free lists.
57 */
58 #define BQUEUES 3 /* number of free buffer queues */
59
60 #define BQ_LOCKED 0 /* super-blocks &c */
61 #define BQ_LRU 1 /* lru, useful buffers */
62 #define BQ_AGE 2 /* rubbish */
63
64 TAILQ_HEAD(bqueues, ubuf) bufqueues[BQUEUES];
65
66 #define HASH_MAX 101
67
68 struct bufhash_struct bufhash[HASH_MAX];
69
70 int maxbufs = BUF_CACHE_SIZE;
71 int nbufs = 0;
72 int cachehits = 0;
73 int cachemisses = 0;
74 int hashmax = 0;
75 off_t locked_queue_bytes = 0;
76
77 /* Simple buffer hash function */
78 static int
79 vl_hash(struct uvnode * vp, daddr_t lbn)
80 {
81 return (int)((unsigned long) vp + lbn) % HASH_MAX;
82 }
83
84 /* Initialize buffer cache */
85 void
86 bufinit(void)
87 {
88 int i;
89
90 for (i = 0; i < BQUEUES; i++) {
91 TAILQ_INIT(&bufqueues[i]);
92 }
93 for (i = 0; i < HASH_MAX; i++)
94 LIST_INIT(&bufhash[HASH_MAX]);
95 }
96
97 /* Print statistics of buffer cache usage */
98 void
99 bufstats(void)
100 {
101 printf("buffer cache: %d hits %d misses (%2.2f%%); hash depth %d\n",
102 cachehits, cachemisses,
103 (cachehits * 100.0) / (cachehits + cachemisses),
104 hashmax);
105 }
106
107 /*
108 * Remove a buffer from the cache.
109 * Caller must remove the buffer from its free list.
110 */
111 void
112 buf_destroy(struct ubuf * bp)
113 {
114 bp->b_flags |= B_NEEDCOMMIT;
115 LIST_REMOVE(bp, b_vnbufs);
116 LIST_REMOVE(bp, b_hash);
117 free(bp->b_data);
118 free(bp);
119 --nbufs;
120 }
121
122 /* Remove a buffer from its free list. */
123 void
124 bremfree(struct ubuf * bp)
125 {
126 struct bqueues *dp = NULL;
127
128 /*
129 * We only calculate the head of the freelist when removing
130 * the last element of the list as that is the only time that
131 * it is needed (e.g. to reset the tail pointer).
132 *
133 * NB: This makes an assumption about how tailq's are implemented.
134 */
135 if (bp->b_flags & B_LOCKED)
136 locked_queue_bytes -= bp->b_bcount;
137 if (TAILQ_NEXT(bp, b_freelist) == NULL) {
138 for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++)
139 if (dp->tqh_last == &bp->b_freelist.tqe_next)
140 break;
141 if (dp == &bufqueues[BQUEUES])
142 errx(1, "bremfree: lost tail");
143 }
144 ++bp->b_vp->v_usecount;
145 TAILQ_REMOVE(dp, bp, b_freelist);
146 }
147
148 /* Return a buffer if it is in the cache, otherwise return NULL. */
149 struct ubuf *
150 incore(struct uvnode * vp, int lbn)
151 {
152 struct ubuf *bp;
153 int hash, depth;
154
155 hash = vl_hash(vp, lbn);
156 /* XXX use a real hash instead. */
157 depth = 0;
158 LIST_FOREACH(bp, &bufhash[hash], b_hash) {
159 if (++depth > hashmax)
160 hashmax = depth;
161 if (bp->b_vp == vp && bp->b_lblkno == lbn) {
162 return bp;
163 }
164 }
165 return NULL;
166 }
167
168 /*
169 * Return a buffer of the given size, lbn and uvnode.
170 * If none is in core, make a new one.
171 */
172 struct ubuf *
173 getblk(struct uvnode * vp, daddr_t lbn, int size)
174 {
175 struct ubuf *bp;
176
177 /*
178 * First check the buffer cache lists.
179 */
180 if ((bp = incore(vp, lbn)) != NULL) {
181 assert(!(bp->b_flags & B_NEEDCOMMIT));
182 assert(!(bp->b_flags & B_BUSY));
183 assert(bp->b_bcount == size);
184 bp->b_flags |= B_BUSY;
185 bremfree(bp);
186 return bp;
187 }
188 /*
189 * Not on the list.
190 * Get a new block of the appropriate size and use that.
191 * If not enough space, free blocks from the AGE and LRU lists
192 * to make room.
193 */
194 while (nbufs >= maxbufs) {
195 bp = TAILQ_FIRST(&bufqueues[BQ_AGE]);
196 if (bp)
197 TAILQ_REMOVE(&bufqueues[BQ_AGE], bp, b_freelist);
198 if (bp == NULL) {
199 bp = TAILQ_FIRST(&bufqueues[BQ_LRU]);
200 if (bp)
201 TAILQ_REMOVE(&bufqueues[BQ_AGE], bp,
202 b_freelist);
203 }
204 if (bp) {
205 if (bp->b_flags & B_DELWRI)
206 VOP_STRATEGY(bp);
207 buf_destroy(bp);
208 }
209 #ifdef DEBUG
210 else {
211 warnx("no free buffers, allocating more than %d",
212 maxbufs);
213 }
214 #endif
215 }
216 ++nbufs;
217 bp = (struct ubuf *) malloc(sizeof(*bp));
218 memset(bp, 0, sizeof(*bp));
219 bp->b_data = malloc(size);
220 memset(bp->b_data, 0, size);
221
222 bp->b_vp = vp;
223 bp->b_blkno = bp->b_lblkno = lbn;
224 bp->b_bcount = size;
225 LIST_INSERT_HEAD(&bufhash[vl_hash(vp, lbn)], bp, b_hash);
226 LIST_INSERT_HEAD(&vp->v_cleanblkhd, bp, b_vnbufs);
227 bp->b_flags = B_BUSY;
228
229 return bp;
230 }
231
232 /* Write a buffer to disk according to its strategy routine. */
233 void
234 bwrite(struct ubuf * bp)
235 {
236 bp->b_flags &= ~(B_READ | B_DONE | B_DELWRI | B_LOCKED);
237 VOP_STRATEGY(bp);
238 bp->b_flags |= B_DONE;
239 reassignbuf(bp, bp->b_vp);
240 brelse(bp);
241 }
242
243 /* Put a buffer back on its free list, clear B_BUSY. */
244 void
245 brelse(struct ubuf * bp)
246 {
247 int age;
248
249 assert(!(bp->b_flags & B_NEEDCOMMIT));
250 assert(bp->b_flags & B_BUSY);
251
252 age = bp->b_flags & B_AGE;
253 bp->b_flags &= ~(B_BUSY | B_AGE);
254 if (bp->b_flags & B_INVAL) {
255 buf_destroy(bp);
256 return;
257 }
258 if (bp->b_flags & B_LOCKED) {
259 locked_queue_bytes += bp->b_bcount;
260 TAILQ_INSERT_TAIL(&bufqueues[BQ_LOCKED], bp, b_freelist);
261 } else if (age) {
262 TAILQ_INSERT_TAIL(&bufqueues[BQ_AGE], bp, b_freelist);
263 } else {
264 TAILQ_INSERT_TAIL(&bufqueues[BQ_LRU], bp, b_freelist);
265 }
266 --bp->b_vp->v_usecount;
267 }
268
269 /* Read the given block from disk, return it B_BUSY. */
270 int
271 bread(struct uvnode * vp, daddr_t lbn, int size, struct ucred * unused,
272 struct ubuf ** bpp)
273 {
274 struct ubuf *bp;
275 daddr_t daddr;
276 int error, count;
277
278 bp = getblk(vp, lbn, size);
279 *bpp = bp;
280 if (bp->b_flags & (B_DELWRI | B_DONE)) {
281 ++cachehits;
282 return 0;
283 }
284 ++cachemisses;
285
286 /*
287 * Not found. Need to find that block's location on disk,
288 * and load it in.
289 */
290 daddr = -1;
291 error = VOP_BMAP(vp, lbn, &daddr);
292 bp->b_blkno = daddr;
293 if (daddr >= 0) {
294 count = pread(vp->v_fd, bp->b_data, bp->b_bcount,
295 dbtob((off_t) daddr));
296 if (count == bp->b_bcount) {
297 bp->b_flags |= B_DONE;
298 return 0;
299 }
300 return -1;
301 }
302 memset(bp->b_data, 0, bp->b_bcount);
303 return 0;
304 }
305
306 /* Move a buffer between dirty and clean block lists. */
307 void
308 reassignbuf(struct ubuf * bp, struct uvnode * vp)
309 {
310 LIST_REMOVE(bp, b_vnbufs);
311 if (bp->b_flags & B_DELWRI) {
312 LIST_INSERT_HEAD(&vp->v_dirtyblkhd, bp, b_vnbufs);
313 } else {
314 LIST_INSERT_HEAD(&vp->v_cleanblkhd, bp, b_vnbufs);
315 }
316 }
317