1 /* $NetBSD: bufcache.c,v 1.7 2005/05/20 18:59:36 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@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 struct bufhash_struct *bufhash; 67 68 #define HASH_MAX 1024 69 int hashmax = HASH_MAX; 70 int hashmask = (HASH_MAX - 1); 71 72 int maxbufs = BUF_CACHE_SIZE; 73 int nbufs = 0; 74 int cachehits = 0; 75 int cachemisses = 0; 76 int max_depth = 0; 77 off_t locked_queue_bytes = 0; 78 int locked_queue_count = 0; 79 80 /* Simple buffer hash function */ 81 static int 82 vl_hash(struct uvnode * vp, daddr_t lbn) 83 { 84 return (int)((unsigned long) vp + lbn) & hashmask; 85 } 86 87 /* Initialize buffer cache */ 88 void 89 bufinit(int max) 90 { 91 int i; 92 93 if (max) { 94 for (hashmax = 1; hashmax < max; hashmax <<= 1) 95 ; 96 hashmask = hashmax - 1; 97 } 98 99 for (i = 0; i < BQUEUES; i++) { 100 TAILQ_INIT(&bufqueues[i]); 101 } 102 bufhash = (struct bufhash_struct *)malloc(hashmax * sizeof(*bufhash)); 103 for (i = 0; i < hashmax; i++) 104 LIST_INIT(&bufhash[i]); 105 } 106 107 /* Widen the hash table. */ 108 void bufrehash(int max) 109 { 110 int i, newhashmax, newhashmask; 111 struct ubuf *bp, *nbp; 112 struct bufhash_struct *np; 113 114 if (max < 0 || max < hashmax) 115 return; 116 117 /* Round up to a power of two */ 118 for (newhashmax = 1; newhashmax < max; newhashmax <<= 1) 119 ; 120 newhashmask = newhashmax - 1; 121 122 /* Allocate new empty hash table, if we can */ 123 np = (struct bufhash_struct *)malloc(newhashmax * sizeof(*bufhash)); 124 if (np == NULL) 125 return; 126 for (i = 0; i < newhashmax; i++) 127 LIST_INIT(&np[i]); 128 129 /* Now reassign all existing buffers to their new hash chains. */ 130 for (i = 0; i < hashmax; i++) { 131 bp = LIST_FIRST(&bufhash[i]); 132 while(bp) { 133 nbp = LIST_NEXT(bp, b_hash); 134 LIST_REMOVE(bp, b_hash); 135 bp->b_hashval = vl_hash(bp->b_vp, bp->b_lblkno); 136 LIST_INSERT_HEAD(&np[bp->b_hashval], bp, b_hash); 137 bp = nbp; 138 } 139 } 140 141 /* Switch over and clean up */ 142 free(bufhash); 143 bufhash = np; 144 hashmax = newhashmax; 145 hashmask = newhashmask; 146 } 147 148 /* Print statistics of buffer cache usage */ 149 void 150 bufstats(void) 151 { 152 printf("buffer cache: %d hits %d misses (%2.2f%%); hash width %d, depth %d\n", 153 cachehits, cachemisses, 154 (cachehits * 100.0) / (cachehits + cachemisses), 155 hashmax, max_depth); 156 } 157 158 /* 159 * Remove a buffer from the cache. 160 * Caller must remove the buffer from its free list. 161 */ 162 void 163 buf_destroy(struct ubuf * bp) 164 { 165 bp->b_flags |= B_NEEDCOMMIT; 166 LIST_REMOVE(bp, b_vnbufs); 167 LIST_REMOVE(bp, b_hash); 168 if (!(bp->b_flags & B_DONTFREE)) 169 free(bp->b_data); 170 free(bp); 171 --nbufs; 172 } 173 174 /* Remove a buffer from its free list. */ 175 void 176 bremfree(struct ubuf * bp) 177 { 178 struct bqueues *dp = NULL; 179 180 /* 181 * We only calculate the head of the freelist when removing 182 * the last element of the list as that is the only time that 183 * it is needed (e.g. to reset the tail pointer). 184 * 185 * NB: This makes an assumption about how tailq's are implemented. 186 */ 187 if (bp->b_flags & B_LOCKED) { 188 locked_queue_bytes -= bp->b_bcount; 189 --locked_queue_count; 190 } 191 if (TAILQ_NEXT(bp, b_freelist) == NULL) { 192 for (dp = bufqueues; dp < &bufqueues[BQUEUES]; dp++) 193 if (dp->tqh_last == &bp->b_freelist.tqe_next) 194 break; 195 if (dp == &bufqueues[BQUEUES]) 196 errx(1, "bremfree: lost tail"); 197 } 198 ++bp->b_vp->v_usecount; 199 TAILQ_REMOVE(dp, bp, b_freelist); 200 } 201 202 /* Return a buffer if it is in the cache, otherwise return NULL. */ 203 struct ubuf * 204 incore(struct uvnode * vp, int lbn) 205 { 206 struct ubuf *bp; 207 int hash, depth; 208 209 hash = vl_hash(vp, lbn); 210 /* XXX use a real hash instead. */ 211 depth = 0; 212 LIST_FOREACH(bp, &bufhash[hash], b_hash) { 213 if (++depth > max_depth) 214 max_depth = depth; 215 assert(depth <= nbufs); 216 if (bp->b_vp == vp && bp->b_lblkno == lbn) { 217 return bp; 218 } 219 } 220 return NULL; 221 } 222 223 /* 224 * Return a buffer of the given size, lbn and uvnode. 225 * If none is in core, make a new one. 226 */ 227 struct ubuf * 228 getblk(struct uvnode * vp, daddr_t lbn, int size) 229 { 230 struct ubuf *bp; 231 #ifdef DEBUG 232 static int warned; 233 #endif 234 235 /* 236 * First check the buffer cache lists. 237 * We might sometimes need to resize a buffer. If we are growing 238 * the buffer, its contents are invalid; but shrinking is okay. 239 */ 240 if ((bp = incore(vp, lbn)) != NULL) { 241 assert(!(bp->b_flags & B_NEEDCOMMIT)); 242 assert(!(bp->b_flags & B_BUSY)); 243 bp->b_flags |= B_BUSY; 244 bremfree(bp); 245 if (bp->b_bcount == size) 246 return bp; 247 else if (bp->b_bcount > size) { 248 assert(!(bp->b_flags & B_DELWRI)); 249 bp->b_bcount = size; 250 bp->b_data = realloc(bp->b_data, size); 251 return bp; 252 } 253 254 buf_destroy(bp); 255 bp = NULL; 256 } 257 258 /* 259 * Not on the list. 260 * Get a new block of the appropriate size and use that. 261 * If not enough space, free blocks from the AGE and LRU lists 262 * to make room. 263 */ 264 while (nbufs >= maxbufs + locked_queue_count) { 265 bp = TAILQ_FIRST(&bufqueues[BQ_AGE]); 266 if (bp) 267 TAILQ_REMOVE(&bufqueues[BQ_AGE], bp, b_freelist); 268 if (bp == NULL) { 269 bp = TAILQ_FIRST(&bufqueues[BQ_LRU]); 270 if (bp) 271 TAILQ_REMOVE(&bufqueues[BQ_LRU], bp, 272 b_freelist); 273 } 274 if (bp) { 275 if (bp->b_flags & B_DELWRI) 276 VOP_STRATEGY(bp); 277 buf_destroy(bp); 278 break; 279 } 280 #ifdef DEBUG 281 else if (!warned) { 282 warnx("allocating more than %d buffers", maxbufs); 283 ++warned; 284 } 285 #endif 286 break; 287 } 288 ++nbufs; 289 bp = (struct ubuf *) malloc(sizeof(*bp)); 290 memset(bp, 0, sizeof(*bp)); 291 bp->b_data = malloc(size); 292 memset(bp->b_data, 0, size); 293 294 bp->b_vp = vp; 295 bp->b_blkno = bp->b_lblkno = lbn; 296 bp->b_bcount = size; 297 bp->b_hashval = vl_hash(vp, lbn); 298 LIST_INSERT_HEAD(&bufhash[bp->b_hashval], bp, b_hash); 299 LIST_INSERT_HEAD(&vp->v_cleanblkhd, bp, b_vnbufs); 300 bp->b_flags = B_BUSY; 301 302 return bp; 303 } 304 305 /* Write a buffer to disk according to its strategy routine. */ 306 void 307 bwrite(struct ubuf * bp) 308 { 309 bp->b_flags &= ~(B_READ | B_DONE | B_DELWRI | B_LOCKED); 310 VOP_STRATEGY(bp); 311 bp->b_flags |= B_DONE; 312 reassignbuf(bp, bp->b_vp); 313 brelse(bp); 314 } 315 316 /* Put a buffer back on its free list, clear B_BUSY. */ 317 void 318 brelse(struct ubuf * bp) 319 { 320 int age; 321 322 assert(!(bp->b_flags & B_NEEDCOMMIT)); 323 assert(bp->b_flags & B_BUSY); 324 325 age = bp->b_flags & B_AGE; 326 bp->b_flags &= ~(B_BUSY | B_AGE); 327 if (bp->b_flags & B_INVAL) { 328 buf_destroy(bp); 329 return; 330 } 331 if (bp->b_flags & B_LOCKED) { 332 locked_queue_bytes += bp->b_bcount; 333 ++locked_queue_count; 334 TAILQ_INSERT_TAIL(&bufqueues[BQ_LOCKED], bp, b_freelist); 335 } else if (age) { 336 TAILQ_INSERT_TAIL(&bufqueues[BQ_AGE], bp, b_freelist); 337 } else { 338 TAILQ_INSERT_TAIL(&bufqueues[BQ_LRU], bp, b_freelist); 339 } 340 --bp->b_vp->v_usecount; 341 342 /* Move to the front of the hash chain */ 343 if (LIST_FIRST(&bufhash[bp->b_hashval]) != bp) { 344 LIST_REMOVE(bp, b_hash); 345 LIST_INSERT_HEAD(&bufhash[bp->b_hashval], bp, b_hash); 346 } 347 } 348 349 /* Read the given block from disk, return it B_BUSY. */ 350 int 351 bread(struct uvnode * vp, daddr_t lbn, int size, struct ucred * unused, 352 struct ubuf ** bpp) 353 { 354 struct ubuf *bp; 355 daddr_t daddr; 356 int error; 357 358 bp = getblk(vp, lbn, size); 359 *bpp = bp; 360 if (bp->b_flags & (B_DELWRI | B_DONE)) { 361 ++cachehits; 362 return 0; 363 } 364 ++cachemisses; 365 366 /* 367 * Not found. Need to find that block's location on disk, 368 * and load it in. 369 */ 370 daddr = -1; 371 error = VOP_BMAP(vp, lbn, &daddr); 372 bp->b_blkno = daddr; 373 if (daddr >= 0) { 374 bp->b_flags |= B_READ; 375 return VOP_STRATEGY(bp); 376 } 377 memset(bp->b_data, 0, bp->b_bcount); 378 return 0; 379 } 380 381 /* Move a buffer between dirty and clean block lists. */ 382 void 383 reassignbuf(struct ubuf * bp, struct uvnode * vp) 384 { 385 LIST_REMOVE(bp, b_vnbufs); 386 if (bp->b_flags & B_DELWRI) { 387 LIST_INSERT_HEAD(&vp->v_dirtyblkhd, bp, b_vnbufs); 388 } else { 389 LIST_INSERT_HEAD(&vp->v_cleanblkhd, bp, b_vnbufs); 390 } 391 } 392 393 #ifdef DEBUG 394 void 395 dump_free_lists(void) 396 { 397 struct ubuf *bp; 398 int i; 399 400 for (i = 0; i <= BQ_LOCKED; i++) { 401 printf("==> free list %d:\n", i); 402 TAILQ_FOREACH(bp, &bufqueues[i], b_freelist) { 403 printf("vp %p lbn %" PRId64 " flags %lx\n", 404 bp->b_vp, bp->b_lblkno, bp->b_flags); 405 } 406 } 407 } 408 #endif 409