subr_disk.c revision 1.26
1 /* $NetBSD: subr_disk.c,v 1.26 2000/01/21 23:20:51 thorpej Exp $ */ 2 3 /*- 4 * Copyright (c) 1996, 1997, 1999, 2000 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the NetBSD 22 * Foundation, Inc. and its contributors. 23 * 4. Neither the name of The NetBSD Foundation nor the names of its 24 * contributors may be used to endorse or promote products derived 25 * from this software without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 37 * POSSIBILITY OF SUCH DAMAGE. 38 */ 39 40 /* 41 * Copyright (c) 1982, 1986, 1988, 1993 42 * The Regents of the University of California. All rights reserved. 43 * (c) UNIX System Laboratories, Inc. 44 * All or some portions of this file are derived from material licensed 45 * to the University of California by American Telephone and Telegraph 46 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 47 * the permission of UNIX System Laboratories, Inc. 48 * 49 * Redistribution and use in source and binary forms, with or without 50 * modification, are permitted provided that the following conditions 51 * are met: 52 * 1. Redistributions of source code must retain the above copyright 53 * notice, this list of conditions and the following disclaimer. 54 * 2. Redistributions in binary form must reproduce the above copyright 55 * notice, this list of conditions and the following disclaimer in the 56 * documentation and/or other materials provided with the distribution. 57 * 3. All advertising materials mentioning features or use of this software 58 * must display the following acknowledgement: 59 * This product includes software developed by the University of 60 * California, Berkeley and its contributors. 61 * 4. Neither the name of the University nor the names of its contributors 62 * may be used to endorse or promote products derived from this software 63 * without specific prior written permission. 64 * 65 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 66 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 67 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 68 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 69 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 70 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 71 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 72 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 73 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 74 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 75 * SUCH DAMAGE. 76 * 77 * @(#)ufs_disksubr.c 8.5 (Berkeley) 1/21/94 78 */ 79 80 #include <sys/param.h> 81 #include <sys/systm.h> 82 #include <sys/kernel.h> 83 #include <sys/malloc.h> 84 #include <sys/buf.h> 85 #include <sys/syslog.h> 86 #include <sys/time.h> 87 #include <sys/disklabel.h> 88 #include <sys/disk.h> 89 90 /* 91 * A global list of all disks attached to the system. May grow or 92 * shrink over time. 93 */ 94 struct disklist_head disklist; /* TAILQ_HEAD */ 95 int disk_count; /* number of drives in global disklist */ 96 97 /* 98 * Seek sort for disks. We depend on the driver which calls us using b_resid 99 * as the current cylinder number. 100 * 101 * The argument bufq is an I/O queue for the device, on which there are 102 * actually two queues, sorted in ascending cylinder order. The first 103 * queue holds those requests which are positioned after the current 104 * cylinder (in the first request); the second holds requests which came 105 * in after their cylinder number was passed. Thus we implement a one-way 106 * scan, retracting after reaching the end of the drive to the first request 107 * on the second queue, at which time it becomes the first queue. 108 * 109 * A one-way scan is natural because of the way UNIX read-ahead blocks are 110 * allocated. 111 * 112 * This is further adjusted by any `barriers' which may exist in the queue. 113 * The bufq points to the last such ordered request. 114 */ 115 void 116 disksort_cylinder(bufq, bp) 117 struct buf_queue *bufq; 118 struct buf *bp; 119 { 120 struct buf *bq, *nbq; 121 122 /* 123 * If there are ordered requests on the queue, we must start 124 * the elevator sort after the last of these. 125 */ 126 if ((bq = bufq->bq_barrier) == NULL) 127 bq = BUFQ_FIRST(bufq); 128 129 /* 130 * If the queue is empty, of if it's an ordered request, 131 * it's easy; we just go on the end. 132 */ 133 if (bq == NULL || (bp->b_flags & B_ORDERED) != 0) { 134 BUFQ_INSERT_TAIL(bufq, bp); 135 return; 136 } 137 138 /* 139 * If we lie after the first (currently active) request, then we 140 * must locate the second request list and add ourselves to it. 141 */ 142 if (bp->b_cylinder < bq->b_cylinder) { 143 while ((nbq = BUFQ_NEXT(bq)) != NULL) { 144 /* 145 * Check for an ``inversion'' in the normally ascending 146 * cylinder numbers, indicating the start of the second 147 * request list. 148 */ 149 if (nbq->b_cylinder < bq->b_cylinder) { 150 /* 151 * Search the second request list for the first 152 * request at a larger cylinder number. We go 153 * before that; if there is no such request, we 154 * go at end. 155 */ 156 do { 157 if (bp->b_cylinder < nbq->b_cylinder) 158 goto insert; 159 if (bp->b_cylinder == nbq->b_cylinder && 160 bp->b_blkno < nbq->b_blkno) 161 goto insert; 162 bq = nbq; 163 } while ((nbq = BUFQ_NEXT(bq)) != NULL); 164 goto insert; /* after last */ 165 } 166 bq = BUFQ_NEXT(bq); 167 } 168 /* 169 * No inversions... we will go after the last, and 170 * be the first request in the second request list. 171 */ 172 goto insert; 173 } 174 /* 175 * Request is at/after the current request... 176 * sort in the first request list. 177 */ 178 while ((nbq = BUFQ_NEXT(bq)) != NULL) { 179 /* 180 * We want to go after the current request if there is an 181 * inversion after it (i.e. it is the end of the first 182 * request list), or if the next request is a larger cylinder 183 * than our request. 184 */ 185 if (nbq->b_cylinder < bq->b_cylinder || 186 bp->b_cylinder < nbq->b_cylinder || 187 (bp->b_cylinder == nbq->b_cylinder && 188 bp->b_blkno < nbq->b_blkno)) 189 goto insert; 190 bq = nbq; 191 } 192 /* 193 * Neither a second list nor a larger request... we go at the end of 194 * the first list, which is the same as the end of the whole schebang. 195 */ 196 insert: BUFQ_INSERT_AFTER(bufq, bq, bp); 197 } 198 199 /* 200 * Seek sort for disks. This version sorts based on b_blkno, which 201 * indicates the block number. 202 * 203 * As before, there are actually two queues, sorted in ascendening block 204 * order. The first queue holds those requests which are positioned after 205 * the current block (in the first request); the second holds requests which 206 * came in after their block number was passed. Thus we implement a one-way 207 * scan, retracting after reaching the end of the driver to the first request 208 * on the second queue, at which time it becomes the first queue. 209 * 210 * A one-way scan is natural because of the way UNIX read-ahead blocks are 211 * allocated. 212 * 213 * This is further adjusted by any `barriers' which may exist in the queue. 214 * The bufq points to the last such ordered request. 215 */ 216 void 217 disksort_blkno(bufq, bp) 218 struct buf_queue *bufq; 219 struct buf *bp; 220 { 221 struct buf *bq, *nbq; 222 223 /* 224 * If there are ordered requests on the queue, we must start 225 * the elevator sort after the last of these. 226 */ 227 if ((bq = bufq->bq_barrier) == NULL) 228 bq = BUFQ_FIRST(bufq); 229 230 /* 231 * If the queue is empty, or if it's an ordered request, 232 * it's easy; we just go on the end. 233 */ 234 if (bq == NULL || (bp->b_flags & B_ORDERED) != 0) { 235 BUFQ_INSERT_TAIL(bufq, bp); 236 return; 237 } 238 239 /* 240 * If we lie after the first (currently active) request, then we 241 * must locate the second request list and add ourselves to it. 242 */ 243 if (bp->b_blkno < bq->b_blkno) { 244 while ((nbq = BUFQ_NEXT(bq)) != NULL) { 245 /* 246 * Check for an ``inversion'' in the normally ascending 247 * block numbers, indicating the start of the second 248 * request list. 249 */ 250 if (nbq->b_blkno < bq->b_blkno) { 251 /* 252 * Search the second request list for the first 253 * request at a larger block number. We go 254 * after that; if there is no such request, we 255 * go at the end. 256 */ 257 do { 258 if (bp->b_blkno < nbq->b_blkno) 259 goto insert; 260 bq = nbq; 261 } while ((nbq = BUFQ_NEXT(bq)) != NULL); 262 goto insert; /* after last */ 263 } 264 bq = BUFQ_NEXT(bq); 265 } 266 /* 267 * No inversions... we will go after the last, and 268 * be the first request in the second request list. 269 */ 270 goto insert; 271 } 272 /* 273 * Request is at/after the current request... 274 * sort in the first request list. 275 */ 276 while ((nbq = BUFQ_NEXT(bq)) != NULL) { 277 /* 278 * We want to go after the current request if there is an 279 * inversion after it (i.e. it is the end of the first 280 * request list), or if the next request is a larger cylinder 281 * than our request. 282 */ 283 if (nbq->b_blkno < bq->b_blkno || 284 bp->b_blkno < nbq->b_blkno) 285 goto insert; 286 bq = nbq; 287 } 288 /* 289 * Neither a second list nor a larger request... we go at the end of 290 * the first list, which is the same as the end of the whole schebang. 291 */ 292 insert: BUFQ_INSERT_AFTER(bufq, bq, bp); 293 } 294 295 /* 296 * Seek non-sort for disks. This version simply inserts requests at 297 * the tail of the queue. 298 */ 299 void 300 disksort_tail(bufq, bp) 301 struct buf_queue *bufq; 302 struct buf *bp; 303 { 304 305 BUFQ_INSERT_TAIL(bufq, bp); 306 } 307 308 /* 309 * Compute checksum for disk label. 310 */ 311 u_int 312 dkcksum(lp) 313 register struct disklabel *lp; 314 { 315 register u_short *start, *end; 316 register u_short sum = 0; 317 318 start = (u_short *)lp; 319 end = (u_short *)&lp->d_partitions[lp->d_npartitions]; 320 while (start < end) 321 sum ^= *start++; 322 return (sum); 323 } 324 325 /* 326 * Disk error is the preface to plaintive error messages 327 * about failing disk transfers. It prints messages of the form 328 329 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d) 330 331 * if the offset of the error in the transfer and a disk label 332 * are both available. blkdone should be -1 if the position of the error 333 * is unknown; the disklabel pointer may be null from drivers that have not 334 * been converted to use them. The message is printed with printf 335 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority. 336 * The message should be completed (with at least a newline) with printf 337 * or addlog, respectively. There is no trailing space. 338 */ 339 void 340 diskerr(bp, dname, what, pri, blkdone, lp) 341 register struct buf *bp; 342 char *dname, *what; 343 int pri, blkdone; 344 register struct disklabel *lp; 345 { 346 int unit = DISKUNIT(bp->b_dev), part = DISKPART(bp->b_dev); 347 register void (*pr) __P((const char *, ...)); 348 char partname = 'a' + part; 349 int sn; 350 351 if (pri != LOG_PRINTF) { 352 static const char fmt[] = ""; 353 log(pri, fmt); 354 pr = addlog; 355 } else 356 pr = printf; 357 (*pr)("%s%d%c: %s %sing fsbn ", dname, unit, partname, what, 358 bp->b_flags & B_READ ? "read" : "writ"); 359 sn = bp->b_blkno; 360 if (bp->b_bcount <= DEV_BSIZE) 361 (*pr)("%d", sn); 362 else { 363 if (blkdone >= 0) { 364 sn += blkdone; 365 (*pr)("%d of ", sn); 366 } 367 (*pr)("%d-%d", bp->b_blkno, 368 bp->b_blkno + (bp->b_bcount - 1) / DEV_BSIZE); 369 } 370 if (lp && (blkdone >= 0 || bp->b_bcount <= lp->d_secsize)) { 371 sn += lp->d_partitions[part].p_offset; 372 (*pr)(" (%s%d bn %d; cn %d", dname, unit, sn, 373 sn / lp->d_secpercyl); 374 sn %= lp->d_secpercyl; 375 (*pr)(" tn %d sn %d)", sn / lp->d_nsectors, sn % lp->d_nsectors); 376 } 377 } 378 379 /* 380 * Initialize the disklist. Called by main() before autoconfiguration. 381 */ 382 void 383 disk_init() 384 { 385 386 TAILQ_INIT(&disklist); 387 disk_count = 0; 388 } 389 390 /* 391 * Searches the disklist for the disk corresponding to the 392 * name provided. 393 */ 394 struct disk * 395 disk_find(name) 396 char *name; 397 { 398 struct disk *diskp; 399 400 if ((name == NULL) || (disk_count <= 0)) 401 return (NULL); 402 403 for (diskp = disklist.tqh_first; diskp != NULL; 404 diskp = diskp->dk_link.tqe_next) 405 if (strcmp(diskp->dk_name, name) == 0) 406 return (diskp); 407 408 return (NULL); 409 } 410 411 /* 412 * Attach a disk. 413 */ 414 void 415 disk_attach(diskp) 416 struct disk *diskp; 417 { 418 int s; 419 420 /* 421 * Allocate and initialize the disklabel structures. Note that 422 * it's not safe to sleep here, since we're probably going to be 423 * called during autoconfiguration. 424 */ 425 diskp->dk_label = malloc(sizeof(struct disklabel), M_DEVBUF, M_NOWAIT); 426 diskp->dk_cpulabel = malloc(sizeof(struct cpu_disklabel), M_DEVBUF, 427 M_NOWAIT); 428 if ((diskp->dk_label == NULL) || (diskp->dk_cpulabel == NULL)) 429 panic("disk_attach: can't allocate storage for disklabel"); 430 431 memset(diskp->dk_label, 0, sizeof(struct disklabel)); 432 memset(diskp->dk_cpulabel, 0, sizeof(struct cpu_disklabel)); 433 434 /* 435 * Set the attached timestamp. 436 */ 437 s = splclock(); 438 diskp->dk_attachtime = mono_time; 439 splx(s); 440 441 /* 442 * Link into the disklist. 443 */ 444 TAILQ_INSERT_TAIL(&disklist, diskp, dk_link); 445 ++disk_count; 446 } 447 448 /* 449 * Detach a disk. 450 */ 451 void 452 disk_detach(diskp) 453 struct disk *diskp; 454 { 455 456 /* 457 * Remove from the disklist. 458 */ 459 if (--disk_count < 0) 460 panic("disk_detach: disk_count < 0"); 461 TAILQ_REMOVE(&disklist, diskp, dk_link); 462 463 /* 464 * Free the space used by the disklabel structures. 465 */ 466 free(diskp->dk_label, M_DEVBUF); 467 free(diskp->dk_cpulabel, M_DEVBUF); 468 } 469 470 /* 471 * Increment a disk's busy counter. If the counter is going from 472 * 0 to 1, set the timestamp. 473 */ 474 void 475 disk_busy(diskp) 476 struct disk *diskp; 477 { 478 int s; 479 480 /* 481 * XXX We'd like to use something as accurate as microtime(), 482 * but that doesn't depend on the system TOD clock. 483 */ 484 if (diskp->dk_busy++ == 0) { 485 s = splclock(); 486 diskp->dk_timestamp = mono_time; 487 splx(s); 488 } 489 } 490 491 /* 492 * Decrement a disk's busy counter, increment the byte count, total busy 493 * time, and reset the timestamp. 494 */ 495 void 496 disk_unbusy(diskp, bcount) 497 struct disk *diskp; 498 long bcount; 499 { 500 int s; 501 struct timeval dv_time, diff_time; 502 503 if (diskp->dk_busy-- == 0) { 504 printf("%s: dk_busy < 0\n", diskp->dk_name); 505 panic("disk_unbusy"); 506 } 507 508 s = splclock(); 509 dv_time = mono_time; 510 splx(s); 511 512 timersub(&dv_time, &diskp->dk_timestamp, &diff_time); 513 timeradd(&diskp->dk_time, &diff_time, &diskp->dk_time); 514 515 diskp->dk_timestamp = dv_time; 516 if (bcount > 0) { 517 diskp->dk_bytes += bcount; 518 diskp->dk_xfer++; 519 } 520 } 521 522 /* 523 * Reset the metrics counters on the given disk. Note that we cannot 524 * reset the busy counter, as it may case a panic in disk_unbusy(). 525 * We also must avoid playing with the timestamp information, as it 526 * may skew any pending transfer results. 527 */ 528 void 529 disk_resetstat(diskp) 530 struct disk *diskp; 531 { 532 int s = splbio(), t; 533 534 diskp->dk_xfer = 0; 535 diskp->dk_bytes = 0; 536 537 t = splclock(); 538 diskp->dk_attachtime = mono_time; 539 splx(t); 540 541 timerclear(&diskp->dk_time); 542 543 splx(s); 544 } 545
Indexes created Thu Sep 25 16:09:42 GMT 2025