1 /* $NetBSD: subr_disk.c,v 1.28 2000/02/07 20:16:58 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 (bp->b_cylinder == bq->b_cylinder && 144 bp->b_rawblkno < bq->b_rawblkno)) { 145 while ((nbq = BUFQ_NEXT(bq)) != NULL) { 146 /* 147 * Check for an ``inversion'' in the normally ascending 148 * cylinder numbers, indicating the start of the second 149 * request list. 150 */ 151 if (nbq->b_cylinder < bq->b_cylinder) { 152 /* 153 * Search the second request list for the first 154 * request at a larger cylinder number. We go 155 * before that; if there is no such request, we 156 * go at end. 157 */ 158 do { 159 if (bp->b_cylinder < nbq->b_cylinder) 160 goto insert; 161 if (bp->b_cylinder == nbq->b_cylinder && 162 bp->b_rawblkno < nbq->b_rawblkno) 163 goto insert; 164 bq = nbq; 165 } while ((nbq = BUFQ_NEXT(bq)) != NULL); 166 goto insert; /* after last */ 167 } 168 bq = BUFQ_NEXT(bq); 169 } 170 /* 171 * No inversions... we will go after the last, and 172 * be the first request in the second request list. 173 */ 174 goto insert; 175 } 176 /* 177 * Request is at/after the current request... 178 * sort in the first request list. 179 */ 180 while ((nbq = BUFQ_NEXT(bq)) != NULL) { 181 /* 182 * We want to go after the current request if there is an 183 * inversion after it (i.e. it is the end of the first 184 * request list), or if the next request is a larger cylinder 185 * than our request. 186 */ 187 if (nbq->b_cylinder < bq->b_cylinder || 188 bp->b_cylinder < nbq->b_cylinder || 189 (bp->b_cylinder == nbq->b_cylinder && 190 bp->b_rawblkno < nbq->b_rawblkno)) 191 goto insert; 192 bq = nbq; 193 } 194 /* 195 * Neither a second list nor a larger request... we go at the end of 196 * the first list, which is the same as the end of the whole schebang. 197 */ 198 insert: BUFQ_INSERT_AFTER(bufq, bq, bp); 199 } 200 201 /* 202 * Seek sort for disks. This version sorts based on b_rawblkno, which 203 * indicates the block number. 204 * 205 * As before, there are actually two queues, sorted in ascendening block 206 * order. The first queue holds those requests which are positioned after 207 * the current block (in the first request); the second holds requests which 208 * came in after their block number was passed. Thus we implement a one-way 209 * scan, retracting after reaching the end of the driver to the first request 210 * on the second queue, at which time it becomes the first queue. 211 * 212 * A one-way scan is natural because of the way UNIX read-ahead blocks are 213 * allocated. 214 * 215 * This is further adjusted by any `barriers' which may exist in the queue. 216 * The bufq points to the last such ordered request. 217 */ 218 void 219 disksort_blkno(bufq, bp) 220 struct buf_queue *bufq; 221 struct buf *bp; 222 { 223 struct buf *bq, *nbq; 224 225 /* 226 * If there are ordered requests on the queue, we must start 227 * the elevator sort after the last of these. 228 */ 229 if ((bq = bufq->bq_barrier) == NULL) 230 bq = BUFQ_FIRST(bufq); 231 232 /* 233 * If the queue is empty, or if it's an ordered request, 234 * it's easy; we just go on the end. 235 */ 236 if (bq == NULL || (bp->b_flags & B_ORDERED) != 0) { 237 BUFQ_INSERT_TAIL(bufq, bp); 238 return; 239 } 240 241 /* 242 * If we lie after the first (currently active) request, then we 243 * must locate the second request list and add ourselves to it. 244 */ 245 if (bp->b_rawblkno < bq->b_rawblkno) { 246 while ((nbq = BUFQ_NEXT(bq)) != NULL) { 247 /* 248 * Check for an ``inversion'' in the normally ascending 249 * block numbers, indicating the start of the second 250 * request list. 251 */ 252 if (nbq->b_rawblkno < bq->b_rawblkno) { 253 /* 254 * Search the second request list for the first 255 * request at a larger block number. We go 256 * after that; if there is no such request, we 257 * go at the end. 258 */ 259 do { 260 if (bp->b_rawblkno < nbq->b_rawblkno) 261 goto insert; 262 bq = nbq; 263 } while ((nbq = BUFQ_NEXT(bq)) != NULL); 264 goto insert; /* after last */ 265 } 266 bq = BUFQ_NEXT(bq); 267 } 268 /* 269 * No inversions... we will go after the last, and 270 * be the first request in the second request list. 271 */ 272 goto insert; 273 } 274 /* 275 * Request is at/after the current request... 276 * sort in the first request list. 277 */ 278 while ((nbq = BUFQ_NEXT(bq)) != NULL) { 279 /* 280 * We want to go after the current request if there is an 281 * inversion after it (i.e. it is the end of the first 282 * request list), or if the next request is a larger cylinder 283 * than our request. 284 */ 285 if (nbq->b_rawblkno < bq->b_rawblkno || 286 bp->b_rawblkno < nbq->b_rawblkno) 287 goto insert; 288 bq = nbq; 289 } 290 /* 291 * Neither a second list nor a larger request... we go at the end of 292 * the first list, which is the same as the end of the whole schebang. 293 */ 294 insert: BUFQ_INSERT_AFTER(bufq, bq, bp); 295 } 296 297 /* 298 * Seek non-sort for disks. This version simply inserts requests at 299 * the tail of the queue. 300 */ 301 void 302 disksort_tail(bufq, bp) 303 struct buf_queue *bufq; 304 struct buf *bp; 305 { 306 307 BUFQ_INSERT_TAIL(bufq, bp); 308 } 309 310 /* 311 * Compute checksum for disk label. 312 */ 313 u_int 314 dkcksum(lp) 315 register struct disklabel *lp; 316 { 317 register u_short *start, *end; 318 register u_short sum = 0; 319 320 start = (u_short *)lp; 321 end = (u_short *)&lp->d_partitions[lp->d_npartitions]; 322 while (start < end) 323 sum ^= *start++; 324 return (sum); 325 } 326 327 /* 328 * Disk error is the preface to plaintive error messages 329 * about failing disk transfers. It prints messages of the form 330 331 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d) 332 333 * if the offset of the error in the transfer and a disk label 334 * are both available. blkdone should be -1 if the position of the error 335 * is unknown; the disklabel pointer may be null from drivers that have not 336 * been converted to use them. The message is printed with printf 337 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority. 338 * The message should be completed (with at least a newline) with printf 339 * or addlog, respectively. There is no trailing space. 340 */ 341 void 342 diskerr(bp, dname, what, pri, blkdone, lp) 343 register struct buf *bp; 344 char *dname, *what; 345 int pri, blkdone; 346 register struct disklabel *lp; 347 { 348 int unit = DISKUNIT(bp->b_dev), part = DISKPART(bp->b_dev); 349 register void (*pr) __P((const char *, ...)); 350 char partname = 'a' + part; 351 int sn; 352 353 if (pri != LOG_PRINTF) { 354 static const char fmt[] = ""; 355 log(pri, fmt); 356 pr = addlog; 357 } else 358 pr = printf; 359 (*pr)("%s%d%c: %s %sing fsbn ", dname, unit, partname, what, 360 bp->b_flags & B_READ ? "read" : "writ"); 361 sn = bp->b_blkno; 362 if (bp->b_bcount <= DEV_BSIZE) 363 (*pr)("%d", sn); 364 else { 365 if (blkdone >= 0) { 366 sn += blkdone; 367 (*pr)("%d of ", sn); 368 } 369 (*pr)("%d-%d", bp->b_blkno, 370 bp->b_blkno + (bp->b_bcount - 1) / DEV_BSIZE); 371 } 372 if (lp && (blkdone >= 0 || bp->b_bcount <= lp->d_secsize)) { 373 sn += lp->d_partitions[part].p_offset; 374 (*pr)(" (%s%d bn %d; cn %d", dname, unit, sn, 375 sn / lp->d_secpercyl); 376 sn %= lp->d_secpercyl; 377 (*pr)(" tn %d sn %d)", sn / lp->d_nsectors, sn % lp->d_nsectors); 378 } 379 } 380 381 /* 382 * Initialize the disklist. Called by main() before autoconfiguration. 383 */ 384 void 385 disk_init() 386 { 387 388 TAILQ_INIT(&disklist); 389 disk_count = 0; 390 } 391 392 /* 393 * Searches the disklist for the disk corresponding to the 394 * name provided. 395 */ 396 struct disk * 397 disk_find(name) 398 char *name; 399 { 400 struct disk *diskp; 401 402 if ((name == NULL) || (disk_count <= 0)) 403 return (NULL); 404 405 for (diskp = disklist.tqh_first; diskp != NULL; 406 diskp = diskp->dk_link.tqe_next) 407 if (strcmp(diskp->dk_name, name) == 0) 408 return (diskp); 409 410 return (NULL); 411 } 412 413 /* 414 * Attach a disk. 415 */ 416 void 417 disk_attach(diskp) 418 struct disk *diskp; 419 { 420 int s; 421 422 /* 423 * Allocate and initialize the disklabel structures. Note that 424 * it's not safe to sleep here, since we're probably going to be 425 * called during autoconfiguration. 426 */ 427 diskp->dk_label = malloc(sizeof(struct disklabel), M_DEVBUF, M_NOWAIT); 428 diskp->dk_cpulabel = malloc(sizeof(struct cpu_disklabel), M_DEVBUF, 429 M_NOWAIT); 430 if ((diskp->dk_label == NULL) || (diskp->dk_cpulabel == NULL)) 431 panic("disk_attach: can't allocate storage for disklabel"); 432 433 memset(diskp->dk_label, 0, sizeof(struct disklabel)); 434 memset(diskp->dk_cpulabel, 0, sizeof(struct cpu_disklabel)); 435 436 /* 437 * Set the attached timestamp. 438 */ 439 s = splclock(); 440 diskp->dk_attachtime = mono_time; 441 splx(s); 442 443 /* 444 * Link into the disklist. 445 */ 446 TAILQ_INSERT_TAIL(&disklist, diskp, dk_link); 447 ++disk_count; 448 } 449 450 /* 451 * Detach a disk. 452 */ 453 void 454 disk_detach(diskp) 455 struct disk *diskp; 456 { 457 458 /* 459 * Remove from the disklist. 460 */ 461 if (--disk_count < 0) 462 panic("disk_detach: disk_count < 0"); 463 TAILQ_REMOVE(&disklist, diskp, dk_link); 464 465 /* 466 * Free the space used by the disklabel structures. 467 */ 468 free(diskp->dk_label, M_DEVBUF); 469 free(diskp->dk_cpulabel, M_DEVBUF); 470 } 471 472 /* 473 * Increment a disk's busy counter. If the counter is going from 474 * 0 to 1, set the timestamp. 475 */ 476 void 477 disk_busy(diskp) 478 struct disk *diskp; 479 { 480 int s; 481 482 /* 483 * XXX We'd like to use something as accurate as microtime(), 484 * but that doesn't depend on the system TOD clock. 485 */ 486 if (diskp->dk_busy++ == 0) { 487 s = splclock(); 488 diskp->dk_timestamp = mono_time; 489 splx(s); 490 } 491 } 492 493 /* 494 * Decrement a disk's busy counter, increment the byte count, total busy 495 * time, and reset the timestamp. 496 */ 497 void 498 disk_unbusy(diskp, bcount) 499 struct disk *diskp; 500 long bcount; 501 { 502 int s; 503 struct timeval dv_time, diff_time; 504 505 if (diskp->dk_busy-- == 0) { 506 printf("%s: dk_busy < 0\n", diskp->dk_name); 507 panic("disk_unbusy"); 508 } 509 510 s = splclock(); 511 dv_time = mono_time; 512 splx(s); 513 514 timersub(&dv_time, &diskp->dk_timestamp, &diff_time); 515 timeradd(&diskp->dk_time, &diff_time, &diskp->dk_time); 516 517 diskp->dk_timestamp = dv_time; 518 if (bcount > 0) { 519 diskp->dk_bytes += bcount; 520 diskp->dk_xfer++; 521 } 522 } 523 524 /* 525 * Reset the metrics counters on the given disk. Note that we cannot 526 * reset the busy counter, as it may case a panic in disk_unbusy(). 527 * We also must avoid playing with the timestamp information, as it 528 * may skew any pending transfer results. 529 */ 530 void 531 disk_resetstat(diskp) 532 struct disk *diskp; 533 { 534 int s = splbio(), t; 535 536 diskp->dk_xfer = 0; 537 diskp->dk_bytes = 0; 538 539 t = splclock(); 540 diskp->dk_attachtime = mono_time; 541 splx(t); 542 543 timerclear(&diskp->dk_time); 544 545 splx(s); 546 } 547