1 /* 2 * Copyright (c) 2003,2004,2009 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@backplane.com> 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 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 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * ---------------------------------------------------------------------------- 35 * "THE BEER-WARE LICENSE" (Revision 42): 36 * <phk@FreeBSD.ORG> wrote this file. As long as you retain this notice you 37 * can do whatever you want with this stuff. If we meet some day, and you think 38 * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp 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 * $FreeBSD: src/sys/kern/subr_disk.c,v 1.20.2.6 2001/10/05 07:14:57 peter Exp $ 79 * $FreeBSD: src/sys/ufs/ufs/ufs_disksubr.c,v 1.44.2.3 2001/03/05 05:42:19 obrien Exp $ 80 * $DragonFly: src/sys/kern/subr_disk.c,v 1.40 2008/06/05 18:06:32 swildner Exp $ 81 */ 82 83 #include <sys/param.h> 84 #include <sys/systm.h> 85 #include <sys/kernel.h> 86 #include <sys/proc.h> 87 #include <sys/sysctl.h> 88 #include <sys/buf.h> 89 #include <sys/conf.h> 90 #include <sys/disklabel.h> 91 #include <sys/disklabel32.h> 92 #include <sys/disklabel64.h> 93 #include <sys/diskslice.h> 94 #include <sys/diskmbr.h> 95 #include <sys/disk.h> 96 #include <sys/kerneldump.h> 97 #include <sys/malloc.h> 98 #include <sys/sysctl.h> 99 #include <machine/md_var.h> 100 #include <sys/ctype.h> 101 #include <sys/syslog.h> 102 #include <sys/device.h> 103 #include <sys/msgport.h> 104 #include <sys/msgport2.h> 105 #include <sys/buf2.h> 106 #include <sys/devfs.h> 107 #include <sys/thread.h> 108 #include <sys/thread2.h> 109 #include <sys/dsched.h> 110 #include <sys/queue.h> 111 #include <sys/lock.h> 112 113 static MALLOC_DEFINE(M_DISK, "disk", "disk data"); 114 static int disk_debug_enable = 0; 115 116 static void disk_msg_autofree_reply(lwkt_port_t, lwkt_msg_t); 117 static void disk_msg_core(void *); 118 static int disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe); 119 static void disk_probe(struct disk *dp, int reprobe); 120 static void _setdiskinfo(struct disk *disk, struct disk_info *info); 121 static void bioqwritereorder(struct bio_queue_head *bioq); 122 static void disk_cleanserial(char *serno); 123 124 static d_open_t diskopen; 125 static d_close_t diskclose; 126 static d_ioctl_t diskioctl; 127 static d_strategy_t diskstrategy; 128 static d_psize_t diskpsize; 129 static d_clone_t diskclone; 130 static d_dump_t diskdump; 131 132 static LIST_HEAD(, disk) disklist = LIST_HEAD_INITIALIZER(&disklist); 133 static struct lwkt_token disklist_token; 134 135 static struct dev_ops disk_ops = { 136 { "disk", 0, D_DISK }, 137 .d_open = diskopen, 138 .d_close = diskclose, 139 .d_read = physread, 140 .d_write = physwrite, 141 .d_ioctl = diskioctl, 142 .d_strategy = diskstrategy, 143 .d_dump = diskdump, 144 .d_psize = diskpsize, 145 .d_clone = diskclone 146 }; 147 148 static struct objcache *disk_msg_cache; 149 150 struct objcache_malloc_args disk_msg_malloc_args = { 151 sizeof(struct disk_msg), M_DISK }; 152 153 static struct lwkt_port disk_dispose_port; 154 static struct lwkt_port disk_msg_port; 155 156 static int 157 disk_debug(int level, char *fmt, ...) 158 { 159 __va_list ap; 160 161 __va_start(ap, fmt); 162 if (level <= disk_debug_enable) 163 kvprintf(fmt, ap); 164 __va_end(ap); 165 166 return 0; 167 } 168 169 static int 170 disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe) 171 { 172 struct disk_info *info = &dp->d_info; 173 struct diskslice *sp = &dp->d_slice->dss_slices[slice]; 174 disklabel_ops_t ops; 175 struct partinfo part; 176 const char *msg; 177 cdev_t ndev; 178 int sno; 179 u_int i; 180 181 disk_debug(2, 182 "disk_probe_slice (begin): %s (%s)\n", 183 dev->si_name, dp->d_cdev->si_name); 184 185 sno = slice ? slice - 1 : 0; 186 187 ops = &disklabel32_ops; 188 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info); 189 if (msg && !strcmp(msg, "no disk label")) { 190 ops = &disklabel64_ops; 191 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info); 192 } 193 if (msg == NULL) { 194 if (slice != WHOLE_DISK_SLICE) 195 ops->op_adjust_label_reserved(dp->d_slice, slice, sp); 196 else 197 sp->ds_reserved = 0; 198 199 sp->ds_ops = ops; 200 for (i = 0; i < ops->op_getnumparts(sp->ds_label); i++) { 201 ops->op_loadpartinfo(sp->ds_label, i, &part); 202 if (part.fstype) { 203 if (reprobe && 204 (ndev = devfs_find_device_by_name("%s%c", 205 dev->si_name, 'a' + i)) 206 ) { 207 /* 208 * Device already exists and 209 * is still valid. 210 */ 211 ndev->si_flags |= SI_REPROBE_TEST; 212 } else { 213 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops, 214 dkmakeminor(dkunit(dp->d_cdev), 215 slice, i), 216 UID_ROOT, GID_OPERATOR, 0640, 217 "%s%c", dev->si_name, 'a'+ i); 218 ndev->si_disk = dp; 219 if (dp->d_info.d_serialno) { 220 make_dev_alias(ndev, 221 "serno/%s.s%d%c", 222 dp->d_info.d_serialno, 223 sno, 'a' + i); 224 } 225 ndev->si_flags |= SI_REPROBE_TEST; 226 } 227 } 228 } 229 } else if (info->d_dsflags & DSO_COMPATLABEL) { 230 msg = NULL; 231 if (sp->ds_size >= 0x100000000ULL) 232 ops = &disklabel64_ops; 233 else 234 ops = &disklabel32_ops; 235 sp->ds_label = ops->op_clone_label(info, sp); 236 } else { 237 if (sp->ds_type == DOSPTYP_386BSD || /* XXX */ 238 sp->ds_type == DOSPTYP_NETBSD || 239 sp->ds_type == DOSPTYP_OPENBSD) { 240 log(LOG_WARNING, "%s: cannot find label (%s)\n", 241 dev->si_name, msg); 242 } 243 } 244 245 if (msg == NULL) { 246 sp->ds_wlabel = FALSE; 247 } 248 249 return (msg ? EINVAL : 0); 250 } 251 252 /* 253 * This routine is only called for newly minted drives or to reprobe 254 * a drive with no open slices. disk_probe_slice() is called directly 255 * when reprobing partition changes within slices. 256 */ 257 static void 258 disk_probe(struct disk *dp, int reprobe) 259 { 260 struct disk_info *info = &dp->d_info; 261 cdev_t dev = dp->d_cdev; 262 cdev_t ndev; 263 int error, i, sno; 264 struct diskslices *osp; 265 struct diskslice *sp; 266 267 KKASSERT (info->d_media_blksize != 0); 268 269 osp = dp->d_slice; 270 dp->d_slice = dsmakeslicestruct(BASE_SLICE, info); 271 disk_debug(1, "disk_probe (begin): %s\n", dp->d_cdev->si_name); 272 273 error = mbrinit(dev, info, &(dp->d_slice)); 274 if (error) { 275 dsgone(&osp); 276 return; 277 } 278 279 for (i = 0; i < dp->d_slice->dss_nslices; i++) { 280 /* 281 * Ignore the whole-disk slice, it has already been created. 282 */ 283 if (i == WHOLE_DISK_SLICE) 284 continue; 285 sp = &dp->d_slice->dss_slices[i]; 286 287 /* 288 * Handle s0. s0 is a compatibility slice if there are no 289 * other slices and it has not otherwise been set up, else 290 * we ignore it. 291 */ 292 if (i == COMPATIBILITY_SLICE) { 293 sno = 0; 294 if (sp->ds_type == 0 && 295 dp->d_slice->dss_nslices == BASE_SLICE) { 296 sp->ds_size = info->d_media_blocks; 297 sp->ds_reserved = 0; 298 } 299 } else { 300 sno = i - 1; 301 sp->ds_reserved = 0; 302 } 303 304 /* 305 * Ignore 0-length slices 306 */ 307 if (sp->ds_size == 0) 308 continue; 309 310 if (reprobe && 311 (ndev = devfs_find_device_by_name("%ss%d", 312 dev->si_name, sno))) { 313 /* 314 * Device already exists and is still valid 315 */ 316 ndev->si_flags |= SI_REPROBE_TEST; 317 } else { 318 /* 319 * Else create new device 320 */ 321 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops, 322 dkmakewholeslice(dkunit(dev), i), 323 UID_ROOT, GID_OPERATOR, 0640, 324 "%ss%d", dev->si_name, sno); 325 if (dp->d_info.d_serialno) { 326 make_dev_alias(ndev, "serno/%s.s%d", 327 dp->d_info.d_serialno, sno); 328 } 329 ndev->si_disk = dp; 330 ndev->si_flags |= SI_REPROBE_TEST; 331 } 332 sp->ds_dev = ndev; 333 334 /* 335 * Probe appropriate slices for a disklabel 336 * 337 * XXX slice type 1 used by our gpt probe code. 338 * XXX slice type 0 used by mbr compat slice. 339 */ 340 if (sp->ds_type == DOSPTYP_386BSD || 341 sp->ds_type == DOSPTYP_NETBSD || 342 sp->ds_type == DOSPTYP_OPENBSD || 343 sp->ds_type == 0 || 344 sp->ds_type == 1) { 345 if (dp->d_slice->dss_first_bsd_slice == 0) 346 dp->d_slice->dss_first_bsd_slice = i; 347 disk_probe_slice(dp, ndev, i, reprobe); 348 } 349 } 350 dsgone(&osp); 351 disk_debug(1, "disk_probe (end): %s\n", dp->d_cdev->si_name); 352 } 353 354 355 static void 356 disk_msg_core(void *arg) 357 { 358 struct disk *dp; 359 struct diskslice *sp; 360 lwkt_tokref ilock; 361 disk_msg_t msg; 362 int run; 363 364 lwkt_initport_thread(&disk_msg_port, curthread); 365 wakeup(curthread); 366 run = 1; 367 368 while (run) { 369 msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0); 370 371 switch (msg->hdr.u.ms_result) { 372 case DISK_DISK_PROBE: 373 dp = (struct disk *)msg->load; 374 disk_debug(1, 375 "DISK_DISK_PROBE: %s\n", 376 dp->d_cdev->si_name); 377 disk_probe(dp, 0); 378 break; 379 case DISK_DISK_DESTROY: 380 dp = (struct disk *)msg->load; 381 disk_debug(1, 382 "DISK_DISK_DESTROY: %s\n", 383 dp->d_cdev->si_name); 384 devfs_destroy_subnames(dp->d_cdev->si_name); 385 devfs_destroy_dev(dp->d_cdev); 386 lwkt_gettoken(&ilock, &disklist_token); 387 LIST_REMOVE(dp, d_list); 388 lwkt_reltoken(&ilock); 389 if (dp->d_info.d_serialno) { 390 kfree(dp->d_info.d_serialno, M_TEMP); 391 dp->d_info.d_serialno = NULL; 392 } 393 break; 394 case DISK_UNPROBE: 395 dp = (struct disk *)msg->load; 396 disk_debug(1, 397 "DISK_DISK_UNPROBE: %s\n", 398 dp->d_cdev->si_name); 399 devfs_destroy_subnames(dp->d_cdev->si_name); 400 break; 401 case DISK_SLICE_REPROBE: 402 dp = (struct disk *)msg->load; 403 sp = (struct diskslice *)msg->load2; 404 devfs_clr_subnames_flag(sp->ds_dev->si_name, 405 SI_REPROBE_TEST); 406 disk_debug(1, 407 "DISK_SLICE_REPROBE: %s\n", 408 sp->ds_dev->si_name); 409 disk_probe_slice(dp, sp->ds_dev, 410 dkslice(sp->ds_dev), 1); 411 devfs_destroy_subnames_without_flag( 412 sp->ds_dev->si_name, SI_REPROBE_TEST); 413 break; 414 case DISK_DISK_REPROBE: 415 dp = (struct disk *)msg->load; 416 devfs_clr_subnames_flag(dp->d_cdev->si_name, SI_REPROBE_TEST); 417 disk_debug(1, 418 "DISK_DISK_REPROBE: %s\n", 419 dp->d_cdev->si_name); 420 disk_probe(dp, 1); 421 devfs_destroy_subnames_without_flag( 422 dp->d_cdev->si_name, SI_REPROBE_TEST); 423 break; 424 case DISK_SYNC: 425 disk_debug(1, "DISK_SYNC\n"); 426 break; 427 default: 428 devfs_debug(DEVFS_DEBUG_WARNING, 429 "disk_msg_core: unknown message " 430 "received at core\n"); 431 break; 432 } 433 lwkt_replymsg(&msg->hdr, 0); 434 } 435 lwkt_exit(); 436 } 437 438 439 /* 440 * Acts as a message drain. Any message that is replied to here gets 441 * destroyed and the memory freed. 442 */ 443 static void 444 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg) 445 { 446 objcache_put(disk_msg_cache, msg); 447 } 448 449 450 void 451 disk_msg_send(uint32_t cmd, void *load, void *load2) 452 { 453 disk_msg_t disk_msg; 454 lwkt_port_t port = &disk_msg_port; 455 456 disk_msg = objcache_get(disk_msg_cache, M_WAITOK); 457 458 lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0); 459 460 disk_msg->hdr.u.ms_result = cmd; 461 disk_msg->load = load; 462 disk_msg->load2 = load2; 463 KKASSERT(port); 464 lwkt_sendmsg(port, &disk_msg->hdr); 465 } 466 467 void 468 disk_msg_send_sync(uint32_t cmd, void *load, void *load2) 469 { 470 struct lwkt_port rep_port; 471 disk_msg_t disk_msg; 472 lwkt_port_t port; 473 474 disk_msg = objcache_get(disk_msg_cache, M_WAITOK); 475 port = &disk_msg_port; 476 477 /* XXX could probably use curthread's built-in msgport */ 478 lwkt_initport_thread(&rep_port, curthread); 479 lwkt_initmsg(&disk_msg->hdr, &rep_port, 0); 480 481 disk_msg->hdr.u.ms_result = cmd; 482 disk_msg->load = load; 483 disk_msg->load2 = load2; 484 485 lwkt_sendmsg(port, &disk_msg->hdr); 486 lwkt_waitmsg(&disk_msg->hdr, 0); 487 objcache_put(disk_msg_cache, disk_msg); 488 } 489 490 /* 491 * Create a raw device for the dev_ops template (which is returned). Also 492 * create a slice and unit managed disk and overload the user visible 493 * device space with it. 494 * 495 * NOTE: The returned raw device is NOT a slice and unit managed device. 496 * It is an actual raw device representing the raw disk as specified by 497 * the passed dev_ops. The disk layer not only returns such a raw device, 498 * it also uses it internally when passing (modified) commands through. 499 */ 500 cdev_t 501 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops) 502 { 503 lwkt_tokref ilock; 504 cdev_t rawdev; 505 506 disk_debug(1, 507 "disk_create (begin): %s%d\n", 508 raw_ops->head.name, unit); 509 510 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit), 511 UID_ROOT, GID_OPERATOR, 0640, 512 "%s%d", raw_ops->head.name, unit); 513 514 bzero(dp, sizeof(*dp)); 515 516 dp->d_rawdev = rawdev; 517 dp->d_raw_ops = raw_ops; 518 dp->d_dev_ops = &disk_ops; 519 dp->d_cdev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops, 520 dkmakewholedisk(unit), 521 UID_ROOT, GID_OPERATOR, 0640, 522 "%s%d", raw_ops->head.name, unit); 523 524 dp->d_cdev->si_disk = dp; 525 526 dsched_disk_create_callback(dp, raw_ops->head.name, unit); 527 528 lwkt_gettoken(&ilock, &disklist_token); 529 LIST_INSERT_HEAD(&disklist, dp, d_list); 530 lwkt_reltoken(&ilock); 531 532 disk_debug(1, 533 "disk_create (end): %s%d\n", 534 raw_ops->head.name, unit); 535 536 return (dp->d_rawdev); 537 } 538 539 540 static void 541 _setdiskinfo(struct disk *disk, struct disk_info *info) 542 { 543 char *oldserialno; 544 545 oldserialno = disk->d_info.d_serialno; 546 bcopy(info, &disk->d_info, sizeof(disk->d_info)); 547 info = &disk->d_info; 548 549 disk_debug(1, 550 "_setdiskinfo: %s\n", 551 disk->d_cdev->si_name); 552 553 /* 554 * The serial number is duplicated so the caller can throw 555 * their copy away. 556 */ 557 if (info->d_serialno && info->d_serialno[0]) { 558 info->d_serialno = kstrdup(info->d_serialno, M_TEMP); 559 disk_cleanserial(info->d_serialno); 560 if (disk->d_cdev) { 561 make_dev_alias(disk->d_cdev, "serno/%s", 562 info->d_serialno); 563 } 564 } else { 565 info->d_serialno = NULL; 566 } 567 if (oldserialno) 568 kfree(oldserialno, M_TEMP); 569 570 dsched_disk_update_callback(disk, info); 571 572 /* 573 * The caller may set d_media_size or d_media_blocks and we 574 * calculate the other. 575 */ 576 KKASSERT(info->d_media_size == 0 || info->d_media_blksize == 0); 577 if (info->d_media_size == 0 && info->d_media_blocks) { 578 info->d_media_size = (u_int64_t)info->d_media_blocks * 579 info->d_media_blksize; 580 } else if (info->d_media_size && info->d_media_blocks == 0 && 581 info->d_media_blksize) { 582 info->d_media_blocks = info->d_media_size / 583 info->d_media_blksize; 584 } 585 586 /* 587 * The si_* fields for rawdev are not set until after the 588 * disk_create() call, so someone using the cooked version 589 * of the raw device (i.e. da0s0) will not get the right 590 * si_iosize_max unless we fix it up here. 591 */ 592 if (disk->d_cdev && disk->d_rawdev && 593 disk->d_cdev->si_iosize_max == 0) { 594 disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max; 595 disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys; 596 disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best; 597 } 598 } 599 600 /* 601 * Disk drivers must call this routine when media parameters are available 602 * or have changed. 603 */ 604 void 605 disk_setdiskinfo(struct disk *disk, struct disk_info *info) 606 { 607 _setdiskinfo(disk, info); 608 disk_msg_send(DISK_DISK_PROBE, disk, NULL); 609 disk_debug(1, 610 "disk_setdiskinfo: sent probe for %s\n", 611 disk->d_cdev->si_name); 612 } 613 614 void 615 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info) 616 { 617 _setdiskinfo(disk, info); 618 disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL); 619 disk_debug(1, 620 "disk_setdiskinfo_sync: sent probe for %s\n", 621 disk->d_cdev->si_name); 622 } 623 624 /* 625 * This routine is called when an adapter detaches. The higher level 626 * managed disk device is destroyed while the lower level raw device is 627 * released. 628 */ 629 void 630 disk_destroy(struct disk *disk) 631 { 632 dsched_disk_destroy_callback(disk); 633 disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL); 634 return; 635 } 636 637 int 638 disk_dumpcheck(cdev_t dev, u_int64_t *size, u_int64_t *blkno, u_int32_t *secsize) 639 { 640 struct partinfo pinfo; 641 int error; 642 643 bzero(&pinfo, sizeof(pinfo)); 644 error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0, 645 proc0.p_ucred, NULL); 646 if (error) 647 return (error); 648 649 if (pinfo.media_blksize == 0) 650 return (ENXIO); 651 652 if (blkno) /* XXX: make sure this reserved stuff is right */ 653 *blkno = pinfo.reserved_blocks + 654 pinfo.media_offset / pinfo.media_blksize; 655 if (secsize) 656 *secsize = pinfo.media_blksize; 657 if (size) 658 *size = (pinfo.media_blocks - pinfo.reserved_blocks); 659 660 return (0); 661 } 662 663 int 664 disk_dumpconf(cdev_t dev, u_int onoff) 665 { 666 struct dumperinfo di; 667 u_int64_t size, blkno; 668 u_int32_t secsize; 669 int error; 670 671 if (!onoff) 672 return set_dumper(NULL); 673 674 error = disk_dumpcheck(dev, &size, &blkno, &secsize); 675 676 if (error) 677 return ENXIO; 678 679 bzero(&di, sizeof(struct dumperinfo)); 680 di.dumper = diskdump; 681 di.priv = dev; 682 di.blocksize = secsize; 683 di.mediaoffset = blkno * DEV_BSIZE; 684 di.mediasize = size * DEV_BSIZE; 685 686 return set_dumper(&di); 687 } 688 689 void 690 disk_unprobe(struct disk *disk) 691 { 692 if (disk == NULL) 693 return; 694 695 disk_msg_send_sync(DISK_UNPROBE, disk, NULL); 696 } 697 698 void 699 disk_invalidate (struct disk *disk) 700 { 701 dsgone(&disk->d_slice); 702 } 703 704 struct disk * 705 disk_enumerate(struct disk *disk) 706 { 707 struct disk *dp; 708 lwkt_tokref ilock; 709 710 lwkt_gettoken(&ilock, &disklist_token); 711 if (!disk) 712 dp = (LIST_FIRST(&disklist)); 713 else 714 dp = (LIST_NEXT(disk, d_list)); 715 lwkt_reltoken(&ilock); 716 717 return dp; 718 } 719 720 static 721 int 722 sysctl_disks(SYSCTL_HANDLER_ARGS) 723 { 724 struct disk *disk; 725 int error, first; 726 727 disk = NULL; 728 first = 1; 729 730 while ((disk = disk_enumerate(disk))) { 731 if (!first) { 732 error = SYSCTL_OUT(req, " ", 1); 733 if (error) 734 return error; 735 } else { 736 first = 0; 737 } 738 error = SYSCTL_OUT(req, disk->d_rawdev->si_name, 739 strlen(disk->d_rawdev->si_name)); 740 if (error) 741 return error; 742 } 743 error = SYSCTL_OUT(req, "", 1); 744 return error; 745 } 746 747 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0, 748 sysctl_disks, "A", "names of available disks"); 749 750 /* 751 * Open a disk device or partition. 752 */ 753 static 754 int 755 diskopen(struct dev_open_args *ap) 756 { 757 cdev_t dev = ap->a_head.a_dev; 758 struct disk *dp; 759 int error; 760 761 /* 762 * dp can't be NULL here XXX. 763 * 764 * d_slice will be NULL if setdiskinfo() has not been called yet. 765 * setdiskinfo() is typically called whether the disk is present 766 * or not (e.g. CD), but the base disk device is created first 767 * and there may be a race. 768 */ 769 dp = dev->si_disk; 770 if (dp == NULL || dp->d_slice == NULL) 771 return (ENXIO); 772 error = 0; 773 774 /* 775 * Deal with open races 776 */ 777 while (dp->d_flags & DISKFLAG_LOCK) { 778 dp->d_flags |= DISKFLAG_WANTED; 779 error = tsleep(dp, PCATCH, "diskopen", hz); 780 if (error) 781 return (error); 782 } 783 dp->d_flags |= DISKFLAG_LOCK; 784 785 /* 786 * Open the underlying raw device. 787 */ 788 if (!dsisopen(dp->d_slice)) { 789 #if 0 790 if (!pdev->si_iosize_max) 791 pdev->si_iosize_max = dev->si_iosize_max; 792 #endif 793 error = dev_dopen(dp->d_rawdev, ap->a_oflags, 794 ap->a_devtype, ap->a_cred); 795 } 796 #if 0 797 /* 798 * Inherit properties from the underlying device now that it is 799 * open. 800 */ 801 dev_dclone(dev); 802 #endif 803 804 if (error) 805 goto out; 806 error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags, 807 &dp->d_slice, &dp->d_info); 808 if (!dsisopen(dp->d_slice)) { 809 dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype); 810 } 811 out: 812 dp->d_flags &= ~DISKFLAG_LOCK; 813 if (dp->d_flags & DISKFLAG_WANTED) { 814 dp->d_flags &= ~DISKFLAG_WANTED; 815 wakeup(dp); 816 } 817 818 return(error); 819 } 820 821 /* 822 * Close a disk device or partition 823 */ 824 static 825 int 826 diskclose(struct dev_close_args *ap) 827 { 828 cdev_t dev = ap->a_head.a_dev; 829 struct disk *dp; 830 int error; 831 832 error = 0; 833 dp = dev->si_disk; 834 835 dsclose(dev, ap->a_devtype, dp->d_slice); 836 if (!dsisopen(dp->d_slice)) { 837 error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype); 838 } 839 return (error); 840 } 841 842 /* 843 * First execute the ioctl on the disk device, and if it isn't supported 844 * try running it on the backing device. 845 */ 846 static 847 int 848 diskioctl(struct dev_ioctl_args *ap) 849 { 850 cdev_t dev = ap->a_head.a_dev; 851 struct disk *dp; 852 int error; 853 u_int u; 854 855 dp = dev->si_disk; 856 if (dp == NULL) 857 return (ENXIO); 858 859 devfs_debug(DEVFS_DEBUG_DEBUG, 860 "diskioctl: cmd is: %lx (name: %s)\n", 861 ap->a_cmd, dev->si_name); 862 devfs_debug(DEVFS_DEBUG_DEBUG, 863 "diskioctl: &dp->d_slice is: %p, %p\n", 864 &dp->d_slice, dp->d_slice); 865 866 if (ap->a_cmd == DIOCGKERNELDUMP) { 867 u = *(u_int *)ap->a_data; 868 return disk_dumpconf(dev, u); 869 } 870 871 error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag, 872 &dp->d_slice, &dp->d_info); 873 874 if (error == ENOIOCTL) { 875 error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data, 876 ap->a_fflag, ap->a_cred, NULL); 877 } 878 return (error); 879 } 880 881 /* 882 * Execute strategy routine 883 */ 884 static 885 int 886 diskstrategy(struct dev_strategy_args *ap) 887 { 888 cdev_t dev = ap->a_head.a_dev; 889 struct bio *bio = ap->a_bio; 890 struct bio *nbio; 891 struct disk *dp; 892 893 dp = dev->si_disk; 894 895 if (dp == NULL) { 896 bio->bio_buf->b_error = ENXIO; 897 bio->bio_buf->b_flags |= B_ERROR; 898 biodone(bio); 899 return(0); 900 } 901 KKASSERT(dev->si_disk == dp); 902 903 /* 904 * The dscheck() function will also transform the slice relative 905 * block number i.e. bio->bio_offset into a block number that can be 906 * passed directly to the underlying raw device. If dscheck() 907 * returns NULL it will have handled the bio for us (e.g. EOF 908 * or error due to being beyond the device size). 909 */ 910 if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) { 911 dsched_queue(dp, nbio); 912 } else { 913 biodone(bio); 914 } 915 return(0); 916 } 917 918 /* 919 * Return the partition size in ?blocks? 920 */ 921 static 922 int 923 diskpsize(struct dev_psize_args *ap) 924 { 925 cdev_t dev = ap->a_head.a_dev; 926 struct disk *dp; 927 928 dp = dev->si_disk; 929 if (dp == NULL) 930 return(ENODEV); 931 ap->a_result = dssize(dev, &dp->d_slice); 932 return(0); 933 } 934 935 /* 936 * When new device entries are instantiated, make sure they inherit our 937 * si_disk structure and block and iosize limits from the raw device. 938 * 939 * This routine is always called synchronously in the context of the 940 * client. 941 * 942 * XXX The various io and block size constraints are not always initialized 943 * properly by devices. 944 */ 945 static 946 int 947 diskclone(struct dev_clone_args *ap) 948 { 949 cdev_t dev = ap->a_head.a_dev; 950 struct disk *dp; 951 dp = dev->si_disk; 952 953 KKASSERT(dp != NULL); 954 dev->si_disk = dp; 955 dev->si_iosize_max = dp->d_rawdev->si_iosize_max; 956 dev->si_bsize_phys = dp->d_rawdev->si_bsize_phys; 957 dev->si_bsize_best = dp->d_rawdev->si_bsize_best; 958 return(0); 959 } 960 961 int 962 diskdump(struct dev_dump_args *ap) 963 { 964 cdev_t dev = ap->a_head.a_dev; 965 struct disk *dp = dev->si_disk; 966 u_int64_t size, offset; 967 int error; 968 969 error = disk_dumpcheck(dev, &size, &ap->a_blkno, &ap->a_secsize); 970 /* XXX: this should probably go in disk_dumpcheck somehow */ 971 if (ap->a_length != 0) { 972 size *= DEV_BSIZE; 973 offset = ap->a_blkno * DEV_BSIZE; 974 if ((ap->a_offset < offset) || 975 (ap->a_offset + ap->a_length - offset > size)) { 976 kprintf("Attempt to write outside dump device boundaries.\n"); 977 error = ENOSPC; 978 } 979 } 980 981 if (error == 0) { 982 ap->a_head.a_dev = dp->d_rawdev; 983 error = dev_doperate(&ap->a_head); 984 } 985 986 return(error); 987 } 988 989 990 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD, 991 0, sizeof(struct diskslices), "sizeof(struct diskslices)"); 992 993 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD, 994 0, sizeof(struct disk), "sizeof(struct disk)"); 995 996 /* 997 * Reorder interval for burst write allowance and minor write 998 * allowance. 999 * 1000 * We always want to trickle some writes in to make use of the 1001 * disk's zone cache. Bursting occurs on a longer interval and only 1002 * runningbufspace is well over the hirunningspace limit. 1003 */ 1004 int bioq_reorder_burst_interval = 60; /* should be multiple of minor */ 1005 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_interval, 1006 CTLFLAG_RW, &bioq_reorder_burst_interval, 0, ""); 1007 int bioq_reorder_minor_interval = 5; 1008 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_interval, 1009 CTLFLAG_RW, &bioq_reorder_minor_interval, 0, ""); 1010 1011 int bioq_reorder_burst_bytes = 3000000; 1012 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_bytes, 1013 CTLFLAG_RW, &bioq_reorder_burst_bytes, 0, ""); 1014 int bioq_reorder_minor_bytes = 262144; 1015 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_bytes, 1016 CTLFLAG_RW, &bioq_reorder_minor_bytes, 0, ""); 1017 1018 1019 /* 1020 * Order I/Os. Generally speaking this code is designed to make better 1021 * use of drive zone caches. A drive zone cache can typically track linear 1022 * reads or writes for around 16 zones simultaniously. 1023 * 1024 * Read prioritization issues: It is possible for hundreds of megabytes worth 1025 * of writes to be queued asynchronously. This creates a huge bottleneck 1026 * for reads which reduce read bandwidth to a trickle. 1027 * 1028 * To solve this problem we generally reorder reads before writes. 1029 * 1030 * However, a large number of random reads can also starve writes and 1031 * make poor use of the drive zone cache so we allow writes to trickle 1032 * in every N reads. 1033 */ 1034 void 1035 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio) 1036 { 1037 /* 1038 * The BIO wants to be ordered. Adding to the tail also 1039 * causes transition to be set to NULL, forcing the ordering 1040 * of all prior I/O's. 1041 */ 1042 if (bio->bio_buf->b_flags & B_ORDERED) { 1043 bioq_insert_tail(bioq, bio); 1044 return; 1045 } 1046 1047 switch(bio->bio_buf->b_cmd) { 1048 case BUF_CMD_READ: 1049 if (bioq->transition) { 1050 /* 1051 * Insert before the first write. Bleedover writes 1052 * based on reorder intervals to prevent starvation. 1053 */ 1054 TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act); 1055 ++bioq->reorder; 1056 if (bioq->reorder % bioq_reorder_minor_interval == 0) { 1057 bioqwritereorder(bioq); 1058 if (bioq->reorder >= 1059 bioq_reorder_burst_interval) { 1060 bioq->reorder = 0; 1061 } 1062 } 1063 } else { 1064 /* 1065 * No writes queued (or ordering was forced), 1066 * insert at tail. 1067 */ 1068 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act); 1069 } 1070 break; 1071 case BUF_CMD_WRITE: 1072 /* 1073 * Writes are always appended. If no writes were previously 1074 * queued or an ordered tail insertion occured the transition 1075 * field will be NULL. 1076 */ 1077 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act); 1078 if (bioq->transition == NULL) 1079 bioq->transition = bio; 1080 break; 1081 default: 1082 /* 1083 * All other request types are forced to be ordered. 1084 */ 1085 bioq_insert_tail(bioq, bio); 1086 break; 1087 } 1088 } 1089 1090 /* 1091 * Move the read-write transition point to prevent reads from 1092 * completely starving our writes. This brings a number of writes into 1093 * the fold every N reads. 1094 * 1095 * We bring a few linear writes into the fold on a minor interval 1096 * and we bring a non-linear burst of writes into the fold on a major 1097 * interval. Bursting only occurs if runningbufspace is really high 1098 * (typically from syncs, fsyncs, or HAMMER flushes). 1099 */ 1100 static 1101 void 1102 bioqwritereorder(struct bio_queue_head *bioq) 1103 { 1104 struct bio *bio; 1105 off_t next_offset; 1106 size_t left; 1107 size_t n; 1108 int check_off; 1109 1110 if (bioq->reorder < bioq_reorder_burst_interval || 1111 !buf_runningbufspace_severe()) { 1112 left = (size_t)bioq_reorder_minor_bytes; 1113 check_off = 1; 1114 } else { 1115 left = (size_t)bioq_reorder_burst_bytes; 1116 check_off = 0; 1117 } 1118 1119 next_offset = bioq->transition->bio_offset; 1120 while ((bio = bioq->transition) != NULL && 1121 (check_off == 0 || next_offset == bio->bio_offset) 1122 ) { 1123 n = bio->bio_buf->b_bcount; 1124 next_offset = bio->bio_offset + n; 1125 bioq->transition = TAILQ_NEXT(bio, bio_act); 1126 if (left < n) 1127 break; 1128 left -= n; 1129 } 1130 } 1131 1132 /* 1133 * Disk error is the preface to plaintive error messages 1134 * about failing disk transfers. It prints messages of the form 1135 1136 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d) 1137 1138 * if the offset of the error in the transfer and a disk label 1139 * are both available. blkdone should be -1 if the position of the error 1140 * is unknown; the disklabel pointer may be null from drivers that have not 1141 * been converted to use them. The message is printed with kprintf 1142 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority. 1143 * The message should be completed (with at least a newline) with kprintf 1144 * or log(-1, ...), respectively. There is no trailing space. 1145 */ 1146 void 1147 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt) 1148 { 1149 struct buf *bp = bio->bio_buf; 1150 const char *term; 1151 1152 switch(bp->b_cmd) { 1153 case BUF_CMD_READ: 1154 term = "read"; 1155 break; 1156 case BUF_CMD_WRITE: 1157 term = "write"; 1158 break; 1159 default: 1160 term = "access"; 1161 break; 1162 } 1163 kprintf("%s: %s %sing ", dev->si_name, what, term); 1164 kprintf("offset %012llx for %d", 1165 (long long)bio->bio_offset, 1166 bp->b_bcount); 1167 1168 if (donecnt) 1169 kprintf(" (%d bytes completed)", donecnt); 1170 } 1171 1172 /* 1173 * Locate a disk device 1174 */ 1175 cdev_t 1176 disk_locate(const char *devname) 1177 { 1178 return devfs_find_device_by_name(devname); 1179 } 1180 1181 void 1182 disk_config(void *arg) 1183 { 1184 disk_msg_send_sync(DISK_SYNC, NULL, NULL); 1185 } 1186 1187 static void 1188 disk_init(void) 1189 { 1190 struct thread* td_core; 1191 1192 disk_msg_cache = objcache_create("disk-msg-cache", 0, 0, 1193 NULL, NULL, NULL, 1194 objcache_malloc_alloc, 1195 objcache_malloc_free, 1196 &disk_msg_malloc_args); 1197 1198 lwkt_token_init(&disklist_token); 1199 1200 /* 1201 * Initialize the reply-only port which acts as a message drain 1202 */ 1203 lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply); 1204 1205 lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL, 1206 0, 0, "disk_msg_core"); 1207 1208 tsleep(td_core, 0, "diskcore", 0); 1209 } 1210 1211 static void 1212 disk_uninit(void) 1213 { 1214 objcache_destroy(disk_msg_cache); 1215 } 1216 1217 /* 1218 * Clean out illegal characters in serial numbers. 1219 */ 1220 static void 1221 disk_cleanserial(char *serno) 1222 { 1223 char c; 1224 1225 while ((c = *serno) != 0) { 1226 if (c >= 'a' && c <= 'z') 1227 ; 1228 else if (c >= 'A' && c <= 'Z') 1229 ; 1230 else if (c >= '0' && c <= '9') 1231 ; 1232 else if (c == '-' || c == '@' || c == '+' || c == '.') 1233 ; 1234 else 1235 c = '_'; 1236 *serno++= c; 1237 } 1238 } 1239 1240 TUNABLE_INT("kern.disk_debug", &disk_debug_enable); 1241 SYSCTL_INT(_kern, OID_AUTO, disk_debug, CTLFLAG_RW, &disk_debug_enable, 1242 0, "Enable subr_disk debugging"); 1243 1244 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL); 1245 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL); 1246