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 * and Alex Hornung <ahornung@gmail.com> 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 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 16 * the documentation and/or other materials provided with the 17 * distribution. 18 * 3. Neither the name of The DragonFly Project nor the names of its 19 * contributors may be used to endorse or promote products derived 20 * from this software without specific, prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 25 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 26 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 27 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 28 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 29 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 30 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 31 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 32 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 * 35 * ---------------------------------------------------------------------------- 36 * "THE BEER-WARE LICENSE" (Revision 42): 37 * <phk@FreeBSD.ORG> wrote this file. As long as you retain this notice you 38 * can do whatever you want with this stuff. If we meet some day, and you think 39 * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp 40 * ---------------------------------------------------------------------------- 41 * 42 * Copyright (c) 1982, 1986, 1988, 1993 43 * The Regents of the University of California. All rights reserved. 44 * (c) UNIX System Laboratories, Inc. 45 * All or some portions of this file are derived from material licensed 46 * to the University of California by American Telephone and Telegraph 47 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 48 * the permission of UNIX System Laboratories, Inc. 49 * 50 * Redistribution and use in source and binary forms, with or without 51 * modification, are permitted provided that the following conditions 52 * are met: 53 * 1. Redistributions of source code must retain the above copyright 54 * notice, this list of conditions and the following disclaimer. 55 * 2. Redistributions in binary form must reproduce the above copyright 56 * notice, this list of conditions and the following disclaimer in the 57 * documentation and/or other materials provided with the distribution. 58 * 3. Neither the name of the University nor the names of its contributors 59 * may be used to endorse or promote products derived from this software 60 * without specific prior written permission. 61 * 62 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 63 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 64 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 65 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 66 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 67 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 68 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 69 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 70 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 71 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 72 * SUCH DAMAGE. 73 * 74 * @(#)ufs_disksubr.c 8.5 (Berkeley) 1/21/94 75 * $FreeBSD: src/sys/kern/subr_disk.c,v 1.20.2.6 2001/10/05 07:14:57 peter Exp $ 76 * $FreeBSD: src/sys/ufs/ufs/ufs_disksubr.c,v 1.44.2.3 2001/03/05 05:42:19 obrien Exp $ 77 */ 78 79 #include <sys/param.h> 80 #include <sys/systm.h> 81 #include <sys/kernel.h> 82 #include <sys/proc.h> 83 #include <sys/sysctl.h> 84 #include <sys/buf.h> 85 #include <sys/conf.h> 86 #include <sys/disklabel.h> 87 #include <sys/disklabel32.h> 88 #include <sys/disklabel64.h> 89 #include <sys/diskslice.h> 90 #include <sys/diskmbr.h> 91 #include <sys/disk.h> 92 #include <sys/kerneldump.h> 93 #include <sys/malloc.h> 94 #include <machine/md_var.h> 95 #include <sys/ctype.h> 96 #include <sys/syslog.h> 97 #include <sys/device.h> 98 #include <sys/msgport.h> 99 #include <sys/devfs.h> 100 #include <sys/thread.h> 101 #include <sys/dsched.h> 102 #include <sys/queue.h> 103 #include <sys/lock.h> 104 #include <sys/udev.h> 105 #include <sys/uuid.h> 106 107 #include <sys/buf2.h> 108 #include <sys/msgport2.h> 109 #include <sys/thread2.h> 110 111 static MALLOC_DEFINE(M_DISK, "disk", "disk data"); 112 static int disk_debug_enable = 0; 113 114 static void disk_msg_autofree_reply(lwkt_port_t, lwkt_msg_t); 115 static void disk_msg_core(void *); 116 static int disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe); 117 static void disk_probe(struct disk *dp, int reprobe); 118 static void _setdiskinfo(struct disk *disk, struct disk_info *info); 119 static void bioqwritereorder(struct bio_queue_head *bioq); 120 static void disk_cleanserial(char *serno); 121 static int disk_debug(int, char *, ...) __printflike(2, 3); 122 static cdev_t _disk_create_named(const char *name, int unit, struct disk *dp, 123 struct dev_ops *raw_ops, int clone); 124 125 static d_open_t diskopen; 126 static d_close_t diskclose; 127 static d_ioctl_t diskioctl; 128 static d_strategy_t diskstrategy; 129 static d_psize_t diskpsize; 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 static struct lwkt_token ds_token; 135 136 static struct dev_ops disk_ops = { 137 { "disk", 0, D_DISK | D_MPSAFE | D_TRACKCLOSE }, 138 .d_open = diskopen, 139 .d_close = diskclose, 140 .d_read = physread, 141 .d_write = physwrite, 142 .d_ioctl = diskioctl, 143 .d_strategy = diskstrategy, 144 .d_dump = diskdump, 145 .d_psize = diskpsize, 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 char uuid_buf[128]; 178 cdev_t ndev; 179 int sno; 180 u_int i; 181 182 disk_debug(2, "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 } 193 194 if (msg == NULL) { 195 if (slice != WHOLE_DISK_SLICE) 196 ops->op_adjust_label_reserved(dp->d_slice, slice, sp); 197 else 198 sp->ds_reserved = 0; 199 200 sp->ds_ops = ops; 201 for (i = 0; i < ops->op_getnumparts(sp->ds_label); i++) { 202 ops->op_loadpartinfo(sp->ds_label, i, &part); 203 if (part.fstype) { 204 if (reprobe && 205 (ndev = devfs_find_device_by_name("%s%c", 206 dev->si_name, 'a' + i)) 207 ) { 208 /* 209 * Device already exists and 210 * is still valid. 211 */ 212 ndev->si_flags |= SI_REPROBE_TEST; 213 214 /* 215 * Destroy old UUID alias 216 */ 217 destroy_dev_alias(ndev, "part-by-uuid/*"); 218 219 /* Create UUID alias */ 220 if (!kuuid_is_nil(&part.storage_uuid)) { 221 snprintf_uuid(uuid_buf, 222 sizeof(uuid_buf), 223 &part.storage_uuid); 224 make_dev_alias(ndev, 225 "part-by-uuid/%s", 226 uuid_buf); 227 udev_dict_set_cstr(ndev, "uuid", uuid_buf); 228 } 229 } else { 230 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops, 231 dkmakeminor(dkunit(dp->d_cdev), 232 slice, i), 233 UID_ROOT, GID_OPERATOR, 0640, 234 "%s%c", dev->si_name, 'a'+ i); 235 ndev->si_parent = dev; 236 ndev->si_iosize_max = dev->si_iosize_max; 237 ndev->si_disk = dp; 238 udev_dict_set_cstr(ndev, "subsystem", "disk"); 239 /* Inherit parent's disk type */ 240 if (dp->d_disktype) { 241 udev_dict_set_cstr(ndev, "disk-type", 242 __DECONST(char *, dp->d_disktype)); 243 } 244 245 /* Create serno alias */ 246 if (dp->d_info.d_serialno) { 247 make_dev_alias(ndev, 248 "serno/%s.s%d%c", 249 dp->d_info.d_serialno, 250 sno, 'a' + i); 251 } 252 253 /* Create UUID alias */ 254 if (!kuuid_is_nil(&part.storage_uuid)) { 255 snprintf_uuid(uuid_buf, 256 sizeof(uuid_buf), 257 &part.storage_uuid); 258 make_dev_alias(ndev, 259 "part-by-uuid/%s", 260 uuid_buf); 261 udev_dict_set_cstr(ndev, "uuid", uuid_buf); 262 } 263 ndev->si_flags |= SI_REPROBE_TEST; 264 } 265 } 266 } 267 } else if (info->d_dsflags & DSO_COMPATLABEL) { 268 msg = NULL; 269 if (sp->ds_size >= 0x100000000ULL) 270 ops = &disklabel64_ops; 271 else 272 ops = &disklabel32_ops; 273 sp->ds_label = ops->op_clone_label(info, sp); 274 } else { 275 if (sp->ds_type == DOSPTYP_386BSD || /* XXX */ 276 sp->ds_type == DOSPTYP_NETBSD || 277 sp->ds_type == DOSPTYP_OPENBSD) { 278 log(LOG_WARNING, "%s: cannot find label (%s)\n", 279 dev->si_name, msg); 280 } 281 282 if (sp->ds_label.opaque != NULL && sp->ds_ops != NULL) { 283 /* Clear out old label - it's not around anymore */ 284 disk_debug(2, 285 "disk_probe_slice: clear out old diskabel on %s\n", 286 dev->si_name); 287 288 sp->ds_ops->op_freedisklabel(&sp->ds_label); 289 sp->ds_ops = NULL; 290 } 291 } 292 293 if (msg == NULL) { 294 sp->ds_wlabel = FALSE; 295 } 296 297 return (msg ? EINVAL : 0); 298 } 299 300 /* 301 * This routine is only called for newly minted drives or to reprobe 302 * a drive with no open slices. disk_probe_slice() is called directly 303 * when reprobing partition changes within slices. 304 */ 305 static void 306 disk_probe(struct disk *dp, int reprobe) 307 { 308 struct disk_info *info = &dp->d_info; 309 cdev_t dev = dp->d_cdev; 310 cdev_t ndev; 311 int error, i, sno; 312 struct diskslices *osp; 313 struct diskslice *sp; 314 char uuid_buf[128]; 315 316 KKASSERT (info->d_media_blksize != 0); 317 318 osp = dp->d_slice; 319 dp->d_slice = dsmakeslicestruct(BASE_SLICE, info); 320 disk_debug(1, "disk_probe (begin): %s\n", dp->d_cdev->si_name); 321 322 error = mbrinit(dev, info, &(dp->d_slice)); 323 if (error) { 324 dsgone(&osp); 325 return; 326 } 327 328 for (i = 0; i < dp->d_slice->dss_nslices; i++) { 329 /* 330 * Ignore the whole-disk slice, it has already been created. 331 */ 332 if (i == WHOLE_DISK_SLICE) 333 continue; 334 335 #if 1 336 /* 337 * Ignore the compatibility slice s0 if it's a device mapper 338 * volume. 339 */ 340 if ((i == COMPATIBILITY_SLICE) && 341 (info->d_dsflags & DSO_DEVICEMAPPER)) 342 continue; 343 #endif 344 345 sp = &dp->d_slice->dss_slices[i]; 346 347 /* 348 * Handle s0. s0 is a compatibility slice if there are no 349 * other slices and it has not otherwise been set up, else 350 * we ignore it. 351 */ 352 if (i == COMPATIBILITY_SLICE) { 353 sno = 0; 354 if (sp->ds_type == 0 && 355 dp->d_slice->dss_nslices == BASE_SLICE) { 356 sp->ds_size = info->d_media_blocks; 357 sp->ds_reserved = 0; 358 } 359 } else { 360 sno = i - 1; 361 sp->ds_reserved = 0; 362 } 363 364 /* 365 * Ignore 0-length slices 366 */ 367 if (sp->ds_size == 0) 368 continue; 369 370 if (reprobe && 371 (ndev = devfs_find_device_by_name("%ss%d", 372 dev->si_name, sno))) { 373 /* 374 * Device already exists and is still valid 375 */ 376 ndev->si_flags |= SI_REPROBE_TEST; 377 378 /* 379 * Destroy old UUID alias 380 */ 381 destroy_dev_alias(ndev, "slice-by-uuid/*"); 382 383 /* Create UUID alias */ 384 if (!kuuid_is_nil(&sp->ds_stor_uuid)) { 385 snprintf_uuid(uuid_buf, sizeof(uuid_buf), 386 &sp->ds_stor_uuid); 387 make_dev_alias(ndev, "slice-by-uuid/%s", 388 uuid_buf); 389 } 390 } else { 391 /* 392 * Else create new device 393 */ 394 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops, 395 dkmakewholeslice(dkunit(dev), i), 396 UID_ROOT, GID_OPERATOR, 0640, 397 (info->d_dsflags & DSO_DEVICEMAPPER)? 398 "%s.s%d" : "%ss%d", dev->si_name, sno); 399 ndev->si_parent = dev; 400 ndev->si_iosize_max = dev->si_iosize_max; 401 udev_dict_set_cstr(ndev, "subsystem", "disk"); 402 /* Inherit parent's disk type */ 403 if (dp->d_disktype) { 404 udev_dict_set_cstr(ndev, "disk-type", 405 __DECONST(char *, dp->d_disktype)); 406 } 407 408 /* Create serno alias */ 409 if (dp->d_info.d_serialno) { 410 make_dev_alias(ndev, "serno/%s.s%d", 411 dp->d_info.d_serialno, sno); 412 } 413 414 /* Create UUID alias */ 415 if (!kuuid_is_nil(&sp->ds_stor_uuid)) { 416 snprintf_uuid(uuid_buf, sizeof(uuid_buf), 417 &sp->ds_stor_uuid); 418 make_dev_alias(ndev, "slice-by-uuid/%s", 419 uuid_buf); 420 } 421 422 ndev->si_disk = dp; 423 ndev->si_flags |= SI_REPROBE_TEST; 424 } 425 sp->ds_dev = ndev; 426 427 /* 428 * Probe appropriate slices for a disklabel 429 * 430 * XXX slice type 1 used by our gpt probe code. 431 * XXX slice type 0 used by mbr compat slice. 432 */ 433 if (sp->ds_type == DOSPTYP_386BSD || 434 sp->ds_type == DOSPTYP_NETBSD || 435 sp->ds_type == DOSPTYP_OPENBSD || 436 sp->ds_type == 0 || 437 sp->ds_type == 1) { 438 if (dp->d_slice->dss_first_bsd_slice == 0) 439 dp->d_slice->dss_first_bsd_slice = i; 440 disk_probe_slice(dp, ndev, i, reprobe); 441 } 442 } 443 dsgone(&osp); 444 disk_debug(1, "disk_probe (end): %s\n", dp->d_cdev->si_name); 445 } 446 447 448 static void 449 disk_msg_core(void *arg) 450 { 451 struct disk *dp; 452 struct diskslice *sp; 453 disk_msg_t msg; 454 int run; 455 456 lwkt_gettoken(&disklist_token); 457 lwkt_initport_thread(&disk_msg_port, curthread); 458 wakeup(curthread); /* synchronous startup */ 459 lwkt_reltoken(&disklist_token); 460 461 lwkt_gettoken(&ds_token); 462 run = 1; 463 464 while (run) { 465 msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0); 466 467 switch (msg->hdr.u.ms_result) { 468 case DISK_DISK_PROBE: 469 dp = (struct disk *)msg->load; 470 disk_debug(1, 471 "DISK_DISK_PROBE: %s\n", 472 dp->d_cdev->si_name); 473 disk_iocom_update(dp); 474 disk_probe(dp, 0); 475 break; 476 case DISK_DISK_DESTROY: 477 dp = (struct disk *)msg->load; 478 disk_debug(1, 479 "DISK_DISK_DESTROY: %s\n", 480 dp->d_cdev->si_name); 481 disk_iocom_uninit(dp); 482 483 /* 484 * Interlock against struct disk enumerations. 485 * Wait for enumerations to complete then remove 486 * the dp from the list before tearing it down. 487 * This avoids numerous races. 488 */ 489 lwkt_gettoken(&disklist_token); 490 while (dp->d_refs) 491 tsleep(&dp->d_refs, 0, "diskdel", hz / 10); 492 LIST_REMOVE(dp, d_list); 493 494 dsched_disk_destroy(dp); 495 devfs_destroy_related(dp->d_cdev); 496 destroy_dev(dp->d_cdev); 497 destroy_only_dev(dp->d_rawdev); 498 499 lwkt_reltoken(&disklist_token); 500 501 if (dp->d_info.d_serialno) { 502 kfree(dp->d_info.d_serialno, M_TEMP); 503 dp->d_info.d_serialno = NULL; 504 } 505 break; 506 case DISK_UNPROBE: 507 dp = (struct disk *)msg->load; 508 disk_debug(1, 509 "DISK_DISK_UNPROBE: %s\n", 510 dp->d_cdev->si_name); 511 devfs_destroy_related(dp->d_cdev); 512 break; 513 case DISK_SLICE_REPROBE: 514 dp = (struct disk *)msg->load; 515 sp = (struct diskslice *)msg->load2; 516 devfs_clr_related_flag(sp->ds_dev, 517 SI_REPROBE_TEST); 518 disk_debug(1, 519 "DISK_SLICE_REPROBE: %s\n", 520 sp->ds_dev->si_name); 521 disk_probe_slice(dp, sp->ds_dev, 522 dkslice(sp->ds_dev), 1); 523 devfs_destroy_related_without_flag( 524 sp->ds_dev, SI_REPROBE_TEST); 525 break; 526 case DISK_DISK_REPROBE: 527 dp = (struct disk *)msg->load; 528 devfs_clr_related_flag(dp->d_cdev, SI_REPROBE_TEST); 529 disk_debug(1, 530 "DISK_DISK_REPROBE: %s\n", 531 dp->d_cdev->si_name); 532 disk_probe(dp, 1); 533 devfs_destroy_related_without_flag( 534 dp->d_cdev, SI_REPROBE_TEST); 535 break; 536 case DISK_SYNC: 537 disk_debug(1, "DISK_SYNC\n"); 538 break; 539 default: 540 devfs_debug(DEVFS_DEBUG_WARNING, 541 "disk_msg_core: unknown message " 542 "received at core\n"); 543 break; 544 } 545 lwkt_replymsg(&msg->hdr, 0); 546 } 547 lwkt_reltoken(&ds_token); 548 lwkt_exit(); 549 } 550 551 552 /* 553 * Acts as a message drain. Any message that is replied to here gets 554 * destroyed and the memory freed. 555 */ 556 static void 557 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg) 558 { 559 objcache_put(disk_msg_cache, msg); 560 } 561 562 563 void 564 disk_msg_send(uint32_t cmd, void *load, void *load2) 565 { 566 disk_msg_t disk_msg; 567 lwkt_port_t port = &disk_msg_port; 568 569 disk_msg = objcache_get(disk_msg_cache, M_WAITOK); 570 571 lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0); 572 573 disk_msg->hdr.u.ms_result = cmd; 574 disk_msg->load = load; 575 disk_msg->load2 = load2; 576 KKASSERT(port); 577 lwkt_sendmsg(port, &disk_msg->hdr); 578 } 579 580 void 581 disk_msg_send_sync(uint32_t cmd, void *load, void *load2) 582 { 583 struct lwkt_port rep_port; 584 disk_msg_t disk_msg; 585 lwkt_port_t port; 586 587 disk_msg = objcache_get(disk_msg_cache, M_WAITOK); 588 port = &disk_msg_port; 589 590 /* XXX could probably use curthread's built-in msgport */ 591 lwkt_initport_thread(&rep_port, curthread); 592 lwkt_initmsg(&disk_msg->hdr, &rep_port, 0); 593 594 disk_msg->hdr.u.ms_result = cmd; 595 disk_msg->load = load; 596 disk_msg->load2 = load2; 597 598 lwkt_domsg(port, &disk_msg->hdr, 0); 599 objcache_put(disk_msg_cache, disk_msg); 600 } 601 602 /* 603 * Create a raw device for the dev_ops template (which is returned). Also 604 * create a slice and unit managed disk and overload the user visible 605 * device space with it. 606 * 607 * NOTE: The returned raw device is NOT a slice and unit managed device. 608 * It is an actual raw device representing the raw disk as specified by 609 * the passed dev_ops. The disk layer not only returns such a raw device, 610 * it also uses it internally when passing (modified) commands through. 611 */ 612 cdev_t 613 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops) 614 { 615 return _disk_create_named(NULL, unit, dp, raw_ops, 0); 616 } 617 618 cdev_t 619 disk_create_clone(int unit, struct disk *dp, 620 struct dev_ops *raw_ops) 621 { 622 return _disk_create_named(NULL, unit, dp, raw_ops, 1); 623 } 624 625 cdev_t 626 disk_create_named(const char *name, int unit, struct disk *dp, 627 struct dev_ops *raw_ops) 628 { 629 return _disk_create_named(name, unit, dp, raw_ops, 0); 630 } 631 632 cdev_t 633 disk_create_named_clone(const char *name, int unit, struct disk *dp, 634 struct dev_ops *raw_ops) 635 { 636 return _disk_create_named(name, unit, dp, raw_ops, 1); 637 } 638 639 static cdev_t 640 _disk_create_named(const char *name, int unit, struct disk *dp, 641 struct dev_ops *raw_ops, int clone) 642 { 643 cdev_t rawdev; 644 645 disk_debug(1, "disk_create (begin): %s%d\n", name, unit); 646 647 if (name) { 648 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit), 649 UID_ROOT, GID_OPERATOR, 0640, "%s", name); 650 } else { 651 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit), 652 UID_ROOT, GID_OPERATOR, 0640, 653 "%s%d", raw_ops->head.name, unit); 654 } 655 656 bzero(dp, sizeof(*dp)); 657 658 dp->d_rawdev = rawdev; 659 dp->d_raw_ops = raw_ops; 660 dp->d_dev_ops = &disk_ops; 661 662 if (name) { 663 if (clone) { 664 dp->d_cdev = make_only_dev_covering( 665 &disk_ops, dp->d_rawdev->si_ops, 666 dkmakewholedisk(unit), 667 UID_ROOT, GID_OPERATOR, 0640, 668 "%s", name); 669 } else { 670 dp->d_cdev = make_dev_covering( 671 &disk_ops, dp->d_rawdev->si_ops, 672 dkmakewholedisk(unit), 673 UID_ROOT, GID_OPERATOR, 0640, 674 "%s", name); 675 } 676 } else { 677 if (clone) { 678 dp->d_cdev = make_only_dev_covering( 679 &disk_ops, dp->d_rawdev->si_ops, 680 dkmakewholedisk(unit), 681 UID_ROOT, GID_OPERATOR, 0640, 682 "%s%d", raw_ops->head.name, unit); 683 } else { 684 dp->d_cdev = make_dev_covering( 685 &disk_ops, dp->d_rawdev->si_ops, 686 dkmakewholedisk(unit), 687 UID_ROOT, GID_OPERATOR, 0640, 688 "%s%d", raw_ops->head.name, unit); 689 } 690 } 691 692 udev_dict_set_cstr(dp->d_cdev, "subsystem", "disk"); 693 dp->d_cdev->si_disk = dp; 694 695 if (name) 696 dsched_disk_create(dp, name, unit); 697 else 698 dsched_disk_create(dp, raw_ops->head.name, unit); 699 700 lwkt_gettoken(&disklist_token); 701 LIST_INSERT_HEAD(&disklist, dp, d_list); 702 lwkt_reltoken(&disklist_token); 703 704 disk_iocom_init(dp); 705 706 disk_debug(1, "disk_create (end): %s%d\n", 707 (name != NULL)?(name):(raw_ops->head.name), unit); 708 709 return (dp->d_rawdev); 710 } 711 712 int 713 disk_setdisktype(struct disk *disk, const char *type) 714 { 715 int error; 716 717 KKASSERT(disk != NULL); 718 719 disk->d_disktype = type; 720 error = udev_dict_set_cstr(disk->d_cdev, "disk-type", 721 __DECONST(char *, type)); 722 return error; 723 } 724 725 int 726 disk_getopencount(struct disk *disk) 727 { 728 return disk->d_opencount; 729 } 730 731 static void 732 _setdiskinfo(struct disk *disk, struct disk_info *info) 733 { 734 char *oldserialno; 735 736 oldserialno = disk->d_info.d_serialno; 737 bcopy(info, &disk->d_info, sizeof(disk->d_info)); 738 info = &disk->d_info; 739 740 disk_debug(1, "_setdiskinfo: %s\n", disk->d_cdev->si_name); 741 742 /* 743 * The serial number is duplicated so the caller can throw 744 * their copy away. 745 */ 746 if (info->d_serialno && info->d_serialno[0] && 747 (info->d_serialno[0] != ' ' || strlen(info->d_serialno) > 1)) { 748 info->d_serialno = kstrdup(info->d_serialno, M_TEMP); 749 disk_cleanserial(info->d_serialno); 750 if (disk->d_cdev) { 751 make_dev_alias(disk->d_cdev, "serno/%s", 752 info->d_serialno); 753 } 754 } else { 755 info->d_serialno = NULL; 756 } 757 if (oldserialno) 758 kfree(oldserialno, M_TEMP); 759 760 dsched_disk_update(disk, info); 761 762 /* 763 * The caller may set d_media_size or d_media_blocks and we 764 * calculate the other. 765 */ 766 KKASSERT(info->d_media_size == 0 || info->d_media_blocks == 0); 767 if (info->d_media_size == 0 && info->d_media_blocks) { 768 info->d_media_size = (u_int64_t)info->d_media_blocks * 769 info->d_media_blksize; 770 } else if (info->d_media_size && info->d_media_blocks == 0 && 771 info->d_media_blksize) { 772 info->d_media_blocks = info->d_media_size / 773 info->d_media_blksize; 774 } 775 776 /* 777 * The si_* fields for rawdev are not set until after the 778 * disk_create() call, so someone using the cooked version 779 * of the raw device (i.e. da0s0) will not get the right 780 * si_iosize_max unless we fix it up here. 781 */ 782 if (disk->d_cdev && disk->d_rawdev && 783 disk->d_cdev->si_iosize_max == 0) { 784 disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max; 785 disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys; 786 disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best; 787 } 788 789 /* Add the serial number to the udev_dictionary */ 790 if (info->d_serialno) 791 udev_dict_set_cstr(disk->d_cdev, "serno", info->d_serialno); 792 } 793 794 /* 795 * Disk drivers must call this routine when media parameters are available 796 * or have changed. 797 */ 798 void 799 disk_setdiskinfo(struct disk *disk, struct disk_info *info) 800 { 801 _setdiskinfo(disk, info); 802 disk_msg_send(DISK_DISK_PROBE, disk, NULL); 803 disk_debug(1, "disk_setdiskinfo: sent probe for %s\n", 804 disk->d_cdev->si_name); 805 } 806 807 void 808 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info) 809 { 810 _setdiskinfo(disk, info); 811 disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL); 812 disk_debug(1, "disk_setdiskinfo_sync: sent probe for %s\n", 813 disk->d_cdev->si_name); 814 } 815 816 /* 817 * This routine is called when an adapter detaches. The higher level 818 * managed disk device is destroyed while the lower level raw device is 819 * released. 820 */ 821 void 822 disk_destroy(struct disk *disk) 823 { 824 disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL); 825 return; 826 } 827 828 int 829 disk_dumpcheck(cdev_t dev, u_int64_t *size, 830 u_int64_t *blkno, u_int32_t *secsize) 831 { 832 struct partinfo pinfo; 833 int error; 834 835 if (size) 836 *size = 0; /* avoid gcc warnings */ 837 if (secsize) 838 *secsize = 512; /* avoid gcc warnings */ 839 bzero(&pinfo, sizeof(pinfo)); 840 841 error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0, 842 proc0.p_ucred, NULL, NULL); 843 if (error) 844 return (error); 845 846 if (pinfo.media_blksize == 0) 847 return (ENXIO); 848 849 if (blkno) /* XXX: make sure this reserved stuff is right */ 850 *blkno = pinfo.reserved_blocks + 851 pinfo.media_offset / pinfo.media_blksize; 852 if (secsize) 853 *secsize = pinfo.media_blksize; 854 if (size) 855 *size = (pinfo.media_blocks - pinfo.reserved_blocks); 856 857 return (0); 858 } 859 860 int 861 disk_dumpconf(cdev_t dev, u_int onoff) 862 { 863 struct dumperinfo di; 864 u_int64_t size, blkno; 865 u_int32_t secsize; 866 int error; 867 868 if (!onoff) 869 return set_dumper(NULL); 870 871 error = disk_dumpcheck(dev, &size, &blkno, &secsize); 872 873 if (error) 874 return ENXIO; 875 876 bzero(&di, sizeof(struct dumperinfo)); 877 di.dumper = diskdump; 878 di.priv = dev; 879 di.blocksize = secsize; 880 di.maxiosize = dev->si_iosize_max; 881 di.mediaoffset = blkno * DEV_BSIZE; 882 di.mediasize = size * DEV_BSIZE; 883 884 return set_dumper(&di); 885 } 886 887 void 888 disk_unprobe(struct disk *disk) 889 { 890 if (disk == NULL) 891 return; 892 893 disk_msg_send_sync(DISK_UNPROBE, disk, NULL); 894 } 895 896 void 897 disk_invalidate (struct disk *disk) 898 { 899 dsgone(&disk->d_slice); 900 } 901 902 /* 903 * Enumerate disks, pass a marker and an initial NULL dp to initialize, 904 * then loop with the previously returned dp. 905 * 906 * The returned dp will be referenced, preventing its destruction. When 907 * you pass the returned dp back into the loop the ref is dropped. 908 * 909 * WARNING: If terminating your loop early you must call 910 * disk_enumerate_stop(). 911 */ 912 struct disk * 913 disk_enumerate(struct disk *marker, struct disk *dp) 914 { 915 lwkt_gettoken(&disklist_token); 916 if (dp) { 917 --dp->d_refs; 918 dp = LIST_NEXT(marker, d_list); 919 LIST_REMOVE(marker, d_list); 920 } else { 921 bzero(marker, sizeof(*marker)); 922 marker->d_flags = DISKFLAG_MARKER; 923 dp = LIST_FIRST(&disklist); 924 } 925 while (dp) { 926 if ((dp->d_flags & DISKFLAG_MARKER) == 0) 927 break; 928 dp = LIST_NEXT(dp, d_list); 929 } 930 if (dp) { 931 ++dp->d_refs; 932 LIST_INSERT_AFTER(dp, marker, d_list); 933 } 934 lwkt_reltoken(&disklist_token); 935 return (dp); 936 } 937 938 /* 939 * Terminate an enumeration early. Do not call this function if the 940 * enumeration ended normally. dp can be NULL, indicating that you 941 * wish to retain the ref count on dp. 942 * 943 * This function removes the marker. 944 */ 945 void 946 disk_enumerate_stop(struct disk *marker, struct disk *dp) 947 { 948 lwkt_gettoken(&disklist_token); 949 LIST_REMOVE(marker, d_list); 950 if (dp) 951 --dp->d_refs; 952 lwkt_reltoken(&disklist_token); 953 } 954 955 static 956 int 957 sysctl_disks(SYSCTL_HANDLER_ARGS) 958 { 959 struct disk marker; 960 struct disk *dp; 961 int error, first; 962 963 first = 1; 964 error = 0; 965 dp = NULL; 966 967 while ((dp = disk_enumerate(&marker, dp))) { 968 if (!first) { 969 error = SYSCTL_OUT(req, " ", 1); 970 if (error) { 971 disk_enumerate_stop(&marker, dp); 972 break; 973 } 974 } else { 975 first = 0; 976 } 977 error = SYSCTL_OUT(req, dp->d_rawdev->si_name, 978 strlen(dp->d_rawdev->si_name)); 979 if (error) { 980 disk_enumerate_stop(&marker, dp); 981 break; 982 } 983 } 984 if (error == 0) 985 error = SYSCTL_OUT(req, "", 1); 986 return error; 987 } 988 989 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0, 990 sysctl_disks, "A", "names of available disks"); 991 992 /* 993 * Open a disk device or partition. 994 */ 995 static 996 int 997 diskopen(struct dev_open_args *ap) 998 { 999 cdev_t dev = ap->a_head.a_dev; 1000 struct disk *dp; 1001 int error; 1002 1003 /* 1004 * dp can't be NULL here XXX. 1005 * 1006 * d_slice will be NULL if setdiskinfo() has not been called yet. 1007 * setdiskinfo() is typically called whether the disk is present 1008 * or not (e.g. CD), but the base disk device is created first 1009 * and there may be a race. 1010 */ 1011 dp = dev->si_disk; 1012 if (dp == NULL || dp->d_slice == NULL) 1013 return (ENXIO); 1014 error = 0; 1015 1016 /* 1017 * Deal with open races 1018 */ 1019 lwkt_gettoken(&ds_token); 1020 while (dp->d_flags & DISKFLAG_LOCK) { 1021 dp->d_flags |= DISKFLAG_WANTED; 1022 error = tsleep(dp, PCATCH, "diskopen", hz); 1023 if (error) { 1024 lwkt_reltoken(&ds_token); 1025 return (error); 1026 } 1027 } 1028 dp->d_flags |= DISKFLAG_LOCK; 1029 1030 /* 1031 * Open the underlying raw device. 1032 */ 1033 if (!dsisopen(dp->d_slice)) { 1034 #if 0 1035 if (!pdev->si_iosize_max) 1036 pdev->si_iosize_max = dev->si_iosize_max; 1037 #endif 1038 error = dev_dopen(dp->d_rawdev, ap->a_oflags, 1039 ap->a_devtype, ap->a_cred, NULL); 1040 } 1041 1042 if (error) 1043 goto out; 1044 error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags, 1045 &dp->d_slice, &dp->d_info); 1046 if (!dsisopen(dp->d_slice)) { 1047 dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype, NULL); 1048 } 1049 out: 1050 dp->d_flags &= ~DISKFLAG_LOCK; 1051 if (dp->d_flags & DISKFLAG_WANTED) { 1052 dp->d_flags &= ~DISKFLAG_WANTED; 1053 wakeup(dp); 1054 } 1055 lwkt_reltoken(&ds_token); 1056 1057 KKASSERT(dp->d_opencount >= 0); 1058 /* If the open was successful, bump open count */ 1059 if (error == 0) 1060 atomic_add_int(&dp->d_opencount, 1); 1061 1062 return(error); 1063 } 1064 1065 /* 1066 * Close a disk device or partition 1067 */ 1068 static 1069 int 1070 diskclose(struct dev_close_args *ap) 1071 { 1072 cdev_t dev = ap->a_head.a_dev; 1073 struct disk *dp; 1074 int error; 1075 int lcount; 1076 1077 error = 0; 1078 dp = dev->si_disk; 1079 1080 /* 1081 * The cdev_t represents the disk/slice/part. The shared 1082 * dp structure governs all cdevs associated with the disk. 1083 * 1084 * As a safety only close the underlying raw device on the last 1085 * close the disk device if our tracking of the slices/partitions 1086 * also indicates nothing is open. 1087 */ 1088 KKASSERT(dp->d_opencount >= 1); 1089 lcount = atomic_fetchadd_int(&dp->d_opencount, -1); 1090 1091 lwkt_gettoken(&ds_token); 1092 dsclose(dev, ap->a_devtype, dp->d_slice); 1093 if (lcount <= 1 && !dsisopen(dp->d_slice)) { 1094 error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype, NULL); 1095 } 1096 lwkt_reltoken(&ds_token); 1097 1098 return (error); 1099 } 1100 1101 /* 1102 * First execute the ioctl on the disk device, and if it isn't supported 1103 * try running it on the backing device. 1104 */ 1105 static 1106 int 1107 diskioctl(struct dev_ioctl_args *ap) 1108 { 1109 cdev_t dev = ap->a_head.a_dev; 1110 struct disk *dp; 1111 int error; 1112 u_int u; 1113 1114 dp = dev->si_disk; 1115 if (dp == NULL) 1116 return (ENXIO); 1117 1118 devfs_debug(DEVFS_DEBUG_DEBUG, 1119 "diskioctl: cmd is: %lx (name: %s)\n", 1120 ap->a_cmd, dev->si_name); 1121 devfs_debug(DEVFS_DEBUG_DEBUG, 1122 "diskioctl: &dp->d_slice is: %p, %p\n", 1123 &dp->d_slice, dp->d_slice); 1124 1125 if (ap->a_cmd == DIOCGKERNELDUMP) { 1126 u = *(u_int *)ap->a_data; 1127 return disk_dumpconf(dev, u); 1128 } 1129 1130 if (ap->a_cmd == DIOCRECLUSTER && dev == dp->d_cdev) { 1131 error = disk_iocom_ioctl(dp, ap->a_cmd, ap->a_data); 1132 return error; 1133 } 1134 1135 if (&dp->d_slice == NULL || dp->d_slice == NULL || 1136 ((dp->d_info.d_dsflags & DSO_DEVICEMAPPER) && 1137 dkslice(dev) == WHOLE_DISK_SLICE)) { 1138 error = ENOIOCTL; 1139 } else { 1140 lwkt_gettoken(&ds_token); 1141 error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag, 1142 &dp->d_slice, &dp->d_info); 1143 lwkt_reltoken(&ds_token); 1144 } 1145 1146 if (error == ENOIOCTL) { 1147 error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data, 1148 ap->a_fflag, ap->a_cred, NULL, NULL); 1149 } 1150 return (error); 1151 } 1152 1153 /* 1154 * Execute strategy routine 1155 */ 1156 static 1157 int 1158 diskstrategy(struct dev_strategy_args *ap) 1159 { 1160 cdev_t dev = ap->a_head.a_dev; 1161 struct bio *bio = ap->a_bio; 1162 struct bio *nbio; 1163 struct disk *dp; 1164 1165 dp = dev->si_disk; 1166 1167 if (dp == NULL) { 1168 bio->bio_buf->b_error = ENXIO; 1169 bio->bio_buf->b_flags |= B_ERROR; 1170 biodone(bio); 1171 return(0); 1172 } 1173 KKASSERT(dev->si_disk == dp); 1174 1175 /* 1176 * The dscheck() function will also transform the slice relative 1177 * block number i.e. bio->bio_offset into a block number that can be 1178 * passed directly to the underlying raw device. If dscheck() 1179 * returns NULL it will have handled the bio for us (e.g. EOF 1180 * or error due to being beyond the device size). 1181 */ 1182 if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) { 1183 dev_dstrategy(dp->d_rawdev, nbio); 1184 } else { 1185 biodone(bio); 1186 } 1187 return(0); 1188 } 1189 1190 /* 1191 * Return the partition size in ?blocks? 1192 */ 1193 static 1194 int 1195 diskpsize(struct dev_psize_args *ap) 1196 { 1197 cdev_t dev = ap->a_head.a_dev; 1198 struct disk *dp; 1199 1200 dp = dev->si_disk; 1201 if (dp == NULL) 1202 return(ENODEV); 1203 1204 ap->a_result = dssize(dev, &dp->d_slice); 1205 1206 if ((ap->a_result == -1) && 1207 (dp->d_info.d_dsflags & DSO_RAWPSIZE)) { 1208 ap->a_head.a_dev = dp->d_rawdev; 1209 return dev_doperate(&ap->a_head); 1210 } 1211 return(0); 1212 } 1213 1214 static int 1215 diskdump(struct dev_dump_args *ap) 1216 { 1217 cdev_t dev = ap->a_head.a_dev; 1218 struct disk *dp = dev->si_disk; 1219 u_int64_t size, offset; 1220 int error; 1221 1222 error = disk_dumpcheck(dev, &size, &ap->a_blkno, &ap->a_secsize); 1223 /* XXX: this should probably go in disk_dumpcheck somehow */ 1224 if (ap->a_length != 0) { 1225 size *= DEV_BSIZE; 1226 offset = ap->a_blkno * DEV_BSIZE; 1227 if ((ap->a_offset < offset) || 1228 (ap->a_offset + ap->a_length - offset > size)) { 1229 kprintf("Attempt to write outside dump " 1230 "device boundaries.\n"); 1231 error = ENOSPC; 1232 } 1233 } 1234 1235 if (error == 0) { 1236 ap->a_head.a_dev = dp->d_rawdev; 1237 error = dev_doperate(&ap->a_head); 1238 } 1239 1240 return(error); 1241 } 1242 1243 1244 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD, 1245 0, sizeof(struct diskslices), "sizeof(struct diskslices)"); 1246 1247 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD, 1248 0, sizeof(struct disk), "sizeof(struct disk)"); 1249 1250 /* 1251 * Reorder interval for burst write allowance and minor write 1252 * allowance. 1253 * 1254 * We always want to trickle some writes in to make use of the 1255 * disk's zone cache. Bursting occurs on a longer interval and only 1256 * runningbufspace is well over the hirunningspace limit. 1257 */ 1258 int bioq_reorder_burst_interval = 60; /* should be multiple of minor */ 1259 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_interval, 1260 CTLFLAG_RW, &bioq_reorder_burst_interval, 0, ""); 1261 int bioq_reorder_minor_interval = 5; 1262 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_interval, 1263 CTLFLAG_RW, &bioq_reorder_minor_interval, 0, ""); 1264 1265 int bioq_reorder_burst_bytes = 3000000; 1266 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_bytes, 1267 CTLFLAG_RW, &bioq_reorder_burst_bytes, 0, ""); 1268 int bioq_reorder_minor_bytes = 262144; 1269 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_bytes, 1270 CTLFLAG_RW, &bioq_reorder_minor_bytes, 0, ""); 1271 1272 1273 /* 1274 * Order I/Os. Generally speaking this code is designed to make better 1275 * use of drive zone caches. A drive zone cache can typically track linear 1276 * reads or writes for around 16 zones simultaniously. 1277 * 1278 * Read prioritization issues: It is possible for hundreds of megabytes worth 1279 * of writes to be queued asynchronously. This creates a huge bottleneck 1280 * for reads which reduce read bandwidth to a trickle. 1281 * 1282 * To solve this problem we generally reorder reads before writes. 1283 * 1284 * However, a large number of random reads can also starve writes and 1285 * make poor use of the drive zone cache so we allow writes to trickle 1286 * in every N reads. 1287 */ 1288 void 1289 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio) 1290 { 1291 #if 0 1292 /* 1293 * The BIO wants to be ordered. Adding to the tail also 1294 * causes transition to be set to NULL, forcing the ordering 1295 * of all prior I/O's. 1296 */ 1297 if (bio->bio_buf->b_flags & B_ORDERED) { 1298 bioq_insert_tail(bioq, bio); 1299 return; 1300 } 1301 #endif 1302 1303 switch(bio->bio_buf->b_cmd) { 1304 case BUF_CMD_READ: 1305 if (bioq->transition) { 1306 /* 1307 * Insert before the first write. Bleedover writes 1308 * based on reorder intervals to prevent starvation. 1309 */ 1310 TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act); 1311 ++bioq->reorder; 1312 if (bioq->reorder % bioq_reorder_minor_interval == 0) { 1313 bioqwritereorder(bioq); 1314 if (bioq->reorder >= 1315 bioq_reorder_burst_interval) { 1316 bioq->reorder = 0; 1317 } 1318 } 1319 } else { 1320 /* 1321 * No writes queued (or ordering was forced), 1322 * insert at tail. 1323 */ 1324 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act); 1325 } 1326 break; 1327 case BUF_CMD_WRITE: 1328 /* 1329 * Writes are always appended. If no writes were previously 1330 * queued or an ordered tail insertion occured the transition 1331 * field will be NULL. 1332 */ 1333 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act); 1334 if (bioq->transition == NULL) 1335 bioq->transition = bio; 1336 break; 1337 default: 1338 /* 1339 * All other request types are forced to be ordered. 1340 */ 1341 bioq_insert_tail(bioq, bio); 1342 break; 1343 } 1344 } 1345 1346 /* 1347 * Move the read-write transition point to prevent reads from 1348 * completely starving our writes. This brings a number of writes into 1349 * the fold every N reads. 1350 * 1351 * We bring a few linear writes into the fold on a minor interval 1352 * and we bring a non-linear burst of writes into the fold on a major 1353 * interval. Bursting only occurs if runningbufspace is really high 1354 * (typically from syncs, fsyncs, or HAMMER flushes). 1355 */ 1356 static 1357 void 1358 bioqwritereorder(struct bio_queue_head *bioq) 1359 { 1360 struct bio *bio; 1361 off_t next_offset; 1362 size_t left; 1363 size_t n; 1364 int check_off; 1365 1366 if (bioq->reorder < bioq_reorder_burst_interval || 1367 !buf_runningbufspace_severe()) { 1368 left = (size_t)bioq_reorder_minor_bytes; 1369 check_off = 1; 1370 } else { 1371 left = (size_t)bioq_reorder_burst_bytes; 1372 check_off = 0; 1373 } 1374 1375 next_offset = bioq->transition->bio_offset; 1376 while ((bio = bioq->transition) != NULL && 1377 (check_off == 0 || next_offset == bio->bio_offset) 1378 ) { 1379 n = bio->bio_buf->b_bcount; 1380 next_offset = bio->bio_offset + n; 1381 bioq->transition = TAILQ_NEXT(bio, bio_act); 1382 if (left < n) 1383 break; 1384 left -= n; 1385 } 1386 } 1387 1388 /* 1389 * Bounds checking against the media size, used for the raw partition. 1390 * secsize, mediasize and b_blkno must all be the same units. 1391 * Possibly this has to be DEV_BSIZE (512). 1392 */ 1393 int 1394 bounds_check_with_mediasize(struct bio *bio, int secsize, uint64_t mediasize) 1395 { 1396 struct buf *bp = bio->bio_buf; 1397 int64_t sz; 1398 1399 sz = howmany(bp->b_bcount, secsize); 1400 1401 if (bio->bio_offset/DEV_BSIZE + sz > mediasize) { 1402 sz = mediasize - bio->bio_offset/DEV_BSIZE; 1403 if (sz == 0) { 1404 /* If exactly at end of disk, return EOF. */ 1405 bp->b_resid = bp->b_bcount; 1406 return 0; 1407 } 1408 if (sz < 0) { 1409 /* If past end of disk, return EINVAL. */ 1410 bp->b_error = EINVAL; 1411 return 0; 1412 } 1413 /* Otherwise, truncate request. */ 1414 bp->b_bcount = sz * secsize; 1415 } 1416 1417 return 1; 1418 } 1419 1420 /* 1421 * Disk error is the preface to plaintive error messages 1422 * about failing disk transfers. It prints messages of the form 1423 1424 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d) 1425 1426 * if the offset of the error in the transfer and a disk label 1427 * are both available. blkdone should be -1 if the position of the error 1428 * is unknown; the disklabel pointer may be null from drivers that have not 1429 * been converted to use them. The message is printed with kprintf 1430 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority. 1431 * The message should be completed (with at least a newline) with kprintf 1432 * or log(-1, ...), respectively. There is no trailing space. 1433 */ 1434 void 1435 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt) 1436 { 1437 struct buf *bp = bio->bio_buf; 1438 const char *term; 1439 1440 switch(bp->b_cmd) { 1441 case BUF_CMD_READ: 1442 term = "read"; 1443 break; 1444 case BUF_CMD_WRITE: 1445 term = "write"; 1446 break; 1447 default: 1448 term = "access"; 1449 break; 1450 } 1451 kprintf("%s: %s %sing ", dev->si_name, what, term); 1452 kprintf("offset %012llx for %d", 1453 (long long)bio->bio_offset, 1454 bp->b_bcount); 1455 1456 if (donecnt) 1457 kprintf(" (%d bytes completed)", donecnt); 1458 } 1459 1460 /* 1461 * Locate a disk device 1462 */ 1463 cdev_t 1464 disk_locate(const char *devname) 1465 { 1466 return devfs_find_device_by_name("%s", devname); 1467 } 1468 1469 void 1470 disk_config(void *arg) 1471 { 1472 disk_msg_send_sync(DISK_SYNC, NULL, NULL); 1473 } 1474 1475 static void 1476 disk_init(void) 1477 { 1478 struct thread* td_core; 1479 1480 disk_msg_cache = objcache_create("disk-msg-cache", 0, 0, 1481 NULL, NULL, NULL, 1482 objcache_malloc_alloc, 1483 objcache_malloc_free, 1484 &disk_msg_malloc_args); 1485 1486 lwkt_token_init(&disklist_token, "disks"); 1487 lwkt_token_init(&ds_token, "ds"); 1488 1489 /* 1490 * Initialize the reply-only port which acts as a message drain 1491 */ 1492 lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply); 1493 1494 lwkt_gettoken(&disklist_token); 1495 lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL, 1496 0, -1, "disk_msg_core"); 1497 tsleep(td_core, 0, "diskcore", 0); 1498 lwkt_reltoken(&disklist_token); 1499 } 1500 1501 static void 1502 disk_uninit(void) 1503 { 1504 objcache_destroy(disk_msg_cache); 1505 } 1506 1507 /* 1508 * Clean out illegal characters in serial numbers. 1509 */ 1510 static void 1511 disk_cleanserial(char *serno) 1512 { 1513 char c; 1514 1515 while ((c = *serno) != 0) { 1516 if (c >= 'a' && c <= 'z') 1517 ; 1518 else if (c >= 'A' && c <= 'Z') 1519 ; 1520 else if (c >= '0' && c <= '9') 1521 ; 1522 else if (c == '-' || c == '@' || c == '+' || c == '.') 1523 ; 1524 else 1525 c = '_'; 1526 *serno++= c; 1527 } 1528 } 1529 1530 TUNABLE_INT("kern.disk_debug", &disk_debug_enable); 1531 SYSCTL_INT(_kern, OID_AUTO, disk_debug, CTLFLAG_RW, &disk_debug_enable, 1532 0, "Enable subr_disk debugging"); 1533 1534 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL); 1535 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL); 1536