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. All advertising materials mentioning features or use of this software 59 * must display the following acknowledgement: 60 * This product includes software developed by the University of 61 * California, Berkeley and its contributors. 62 * 4. Neither the name of the University nor the names of its contributors 63 * may be used to endorse or promote products derived from this software 64 * without specific prior written permission. 65 * 66 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 67 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 68 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 69 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 70 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 71 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 72 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 73 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 74 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 75 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 76 * SUCH DAMAGE. 77 * 78 * @(#)ufs_disksubr.c 8.5 (Berkeley) 1/21/94 79 * $FreeBSD: src/sys/kern/subr_disk.c,v 1.20.2.6 2001/10/05 07:14:57 peter Exp $ 80 * $FreeBSD: src/sys/ufs/ufs/ufs_disksubr.c,v 1.44.2.3 2001/03/05 05:42:19 obrien 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 <machine/md_var.h> 99 #include <sys/ctype.h> 100 #include <sys/syslog.h> 101 #include <sys/device.h> 102 #include <sys/msgport.h> 103 #include <sys/devfs.h> 104 #include <sys/thread.h> 105 #include <sys/dsched.h> 106 #include <sys/queue.h> 107 #include <sys/lock.h> 108 #include <sys/udev.h> 109 #include <sys/uuid.h> 110 111 #include <sys/buf2.h> 112 #include <sys/mplock2.h> 113 #include <sys/msgport2.h> 114 #include <sys/thread2.h> 115 116 static MALLOC_DEFINE(M_DISK, "disk", "disk data"); 117 static int disk_debug_enable = 0; 118 119 static void disk_msg_autofree_reply(lwkt_port_t, lwkt_msg_t); 120 static void disk_msg_core(void *); 121 static int disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe); 122 static void disk_probe(struct disk *dp, int reprobe); 123 static void _setdiskinfo(struct disk *disk, struct disk_info *info); 124 static void bioqwritereorder(struct bio_queue_head *bioq); 125 static void disk_cleanserial(char *serno); 126 static int disk_debug(int, char *, ...) __printflike(2, 3); 127 static cdev_t _disk_create_named(const char *name, int unit, struct disk *dp, 128 struct dev_ops *raw_ops, int clone); 129 130 static d_open_t diskopen; 131 static d_close_t diskclose; 132 static d_ioctl_t diskioctl; 133 static d_strategy_t diskstrategy; 134 static d_psize_t diskpsize; 135 static d_dump_t diskdump; 136 137 static LIST_HEAD(, disk) disklist = LIST_HEAD_INITIALIZER(&disklist); 138 static struct lwkt_token disklist_token; 139 140 static struct dev_ops disk_ops = { 141 { "disk", 0, D_DISK | D_MPSAFE | D_TRACKCLOSE }, 142 .d_open = diskopen, 143 .d_close = diskclose, 144 .d_read = physread, 145 .d_write = physwrite, 146 .d_ioctl = diskioctl, 147 .d_strategy = diskstrategy, 148 .d_dump = diskdump, 149 .d_psize = diskpsize, 150 }; 151 152 static struct objcache *disk_msg_cache; 153 154 struct objcache_malloc_args disk_msg_malloc_args = { 155 sizeof(struct disk_msg), M_DISK }; 156 157 static struct lwkt_port disk_dispose_port; 158 static struct lwkt_port disk_msg_port; 159 160 static int 161 disk_debug(int level, char *fmt, ...) 162 { 163 __va_list ap; 164 165 __va_start(ap, fmt); 166 if (level <= disk_debug_enable) 167 kvprintf(fmt, ap); 168 __va_end(ap); 169 170 return 0; 171 } 172 173 static int 174 disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe) 175 { 176 struct disk_info *info = &dp->d_info; 177 struct diskslice *sp = &dp->d_slice->dss_slices[slice]; 178 disklabel_ops_t ops; 179 struct partinfo part; 180 const char *msg; 181 char uuid_buf[128]; 182 cdev_t ndev; 183 int sno; 184 u_int i; 185 186 disk_debug(2, "disk_probe_slice (begin): %s (%s)\n", 187 dev->si_name, dp->d_cdev->si_name); 188 189 sno = slice ? slice - 1 : 0; 190 191 ops = &disklabel32_ops; 192 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info); 193 if (msg && !strcmp(msg, "no disk label")) { 194 ops = &disklabel64_ops; 195 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info); 196 } 197 198 if (msg == NULL) { 199 if (slice != WHOLE_DISK_SLICE) 200 ops->op_adjust_label_reserved(dp->d_slice, slice, sp); 201 else 202 sp->ds_reserved = 0; 203 204 sp->ds_ops = ops; 205 for (i = 0; i < ops->op_getnumparts(sp->ds_label); i++) { 206 ops->op_loadpartinfo(sp->ds_label, i, &part); 207 if (part.fstype) { 208 if (reprobe && 209 (ndev = devfs_find_device_by_name("%s%c", 210 dev->si_name, 'a' + i)) 211 ) { 212 /* 213 * Device already exists and 214 * is still valid. 215 */ 216 ndev->si_flags |= SI_REPROBE_TEST; 217 218 /* 219 * Destroy old UUID alias 220 */ 221 destroy_dev_alias(ndev, "part-by-uuid/*"); 222 223 /* Create UUID alias */ 224 if (!kuuid_is_nil(&part.storage_uuid)) { 225 snprintf_uuid(uuid_buf, 226 sizeof(uuid_buf), 227 &part.storage_uuid); 228 make_dev_alias(ndev, 229 "part-by-uuid/%s", 230 uuid_buf); 231 udev_dict_set_cstr(ndev, "uuid", uuid_buf); 232 } 233 } else { 234 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops, 235 dkmakeminor(dkunit(dp->d_cdev), 236 slice, i), 237 UID_ROOT, GID_OPERATOR, 0640, 238 "%s%c", dev->si_name, 'a'+ i); 239 ndev->si_parent = dev; 240 ndev->si_iosize_max = dev->si_iosize_max; 241 ndev->si_disk = dp; 242 udev_dict_set_cstr(ndev, "subsystem", "disk"); 243 /* Inherit parent's disk type */ 244 if (dp->d_disktype) { 245 udev_dict_set_cstr(ndev, "disk-type", 246 __DECONST(char *, dp->d_disktype)); 247 } 248 249 /* Create serno alias */ 250 if (dp->d_info.d_serialno) { 251 make_dev_alias(ndev, 252 "serno/%s.s%d%c", 253 dp->d_info.d_serialno, 254 sno, 'a' + i); 255 } 256 257 /* Create UUID alias */ 258 if (!kuuid_is_nil(&part.storage_uuid)) { 259 snprintf_uuid(uuid_buf, 260 sizeof(uuid_buf), 261 &part.storage_uuid); 262 make_dev_alias(ndev, 263 "part-by-uuid/%s", 264 uuid_buf); 265 udev_dict_set_cstr(ndev, "uuid", uuid_buf); 266 } 267 ndev->si_flags |= SI_REPROBE_TEST; 268 } 269 } 270 } 271 } else if (info->d_dsflags & DSO_COMPATLABEL) { 272 msg = NULL; 273 if (sp->ds_size >= 0x100000000ULL) 274 ops = &disklabel64_ops; 275 else 276 ops = &disklabel32_ops; 277 sp->ds_label = ops->op_clone_label(info, sp); 278 } else { 279 if (sp->ds_type == DOSPTYP_386BSD || /* XXX */ 280 sp->ds_type == DOSPTYP_NETBSD || 281 sp->ds_type == DOSPTYP_OPENBSD) { 282 log(LOG_WARNING, "%s: cannot find label (%s)\n", 283 dev->si_name, msg); 284 } 285 286 if (sp->ds_label.opaque != NULL && sp->ds_ops != NULL) { 287 /* Clear out old label - it's not around anymore */ 288 disk_debug(2, 289 "disk_probe_slice: clear out old diskabel on %s\n", 290 dev->si_name); 291 292 sp->ds_ops->op_freedisklabel(&sp->ds_label); 293 sp->ds_ops = NULL; 294 } 295 } 296 297 if (msg == NULL) { 298 sp->ds_wlabel = FALSE; 299 } 300 301 return (msg ? EINVAL : 0); 302 } 303 304 /* 305 * This routine is only called for newly minted drives or to reprobe 306 * a drive with no open slices. disk_probe_slice() is called directly 307 * when reprobing partition changes within slices. 308 */ 309 static void 310 disk_probe(struct disk *dp, int reprobe) 311 { 312 struct disk_info *info = &dp->d_info; 313 cdev_t dev = dp->d_cdev; 314 cdev_t ndev; 315 int error, i, sno; 316 struct diskslices *osp; 317 struct diskslice *sp; 318 char uuid_buf[128]; 319 320 KKASSERT (info->d_media_blksize != 0); 321 322 osp = dp->d_slice; 323 dp->d_slice = dsmakeslicestruct(BASE_SLICE, info); 324 disk_debug(1, "disk_probe (begin): %s\n", dp->d_cdev->si_name); 325 326 error = mbrinit(dev, info, &(dp->d_slice)); 327 if (error) { 328 dsgone(&osp); 329 return; 330 } 331 332 for (i = 0; i < dp->d_slice->dss_nslices; i++) { 333 /* 334 * Ignore the whole-disk slice, it has already been created. 335 */ 336 if (i == WHOLE_DISK_SLICE) 337 continue; 338 339 #if 1 340 /* 341 * Ignore the compatibility slice s0 if it's a device mapper 342 * volume. 343 */ 344 if ((i == COMPATIBILITY_SLICE) && 345 (info->d_dsflags & DSO_DEVICEMAPPER)) 346 continue; 347 #endif 348 349 sp = &dp->d_slice->dss_slices[i]; 350 351 /* 352 * Handle s0. s0 is a compatibility slice if there are no 353 * other slices and it has not otherwise been set up, else 354 * we ignore it. 355 */ 356 if (i == COMPATIBILITY_SLICE) { 357 sno = 0; 358 if (sp->ds_type == 0 && 359 dp->d_slice->dss_nslices == BASE_SLICE) { 360 sp->ds_size = info->d_media_blocks; 361 sp->ds_reserved = 0; 362 } 363 } else { 364 sno = i - 1; 365 sp->ds_reserved = 0; 366 } 367 368 /* 369 * Ignore 0-length slices 370 */ 371 if (sp->ds_size == 0) 372 continue; 373 374 if (reprobe && 375 (ndev = devfs_find_device_by_name("%ss%d", 376 dev->si_name, sno))) { 377 /* 378 * Device already exists and is still valid 379 */ 380 ndev->si_flags |= SI_REPROBE_TEST; 381 382 /* 383 * Destroy old UUID alias 384 */ 385 destroy_dev_alias(ndev, "slice-by-uuid/*"); 386 387 /* Create UUID alias */ 388 if (!kuuid_is_nil(&sp->ds_stor_uuid)) { 389 snprintf_uuid(uuid_buf, sizeof(uuid_buf), 390 &sp->ds_stor_uuid); 391 make_dev_alias(ndev, "slice-by-uuid/%s", 392 uuid_buf); 393 } 394 } else { 395 /* 396 * Else create new device 397 */ 398 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops, 399 dkmakewholeslice(dkunit(dev), i), 400 UID_ROOT, GID_OPERATOR, 0640, 401 (info->d_dsflags & DSO_DEVICEMAPPER)? 402 "%s.s%d" : "%ss%d", dev->si_name, sno); 403 ndev->si_parent = dev; 404 ndev->si_iosize_max = dev->si_iosize_max; 405 udev_dict_set_cstr(ndev, "subsystem", "disk"); 406 /* Inherit parent's disk type */ 407 if (dp->d_disktype) { 408 udev_dict_set_cstr(ndev, "disk-type", 409 __DECONST(char *, dp->d_disktype)); 410 } 411 412 /* Create serno alias */ 413 if (dp->d_info.d_serialno) { 414 make_dev_alias(ndev, "serno/%s.s%d", 415 dp->d_info.d_serialno, sno); 416 } 417 418 /* Create UUID alias */ 419 if (!kuuid_is_nil(&sp->ds_stor_uuid)) { 420 snprintf_uuid(uuid_buf, sizeof(uuid_buf), 421 &sp->ds_stor_uuid); 422 make_dev_alias(ndev, "slice-by-uuid/%s", 423 uuid_buf); 424 } 425 426 ndev->si_disk = dp; 427 ndev->si_flags |= SI_REPROBE_TEST; 428 } 429 sp->ds_dev = ndev; 430 431 /* 432 * Probe appropriate slices for a disklabel 433 * 434 * XXX slice type 1 used by our gpt probe code. 435 * XXX slice type 0 used by mbr compat slice. 436 */ 437 if (sp->ds_type == DOSPTYP_386BSD || 438 sp->ds_type == DOSPTYP_NETBSD || 439 sp->ds_type == DOSPTYP_OPENBSD || 440 sp->ds_type == 0 || 441 sp->ds_type == 1) { 442 if (dp->d_slice->dss_first_bsd_slice == 0) 443 dp->d_slice->dss_first_bsd_slice = i; 444 disk_probe_slice(dp, ndev, i, reprobe); 445 } 446 } 447 dsgone(&osp); 448 disk_debug(1, "disk_probe (end): %s\n", dp->d_cdev->si_name); 449 } 450 451 452 static void 453 disk_msg_core(void *arg) 454 { 455 struct disk *dp; 456 struct diskslice *sp; 457 disk_msg_t msg; 458 int run; 459 460 lwkt_gettoken(&disklist_token); 461 lwkt_initport_thread(&disk_msg_port, curthread); 462 wakeup(curthread); /* synchronous startup */ 463 lwkt_reltoken(&disklist_token); 464 465 get_mplock(); /* not mpsafe yet? */ 466 run = 1; 467 468 while (run) { 469 msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0); 470 471 switch (msg->hdr.u.ms_result) { 472 case DISK_DISK_PROBE: 473 dp = (struct disk *)msg->load; 474 disk_debug(1, 475 "DISK_DISK_PROBE: %s\n", 476 dp->d_cdev->si_name); 477 disk_iocom_update(dp); 478 disk_probe(dp, 0); 479 break; 480 case DISK_DISK_DESTROY: 481 dp = (struct disk *)msg->load; 482 disk_debug(1, 483 "DISK_DISK_DESTROY: %s\n", 484 dp->d_cdev->si_name); 485 disk_iocom_uninit(dp); 486 487 /* 488 * Interlock against struct disk enumerations. 489 * Wait for enumerations to complete then remove 490 * the dp from the list before tearing it down. 491 * 492 * This avoids races against e.g. 493 * dsched_thread_io_alloc(). 494 */ 495 lwkt_gettoken(&disklist_token); 496 while (dp->d_refs) 497 tsleep(&dp->d_refs, 0, "diskdel", hz / 10); 498 LIST_REMOVE(dp, d_list); 499 500 dsched_disk_destroy_callback(dp); 501 devfs_destroy_related(dp->d_cdev); 502 destroy_dev(dp->d_cdev); 503 destroy_only_dev(dp->d_rawdev); 504 505 lwkt_reltoken(&disklist_token); 506 507 if (dp->d_info.d_serialno) { 508 kfree(dp->d_info.d_serialno, M_TEMP); 509 dp->d_info.d_serialno = NULL; 510 } 511 break; 512 case DISK_UNPROBE: 513 dp = (struct disk *)msg->load; 514 disk_debug(1, 515 "DISK_DISK_UNPROBE: %s\n", 516 dp->d_cdev->si_name); 517 devfs_destroy_related(dp->d_cdev); 518 break; 519 case DISK_SLICE_REPROBE: 520 dp = (struct disk *)msg->load; 521 sp = (struct diskslice *)msg->load2; 522 devfs_clr_related_flag(sp->ds_dev, 523 SI_REPROBE_TEST); 524 disk_debug(1, 525 "DISK_SLICE_REPROBE: %s\n", 526 sp->ds_dev->si_name); 527 disk_probe_slice(dp, sp->ds_dev, 528 dkslice(sp->ds_dev), 1); 529 devfs_destroy_related_without_flag( 530 sp->ds_dev, SI_REPROBE_TEST); 531 break; 532 case DISK_DISK_REPROBE: 533 dp = (struct disk *)msg->load; 534 devfs_clr_related_flag(dp->d_cdev, SI_REPROBE_TEST); 535 disk_debug(1, 536 "DISK_DISK_REPROBE: %s\n", 537 dp->d_cdev->si_name); 538 disk_probe(dp, 1); 539 devfs_destroy_related_without_flag( 540 dp->d_cdev, SI_REPROBE_TEST); 541 break; 542 case DISK_SYNC: 543 disk_debug(1, "DISK_SYNC\n"); 544 break; 545 default: 546 devfs_debug(DEVFS_DEBUG_WARNING, 547 "disk_msg_core: unknown message " 548 "received at core\n"); 549 break; 550 } 551 lwkt_replymsg(&msg->hdr, 0); 552 } 553 lwkt_exit(); 554 } 555 556 557 /* 558 * Acts as a message drain. Any message that is replied to here gets 559 * destroyed and the memory freed. 560 */ 561 static void 562 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg) 563 { 564 objcache_put(disk_msg_cache, msg); 565 } 566 567 568 void 569 disk_msg_send(uint32_t cmd, void *load, void *load2) 570 { 571 disk_msg_t disk_msg; 572 lwkt_port_t port = &disk_msg_port; 573 574 disk_msg = objcache_get(disk_msg_cache, M_WAITOK); 575 576 lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0); 577 578 disk_msg->hdr.u.ms_result = cmd; 579 disk_msg->load = load; 580 disk_msg->load2 = load2; 581 KKASSERT(port); 582 lwkt_sendmsg(port, &disk_msg->hdr); 583 } 584 585 void 586 disk_msg_send_sync(uint32_t cmd, void *load, void *load2) 587 { 588 struct lwkt_port rep_port; 589 disk_msg_t disk_msg; 590 lwkt_port_t port; 591 592 disk_msg = objcache_get(disk_msg_cache, M_WAITOK); 593 port = &disk_msg_port; 594 595 /* XXX could probably use curthread's built-in msgport */ 596 lwkt_initport_thread(&rep_port, curthread); 597 lwkt_initmsg(&disk_msg->hdr, &rep_port, 0); 598 599 disk_msg->hdr.u.ms_result = cmd; 600 disk_msg->load = load; 601 disk_msg->load2 = load2; 602 603 lwkt_sendmsg(port, &disk_msg->hdr); 604 lwkt_waitmsg(&disk_msg->hdr, 0); 605 objcache_put(disk_msg_cache, disk_msg); 606 } 607 608 /* 609 * Create a raw device for the dev_ops template (which is returned). Also 610 * create a slice and unit managed disk and overload the user visible 611 * device space with it. 612 * 613 * NOTE: The returned raw device is NOT a slice and unit managed device. 614 * It is an actual raw device representing the raw disk as specified by 615 * the passed dev_ops. The disk layer not only returns such a raw device, 616 * it also uses it internally when passing (modified) commands through. 617 */ 618 cdev_t 619 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops) 620 { 621 return _disk_create_named(NULL, unit, dp, raw_ops, 0); 622 } 623 624 cdev_t 625 disk_create_clone(int unit, struct disk *dp, 626 struct dev_ops *raw_ops) 627 { 628 return _disk_create_named(NULL, unit, dp, raw_ops, 1); 629 } 630 631 cdev_t 632 disk_create_named(const char *name, int unit, struct disk *dp, 633 struct dev_ops *raw_ops) 634 { 635 return _disk_create_named(name, unit, dp, raw_ops, 0); 636 } 637 638 cdev_t 639 disk_create_named_clone(const char *name, int unit, struct disk *dp, 640 struct dev_ops *raw_ops) 641 { 642 return _disk_create_named(name, unit, dp, raw_ops, 1); 643 } 644 645 static cdev_t 646 _disk_create_named(const char *name, int unit, struct disk *dp, 647 struct dev_ops *raw_ops, int clone) 648 { 649 cdev_t rawdev; 650 651 disk_debug(1, "disk_create (begin): %s%d\n", name, unit); 652 653 if (name) { 654 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit), 655 UID_ROOT, GID_OPERATOR, 0640, "%s", name); 656 } else { 657 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit), 658 UID_ROOT, GID_OPERATOR, 0640, 659 "%s%d", raw_ops->head.name, unit); 660 } 661 662 bzero(dp, sizeof(*dp)); 663 664 dp->d_rawdev = rawdev; 665 dp->d_raw_ops = raw_ops; 666 dp->d_dev_ops = &disk_ops; 667 668 if (name) { 669 if (clone) { 670 dp->d_cdev = make_only_dev_covering( 671 &disk_ops, dp->d_rawdev->si_ops, 672 dkmakewholedisk(unit), 673 UID_ROOT, GID_OPERATOR, 0640, 674 "%s", name); 675 } else { 676 dp->d_cdev = make_dev_covering( 677 &disk_ops, dp->d_rawdev->si_ops, 678 dkmakewholedisk(unit), 679 UID_ROOT, GID_OPERATOR, 0640, 680 "%s", name); 681 } 682 } else { 683 if (clone) { 684 dp->d_cdev = make_only_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 } else { 690 dp->d_cdev = make_dev_covering( 691 &disk_ops, dp->d_rawdev->si_ops, 692 dkmakewholedisk(unit), 693 UID_ROOT, GID_OPERATOR, 0640, 694 "%s%d", raw_ops->head.name, unit); 695 } 696 } 697 698 udev_dict_set_cstr(dp->d_cdev, "subsystem", "disk"); 699 dp->d_cdev->si_disk = dp; 700 701 if (name) 702 dsched_disk_create_callback(dp, name, unit); 703 else 704 dsched_disk_create_callback(dp, raw_ops->head.name, unit); 705 706 lwkt_gettoken(&disklist_token); 707 LIST_INSERT_HEAD(&disklist, dp, d_list); 708 lwkt_reltoken(&disklist_token); 709 710 disk_iocom_init(dp); 711 712 disk_debug(1, "disk_create (end): %s%d\n", 713 (name != NULL)?(name):(raw_ops->head.name), unit); 714 715 return (dp->d_rawdev); 716 } 717 718 int 719 disk_setdisktype(struct disk *disk, const char *type) 720 { 721 int error; 722 723 KKASSERT(disk != NULL); 724 725 disk->d_disktype = type; 726 error = udev_dict_set_cstr(disk->d_cdev, "disk-type", 727 __DECONST(char *, type)); 728 return error; 729 } 730 731 int 732 disk_getopencount(struct disk *disk) 733 { 734 return disk->d_opencount; 735 } 736 737 static void 738 _setdiskinfo(struct disk *disk, struct disk_info *info) 739 { 740 char *oldserialno; 741 742 oldserialno = disk->d_info.d_serialno; 743 bcopy(info, &disk->d_info, sizeof(disk->d_info)); 744 info = &disk->d_info; 745 746 disk_debug(1, "_setdiskinfo: %s\n", disk->d_cdev->si_name); 747 748 /* 749 * The serial number is duplicated so the caller can throw 750 * their copy away. 751 */ 752 if (info->d_serialno && info->d_serialno[0] && 753 (info->d_serialno[0] != ' ' || strlen(info->d_serialno) > 1)) { 754 info->d_serialno = kstrdup(info->d_serialno, M_TEMP); 755 disk_cleanserial(info->d_serialno); 756 if (disk->d_cdev) { 757 make_dev_alias(disk->d_cdev, "serno/%s", 758 info->d_serialno); 759 } 760 } else { 761 info->d_serialno = NULL; 762 } 763 if (oldserialno) 764 kfree(oldserialno, M_TEMP); 765 766 dsched_disk_update_callback(disk, info); 767 768 /* 769 * The caller may set d_media_size or d_media_blocks and we 770 * calculate the other. 771 */ 772 KKASSERT(info->d_media_size == 0 || info->d_media_blocks == 0); 773 if (info->d_media_size == 0 && info->d_media_blocks) { 774 info->d_media_size = (u_int64_t)info->d_media_blocks * 775 info->d_media_blksize; 776 } else if (info->d_media_size && info->d_media_blocks == 0 && 777 info->d_media_blksize) { 778 info->d_media_blocks = info->d_media_size / 779 info->d_media_blksize; 780 } 781 782 /* 783 * The si_* fields for rawdev are not set until after the 784 * disk_create() call, so someone using the cooked version 785 * of the raw device (i.e. da0s0) will not get the right 786 * si_iosize_max unless we fix it up here. 787 */ 788 if (disk->d_cdev && disk->d_rawdev && 789 disk->d_cdev->si_iosize_max == 0) { 790 disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max; 791 disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys; 792 disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best; 793 } 794 795 /* Add the serial number to the udev_dictionary */ 796 if (info->d_serialno) 797 udev_dict_set_cstr(disk->d_cdev, "serno", info->d_serialno); 798 } 799 800 /* 801 * Disk drivers must call this routine when media parameters are available 802 * or have changed. 803 */ 804 void 805 disk_setdiskinfo(struct disk *disk, struct disk_info *info) 806 { 807 _setdiskinfo(disk, info); 808 disk_msg_send(DISK_DISK_PROBE, disk, NULL); 809 disk_debug(1, "disk_setdiskinfo: sent probe for %s\n", 810 disk->d_cdev->si_name); 811 } 812 813 void 814 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info) 815 { 816 _setdiskinfo(disk, info); 817 disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL); 818 disk_debug(1, "disk_setdiskinfo_sync: sent probe for %s\n", 819 disk->d_cdev->si_name); 820 } 821 822 /* 823 * This routine is called when an adapter detaches. The higher level 824 * managed disk device is destroyed while the lower level raw device is 825 * released. 826 */ 827 void 828 disk_destroy(struct disk *disk) 829 { 830 disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL); 831 return; 832 } 833 834 int 835 disk_dumpcheck(cdev_t dev, u_int64_t *size, 836 u_int64_t *blkno, u_int32_t *secsize) 837 { 838 struct partinfo pinfo; 839 int error; 840 841 bzero(&pinfo, sizeof(pinfo)); 842 error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0, 843 proc0.p_ucred, NULL); 844 if (error) 845 return (error); 846 847 if (pinfo.media_blksize == 0) 848 return (ENXIO); 849 850 if (blkno) /* XXX: make sure this reserved stuff is right */ 851 *blkno = pinfo.reserved_blocks + 852 pinfo.media_offset / pinfo.media_blksize; 853 if (secsize) 854 *secsize = pinfo.media_blksize; 855 if (size) 856 *size = (pinfo.media_blocks - pinfo.reserved_blocks); 857 858 return (0); 859 } 860 861 int 862 disk_dumpconf(cdev_t dev, u_int onoff) 863 { 864 struct dumperinfo di; 865 u_int64_t size, blkno; 866 u_int32_t secsize; 867 int error; 868 869 if (!onoff) 870 return set_dumper(NULL); 871 872 error = disk_dumpcheck(dev, &size, &blkno, &secsize); 873 874 if (error) 875 return ENXIO; 876 877 bzero(&di, sizeof(struct dumperinfo)); 878 di.dumper = diskdump; 879 di.priv = dev; 880 di.blocksize = secsize; 881 di.maxiosize = dev->si_iosize_max; 882 di.mediaoffset = blkno * DEV_BSIZE; 883 di.mediasize = size * DEV_BSIZE; 884 885 return set_dumper(&di); 886 } 887 888 void 889 disk_unprobe(struct disk *disk) 890 { 891 if (disk == NULL) 892 return; 893 894 disk_msg_send_sync(DISK_UNPROBE, disk, NULL); 895 } 896 897 void 898 disk_invalidate (struct disk *disk) 899 { 900 dsgone(&disk->d_slice); 901 } 902 903 /* 904 * Enumerate disks, pass a marker and an initial NULL dp to initialize, 905 * then loop with the previously returned dp. 906 * 907 * The returned dp will be referenced, preventing its destruction. When 908 * you pass the returned dp back into the loop the ref is dropped. 909 * 910 * WARNING: If terminating your loop early you must call 911 * disk_enumerate_stop(). 912 */ 913 struct disk * 914 disk_enumerate(struct disk *marker, struct disk *dp) 915 { 916 lwkt_gettoken(&disklist_token); 917 if (dp) { 918 --dp->d_refs; 919 dp = LIST_NEXT(marker, d_list); 920 LIST_REMOVE(marker, d_list); 921 } else { 922 bzero(marker, sizeof(*marker)); 923 marker->d_flags = DISKFLAG_MARKER; 924 dp = LIST_FIRST(&disklist); 925 } 926 while (dp) { 927 if ((dp->d_flags & DISKFLAG_MARKER) == 0) 928 break; 929 dp = LIST_NEXT(dp, d_list); 930 } 931 if (dp) { 932 ++dp->d_refs; 933 LIST_INSERT_AFTER(dp, marker, d_list); 934 } 935 lwkt_reltoken(&disklist_token); 936 return (dp); 937 } 938 939 /* 940 * Terminate an enumeration early. Do not call this function if the 941 * enumeration ended normally. dp can be NULL, indicating that you 942 * wish to retain the ref count on dp. 943 * 944 * This function removes the marker. 945 */ 946 void 947 disk_enumerate_stop(struct disk *marker, struct disk *dp) 948 { 949 lwkt_gettoken(&disklist_token); 950 LIST_REMOVE(marker, d_list); 951 if (dp) 952 --dp->d_refs; 953 lwkt_reltoken(&disklist_token); 954 } 955 956 static 957 int 958 sysctl_disks(SYSCTL_HANDLER_ARGS) 959 { 960 struct disk marker; 961 struct disk *dp; 962 int error, first; 963 964 first = 1; 965 error = 0; 966 dp = NULL; 967 968 while ((dp = disk_enumerate(&marker, dp))) { 969 if (!first) { 970 error = SYSCTL_OUT(req, " ", 1); 971 if (error) { 972 disk_enumerate_stop(&marker, dp); 973 break; 974 } 975 } else { 976 first = 0; 977 } 978 error = SYSCTL_OUT(req, dp->d_rawdev->si_name, 979 strlen(dp->d_rawdev->si_name)); 980 if (error) { 981 disk_enumerate_stop(&marker, dp); 982 break; 983 } 984 } 985 if (error == 0) 986 error = SYSCTL_OUT(req, "", 1); 987 return error; 988 } 989 990 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0, 991 sysctl_disks, "A", "names of available disks"); 992 993 /* 994 * Open a disk device or partition. 995 */ 996 static 997 int 998 diskopen(struct dev_open_args *ap) 999 { 1000 cdev_t dev = ap->a_head.a_dev; 1001 struct disk *dp; 1002 int error; 1003 1004 /* 1005 * dp can't be NULL here XXX. 1006 * 1007 * d_slice will be NULL if setdiskinfo() has not been called yet. 1008 * setdiskinfo() is typically called whether the disk is present 1009 * or not (e.g. CD), but the base disk device is created first 1010 * and there may be a race. 1011 */ 1012 dp = dev->si_disk; 1013 if (dp == NULL || dp->d_slice == NULL) 1014 return (ENXIO); 1015 error = 0; 1016 1017 /* 1018 * Deal with open races 1019 */ 1020 get_mplock(); 1021 while (dp->d_flags & DISKFLAG_LOCK) { 1022 dp->d_flags |= DISKFLAG_WANTED; 1023 error = tsleep(dp, PCATCH, "diskopen", hz); 1024 if (error) { 1025 rel_mplock(); 1026 return (error); 1027 } 1028 } 1029 dp->d_flags |= DISKFLAG_LOCK; 1030 1031 /* 1032 * Open the underlying raw device. 1033 */ 1034 if (!dsisopen(dp->d_slice)) { 1035 #if 0 1036 if (!pdev->si_iosize_max) 1037 pdev->si_iosize_max = dev->si_iosize_max; 1038 #endif 1039 error = dev_dopen(dp->d_rawdev, ap->a_oflags, 1040 ap->a_devtype, ap->a_cred); 1041 } 1042 1043 if (error) 1044 goto out; 1045 error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags, 1046 &dp->d_slice, &dp->d_info); 1047 if (!dsisopen(dp->d_slice)) { 1048 dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype); 1049 } 1050 out: 1051 dp->d_flags &= ~DISKFLAG_LOCK; 1052 if (dp->d_flags & DISKFLAG_WANTED) { 1053 dp->d_flags &= ~DISKFLAG_WANTED; 1054 wakeup(dp); 1055 } 1056 rel_mplock(); 1057 1058 KKASSERT(dp->d_opencount >= 0); 1059 /* If the open was successful, bump open count */ 1060 if (error == 0) 1061 atomic_add_int(&dp->d_opencount, 1); 1062 1063 return(error); 1064 } 1065 1066 /* 1067 * Close a disk device or partition 1068 */ 1069 static 1070 int 1071 diskclose(struct dev_close_args *ap) 1072 { 1073 cdev_t dev = ap->a_head.a_dev; 1074 struct disk *dp; 1075 int error; 1076 int lcount; 1077 1078 error = 0; 1079 dp = dev->si_disk; 1080 1081 /* 1082 * The cdev_t represents the disk/slice/part. The shared 1083 * dp structure governs all cdevs associated with the disk. 1084 * 1085 * As a safety only close the underlying raw device on the last 1086 * close the disk device if our tracking of the slices/partitions 1087 * also indicates nothing is open. 1088 */ 1089 KKASSERT(dp->d_opencount >= 1); 1090 lcount = atomic_fetchadd_int(&dp->d_opencount, -1); 1091 1092 get_mplock(); 1093 dsclose(dev, ap->a_devtype, dp->d_slice); 1094 if (lcount <= 1 && !dsisopen(dp->d_slice)) { 1095 error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype); 1096 } 1097 rel_mplock(); 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 get_mplock(); 1141 error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag, 1142 &dp->d_slice, &dp->d_info); 1143 rel_mplock(); 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); 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 dsched_queue(dp, 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 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 /* 1292 * The BIO wants to be ordered. Adding to the tail also 1293 * causes transition to be set to NULL, forcing the ordering 1294 * of all prior I/O's. 1295 */ 1296 if (bio->bio_buf->b_flags & B_ORDERED) { 1297 bioq_insert_tail(bioq, bio); 1298 return; 1299 } 1300 1301 switch(bio->bio_buf->b_cmd) { 1302 case BUF_CMD_READ: 1303 if (bioq->transition) { 1304 /* 1305 * Insert before the first write. Bleedover writes 1306 * based on reorder intervals to prevent starvation. 1307 */ 1308 TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act); 1309 ++bioq->reorder; 1310 if (bioq->reorder % bioq_reorder_minor_interval == 0) { 1311 bioqwritereorder(bioq); 1312 if (bioq->reorder >= 1313 bioq_reorder_burst_interval) { 1314 bioq->reorder = 0; 1315 } 1316 } 1317 } else { 1318 /* 1319 * No writes queued (or ordering was forced), 1320 * insert at tail. 1321 */ 1322 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act); 1323 } 1324 break; 1325 case BUF_CMD_WRITE: 1326 /* 1327 * Writes are always appended. If no writes were previously 1328 * queued or an ordered tail insertion occured the transition 1329 * field will be NULL. 1330 */ 1331 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act); 1332 if (bioq->transition == NULL) 1333 bioq->transition = bio; 1334 break; 1335 default: 1336 /* 1337 * All other request types are forced to be ordered. 1338 */ 1339 bioq_insert_tail(bioq, bio); 1340 break; 1341 } 1342 } 1343 1344 /* 1345 * Move the read-write transition point to prevent reads from 1346 * completely starving our writes. This brings a number of writes into 1347 * the fold every N reads. 1348 * 1349 * We bring a few linear writes into the fold on a minor interval 1350 * and we bring a non-linear burst of writes into the fold on a major 1351 * interval. Bursting only occurs if runningbufspace is really high 1352 * (typically from syncs, fsyncs, or HAMMER flushes). 1353 */ 1354 static 1355 void 1356 bioqwritereorder(struct bio_queue_head *bioq) 1357 { 1358 struct bio *bio; 1359 off_t next_offset; 1360 size_t left; 1361 size_t n; 1362 int check_off; 1363 1364 if (bioq->reorder < bioq_reorder_burst_interval || 1365 !buf_runningbufspace_severe()) { 1366 left = (size_t)bioq_reorder_minor_bytes; 1367 check_off = 1; 1368 } else { 1369 left = (size_t)bioq_reorder_burst_bytes; 1370 check_off = 0; 1371 } 1372 1373 next_offset = bioq->transition->bio_offset; 1374 while ((bio = bioq->transition) != NULL && 1375 (check_off == 0 || next_offset == bio->bio_offset) 1376 ) { 1377 n = bio->bio_buf->b_bcount; 1378 next_offset = bio->bio_offset + n; 1379 bioq->transition = TAILQ_NEXT(bio, bio_act); 1380 if (left < n) 1381 break; 1382 left -= n; 1383 } 1384 } 1385 1386 /* 1387 * Bounds checking against the media size, used for the raw partition. 1388 * secsize, mediasize and b_blkno must all be the same units. 1389 * Possibly this has to be DEV_BSIZE (512). 1390 */ 1391 int 1392 bounds_check_with_mediasize(struct bio *bio, int secsize, uint64_t mediasize) 1393 { 1394 struct buf *bp = bio->bio_buf; 1395 int64_t sz; 1396 1397 sz = howmany(bp->b_bcount, secsize); 1398 1399 if (bio->bio_offset/DEV_BSIZE + sz > mediasize) { 1400 sz = mediasize - bio->bio_offset/DEV_BSIZE; 1401 if (sz == 0) { 1402 /* If exactly at end of disk, return EOF. */ 1403 bp->b_resid = bp->b_bcount; 1404 return 0; 1405 } 1406 if (sz < 0) { 1407 /* If past end of disk, return EINVAL. */ 1408 bp->b_error = EINVAL; 1409 return 0; 1410 } 1411 /* Otherwise, truncate request. */ 1412 bp->b_bcount = sz * secsize; 1413 } 1414 1415 return 1; 1416 } 1417 1418 /* 1419 * Disk error is the preface to plaintive error messages 1420 * about failing disk transfers. It prints messages of the form 1421 1422 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d) 1423 1424 * if the offset of the error in the transfer and a disk label 1425 * are both available. blkdone should be -1 if the position of the error 1426 * is unknown; the disklabel pointer may be null from drivers that have not 1427 * been converted to use them. The message is printed with kprintf 1428 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority. 1429 * The message should be completed (with at least a newline) with kprintf 1430 * or log(-1, ...), respectively. There is no trailing space. 1431 */ 1432 void 1433 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt) 1434 { 1435 struct buf *bp = bio->bio_buf; 1436 const char *term; 1437 1438 switch(bp->b_cmd) { 1439 case BUF_CMD_READ: 1440 term = "read"; 1441 break; 1442 case BUF_CMD_WRITE: 1443 term = "write"; 1444 break; 1445 default: 1446 term = "access"; 1447 break; 1448 } 1449 kprintf("%s: %s %sing ", dev->si_name, what, term); 1450 kprintf("offset %012llx for %d", 1451 (long long)bio->bio_offset, 1452 bp->b_bcount); 1453 1454 if (donecnt) 1455 kprintf(" (%d bytes completed)", donecnt); 1456 } 1457 1458 /* 1459 * Locate a disk device 1460 */ 1461 cdev_t 1462 disk_locate(const char *devname) 1463 { 1464 return devfs_find_device_by_name("%s", devname); 1465 } 1466 1467 void 1468 disk_config(void *arg) 1469 { 1470 disk_msg_send_sync(DISK_SYNC, NULL, NULL); 1471 } 1472 1473 static void 1474 disk_init(void) 1475 { 1476 struct thread* td_core; 1477 1478 disk_msg_cache = objcache_create("disk-msg-cache", 0, 0, 1479 NULL, NULL, NULL, 1480 objcache_malloc_alloc, 1481 objcache_malloc_free, 1482 &disk_msg_malloc_args); 1483 1484 lwkt_token_init(&disklist_token, "disks"); 1485 1486 /* 1487 * Initialize the reply-only port which acts as a message drain 1488 */ 1489 lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply); 1490 1491 lwkt_gettoken(&disklist_token); 1492 lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL, 1493 0, -1, "disk_msg_core"); 1494 tsleep(td_core, 0, "diskcore", 0); 1495 lwkt_reltoken(&disklist_token); 1496 } 1497 1498 static void 1499 disk_uninit(void) 1500 { 1501 objcache_destroy(disk_msg_cache); 1502 } 1503 1504 /* 1505 * Clean out illegal characters in serial numbers. 1506 */ 1507 static void 1508 disk_cleanserial(char *serno) 1509 { 1510 char c; 1511 1512 while ((c = *serno) != 0) { 1513 if (c >= 'a' && c <= 'z') 1514 ; 1515 else if (c >= 'A' && c <= 'Z') 1516 ; 1517 else if (c >= '0' && c <= '9') 1518 ; 1519 else if (c == '-' || c == '@' || c == '+' || c == '.') 1520 ; 1521 else 1522 c = '_'; 1523 *serno++= c; 1524 } 1525 } 1526 1527 TUNABLE_INT("kern.disk_debug", &disk_debug_enable); 1528 SYSCTL_INT(_kern, OID_AUTO, disk_debug, CTLFLAG_RW, &disk_debug_enable, 1529 0, "Enable subr_disk debugging"); 1530 1531 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL); 1532 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL); 1533