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/mplock2.h> 109 #include <sys/msgport2.h> 110 #include <sys/thread2.h> 111 112 static MALLOC_DEFINE(M_DISK, "disk", "disk data"); 113 static int disk_debug_enable = 0; 114 115 static void disk_msg_autofree_reply(lwkt_port_t, lwkt_msg_t); 116 static void disk_msg_core(void *); 117 static int disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe); 118 static void disk_probe(struct disk *dp, int reprobe); 119 static void _setdiskinfo(struct disk *disk, struct disk_info *info); 120 static void bioqwritereorder(struct bio_queue_head *bioq); 121 static void disk_cleanserial(char *serno); 122 static int disk_debug(int, char *, ...) __printflike(2, 3); 123 static cdev_t _disk_create_named(const char *name, int unit, struct disk *dp, 124 struct dev_ops *raw_ops, int clone); 125 126 static d_open_t diskopen; 127 static d_close_t diskclose; 128 static d_ioctl_t diskioctl; 129 static d_strategy_t diskstrategy; 130 static d_psize_t diskpsize; 131 static d_dump_t diskdump; 132 133 static LIST_HEAD(, disk) disklist = LIST_HEAD_INITIALIZER(&disklist); 134 static struct lwkt_token disklist_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 get_mplock(); /* not mpsafe yet? */ 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_exit(); 548 } 549 550 551 /* 552 * Acts as a message drain. Any message that is replied to here gets 553 * destroyed and the memory freed. 554 */ 555 static void 556 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg) 557 { 558 objcache_put(disk_msg_cache, msg); 559 } 560 561 562 void 563 disk_msg_send(uint32_t cmd, void *load, void *load2) 564 { 565 disk_msg_t disk_msg; 566 lwkt_port_t port = &disk_msg_port; 567 568 disk_msg = objcache_get(disk_msg_cache, M_WAITOK); 569 570 lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0); 571 572 disk_msg->hdr.u.ms_result = cmd; 573 disk_msg->load = load; 574 disk_msg->load2 = load2; 575 KKASSERT(port); 576 lwkt_sendmsg(port, &disk_msg->hdr); 577 } 578 579 void 580 disk_msg_send_sync(uint32_t cmd, void *load, void *load2) 581 { 582 struct lwkt_port rep_port; 583 disk_msg_t disk_msg; 584 lwkt_port_t port; 585 586 disk_msg = objcache_get(disk_msg_cache, M_WAITOK); 587 port = &disk_msg_port; 588 589 /* XXX could probably use curthread's built-in msgport */ 590 lwkt_initport_thread(&rep_port, curthread); 591 lwkt_initmsg(&disk_msg->hdr, &rep_port, 0); 592 593 disk_msg->hdr.u.ms_result = cmd; 594 disk_msg->load = load; 595 disk_msg->load2 = load2; 596 597 lwkt_domsg(port, &disk_msg->hdr, 0); 598 objcache_put(disk_msg_cache, disk_msg); 599 } 600 601 /* 602 * Create a raw device for the dev_ops template (which is returned). Also 603 * create a slice and unit managed disk and overload the user visible 604 * device space with it. 605 * 606 * NOTE: The returned raw device is NOT a slice and unit managed device. 607 * It is an actual raw device representing the raw disk as specified by 608 * the passed dev_ops. The disk layer not only returns such a raw device, 609 * it also uses it internally when passing (modified) commands through. 610 */ 611 cdev_t 612 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops) 613 { 614 return _disk_create_named(NULL, unit, dp, raw_ops, 0); 615 } 616 617 cdev_t 618 disk_create_clone(int unit, struct disk *dp, 619 struct dev_ops *raw_ops) 620 { 621 return _disk_create_named(NULL, unit, dp, raw_ops, 1); 622 } 623 624 cdev_t 625 disk_create_named(const char *name, int unit, struct disk *dp, 626 struct dev_ops *raw_ops) 627 { 628 return _disk_create_named(name, unit, dp, raw_ops, 0); 629 } 630 631 cdev_t 632 disk_create_named_clone(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, 1); 636 } 637 638 static cdev_t 639 _disk_create_named(const char *name, int unit, struct disk *dp, 640 struct dev_ops *raw_ops, int clone) 641 { 642 cdev_t rawdev; 643 644 disk_debug(1, "disk_create (begin): %s%d\n", name, unit); 645 646 if (name) { 647 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit), 648 UID_ROOT, GID_OPERATOR, 0640, "%s", name); 649 } else { 650 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit), 651 UID_ROOT, GID_OPERATOR, 0640, 652 "%s%d", raw_ops->head.name, unit); 653 } 654 655 bzero(dp, sizeof(*dp)); 656 657 dp->d_rawdev = rawdev; 658 dp->d_raw_ops = raw_ops; 659 dp->d_dev_ops = &disk_ops; 660 661 if (name) { 662 if (clone) { 663 dp->d_cdev = make_only_dev_covering( 664 &disk_ops, dp->d_rawdev->si_ops, 665 dkmakewholedisk(unit), 666 UID_ROOT, GID_OPERATOR, 0640, 667 "%s", name); 668 } else { 669 dp->d_cdev = make_dev_covering( 670 &disk_ops, dp->d_rawdev->si_ops, 671 dkmakewholedisk(unit), 672 UID_ROOT, GID_OPERATOR, 0640, 673 "%s", name); 674 } 675 } else { 676 if (clone) { 677 dp->d_cdev = make_only_dev_covering( 678 &disk_ops, dp->d_rawdev->si_ops, 679 dkmakewholedisk(unit), 680 UID_ROOT, GID_OPERATOR, 0640, 681 "%s%d", raw_ops->head.name, unit); 682 } else { 683 dp->d_cdev = make_dev_covering( 684 &disk_ops, dp->d_rawdev->si_ops, 685 dkmakewholedisk(unit), 686 UID_ROOT, GID_OPERATOR, 0640, 687 "%s%d", raw_ops->head.name, unit); 688 } 689 } 690 691 udev_dict_set_cstr(dp->d_cdev, "subsystem", "disk"); 692 dp->d_cdev->si_disk = dp; 693 694 if (name) 695 dsched_disk_create(dp, name, unit); 696 else 697 dsched_disk_create(dp, raw_ops->head.name, unit); 698 699 lwkt_gettoken(&disklist_token); 700 LIST_INSERT_HEAD(&disklist, dp, d_list); 701 lwkt_reltoken(&disklist_token); 702 703 disk_iocom_init(dp); 704 705 disk_debug(1, "disk_create (end): %s%d\n", 706 (name != NULL)?(name):(raw_ops->head.name), unit); 707 708 return (dp->d_rawdev); 709 } 710 711 int 712 disk_setdisktype(struct disk *disk, const char *type) 713 { 714 int error; 715 716 KKASSERT(disk != NULL); 717 718 disk->d_disktype = type; 719 error = udev_dict_set_cstr(disk->d_cdev, "disk-type", 720 __DECONST(char *, type)); 721 return error; 722 } 723 724 int 725 disk_getopencount(struct disk *disk) 726 { 727 return disk->d_opencount; 728 } 729 730 static void 731 _setdiskinfo(struct disk *disk, struct disk_info *info) 732 { 733 char *oldserialno; 734 735 oldserialno = disk->d_info.d_serialno; 736 bcopy(info, &disk->d_info, sizeof(disk->d_info)); 737 info = &disk->d_info; 738 739 disk_debug(1, "_setdiskinfo: %s\n", disk->d_cdev->si_name); 740 741 /* 742 * The serial number is duplicated so the caller can throw 743 * their copy away. 744 */ 745 if (info->d_serialno && info->d_serialno[0] && 746 (info->d_serialno[0] != ' ' || strlen(info->d_serialno) > 1)) { 747 info->d_serialno = kstrdup(info->d_serialno, M_TEMP); 748 disk_cleanserial(info->d_serialno); 749 if (disk->d_cdev) { 750 make_dev_alias(disk->d_cdev, "serno/%s", 751 info->d_serialno); 752 } 753 } else { 754 info->d_serialno = NULL; 755 } 756 if (oldserialno) 757 kfree(oldserialno, M_TEMP); 758 759 dsched_disk_update(disk, info); 760 761 /* 762 * The caller may set d_media_size or d_media_blocks and we 763 * calculate the other. 764 */ 765 KKASSERT(info->d_media_size == 0 || info->d_media_blocks == 0); 766 if (info->d_media_size == 0 && info->d_media_blocks) { 767 info->d_media_size = (u_int64_t)info->d_media_blocks * 768 info->d_media_blksize; 769 } else if (info->d_media_size && info->d_media_blocks == 0 && 770 info->d_media_blksize) { 771 info->d_media_blocks = info->d_media_size / 772 info->d_media_blksize; 773 } 774 775 /* 776 * The si_* fields for rawdev are not set until after the 777 * disk_create() call, so someone using the cooked version 778 * of the raw device (i.e. da0s0) will not get the right 779 * si_iosize_max unless we fix it up here. 780 */ 781 if (disk->d_cdev && disk->d_rawdev && 782 disk->d_cdev->si_iosize_max == 0) { 783 disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max; 784 disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys; 785 disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best; 786 } 787 788 /* Add the serial number to the udev_dictionary */ 789 if (info->d_serialno) 790 udev_dict_set_cstr(disk->d_cdev, "serno", info->d_serialno); 791 } 792 793 /* 794 * Disk drivers must call this routine when media parameters are available 795 * or have changed. 796 */ 797 void 798 disk_setdiskinfo(struct disk *disk, struct disk_info *info) 799 { 800 _setdiskinfo(disk, info); 801 disk_msg_send(DISK_DISK_PROBE, disk, NULL); 802 disk_debug(1, "disk_setdiskinfo: sent probe for %s\n", 803 disk->d_cdev->si_name); 804 } 805 806 void 807 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info) 808 { 809 _setdiskinfo(disk, info); 810 disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL); 811 disk_debug(1, "disk_setdiskinfo_sync: sent probe for %s\n", 812 disk->d_cdev->si_name); 813 } 814 815 /* 816 * This routine is called when an adapter detaches. The higher level 817 * managed disk device is destroyed while the lower level raw device is 818 * released. 819 */ 820 void 821 disk_destroy(struct disk *disk) 822 { 823 disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL); 824 return; 825 } 826 827 int 828 disk_dumpcheck(cdev_t dev, u_int64_t *size, 829 u_int64_t *blkno, u_int32_t *secsize) 830 { 831 struct partinfo pinfo; 832 int error; 833 834 bzero(&pinfo, sizeof(pinfo)); 835 error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0, 836 proc0.p_ucred, NULL, NULL); 837 if (error) 838 return (error); 839 840 if (pinfo.media_blksize == 0) 841 return (ENXIO); 842 843 if (blkno) /* XXX: make sure this reserved stuff is right */ 844 *blkno = pinfo.reserved_blocks + 845 pinfo.media_offset / pinfo.media_blksize; 846 if (secsize) 847 *secsize = pinfo.media_blksize; 848 if (size) 849 *size = (pinfo.media_blocks - pinfo.reserved_blocks); 850 851 return (0); 852 } 853 854 int 855 disk_dumpconf(cdev_t dev, u_int onoff) 856 { 857 struct dumperinfo di; 858 u_int64_t size, blkno; 859 u_int32_t secsize; 860 int error; 861 862 if (!onoff) 863 return set_dumper(NULL); 864 865 error = disk_dumpcheck(dev, &size, &blkno, &secsize); 866 867 if (error) 868 return ENXIO; 869 870 bzero(&di, sizeof(struct dumperinfo)); 871 di.dumper = diskdump; 872 di.priv = dev; 873 di.blocksize = secsize; 874 di.maxiosize = dev->si_iosize_max; 875 di.mediaoffset = blkno * DEV_BSIZE; 876 di.mediasize = size * DEV_BSIZE; 877 878 return set_dumper(&di); 879 } 880 881 void 882 disk_unprobe(struct disk *disk) 883 { 884 if (disk == NULL) 885 return; 886 887 disk_msg_send_sync(DISK_UNPROBE, disk, NULL); 888 } 889 890 void 891 disk_invalidate (struct disk *disk) 892 { 893 dsgone(&disk->d_slice); 894 } 895 896 /* 897 * Enumerate disks, pass a marker and an initial NULL dp to initialize, 898 * then loop with the previously returned dp. 899 * 900 * The returned dp will be referenced, preventing its destruction. When 901 * you pass the returned dp back into the loop the ref is dropped. 902 * 903 * WARNING: If terminating your loop early you must call 904 * disk_enumerate_stop(). 905 */ 906 struct disk * 907 disk_enumerate(struct disk *marker, struct disk *dp) 908 { 909 lwkt_gettoken(&disklist_token); 910 if (dp) { 911 --dp->d_refs; 912 dp = LIST_NEXT(marker, d_list); 913 LIST_REMOVE(marker, d_list); 914 } else { 915 bzero(marker, sizeof(*marker)); 916 marker->d_flags = DISKFLAG_MARKER; 917 dp = LIST_FIRST(&disklist); 918 } 919 while (dp) { 920 if ((dp->d_flags & DISKFLAG_MARKER) == 0) 921 break; 922 dp = LIST_NEXT(dp, d_list); 923 } 924 if (dp) { 925 ++dp->d_refs; 926 LIST_INSERT_AFTER(dp, marker, d_list); 927 } 928 lwkt_reltoken(&disklist_token); 929 return (dp); 930 } 931 932 /* 933 * Terminate an enumeration early. Do not call this function if the 934 * enumeration ended normally. dp can be NULL, indicating that you 935 * wish to retain the ref count on dp. 936 * 937 * This function removes the marker. 938 */ 939 void 940 disk_enumerate_stop(struct disk *marker, struct disk *dp) 941 { 942 lwkt_gettoken(&disklist_token); 943 LIST_REMOVE(marker, d_list); 944 if (dp) 945 --dp->d_refs; 946 lwkt_reltoken(&disklist_token); 947 } 948 949 static 950 int 951 sysctl_disks(SYSCTL_HANDLER_ARGS) 952 { 953 struct disk marker; 954 struct disk *dp; 955 int error, first; 956 957 first = 1; 958 error = 0; 959 dp = NULL; 960 961 while ((dp = disk_enumerate(&marker, dp))) { 962 if (!first) { 963 error = SYSCTL_OUT(req, " ", 1); 964 if (error) { 965 disk_enumerate_stop(&marker, dp); 966 break; 967 } 968 } else { 969 first = 0; 970 } 971 error = SYSCTL_OUT(req, dp->d_rawdev->si_name, 972 strlen(dp->d_rawdev->si_name)); 973 if (error) { 974 disk_enumerate_stop(&marker, dp); 975 break; 976 } 977 } 978 if (error == 0) 979 error = SYSCTL_OUT(req, "", 1); 980 return error; 981 } 982 983 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0, 984 sysctl_disks, "A", "names of available disks"); 985 986 /* 987 * Open a disk device or partition. 988 */ 989 static 990 int 991 diskopen(struct dev_open_args *ap) 992 { 993 cdev_t dev = ap->a_head.a_dev; 994 struct disk *dp; 995 int error; 996 997 /* 998 * dp can't be NULL here XXX. 999 * 1000 * d_slice will be NULL if setdiskinfo() has not been called yet. 1001 * setdiskinfo() is typically called whether the disk is present 1002 * or not (e.g. CD), but the base disk device is created first 1003 * and there may be a race. 1004 */ 1005 dp = dev->si_disk; 1006 if (dp == NULL || dp->d_slice == NULL) 1007 return (ENXIO); 1008 error = 0; 1009 1010 /* 1011 * Deal with open races 1012 */ 1013 get_mplock(); 1014 while (dp->d_flags & DISKFLAG_LOCK) { 1015 dp->d_flags |= DISKFLAG_WANTED; 1016 error = tsleep(dp, PCATCH, "diskopen", hz); 1017 if (error) { 1018 rel_mplock(); 1019 return (error); 1020 } 1021 } 1022 dp->d_flags |= DISKFLAG_LOCK; 1023 1024 /* 1025 * Open the underlying raw device. 1026 */ 1027 if (!dsisopen(dp->d_slice)) { 1028 #if 0 1029 if (!pdev->si_iosize_max) 1030 pdev->si_iosize_max = dev->si_iosize_max; 1031 #endif 1032 error = dev_dopen(dp->d_rawdev, ap->a_oflags, 1033 ap->a_devtype, ap->a_cred, NULL); 1034 } 1035 1036 if (error) 1037 goto out; 1038 error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags, 1039 &dp->d_slice, &dp->d_info); 1040 if (!dsisopen(dp->d_slice)) { 1041 dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype, NULL); 1042 } 1043 out: 1044 dp->d_flags &= ~DISKFLAG_LOCK; 1045 if (dp->d_flags & DISKFLAG_WANTED) { 1046 dp->d_flags &= ~DISKFLAG_WANTED; 1047 wakeup(dp); 1048 } 1049 rel_mplock(); 1050 1051 KKASSERT(dp->d_opencount >= 0); 1052 /* If the open was successful, bump open count */ 1053 if (error == 0) 1054 atomic_add_int(&dp->d_opencount, 1); 1055 1056 return(error); 1057 } 1058 1059 /* 1060 * Close a disk device or partition 1061 */ 1062 static 1063 int 1064 diskclose(struct dev_close_args *ap) 1065 { 1066 cdev_t dev = ap->a_head.a_dev; 1067 struct disk *dp; 1068 int error; 1069 int lcount; 1070 1071 error = 0; 1072 dp = dev->si_disk; 1073 1074 /* 1075 * The cdev_t represents the disk/slice/part. The shared 1076 * dp structure governs all cdevs associated with the disk. 1077 * 1078 * As a safety only close the underlying raw device on the last 1079 * close the disk device if our tracking of the slices/partitions 1080 * also indicates nothing is open. 1081 */ 1082 KKASSERT(dp->d_opencount >= 1); 1083 lcount = atomic_fetchadd_int(&dp->d_opencount, -1); 1084 1085 get_mplock(); 1086 dsclose(dev, ap->a_devtype, dp->d_slice); 1087 if (lcount <= 1 && !dsisopen(dp->d_slice)) { 1088 error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype, NULL); 1089 } 1090 rel_mplock(); 1091 return (error); 1092 } 1093 1094 /* 1095 * First execute the ioctl on the disk device, and if it isn't supported 1096 * try running it on the backing device. 1097 */ 1098 static 1099 int 1100 diskioctl(struct dev_ioctl_args *ap) 1101 { 1102 cdev_t dev = ap->a_head.a_dev; 1103 struct disk *dp; 1104 int error; 1105 u_int u; 1106 1107 dp = dev->si_disk; 1108 if (dp == NULL) 1109 return (ENXIO); 1110 1111 devfs_debug(DEVFS_DEBUG_DEBUG, 1112 "diskioctl: cmd is: %lx (name: %s)\n", 1113 ap->a_cmd, dev->si_name); 1114 devfs_debug(DEVFS_DEBUG_DEBUG, 1115 "diskioctl: &dp->d_slice is: %p, %p\n", 1116 &dp->d_slice, dp->d_slice); 1117 1118 if (ap->a_cmd == DIOCGKERNELDUMP) { 1119 u = *(u_int *)ap->a_data; 1120 return disk_dumpconf(dev, u); 1121 } 1122 1123 if (ap->a_cmd == DIOCRECLUSTER && dev == dp->d_cdev) { 1124 error = disk_iocom_ioctl(dp, ap->a_cmd, ap->a_data); 1125 return error; 1126 } 1127 1128 if (&dp->d_slice == NULL || dp->d_slice == NULL || 1129 ((dp->d_info.d_dsflags & DSO_DEVICEMAPPER) && 1130 dkslice(dev) == WHOLE_DISK_SLICE)) { 1131 error = ENOIOCTL; 1132 } else { 1133 get_mplock(); 1134 error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag, 1135 &dp->d_slice, &dp->d_info); 1136 rel_mplock(); 1137 } 1138 1139 if (error == ENOIOCTL) { 1140 error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data, 1141 ap->a_fflag, ap->a_cred, NULL, NULL); 1142 } 1143 return (error); 1144 } 1145 1146 /* 1147 * Execute strategy routine 1148 */ 1149 static 1150 int 1151 diskstrategy(struct dev_strategy_args *ap) 1152 { 1153 cdev_t dev = ap->a_head.a_dev; 1154 struct bio *bio = ap->a_bio; 1155 struct bio *nbio; 1156 struct disk *dp; 1157 1158 dp = dev->si_disk; 1159 1160 if (dp == NULL) { 1161 bio->bio_buf->b_error = ENXIO; 1162 bio->bio_buf->b_flags |= B_ERROR; 1163 biodone(bio); 1164 return(0); 1165 } 1166 KKASSERT(dev->si_disk == dp); 1167 1168 /* 1169 * The dscheck() function will also transform the slice relative 1170 * block number i.e. bio->bio_offset into a block number that can be 1171 * passed directly to the underlying raw device. If dscheck() 1172 * returns NULL it will have handled the bio for us (e.g. EOF 1173 * or error due to being beyond the device size). 1174 */ 1175 if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) { 1176 dev_dstrategy(dp->d_rawdev, nbio); 1177 } else { 1178 biodone(bio); 1179 } 1180 return(0); 1181 } 1182 1183 /* 1184 * Return the partition size in ?blocks? 1185 */ 1186 static 1187 int 1188 diskpsize(struct dev_psize_args *ap) 1189 { 1190 cdev_t dev = ap->a_head.a_dev; 1191 struct disk *dp; 1192 1193 dp = dev->si_disk; 1194 if (dp == NULL) 1195 return(ENODEV); 1196 1197 ap->a_result = dssize(dev, &dp->d_slice); 1198 1199 if ((ap->a_result == -1) && 1200 (dp->d_info.d_dsflags & DSO_RAWPSIZE)) { 1201 ap->a_head.a_dev = dp->d_rawdev; 1202 return dev_doperate(&ap->a_head); 1203 } 1204 return(0); 1205 } 1206 1207 static int 1208 diskdump(struct dev_dump_args *ap) 1209 { 1210 cdev_t dev = ap->a_head.a_dev; 1211 struct disk *dp = dev->si_disk; 1212 u_int64_t size, offset; 1213 int error; 1214 1215 error = disk_dumpcheck(dev, &size, &ap->a_blkno, &ap->a_secsize); 1216 /* XXX: this should probably go in disk_dumpcheck somehow */ 1217 if (ap->a_length != 0) { 1218 size *= DEV_BSIZE; 1219 offset = ap->a_blkno * DEV_BSIZE; 1220 if ((ap->a_offset < offset) || 1221 (ap->a_offset + ap->a_length - offset > size)) { 1222 kprintf("Attempt to write outside dump " 1223 "device boundaries.\n"); 1224 error = ENOSPC; 1225 } 1226 } 1227 1228 if (error == 0) { 1229 ap->a_head.a_dev = dp->d_rawdev; 1230 error = dev_doperate(&ap->a_head); 1231 } 1232 1233 return(error); 1234 } 1235 1236 1237 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD, 1238 0, sizeof(struct diskslices), "sizeof(struct diskslices)"); 1239 1240 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD, 1241 0, sizeof(struct disk), "sizeof(struct disk)"); 1242 1243 /* 1244 * Reorder interval for burst write allowance and minor write 1245 * allowance. 1246 * 1247 * We always want to trickle some writes in to make use of the 1248 * disk's zone cache. Bursting occurs on a longer interval and only 1249 * runningbufspace is well over the hirunningspace limit. 1250 */ 1251 int bioq_reorder_burst_interval = 60; /* should be multiple of minor */ 1252 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_interval, 1253 CTLFLAG_RW, &bioq_reorder_burst_interval, 0, ""); 1254 int bioq_reorder_minor_interval = 5; 1255 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_interval, 1256 CTLFLAG_RW, &bioq_reorder_minor_interval, 0, ""); 1257 1258 int bioq_reorder_burst_bytes = 3000000; 1259 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_bytes, 1260 CTLFLAG_RW, &bioq_reorder_burst_bytes, 0, ""); 1261 int bioq_reorder_minor_bytes = 262144; 1262 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_bytes, 1263 CTLFLAG_RW, &bioq_reorder_minor_bytes, 0, ""); 1264 1265 1266 /* 1267 * Order I/Os. Generally speaking this code is designed to make better 1268 * use of drive zone caches. A drive zone cache can typically track linear 1269 * reads or writes for around 16 zones simultaniously. 1270 * 1271 * Read prioritization issues: It is possible for hundreds of megabytes worth 1272 * of writes to be queued asynchronously. This creates a huge bottleneck 1273 * for reads which reduce read bandwidth to a trickle. 1274 * 1275 * To solve this problem we generally reorder reads before writes. 1276 * 1277 * However, a large number of random reads can also starve writes and 1278 * make poor use of the drive zone cache so we allow writes to trickle 1279 * in every N reads. 1280 */ 1281 void 1282 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio) 1283 { 1284 /* 1285 * The BIO wants to be ordered. Adding to the tail also 1286 * causes transition to be set to NULL, forcing the ordering 1287 * of all prior I/O's. 1288 */ 1289 if (bio->bio_buf->b_flags & B_ORDERED) { 1290 bioq_insert_tail(bioq, bio); 1291 return; 1292 } 1293 1294 switch(bio->bio_buf->b_cmd) { 1295 case BUF_CMD_READ: 1296 if (bioq->transition) { 1297 /* 1298 * Insert before the first write. Bleedover writes 1299 * based on reorder intervals to prevent starvation. 1300 */ 1301 TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act); 1302 ++bioq->reorder; 1303 if (bioq->reorder % bioq_reorder_minor_interval == 0) { 1304 bioqwritereorder(bioq); 1305 if (bioq->reorder >= 1306 bioq_reorder_burst_interval) { 1307 bioq->reorder = 0; 1308 } 1309 } 1310 } else { 1311 /* 1312 * No writes queued (or ordering was forced), 1313 * insert at tail. 1314 */ 1315 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act); 1316 } 1317 break; 1318 case BUF_CMD_WRITE: 1319 /* 1320 * Writes are always appended. If no writes were previously 1321 * queued or an ordered tail insertion occured the transition 1322 * field will be NULL. 1323 */ 1324 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act); 1325 if (bioq->transition == NULL) 1326 bioq->transition = bio; 1327 break; 1328 default: 1329 /* 1330 * All other request types are forced to be ordered. 1331 */ 1332 bioq_insert_tail(bioq, bio); 1333 break; 1334 } 1335 } 1336 1337 /* 1338 * Move the read-write transition point to prevent reads from 1339 * completely starving our writes. This brings a number of writes into 1340 * the fold every N reads. 1341 * 1342 * We bring a few linear writes into the fold on a minor interval 1343 * and we bring a non-linear burst of writes into the fold on a major 1344 * interval. Bursting only occurs if runningbufspace is really high 1345 * (typically from syncs, fsyncs, or HAMMER flushes). 1346 */ 1347 static 1348 void 1349 bioqwritereorder(struct bio_queue_head *bioq) 1350 { 1351 struct bio *bio; 1352 off_t next_offset; 1353 size_t left; 1354 size_t n; 1355 int check_off; 1356 1357 if (bioq->reorder < bioq_reorder_burst_interval || 1358 !buf_runningbufspace_severe()) { 1359 left = (size_t)bioq_reorder_minor_bytes; 1360 check_off = 1; 1361 } else { 1362 left = (size_t)bioq_reorder_burst_bytes; 1363 check_off = 0; 1364 } 1365 1366 next_offset = bioq->transition->bio_offset; 1367 while ((bio = bioq->transition) != NULL && 1368 (check_off == 0 || next_offset == bio->bio_offset) 1369 ) { 1370 n = bio->bio_buf->b_bcount; 1371 next_offset = bio->bio_offset + n; 1372 bioq->transition = TAILQ_NEXT(bio, bio_act); 1373 if (left < n) 1374 break; 1375 left -= n; 1376 } 1377 } 1378 1379 /* 1380 * Bounds checking against the media size, used for the raw partition. 1381 * secsize, mediasize and b_blkno must all be the same units. 1382 * Possibly this has to be DEV_BSIZE (512). 1383 */ 1384 int 1385 bounds_check_with_mediasize(struct bio *bio, int secsize, uint64_t mediasize) 1386 { 1387 struct buf *bp = bio->bio_buf; 1388 int64_t sz; 1389 1390 sz = howmany(bp->b_bcount, secsize); 1391 1392 if (bio->bio_offset/DEV_BSIZE + sz > mediasize) { 1393 sz = mediasize - bio->bio_offset/DEV_BSIZE; 1394 if (sz == 0) { 1395 /* If exactly at end of disk, return EOF. */ 1396 bp->b_resid = bp->b_bcount; 1397 return 0; 1398 } 1399 if (sz < 0) { 1400 /* If past end of disk, return EINVAL. */ 1401 bp->b_error = EINVAL; 1402 return 0; 1403 } 1404 /* Otherwise, truncate request. */ 1405 bp->b_bcount = sz * secsize; 1406 } 1407 1408 return 1; 1409 } 1410 1411 /* 1412 * Disk error is the preface to plaintive error messages 1413 * about failing disk transfers. It prints messages of the form 1414 1415 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d) 1416 1417 * if the offset of the error in the transfer and a disk label 1418 * are both available. blkdone should be -1 if the position of the error 1419 * is unknown; the disklabel pointer may be null from drivers that have not 1420 * been converted to use them. The message is printed with kprintf 1421 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority. 1422 * The message should be completed (with at least a newline) with kprintf 1423 * or log(-1, ...), respectively. There is no trailing space. 1424 */ 1425 void 1426 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt) 1427 { 1428 struct buf *bp = bio->bio_buf; 1429 const char *term; 1430 1431 switch(bp->b_cmd) { 1432 case BUF_CMD_READ: 1433 term = "read"; 1434 break; 1435 case BUF_CMD_WRITE: 1436 term = "write"; 1437 break; 1438 default: 1439 term = "access"; 1440 break; 1441 } 1442 kprintf("%s: %s %sing ", dev->si_name, what, term); 1443 kprintf("offset %012llx for %d", 1444 (long long)bio->bio_offset, 1445 bp->b_bcount); 1446 1447 if (donecnt) 1448 kprintf(" (%d bytes completed)", donecnt); 1449 } 1450 1451 /* 1452 * Locate a disk device 1453 */ 1454 cdev_t 1455 disk_locate(const char *devname) 1456 { 1457 return devfs_find_device_by_name("%s", devname); 1458 } 1459 1460 void 1461 disk_config(void *arg) 1462 { 1463 disk_msg_send_sync(DISK_SYNC, NULL, NULL); 1464 } 1465 1466 static void 1467 disk_init(void) 1468 { 1469 struct thread* td_core; 1470 1471 disk_msg_cache = objcache_create("disk-msg-cache", 0, 0, 1472 NULL, NULL, NULL, 1473 objcache_malloc_alloc, 1474 objcache_malloc_free, 1475 &disk_msg_malloc_args); 1476 1477 lwkt_token_init(&disklist_token, "disks"); 1478 1479 /* 1480 * Initialize the reply-only port which acts as a message drain 1481 */ 1482 lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply); 1483 1484 lwkt_gettoken(&disklist_token); 1485 lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL, 1486 0, -1, "disk_msg_core"); 1487 tsleep(td_core, 0, "diskcore", 0); 1488 lwkt_reltoken(&disklist_token); 1489 } 1490 1491 static void 1492 disk_uninit(void) 1493 { 1494 objcache_destroy(disk_msg_cache); 1495 } 1496 1497 /* 1498 * Clean out illegal characters in serial numbers. 1499 */ 1500 static void 1501 disk_cleanserial(char *serno) 1502 { 1503 char c; 1504 1505 while ((c = *serno) != 0) { 1506 if (c >= 'a' && c <= 'z') 1507 ; 1508 else if (c >= 'A' && c <= 'Z') 1509 ; 1510 else if (c >= '0' && c <= '9') 1511 ; 1512 else if (c == '-' || c == '@' || c == '+' || c == '.') 1513 ; 1514 else 1515 c = '_'; 1516 *serno++= c; 1517 } 1518 } 1519 1520 TUNABLE_INT("kern.disk_debug", &disk_debug_enable); 1521 SYSCTL_INT(_kern, OID_AUTO, disk_debug, CTLFLAG_RW, &disk_debug_enable, 1522 0, "Enable subr_disk debugging"); 1523 1524 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL); 1525 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL); 1526