1 /* 2 * Copyright (c) 2003,2004,2009 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@backplane.com> 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * ---------------------------------------------------------------------------- 35 * "THE BEER-WARE LICENSE" (Revision 42): 36 * <phk@FreeBSD.ORG> wrote this file. As long as you retain this notice you 37 * can do whatever you want with this stuff. If we meet some day, and you think 38 * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp 39 * ---------------------------------------------------------------------------- 40 * 41 * Copyright (c) 1982, 1986, 1988, 1993 42 * The Regents of the University of California. All rights reserved. 43 * (c) UNIX System Laboratories, Inc. 44 * All or some portions of this file are derived from material licensed 45 * to the University of California by American Telephone and Telegraph 46 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 47 * the permission of UNIX System Laboratories, Inc. 48 * 49 * Redistribution and use in source and binary forms, with or without 50 * modification, are permitted provided that the following conditions 51 * are met: 52 * 1. Redistributions of source code must retain the above copyright 53 * notice, this list of conditions and the following disclaimer. 54 * 2. Redistributions in binary form must reproduce the above copyright 55 * notice, this list of conditions and the following disclaimer in the 56 * documentation and/or other materials provided with the distribution. 57 * 3. All advertising materials mentioning features or use of this software 58 * must display the following acknowledgement: 59 * This product includes software developed by the University of 60 * California, Berkeley and its contributors. 61 * 4. Neither the name of the University nor the names of its contributors 62 * may be used to endorse or promote products derived from this software 63 * without specific prior written permission. 64 * 65 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 66 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 67 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 68 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 69 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 70 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 71 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 72 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 73 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 74 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 75 * SUCH DAMAGE. 76 * 77 * @(#)ufs_disksubr.c 8.5 (Berkeley) 1/21/94 78 * $FreeBSD: src/sys/kern/subr_disk.c,v 1.20.2.6 2001/10/05 07:14:57 peter Exp $ 79 * $FreeBSD: src/sys/ufs/ufs/ufs_disksubr.c,v 1.44.2.3 2001/03/05 05:42:19 obrien Exp $ 80 * $DragonFly: src/sys/kern/subr_disk.c,v 1.40 2008/06/05 18:06:32 swildner Exp $ 81 */ 82 83 #include <sys/param.h> 84 #include <sys/systm.h> 85 #include <sys/kernel.h> 86 #include <sys/proc.h> 87 #include <sys/sysctl.h> 88 #include <sys/buf.h> 89 #include <sys/conf.h> 90 #include <sys/disklabel.h> 91 #include <sys/disklabel32.h> 92 #include <sys/disklabel64.h> 93 #include <sys/diskslice.h> 94 #include <sys/diskmbr.h> 95 #include <sys/disk.h> 96 #include <sys/kerneldump.h> 97 #include <sys/malloc.h> 98 #include <sys/sysctl.h> 99 #include <machine/md_var.h> 100 #include <sys/ctype.h> 101 #include <sys/syslog.h> 102 #include <sys/device.h> 103 #include <sys/msgport.h> 104 #include <sys/devfs.h> 105 #include <sys/thread.h> 106 #include <sys/dsched.h> 107 #include <sys/queue.h> 108 #include <sys/lock.h> 109 #include <sys/udev.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); 126 127 static d_open_t diskopen; 128 static d_close_t diskclose; 129 static d_ioctl_t diskioctl; 130 static d_strategy_t diskstrategy; 131 static d_psize_t diskpsize; 132 static d_clone_t diskclone; 133 static d_dump_t diskdump; 134 135 static LIST_HEAD(, disk) disklist = LIST_HEAD_INITIALIZER(&disklist); 136 static struct lwkt_token disklist_token; 137 138 static struct dev_ops disk_ops = { 139 { "disk", 0, D_DISK | D_MPSAFE }, 140 .d_open = diskopen, 141 .d_close = diskclose, 142 .d_read = physread, 143 .d_write = physwrite, 144 .d_ioctl = diskioctl, 145 .d_strategy = diskstrategy, 146 .d_dump = diskdump, 147 .d_psize = diskpsize, 148 .d_clone = diskclone 149 }; 150 151 static struct objcache *disk_msg_cache; 152 153 struct objcache_malloc_args disk_msg_malloc_args = { 154 sizeof(struct disk_msg), M_DISK }; 155 156 static struct lwkt_port disk_dispose_port; 157 static struct lwkt_port disk_msg_port; 158 159 static int 160 disk_debug(int level, char *fmt, ...) 161 { 162 __va_list ap; 163 164 __va_start(ap, fmt); 165 if (level <= disk_debug_enable) 166 kvprintf(fmt, ap); 167 __va_end(ap); 168 169 return 0; 170 } 171 172 static int 173 disk_probe_slice(struct disk *dp, cdev_t dev, int slice, int reprobe) 174 { 175 struct disk_info *info = &dp->d_info; 176 struct diskslice *sp = &dp->d_slice->dss_slices[slice]; 177 disklabel_ops_t ops; 178 struct partinfo part; 179 const char *msg; 180 cdev_t ndev; 181 int sno; 182 u_int i; 183 184 disk_debug(2, 185 "disk_probe_slice (begin): %s (%s)\n", 186 dev->si_name, dp->d_cdev->si_name); 187 188 sno = slice ? slice - 1 : 0; 189 190 ops = &disklabel32_ops; 191 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info); 192 if (msg && !strcmp(msg, "no disk label")) { 193 ops = &disklabel64_ops; 194 msg = ops->op_readdisklabel(dev, sp, &sp->ds_label, info); 195 } 196 if (msg == NULL) { 197 if (slice != WHOLE_DISK_SLICE) 198 ops->op_adjust_label_reserved(dp->d_slice, slice, sp); 199 else 200 sp->ds_reserved = 0; 201 202 sp->ds_ops = ops; 203 for (i = 0; i < ops->op_getnumparts(sp->ds_label); i++) { 204 ops->op_loadpartinfo(sp->ds_label, i, &part); 205 if (part.fstype) { 206 if (reprobe && 207 (ndev = devfs_find_device_by_name("%s%c", 208 dev->si_name, 'a' + i)) 209 ) { 210 /* 211 * Device already exists and 212 * is still valid. 213 */ 214 ndev->si_flags |= SI_REPROBE_TEST; 215 } else { 216 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops, 217 dkmakeminor(dkunit(dp->d_cdev), 218 slice, i), 219 UID_ROOT, GID_OPERATOR, 0640, 220 "%s%c", dev->si_name, 'a'+ i); 221 ndev->si_disk = dp; 222 udev_dict_set_cstr(ndev, "subsystem", "disk"); 223 /* Inherit parent's disk type */ 224 if (dp->d_disktype) { 225 udev_dict_set_cstr(ndev, "disk-type", 226 __DECONST(char *, dp->d_disktype)); 227 } 228 if (dp->d_info.d_serialno) { 229 make_dev_alias(ndev, 230 "serno/%s.s%d%c", 231 dp->d_info.d_serialno, 232 sno, 'a' + i); 233 } 234 ndev->si_flags |= SI_REPROBE_TEST; 235 } 236 } 237 } 238 } else if (info->d_dsflags & DSO_COMPATLABEL) { 239 msg = NULL; 240 if (sp->ds_size >= 0x100000000ULL) 241 ops = &disklabel64_ops; 242 else 243 ops = &disklabel32_ops; 244 sp->ds_label = ops->op_clone_label(info, sp); 245 } else { 246 if (sp->ds_type == DOSPTYP_386BSD || /* XXX */ 247 sp->ds_type == DOSPTYP_NETBSD || 248 sp->ds_type == DOSPTYP_OPENBSD) { 249 log(LOG_WARNING, "%s: cannot find label (%s)\n", 250 dev->si_name, msg); 251 } 252 } 253 254 if (msg == NULL) { 255 sp->ds_wlabel = FALSE; 256 } 257 258 return (msg ? EINVAL : 0); 259 } 260 261 /* 262 * This routine is only called for newly minted drives or to reprobe 263 * a drive with no open slices. disk_probe_slice() is called directly 264 * when reprobing partition changes within slices. 265 */ 266 static void 267 disk_probe(struct disk *dp, int reprobe) 268 { 269 struct disk_info *info = &dp->d_info; 270 cdev_t dev = dp->d_cdev; 271 cdev_t ndev; 272 int error, i, sno; 273 struct diskslices *osp; 274 struct diskslice *sp; 275 276 KKASSERT (info->d_media_blksize != 0); 277 278 osp = dp->d_slice; 279 dp->d_slice = dsmakeslicestruct(BASE_SLICE, info); 280 disk_debug(1, "disk_probe (begin): %s\n", dp->d_cdev->si_name); 281 282 error = mbrinit(dev, info, &(dp->d_slice)); 283 if (error) { 284 dsgone(&osp); 285 return; 286 } 287 288 for (i = 0; i < dp->d_slice->dss_nslices; i++) { 289 /* 290 * Ignore the whole-disk slice, it has already been created. 291 */ 292 if (i == WHOLE_DISK_SLICE) 293 continue; 294 sp = &dp->d_slice->dss_slices[i]; 295 296 /* 297 * Handle s0. s0 is a compatibility slice if there are no 298 * other slices and it has not otherwise been set up, else 299 * we ignore it. 300 */ 301 if (i == COMPATIBILITY_SLICE) { 302 sno = 0; 303 if (sp->ds_type == 0 && 304 dp->d_slice->dss_nslices == BASE_SLICE) { 305 sp->ds_size = info->d_media_blocks; 306 sp->ds_reserved = 0; 307 } 308 } else { 309 sno = i - 1; 310 sp->ds_reserved = 0; 311 } 312 313 /* 314 * Ignore 0-length slices 315 */ 316 if (sp->ds_size == 0) 317 continue; 318 319 if (reprobe && 320 (ndev = devfs_find_device_by_name("%ss%d", 321 dev->si_name, sno))) { 322 /* 323 * Device already exists and is still valid 324 */ 325 ndev->si_flags |= SI_REPROBE_TEST; 326 } else { 327 /* 328 * Else create new device 329 */ 330 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops, 331 dkmakewholeslice(dkunit(dev), i), 332 UID_ROOT, GID_OPERATOR, 0640, 333 "%ss%d", dev->si_name, sno); 334 udev_dict_set_cstr(ndev, "subsystem", "disk"); 335 /* Inherit parent's disk type */ 336 if (dp->d_disktype) { 337 udev_dict_set_cstr(ndev, "disk-type", 338 __DECONST(char *, dp->d_disktype)); 339 } 340 if (dp->d_info.d_serialno) { 341 make_dev_alias(ndev, "serno/%s.s%d", 342 dp->d_info.d_serialno, sno); 343 } 344 ndev->si_disk = dp; 345 ndev->si_flags |= SI_REPROBE_TEST; 346 } 347 sp->ds_dev = ndev; 348 349 /* 350 * Probe appropriate slices for a disklabel 351 * 352 * XXX slice type 1 used by our gpt probe code. 353 * XXX slice type 0 used by mbr compat slice. 354 */ 355 if (sp->ds_type == DOSPTYP_386BSD || 356 sp->ds_type == DOSPTYP_NETBSD || 357 sp->ds_type == DOSPTYP_OPENBSD || 358 sp->ds_type == 0 || 359 sp->ds_type == 1) { 360 if (dp->d_slice->dss_first_bsd_slice == 0) 361 dp->d_slice->dss_first_bsd_slice = i; 362 disk_probe_slice(dp, ndev, i, reprobe); 363 } 364 } 365 dsgone(&osp); 366 disk_debug(1, "disk_probe (end): %s\n", dp->d_cdev->si_name); 367 } 368 369 370 static void 371 disk_msg_core(void *arg) 372 { 373 struct disk *dp; 374 struct diskslice *sp; 375 disk_msg_t msg; 376 int run; 377 378 lwkt_gettoken(&disklist_token); 379 lwkt_initport_thread(&disk_msg_port, curthread); 380 wakeup(curthread); /* synchronous startup */ 381 lwkt_reltoken(&disklist_token); 382 383 get_mplock(); /* not mpsafe yet? */ 384 run = 1; 385 386 while (run) { 387 msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0); 388 389 switch (msg->hdr.u.ms_result) { 390 case DISK_DISK_PROBE: 391 dp = (struct disk *)msg->load; 392 disk_debug(1, 393 "DISK_DISK_PROBE: %s\n", 394 dp->d_cdev->si_name); 395 disk_probe(dp, 0); 396 break; 397 case DISK_DISK_DESTROY: 398 dp = (struct disk *)msg->load; 399 disk_debug(1, 400 "DISK_DISK_DESTROY: %s\n", 401 dp->d_cdev->si_name); 402 devfs_destroy_subnames(dp->d_cdev->si_name); 403 devfs_destroy_dev(dp->d_cdev); 404 lwkt_gettoken(&disklist_token); 405 LIST_REMOVE(dp, d_list); 406 lwkt_reltoken(&disklist_token); 407 if (dp->d_info.d_serialno) { 408 kfree(dp->d_info.d_serialno, M_TEMP); 409 dp->d_info.d_serialno = NULL; 410 } 411 break; 412 case DISK_UNPROBE: 413 dp = (struct disk *)msg->load; 414 disk_debug(1, 415 "DISK_DISK_UNPROBE: %s\n", 416 dp->d_cdev->si_name); 417 devfs_destroy_subnames(dp->d_cdev->si_name); 418 break; 419 case DISK_SLICE_REPROBE: 420 dp = (struct disk *)msg->load; 421 sp = (struct diskslice *)msg->load2; 422 devfs_clr_subnames_flag(sp->ds_dev->si_name, 423 SI_REPROBE_TEST); 424 disk_debug(1, 425 "DISK_SLICE_REPROBE: %s\n", 426 sp->ds_dev->si_name); 427 disk_probe_slice(dp, sp->ds_dev, 428 dkslice(sp->ds_dev), 1); 429 devfs_destroy_subnames_without_flag( 430 sp->ds_dev->si_name, SI_REPROBE_TEST); 431 break; 432 case DISK_DISK_REPROBE: 433 dp = (struct disk *)msg->load; 434 devfs_clr_subnames_flag(dp->d_cdev->si_name, SI_REPROBE_TEST); 435 disk_debug(1, 436 "DISK_DISK_REPROBE: %s\n", 437 dp->d_cdev->si_name); 438 disk_probe(dp, 1); 439 devfs_destroy_subnames_without_flag( 440 dp->d_cdev->si_name, SI_REPROBE_TEST); 441 break; 442 case DISK_SYNC: 443 disk_debug(1, "DISK_SYNC\n"); 444 break; 445 default: 446 devfs_debug(DEVFS_DEBUG_WARNING, 447 "disk_msg_core: unknown message " 448 "received at core\n"); 449 break; 450 } 451 lwkt_replymsg(&msg->hdr, 0); 452 } 453 lwkt_exit(); 454 } 455 456 457 /* 458 * Acts as a message drain. Any message that is replied to here gets 459 * destroyed and the memory freed. 460 */ 461 static void 462 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg) 463 { 464 objcache_put(disk_msg_cache, msg); 465 } 466 467 468 void 469 disk_msg_send(uint32_t cmd, void *load, void *load2) 470 { 471 disk_msg_t disk_msg; 472 lwkt_port_t port = &disk_msg_port; 473 474 disk_msg = objcache_get(disk_msg_cache, M_WAITOK); 475 476 lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0); 477 478 disk_msg->hdr.u.ms_result = cmd; 479 disk_msg->load = load; 480 disk_msg->load2 = load2; 481 KKASSERT(port); 482 lwkt_sendmsg(port, &disk_msg->hdr); 483 } 484 485 void 486 disk_msg_send_sync(uint32_t cmd, void *load, void *load2) 487 { 488 struct lwkt_port rep_port; 489 disk_msg_t disk_msg; 490 lwkt_port_t port; 491 492 disk_msg = objcache_get(disk_msg_cache, M_WAITOK); 493 port = &disk_msg_port; 494 495 /* XXX could probably use curthread's built-in msgport */ 496 lwkt_initport_thread(&rep_port, curthread); 497 lwkt_initmsg(&disk_msg->hdr, &rep_port, 0); 498 499 disk_msg->hdr.u.ms_result = cmd; 500 disk_msg->load = load; 501 disk_msg->load2 = load2; 502 503 lwkt_sendmsg(port, &disk_msg->hdr); 504 lwkt_waitmsg(&disk_msg->hdr, 0); 505 objcache_put(disk_msg_cache, disk_msg); 506 } 507 508 /* 509 * Create a raw device for the dev_ops template (which is returned). Also 510 * create a slice and unit managed disk and overload the user visible 511 * device space with it. 512 * 513 * NOTE: The returned raw device is NOT a slice and unit managed device. 514 * It is an actual raw device representing the raw disk as specified by 515 * the passed dev_ops. The disk layer not only returns such a raw device, 516 * it also uses it internally when passing (modified) commands through. 517 */ 518 cdev_t 519 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops) 520 { 521 return disk_create_named(NULL, unit, dp, raw_ops); 522 } 523 524 cdev_t 525 disk_create_named(const char *name, int unit, struct disk *dp, struct dev_ops *raw_ops) 526 { 527 cdev_t rawdev; 528 529 if (name == NULL) 530 name = raw_ops->head.name; 531 532 disk_debug(1, "disk_create (begin): %s%d\n", name, unit); 533 534 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit), 535 UID_ROOT, GID_OPERATOR, 0640, 536 "%s%d", name, unit); 537 538 bzero(dp, sizeof(*dp)); 539 540 dp->d_rawdev = rawdev; 541 dp->d_raw_ops = raw_ops; 542 dp->d_dev_ops = &disk_ops; 543 dp->d_cdev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops, 544 dkmakewholedisk(unit), 545 UID_ROOT, GID_OPERATOR, 0640, 546 "%s%d", name, unit); 547 udev_dict_set_cstr(dp->d_cdev, "subsystem", "disk"); 548 dp->d_cdev->si_disk = dp; 549 550 dsched_disk_create_callback(dp, name, unit); 551 552 lwkt_gettoken(&disklist_token); 553 LIST_INSERT_HEAD(&disklist, dp, d_list); 554 lwkt_reltoken(&disklist_token); 555 556 disk_debug(1, "disk_create (end): %s%d\n", name, unit); 557 558 return (dp->d_rawdev); 559 } 560 561 int 562 disk_setdisktype(struct disk *disk, const char *type) 563 { 564 KKASSERT(disk != NULL); 565 566 disk->d_disktype = type; 567 return udev_dict_set_cstr(disk->d_cdev, "disk-type", __DECONST(char *, type)); 568 } 569 570 static void 571 _setdiskinfo(struct disk *disk, struct disk_info *info) 572 { 573 char *oldserialno; 574 575 oldserialno = disk->d_info.d_serialno; 576 bcopy(info, &disk->d_info, sizeof(disk->d_info)); 577 info = &disk->d_info; 578 579 disk_debug(1, 580 "_setdiskinfo: %s\n", 581 disk->d_cdev->si_name); 582 583 /* 584 * The serial number is duplicated so the caller can throw 585 * their copy away. 586 */ 587 if (info->d_serialno && info->d_serialno[0]) { 588 info->d_serialno = kstrdup(info->d_serialno, M_TEMP); 589 disk_cleanserial(info->d_serialno); 590 if (disk->d_cdev) { 591 make_dev_alias(disk->d_cdev, "serno/%s", 592 info->d_serialno); 593 } 594 } else { 595 info->d_serialno = NULL; 596 } 597 if (oldserialno) 598 kfree(oldserialno, M_TEMP); 599 600 dsched_disk_update_callback(disk, info); 601 602 /* 603 * The caller may set d_media_size or d_media_blocks and we 604 * calculate the other. 605 */ 606 KKASSERT(info->d_media_size == 0 || info->d_media_blksize == 0); 607 if (info->d_media_size == 0 && info->d_media_blocks) { 608 info->d_media_size = (u_int64_t)info->d_media_blocks * 609 info->d_media_blksize; 610 } else if (info->d_media_size && info->d_media_blocks == 0 && 611 info->d_media_blksize) { 612 info->d_media_blocks = info->d_media_size / 613 info->d_media_blksize; 614 } 615 616 /* 617 * The si_* fields for rawdev are not set until after the 618 * disk_create() call, so someone using the cooked version 619 * of the raw device (i.e. da0s0) will not get the right 620 * si_iosize_max unless we fix it up here. 621 */ 622 if (disk->d_cdev && disk->d_rawdev && 623 disk->d_cdev->si_iosize_max == 0) { 624 disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max; 625 disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys; 626 disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best; 627 } 628 629 /* Add the serial number to the udev_dictionary */ 630 if (info->d_serialno) 631 udev_dict_set_cstr(disk->d_cdev, "serno", info->d_serialno); 632 } 633 634 /* 635 * Disk drivers must call this routine when media parameters are available 636 * or have changed. 637 */ 638 void 639 disk_setdiskinfo(struct disk *disk, struct disk_info *info) 640 { 641 _setdiskinfo(disk, info); 642 disk_msg_send(DISK_DISK_PROBE, disk, NULL); 643 disk_debug(1, 644 "disk_setdiskinfo: sent probe for %s\n", 645 disk->d_cdev->si_name); 646 } 647 648 void 649 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info) 650 { 651 _setdiskinfo(disk, info); 652 disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL); 653 disk_debug(1, 654 "disk_setdiskinfo_sync: sent probe for %s\n", 655 disk->d_cdev->si_name); 656 } 657 658 /* 659 * This routine is called when an adapter detaches. The higher level 660 * managed disk device is destroyed while the lower level raw device is 661 * released. 662 */ 663 void 664 disk_destroy(struct disk *disk) 665 { 666 dsched_disk_destroy_callback(disk); 667 disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL); 668 return; 669 } 670 671 int 672 disk_dumpcheck(cdev_t dev, u_int64_t *size, u_int64_t *blkno, u_int32_t *secsize) 673 { 674 struct partinfo pinfo; 675 int error; 676 677 bzero(&pinfo, sizeof(pinfo)); 678 error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0, 679 proc0.p_ucred, NULL); 680 if (error) 681 return (error); 682 683 if (pinfo.media_blksize == 0) 684 return (ENXIO); 685 686 if (blkno) /* XXX: make sure this reserved stuff is right */ 687 *blkno = pinfo.reserved_blocks + 688 pinfo.media_offset / pinfo.media_blksize; 689 if (secsize) 690 *secsize = pinfo.media_blksize; 691 if (size) 692 *size = (pinfo.media_blocks - pinfo.reserved_blocks); 693 694 return (0); 695 } 696 697 int 698 disk_dumpconf(cdev_t dev, u_int onoff) 699 { 700 struct dumperinfo di; 701 u_int64_t size, blkno; 702 u_int32_t secsize; 703 int error; 704 705 if (!onoff) 706 return set_dumper(NULL); 707 708 error = disk_dumpcheck(dev, &size, &blkno, &secsize); 709 710 if (error) 711 return ENXIO; 712 713 bzero(&di, sizeof(struct dumperinfo)); 714 di.dumper = diskdump; 715 di.priv = dev; 716 di.blocksize = secsize; 717 di.mediaoffset = blkno * DEV_BSIZE; 718 di.mediasize = size * DEV_BSIZE; 719 720 return set_dumper(&di); 721 } 722 723 void 724 disk_unprobe(struct disk *disk) 725 { 726 if (disk == NULL) 727 return; 728 729 disk_msg_send_sync(DISK_UNPROBE, disk, NULL); 730 } 731 732 void 733 disk_invalidate (struct disk *disk) 734 { 735 dsgone(&disk->d_slice); 736 } 737 738 struct disk * 739 disk_enumerate(struct disk *disk) 740 { 741 struct disk *dp; 742 743 lwkt_gettoken(&disklist_token); 744 if (!disk) 745 dp = (LIST_FIRST(&disklist)); 746 else 747 dp = (LIST_NEXT(disk, d_list)); 748 lwkt_reltoken(&disklist_token); 749 750 return dp; 751 } 752 753 static 754 int 755 sysctl_disks(SYSCTL_HANDLER_ARGS) 756 { 757 struct disk *disk; 758 int error, first; 759 760 disk = NULL; 761 first = 1; 762 763 while ((disk = disk_enumerate(disk))) { 764 if (!first) { 765 error = SYSCTL_OUT(req, " ", 1); 766 if (error) 767 return error; 768 } else { 769 first = 0; 770 } 771 error = SYSCTL_OUT(req, disk->d_rawdev->si_name, 772 strlen(disk->d_rawdev->si_name)); 773 if (error) 774 return error; 775 } 776 error = SYSCTL_OUT(req, "", 1); 777 return error; 778 } 779 780 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0, 781 sysctl_disks, "A", "names of available disks"); 782 783 /* 784 * Open a disk device or partition. 785 */ 786 static 787 int 788 diskopen(struct dev_open_args *ap) 789 { 790 cdev_t dev = ap->a_head.a_dev; 791 struct disk *dp; 792 int error; 793 794 /* 795 * dp can't be NULL here XXX. 796 * 797 * d_slice will be NULL if setdiskinfo() has not been called yet. 798 * setdiskinfo() is typically called whether the disk is present 799 * or not (e.g. CD), but the base disk device is created first 800 * and there may be a race. 801 */ 802 dp = dev->si_disk; 803 if (dp == NULL || dp->d_slice == NULL) 804 return (ENXIO); 805 error = 0; 806 807 /* 808 * Deal with open races 809 */ 810 get_mplock(); 811 while (dp->d_flags & DISKFLAG_LOCK) { 812 dp->d_flags |= DISKFLAG_WANTED; 813 error = tsleep(dp, PCATCH, "diskopen", hz); 814 if (error) { 815 rel_mplock(); 816 return (error); 817 } 818 } 819 dp->d_flags |= DISKFLAG_LOCK; 820 821 /* 822 * Open the underlying raw device. 823 */ 824 if (!dsisopen(dp->d_slice)) { 825 #if 0 826 if (!pdev->si_iosize_max) 827 pdev->si_iosize_max = dev->si_iosize_max; 828 #endif 829 error = dev_dopen(dp->d_rawdev, ap->a_oflags, 830 ap->a_devtype, ap->a_cred); 831 } 832 #if 0 833 /* 834 * Inherit properties from the underlying device now that it is 835 * open. 836 */ 837 dev_dclone(dev); 838 #endif 839 840 if (error) 841 goto out; 842 error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags, 843 &dp->d_slice, &dp->d_info); 844 if (!dsisopen(dp->d_slice)) { 845 dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype); 846 } 847 out: 848 dp->d_flags &= ~DISKFLAG_LOCK; 849 if (dp->d_flags & DISKFLAG_WANTED) { 850 dp->d_flags &= ~DISKFLAG_WANTED; 851 wakeup(dp); 852 } 853 rel_mplock(); 854 855 return(error); 856 } 857 858 /* 859 * Close a disk device or partition 860 */ 861 static 862 int 863 diskclose(struct dev_close_args *ap) 864 { 865 cdev_t dev = ap->a_head.a_dev; 866 struct disk *dp; 867 int error; 868 869 error = 0; 870 dp = dev->si_disk; 871 872 get_mplock(); 873 dsclose(dev, ap->a_devtype, dp->d_slice); 874 if (!dsisopen(dp->d_slice)) { 875 error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype); 876 } 877 rel_mplock(); 878 return (error); 879 } 880 881 /* 882 * First execute the ioctl on the disk device, and if it isn't supported 883 * try running it on the backing device. 884 */ 885 static 886 int 887 diskioctl(struct dev_ioctl_args *ap) 888 { 889 cdev_t dev = ap->a_head.a_dev; 890 struct disk *dp; 891 int error; 892 u_int u; 893 894 dp = dev->si_disk; 895 if (dp == NULL) 896 return (ENXIO); 897 898 devfs_debug(DEVFS_DEBUG_DEBUG, 899 "diskioctl: cmd is: %lx (name: %s)\n", 900 ap->a_cmd, dev->si_name); 901 devfs_debug(DEVFS_DEBUG_DEBUG, 902 "diskioctl: &dp->d_slice is: %p, %p\n", 903 &dp->d_slice, dp->d_slice); 904 905 if (ap->a_cmd == DIOCGKERNELDUMP) { 906 u = *(u_int *)ap->a_data; 907 return disk_dumpconf(dev, u); 908 } 909 910 if (&dp->d_slice == NULL || dp->d_slice == NULL) { 911 error = ENOIOCTL; 912 } else { 913 get_mplock(); 914 error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag, 915 &dp->d_slice, &dp->d_info); 916 rel_mplock(); 917 } 918 919 if (error == ENOIOCTL) { 920 error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data, 921 ap->a_fflag, ap->a_cred, NULL); 922 } 923 return (error); 924 } 925 926 /* 927 * Execute strategy routine 928 */ 929 static 930 int 931 diskstrategy(struct dev_strategy_args *ap) 932 { 933 cdev_t dev = ap->a_head.a_dev; 934 struct bio *bio = ap->a_bio; 935 struct bio *nbio; 936 struct disk *dp; 937 938 dp = dev->si_disk; 939 940 if (dp == NULL) { 941 bio->bio_buf->b_error = ENXIO; 942 bio->bio_buf->b_flags |= B_ERROR; 943 biodone(bio); 944 return(0); 945 } 946 KKASSERT(dev->si_disk == dp); 947 948 /* 949 * The dscheck() function will also transform the slice relative 950 * block number i.e. bio->bio_offset into a block number that can be 951 * passed directly to the underlying raw device. If dscheck() 952 * returns NULL it will have handled the bio for us (e.g. EOF 953 * or error due to being beyond the device size). 954 */ 955 if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) { 956 dsched_queue(dp, nbio); 957 } else { 958 biodone(bio); 959 } 960 return(0); 961 } 962 963 /* 964 * Return the partition size in ?blocks? 965 */ 966 static 967 int 968 diskpsize(struct dev_psize_args *ap) 969 { 970 cdev_t dev = ap->a_head.a_dev; 971 struct disk *dp; 972 973 dp = dev->si_disk; 974 if (dp == NULL) 975 return(ENODEV); 976 ap->a_result = dssize(dev, &dp->d_slice); 977 return(0); 978 } 979 980 /* 981 * When new device entries are instantiated, make sure they inherit our 982 * si_disk structure and block and iosize limits from the raw device. 983 * 984 * This routine is always called synchronously in the context of the 985 * client. 986 * 987 * XXX The various io and block size constraints are not always initialized 988 * properly by devices. 989 */ 990 static 991 int 992 diskclone(struct dev_clone_args *ap) 993 { 994 cdev_t dev = ap->a_head.a_dev; 995 struct disk *dp; 996 dp = dev->si_disk; 997 998 KKASSERT(dp != NULL); 999 dev->si_disk = dp; 1000 dev->si_iosize_max = dp->d_rawdev->si_iosize_max; 1001 dev->si_bsize_phys = dp->d_rawdev->si_bsize_phys; 1002 dev->si_bsize_best = dp->d_rawdev->si_bsize_best; 1003 return(0); 1004 } 1005 1006 int 1007 diskdump(struct dev_dump_args *ap) 1008 { 1009 cdev_t dev = ap->a_head.a_dev; 1010 struct disk *dp = dev->si_disk; 1011 u_int64_t size, offset; 1012 int error; 1013 1014 error = disk_dumpcheck(dev, &size, &ap->a_blkno, &ap->a_secsize); 1015 /* XXX: this should probably go in disk_dumpcheck somehow */ 1016 if (ap->a_length != 0) { 1017 size *= DEV_BSIZE; 1018 offset = ap->a_blkno * DEV_BSIZE; 1019 if ((ap->a_offset < offset) || 1020 (ap->a_offset + ap->a_length - offset > size)) { 1021 kprintf("Attempt to write outside dump device boundaries.\n"); 1022 error = ENOSPC; 1023 } 1024 } 1025 1026 if (error == 0) { 1027 ap->a_head.a_dev = dp->d_rawdev; 1028 error = dev_doperate(&ap->a_head); 1029 } 1030 1031 return(error); 1032 } 1033 1034 1035 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD, 1036 0, sizeof(struct diskslices), "sizeof(struct diskslices)"); 1037 1038 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD, 1039 0, sizeof(struct disk), "sizeof(struct disk)"); 1040 1041 /* 1042 * Reorder interval for burst write allowance and minor write 1043 * allowance. 1044 * 1045 * We always want to trickle some writes in to make use of the 1046 * disk's zone cache. Bursting occurs on a longer interval and only 1047 * runningbufspace is well over the hirunningspace limit. 1048 */ 1049 int bioq_reorder_burst_interval = 60; /* should be multiple of minor */ 1050 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_interval, 1051 CTLFLAG_RW, &bioq_reorder_burst_interval, 0, ""); 1052 int bioq_reorder_minor_interval = 5; 1053 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_interval, 1054 CTLFLAG_RW, &bioq_reorder_minor_interval, 0, ""); 1055 1056 int bioq_reorder_burst_bytes = 3000000; 1057 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_bytes, 1058 CTLFLAG_RW, &bioq_reorder_burst_bytes, 0, ""); 1059 int bioq_reorder_minor_bytes = 262144; 1060 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_bytes, 1061 CTLFLAG_RW, &bioq_reorder_minor_bytes, 0, ""); 1062 1063 1064 /* 1065 * Order I/Os. Generally speaking this code is designed to make better 1066 * use of drive zone caches. A drive zone cache can typically track linear 1067 * reads or writes for around 16 zones simultaniously. 1068 * 1069 * Read prioritization issues: It is possible for hundreds of megabytes worth 1070 * of writes to be queued asynchronously. This creates a huge bottleneck 1071 * for reads which reduce read bandwidth to a trickle. 1072 * 1073 * To solve this problem we generally reorder reads before writes. 1074 * 1075 * However, a large number of random reads can also starve writes and 1076 * make poor use of the drive zone cache so we allow writes to trickle 1077 * in every N reads. 1078 */ 1079 void 1080 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio) 1081 { 1082 /* 1083 * The BIO wants to be ordered. Adding to the tail also 1084 * causes transition to be set to NULL, forcing the ordering 1085 * of all prior I/O's. 1086 */ 1087 if (bio->bio_buf->b_flags & B_ORDERED) { 1088 bioq_insert_tail(bioq, bio); 1089 return; 1090 } 1091 1092 switch(bio->bio_buf->b_cmd) { 1093 case BUF_CMD_READ: 1094 if (bioq->transition) { 1095 /* 1096 * Insert before the first write. Bleedover writes 1097 * based on reorder intervals to prevent starvation. 1098 */ 1099 TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act); 1100 ++bioq->reorder; 1101 if (bioq->reorder % bioq_reorder_minor_interval == 0) { 1102 bioqwritereorder(bioq); 1103 if (bioq->reorder >= 1104 bioq_reorder_burst_interval) { 1105 bioq->reorder = 0; 1106 } 1107 } 1108 } else { 1109 /* 1110 * No writes queued (or ordering was forced), 1111 * insert at tail. 1112 */ 1113 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act); 1114 } 1115 break; 1116 case BUF_CMD_WRITE: 1117 /* 1118 * Writes are always appended. If no writes were previously 1119 * queued or an ordered tail insertion occured the transition 1120 * field will be NULL. 1121 */ 1122 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act); 1123 if (bioq->transition == NULL) 1124 bioq->transition = bio; 1125 break; 1126 default: 1127 /* 1128 * All other request types are forced to be ordered. 1129 */ 1130 bioq_insert_tail(bioq, bio); 1131 break; 1132 } 1133 } 1134 1135 /* 1136 * Move the read-write transition point to prevent reads from 1137 * completely starving our writes. This brings a number of writes into 1138 * the fold every N reads. 1139 * 1140 * We bring a few linear writes into the fold on a minor interval 1141 * and we bring a non-linear burst of writes into the fold on a major 1142 * interval. Bursting only occurs if runningbufspace is really high 1143 * (typically from syncs, fsyncs, or HAMMER flushes). 1144 */ 1145 static 1146 void 1147 bioqwritereorder(struct bio_queue_head *bioq) 1148 { 1149 struct bio *bio; 1150 off_t next_offset; 1151 size_t left; 1152 size_t n; 1153 int check_off; 1154 1155 if (bioq->reorder < bioq_reorder_burst_interval || 1156 !buf_runningbufspace_severe()) { 1157 left = (size_t)bioq_reorder_minor_bytes; 1158 check_off = 1; 1159 } else { 1160 left = (size_t)bioq_reorder_burst_bytes; 1161 check_off = 0; 1162 } 1163 1164 next_offset = bioq->transition->bio_offset; 1165 while ((bio = bioq->transition) != NULL && 1166 (check_off == 0 || next_offset == bio->bio_offset) 1167 ) { 1168 n = bio->bio_buf->b_bcount; 1169 next_offset = bio->bio_offset + n; 1170 bioq->transition = TAILQ_NEXT(bio, bio_act); 1171 if (left < n) 1172 break; 1173 left -= n; 1174 } 1175 } 1176 1177 /* 1178 * Bounds checking against the media size, used for the raw partition. 1179 * secsize, mediasize and b_blkno must all be the same units. 1180 * Possibly this has to be DEV_BSIZE (512). 1181 */ 1182 int 1183 bounds_check_with_mediasize(struct bio *bio, int secsize, uint64_t mediasize) 1184 { 1185 struct buf *bp = bio->bio_buf; 1186 int64_t sz; 1187 1188 sz = howmany(bp->b_bcount, secsize); 1189 1190 if (bio->bio_offset/DEV_BSIZE + sz > mediasize) { 1191 sz = mediasize - bio->bio_offset/DEV_BSIZE; 1192 if (sz == 0) { 1193 /* If exactly at end of disk, return EOF. */ 1194 bp->b_resid = bp->b_bcount; 1195 return 0; 1196 } 1197 if (sz < 0) { 1198 /* If past end of disk, return EINVAL. */ 1199 bp->b_error = EINVAL; 1200 return 0; 1201 } 1202 /* Otherwise, truncate request. */ 1203 bp->b_bcount = sz * secsize; 1204 } 1205 1206 return 1; 1207 } 1208 1209 /* 1210 * Disk error is the preface to plaintive error messages 1211 * about failing disk transfers. It prints messages of the form 1212 1213 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d) 1214 1215 * if the offset of the error in the transfer and a disk label 1216 * are both available. blkdone should be -1 if the position of the error 1217 * is unknown; the disklabel pointer may be null from drivers that have not 1218 * been converted to use them. The message is printed with kprintf 1219 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority. 1220 * The message should be completed (with at least a newline) with kprintf 1221 * or log(-1, ...), respectively. There is no trailing space. 1222 */ 1223 void 1224 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt) 1225 { 1226 struct buf *bp = bio->bio_buf; 1227 const char *term; 1228 1229 switch(bp->b_cmd) { 1230 case BUF_CMD_READ: 1231 term = "read"; 1232 break; 1233 case BUF_CMD_WRITE: 1234 term = "write"; 1235 break; 1236 default: 1237 term = "access"; 1238 break; 1239 } 1240 kprintf("%s: %s %sing ", dev->si_name, what, term); 1241 kprintf("offset %012llx for %d", 1242 (long long)bio->bio_offset, 1243 bp->b_bcount); 1244 1245 if (donecnt) 1246 kprintf(" (%d bytes completed)", donecnt); 1247 } 1248 1249 /* 1250 * Locate a disk device 1251 */ 1252 cdev_t 1253 disk_locate(const char *devname) 1254 { 1255 return devfs_find_device_by_name(devname); 1256 } 1257 1258 void 1259 disk_config(void *arg) 1260 { 1261 disk_msg_send_sync(DISK_SYNC, NULL, NULL); 1262 } 1263 1264 static void 1265 disk_init(void) 1266 { 1267 struct thread* td_core; 1268 1269 disk_msg_cache = objcache_create("disk-msg-cache", 0, 0, 1270 NULL, NULL, NULL, 1271 objcache_malloc_alloc, 1272 objcache_malloc_free, 1273 &disk_msg_malloc_args); 1274 1275 lwkt_token_init(&disklist_token, 1, "disks"); 1276 1277 /* 1278 * Initialize the reply-only port which acts as a message drain 1279 */ 1280 lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply); 1281 1282 lwkt_gettoken(&disklist_token); 1283 lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL, 1284 0, 0, "disk_msg_core"); 1285 tsleep(td_core, 0, "diskcore", 0); 1286 lwkt_reltoken(&disklist_token); 1287 } 1288 1289 static void 1290 disk_uninit(void) 1291 { 1292 objcache_destroy(disk_msg_cache); 1293 } 1294 1295 /* 1296 * Clean out illegal characters in serial numbers. 1297 */ 1298 static void 1299 disk_cleanserial(char *serno) 1300 { 1301 char c; 1302 1303 while ((c = *serno) != 0) { 1304 if (c >= 'a' && c <= 'z') 1305 ; 1306 else if (c >= 'A' && c <= 'Z') 1307 ; 1308 else if (c >= '0' && c <= '9') 1309 ; 1310 else if (c == '-' || c == '@' || c == '+' || c == '.') 1311 ; 1312 else 1313 c = '_'; 1314 *serno++= c; 1315 } 1316 } 1317 1318 TUNABLE_INT("kern.disk_debug", &disk_debug_enable); 1319 SYSCTL_INT(_kern, OID_AUTO, disk_debug, CTLFLAG_RW, &disk_debug_enable, 1320 0, "Enable subr_disk debugging"); 1321 1322 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL); 1323 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL); 1324