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 static int disk_debug(int, char *, ...) __printflike(2, 3); 127 128 static d_open_t diskopen; 129 static d_close_t diskclose; 130 static d_ioctl_t diskioctl; 131 static d_strategy_t diskstrategy; 132 static d_psize_t diskpsize; 133 static d_clone_t diskclone; 134 static d_dump_t diskdump; 135 136 static LIST_HEAD(, disk) disklist = LIST_HEAD_INITIALIZER(&disklist); 137 static struct lwkt_token disklist_token; 138 139 static struct dev_ops disk_ops = { 140 { "disk", 0, D_DISK | D_MPSAFE }, 141 .d_open = diskopen, 142 .d_close = diskclose, 143 .d_read = physread, 144 .d_write = physwrite, 145 .d_ioctl = diskioctl, 146 .d_strategy = diskstrategy, 147 .d_dump = diskdump, 148 .d_psize = diskpsize, 149 .d_clone = diskclone 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 cdev_t ndev; 182 int sno; 183 u_int i; 184 185 disk_debug(2, 186 "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 if (msg == NULL) { 198 if (slice != WHOLE_DISK_SLICE) 199 ops->op_adjust_label_reserved(dp->d_slice, slice, sp); 200 else 201 sp->ds_reserved = 0; 202 203 sp->ds_ops = ops; 204 for (i = 0; i < ops->op_getnumparts(sp->ds_label); i++) { 205 ops->op_loadpartinfo(sp->ds_label, i, &part); 206 if (part.fstype) { 207 if (reprobe && 208 (ndev = devfs_find_device_by_name("%s%c", 209 dev->si_name, 'a' + i)) 210 ) { 211 /* 212 * Device already exists and 213 * is still valid. 214 */ 215 ndev->si_flags |= SI_REPROBE_TEST; 216 } else { 217 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops, 218 dkmakeminor(dkunit(dp->d_cdev), 219 slice, i), 220 UID_ROOT, GID_OPERATOR, 0640, 221 "%s%c", dev->si_name, 'a'+ i); 222 ndev->si_disk = dp; 223 udev_dict_set_cstr(ndev, "subsystem", "disk"); 224 /* Inherit parent's disk type */ 225 if (dp->d_disktype) { 226 udev_dict_set_cstr(ndev, "disk-type", 227 __DECONST(char *, dp->d_disktype)); 228 } 229 if (dp->d_info.d_serialno) { 230 make_dev_alias(ndev, 231 "serno/%s.s%d%c", 232 dp->d_info.d_serialno, 233 sno, 'a' + i); 234 } 235 ndev->si_flags |= SI_REPROBE_TEST; 236 } 237 } 238 } 239 } else if (info->d_dsflags & DSO_COMPATLABEL) { 240 msg = NULL; 241 if (sp->ds_size >= 0x100000000ULL) 242 ops = &disklabel64_ops; 243 else 244 ops = &disklabel32_ops; 245 sp->ds_label = ops->op_clone_label(info, sp); 246 } else { 247 if (sp->ds_type == DOSPTYP_386BSD || /* XXX */ 248 sp->ds_type == DOSPTYP_NETBSD || 249 sp->ds_type == DOSPTYP_OPENBSD) { 250 log(LOG_WARNING, "%s: cannot find label (%s)\n", 251 dev->si_name, msg); 252 } 253 } 254 255 if (msg == NULL) { 256 sp->ds_wlabel = FALSE; 257 } 258 259 return (msg ? EINVAL : 0); 260 } 261 262 /* 263 * This routine is only called for newly minted drives or to reprobe 264 * a drive with no open slices. disk_probe_slice() is called directly 265 * when reprobing partition changes within slices. 266 */ 267 static void 268 disk_probe(struct disk *dp, int reprobe) 269 { 270 struct disk_info *info = &dp->d_info; 271 cdev_t dev = dp->d_cdev; 272 cdev_t ndev; 273 int error, i, sno; 274 struct diskslices *osp; 275 struct diskslice *sp; 276 277 KKASSERT (info->d_media_blksize != 0); 278 279 osp = dp->d_slice; 280 dp->d_slice = dsmakeslicestruct(BASE_SLICE, info); 281 disk_debug(1, "disk_probe (begin): %s\n", dp->d_cdev->si_name); 282 283 error = mbrinit(dev, info, &(dp->d_slice)); 284 if (error) { 285 dsgone(&osp); 286 return; 287 } 288 289 for (i = 0; i < dp->d_slice->dss_nslices; i++) { 290 /* 291 * Ignore the whole-disk slice, it has already been created. 292 */ 293 if (i == WHOLE_DISK_SLICE) 294 continue; 295 sp = &dp->d_slice->dss_slices[i]; 296 297 /* 298 * Handle s0. s0 is a compatibility slice if there are no 299 * other slices and it has not otherwise been set up, else 300 * we ignore it. 301 */ 302 if (i == COMPATIBILITY_SLICE) { 303 sno = 0; 304 if (sp->ds_type == 0 && 305 dp->d_slice->dss_nslices == BASE_SLICE) { 306 sp->ds_size = info->d_media_blocks; 307 sp->ds_reserved = 0; 308 } 309 } else { 310 sno = i - 1; 311 sp->ds_reserved = 0; 312 } 313 314 /* 315 * Ignore 0-length slices 316 */ 317 if (sp->ds_size == 0) 318 continue; 319 320 if (reprobe && 321 (ndev = devfs_find_device_by_name("%ss%d", 322 dev->si_name, sno))) { 323 /* 324 * Device already exists and is still valid 325 */ 326 ndev->si_flags |= SI_REPROBE_TEST; 327 } else { 328 /* 329 * Else create new device 330 */ 331 ndev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops, 332 dkmakewholeslice(dkunit(dev), i), 333 UID_ROOT, GID_OPERATOR, 0640, 334 "%ss%d", dev->si_name, sno); 335 udev_dict_set_cstr(ndev, "subsystem", "disk"); 336 /* Inherit parent's disk type */ 337 if (dp->d_disktype) { 338 udev_dict_set_cstr(ndev, "disk-type", 339 __DECONST(char *, dp->d_disktype)); 340 } 341 if (dp->d_info.d_serialno) { 342 make_dev_alias(ndev, "serno/%s.s%d", 343 dp->d_info.d_serialno, sno); 344 } 345 ndev->si_disk = dp; 346 ndev->si_flags |= SI_REPROBE_TEST; 347 } 348 sp->ds_dev = ndev; 349 350 /* 351 * Probe appropriate slices for a disklabel 352 * 353 * XXX slice type 1 used by our gpt probe code. 354 * XXX slice type 0 used by mbr compat slice. 355 */ 356 if (sp->ds_type == DOSPTYP_386BSD || 357 sp->ds_type == DOSPTYP_NETBSD || 358 sp->ds_type == DOSPTYP_OPENBSD || 359 sp->ds_type == 0 || 360 sp->ds_type == 1) { 361 if (dp->d_slice->dss_first_bsd_slice == 0) 362 dp->d_slice->dss_first_bsd_slice = i; 363 disk_probe_slice(dp, ndev, i, reprobe); 364 } 365 } 366 dsgone(&osp); 367 disk_debug(1, "disk_probe (end): %s\n", dp->d_cdev->si_name); 368 } 369 370 371 static void 372 disk_msg_core(void *arg) 373 { 374 struct disk *dp; 375 struct diskslice *sp; 376 disk_msg_t msg; 377 int run; 378 379 lwkt_gettoken(&disklist_token); 380 lwkt_initport_thread(&disk_msg_port, curthread); 381 wakeup(curthread); /* synchronous startup */ 382 lwkt_reltoken(&disklist_token); 383 384 get_mplock(); /* not mpsafe yet? */ 385 run = 1; 386 387 while (run) { 388 msg = (disk_msg_t)lwkt_waitport(&disk_msg_port, 0); 389 390 switch (msg->hdr.u.ms_result) { 391 case DISK_DISK_PROBE: 392 dp = (struct disk *)msg->load; 393 disk_debug(1, 394 "DISK_DISK_PROBE: %s\n", 395 dp->d_cdev->si_name); 396 disk_probe(dp, 0); 397 break; 398 case DISK_DISK_DESTROY: 399 dp = (struct disk *)msg->load; 400 disk_debug(1, 401 "DISK_DISK_DESTROY: %s\n", 402 dp->d_cdev->si_name); 403 devfs_destroy_subnames(dp->d_cdev->si_name); 404 devfs_destroy_dev(dp->d_cdev); 405 lwkt_gettoken(&disklist_token); 406 LIST_REMOVE(dp, d_list); 407 lwkt_reltoken(&disklist_token); 408 if (dp->d_info.d_serialno) { 409 kfree(dp->d_info.d_serialno, M_TEMP); 410 dp->d_info.d_serialno = NULL; 411 } 412 break; 413 case DISK_UNPROBE: 414 dp = (struct disk *)msg->load; 415 disk_debug(1, 416 "DISK_DISK_UNPROBE: %s\n", 417 dp->d_cdev->si_name); 418 devfs_destroy_subnames(dp->d_cdev->si_name); 419 break; 420 case DISK_SLICE_REPROBE: 421 dp = (struct disk *)msg->load; 422 sp = (struct diskslice *)msg->load2; 423 devfs_clr_subnames_flag(sp->ds_dev->si_name, 424 SI_REPROBE_TEST); 425 disk_debug(1, 426 "DISK_SLICE_REPROBE: %s\n", 427 sp->ds_dev->si_name); 428 disk_probe_slice(dp, sp->ds_dev, 429 dkslice(sp->ds_dev), 1); 430 devfs_destroy_subnames_without_flag( 431 sp->ds_dev->si_name, SI_REPROBE_TEST); 432 break; 433 case DISK_DISK_REPROBE: 434 dp = (struct disk *)msg->load; 435 devfs_clr_subnames_flag(dp->d_cdev->si_name, SI_REPROBE_TEST); 436 disk_debug(1, 437 "DISK_DISK_REPROBE: %s\n", 438 dp->d_cdev->si_name); 439 disk_probe(dp, 1); 440 devfs_destroy_subnames_without_flag( 441 dp->d_cdev->si_name, SI_REPROBE_TEST); 442 break; 443 case DISK_SYNC: 444 disk_debug(1, "DISK_SYNC\n"); 445 break; 446 default: 447 devfs_debug(DEVFS_DEBUG_WARNING, 448 "disk_msg_core: unknown message " 449 "received at core\n"); 450 break; 451 } 452 lwkt_replymsg(&msg->hdr, 0); 453 } 454 lwkt_exit(); 455 } 456 457 458 /* 459 * Acts as a message drain. Any message that is replied to here gets 460 * destroyed and the memory freed. 461 */ 462 static void 463 disk_msg_autofree_reply(lwkt_port_t port, lwkt_msg_t msg) 464 { 465 objcache_put(disk_msg_cache, msg); 466 } 467 468 469 void 470 disk_msg_send(uint32_t cmd, void *load, void *load2) 471 { 472 disk_msg_t disk_msg; 473 lwkt_port_t port = &disk_msg_port; 474 475 disk_msg = objcache_get(disk_msg_cache, M_WAITOK); 476 477 lwkt_initmsg(&disk_msg->hdr, &disk_dispose_port, 0); 478 479 disk_msg->hdr.u.ms_result = cmd; 480 disk_msg->load = load; 481 disk_msg->load2 = load2; 482 KKASSERT(port); 483 lwkt_sendmsg(port, &disk_msg->hdr); 484 } 485 486 void 487 disk_msg_send_sync(uint32_t cmd, void *load, void *load2) 488 { 489 struct lwkt_port rep_port; 490 disk_msg_t disk_msg; 491 lwkt_port_t port; 492 493 disk_msg = objcache_get(disk_msg_cache, M_WAITOK); 494 port = &disk_msg_port; 495 496 /* XXX could probably use curthread's built-in msgport */ 497 lwkt_initport_thread(&rep_port, curthread); 498 lwkt_initmsg(&disk_msg->hdr, &rep_port, 0); 499 500 disk_msg->hdr.u.ms_result = cmd; 501 disk_msg->load = load; 502 disk_msg->load2 = load2; 503 504 lwkt_sendmsg(port, &disk_msg->hdr); 505 lwkt_waitmsg(&disk_msg->hdr, 0); 506 objcache_put(disk_msg_cache, disk_msg); 507 } 508 509 /* 510 * Create a raw device for the dev_ops template (which is returned). Also 511 * create a slice and unit managed disk and overload the user visible 512 * device space with it. 513 * 514 * NOTE: The returned raw device is NOT a slice and unit managed device. 515 * It is an actual raw device representing the raw disk as specified by 516 * the passed dev_ops. The disk layer not only returns such a raw device, 517 * it also uses it internally when passing (modified) commands through. 518 */ 519 cdev_t 520 disk_create(int unit, struct disk *dp, struct dev_ops *raw_ops) 521 { 522 return disk_create_named(NULL, unit, dp, raw_ops); 523 } 524 525 cdev_t 526 disk_create_named(const char *name, int unit, struct disk *dp, struct dev_ops *raw_ops) 527 { 528 cdev_t rawdev; 529 530 if (name == NULL) 531 name = raw_ops->head.name; 532 533 disk_debug(1, "disk_create (begin): %s%d\n", name, unit); 534 535 rawdev = make_only_dev(raw_ops, dkmakewholedisk(unit), 536 UID_ROOT, GID_OPERATOR, 0640, 537 "%s%d", name, unit); 538 539 bzero(dp, sizeof(*dp)); 540 541 dp->d_rawdev = rawdev; 542 dp->d_raw_ops = raw_ops; 543 dp->d_dev_ops = &disk_ops; 544 dp->d_cdev = make_dev_covering(&disk_ops, dp->d_rawdev->si_ops, 545 dkmakewholedisk(unit), 546 UID_ROOT, GID_OPERATOR, 0640, 547 "%s%d", name, unit); 548 udev_dict_set_cstr(dp->d_cdev, "subsystem", "disk"); 549 dp->d_cdev->si_disk = dp; 550 551 dsched_disk_create_callback(dp, name, unit); 552 553 lwkt_gettoken(&disklist_token); 554 LIST_INSERT_HEAD(&disklist, dp, d_list); 555 lwkt_reltoken(&disklist_token); 556 557 disk_debug(1, "disk_create (end): %s%d\n", name, unit); 558 559 return (dp->d_rawdev); 560 } 561 562 int 563 disk_setdisktype(struct disk *disk, const char *type) 564 { 565 KKASSERT(disk != NULL); 566 567 disk->d_disktype = type; 568 return udev_dict_set_cstr(disk->d_cdev, "disk-type", __DECONST(char *, type)); 569 } 570 571 static void 572 _setdiskinfo(struct disk *disk, struct disk_info *info) 573 { 574 char *oldserialno; 575 576 oldserialno = disk->d_info.d_serialno; 577 bcopy(info, &disk->d_info, sizeof(disk->d_info)); 578 info = &disk->d_info; 579 580 disk_debug(1, 581 "_setdiskinfo: %s\n", 582 disk->d_cdev->si_name); 583 584 /* 585 * The serial number is duplicated so the caller can throw 586 * their copy away. 587 */ 588 if (info->d_serialno && info->d_serialno[0]) { 589 info->d_serialno = kstrdup(info->d_serialno, M_TEMP); 590 disk_cleanserial(info->d_serialno); 591 if (disk->d_cdev) { 592 make_dev_alias(disk->d_cdev, "serno/%s", 593 info->d_serialno); 594 } 595 } else { 596 info->d_serialno = NULL; 597 } 598 if (oldserialno) 599 kfree(oldserialno, M_TEMP); 600 601 dsched_disk_update_callback(disk, info); 602 603 /* 604 * The caller may set d_media_size or d_media_blocks and we 605 * calculate the other. 606 */ 607 KKASSERT(info->d_media_size == 0 || info->d_media_blksize == 0); 608 if (info->d_media_size == 0 && info->d_media_blocks) { 609 info->d_media_size = (u_int64_t)info->d_media_blocks * 610 info->d_media_blksize; 611 } else if (info->d_media_size && info->d_media_blocks == 0 && 612 info->d_media_blksize) { 613 info->d_media_blocks = info->d_media_size / 614 info->d_media_blksize; 615 } 616 617 /* 618 * The si_* fields for rawdev are not set until after the 619 * disk_create() call, so someone using the cooked version 620 * of the raw device (i.e. da0s0) will not get the right 621 * si_iosize_max unless we fix it up here. 622 */ 623 if (disk->d_cdev && disk->d_rawdev && 624 disk->d_cdev->si_iosize_max == 0) { 625 disk->d_cdev->si_iosize_max = disk->d_rawdev->si_iosize_max; 626 disk->d_cdev->si_bsize_phys = disk->d_rawdev->si_bsize_phys; 627 disk->d_cdev->si_bsize_best = disk->d_rawdev->si_bsize_best; 628 } 629 630 /* Add the serial number to the udev_dictionary */ 631 if (info->d_serialno) 632 udev_dict_set_cstr(disk->d_cdev, "serno", info->d_serialno); 633 } 634 635 /* 636 * Disk drivers must call this routine when media parameters are available 637 * or have changed. 638 */ 639 void 640 disk_setdiskinfo(struct disk *disk, struct disk_info *info) 641 { 642 _setdiskinfo(disk, info); 643 disk_msg_send(DISK_DISK_PROBE, disk, NULL); 644 disk_debug(1, 645 "disk_setdiskinfo: sent probe for %s\n", 646 disk->d_cdev->si_name); 647 } 648 649 void 650 disk_setdiskinfo_sync(struct disk *disk, struct disk_info *info) 651 { 652 _setdiskinfo(disk, info); 653 disk_msg_send_sync(DISK_DISK_PROBE, disk, NULL); 654 disk_debug(1, 655 "disk_setdiskinfo_sync: sent probe for %s\n", 656 disk->d_cdev->si_name); 657 } 658 659 /* 660 * This routine is called when an adapter detaches. The higher level 661 * managed disk device is destroyed while the lower level raw device is 662 * released. 663 */ 664 void 665 disk_destroy(struct disk *disk) 666 { 667 dsched_disk_destroy_callback(disk); 668 disk_msg_send_sync(DISK_DISK_DESTROY, disk, NULL); 669 return; 670 } 671 672 int 673 disk_dumpcheck(cdev_t dev, u_int64_t *size, u_int64_t *blkno, u_int32_t *secsize) 674 { 675 struct partinfo pinfo; 676 int error; 677 678 bzero(&pinfo, sizeof(pinfo)); 679 error = dev_dioctl(dev, DIOCGPART, (void *)&pinfo, 0, 680 proc0.p_ucred, NULL); 681 if (error) 682 return (error); 683 684 if (pinfo.media_blksize == 0) 685 return (ENXIO); 686 687 if (blkno) /* XXX: make sure this reserved stuff is right */ 688 *blkno = pinfo.reserved_blocks + 689 pinfo.media_offset / pinfo.media_blksize; 690 if (secsize) 691 *secsize = pinfo.media_blksize; 692 if (size) 693 *size = (pinfo.media_blocks - pinfo.reserved_blocks); 694 695 return (0); 696 } 697 698 int 699 disk_dumpconf(cdev_t dev, u_int onoff) 700 { 701 struct dumperinfo di; 702 u_int64_t size, blkno; 703 u_int32_t secsize; 704 int error; 705 706 if (!onoff) 707 return set_dumper(NULL); 708 709 error = disk_dumpcheck(dev, &size, &blkno, &secsize); 710 711 if (error) 712 return ENXIO; 713 714 bzero(&di, sizeof(struct dumperinfo)); 715 di.dumper = diskdump; 716 di.priv = dev; 717 di.blocksize = secsize; 718 di.mediaoffset = blkno * DEV_BSIZE; 719 di.mediasize = size * DEV_BSIZE; 720 721 return set_dumper(&di); 722 } 723 724 void 725 disk_unprobe(struct disk *disk) 726 { 727 if (disk == NULL) 728 return; 729 730 disk_msg_send_sync(DISK_UNPROBE, disk, NULL); 731 } 732 733 void 734 disk_invalidate (struct disk *disk) 735 { 736 dsgone(&disk->d_slice); 737 } 738 739 struct disk * 740 disk_enumerate(struct disk *disk) 741 { 742 struct disk *dp; 743 744 lwkt_gettoken(&disklist_token); 745 if (!disk) 746 dp = (LIST_FIRST(&disklist)); 747 else 748 dp = (LIST_NEXT(disk, d_list)); 749 lwkt_reltoken(&disklist_token); 750 751 return dp; 752 } 753 754 static 755 int 756 sysctl_disks(SYSCTL_HANDLER_ARGS) 757 { 758 struct disk *disk; 759 int error, first; 760 761 disk = NULL; 762 first = 1; 763 764 while ((disk = disk_enumerate(disk))) { 765 if (!first) { 766 error = SYSCTL_OUT(req, " ", 1); 767 if (error) 768 return error; 769 } else { 770 first = 0; 771 } 772 error = SYSCTL_OUT(req, disk->d_rawdev->si_name, 773 strlen(disk->d_rawdev->si_name)); 774 if (error) 775 return error; 776 } 777 error = SYSCTL_OUT(req, "", 1); 778 return error; 779 } 780 781 SYSCTL_PROC(_kern, OID_AUTO, disks, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0, 782 sysctl_disks, "A", "names of available disks"); 783 784 /* 785 * Open a disk device or partition. 786 */ 787 static 788 int 789 diskopen(struct dev_open_args *ap) 790 { 791 cdev_t dev = ap->a_head.a_dev; 792 struct disk *dp; 793 int error; 794 795 /* 796 * dp can't be NULL here XXX. 797 * 798 * d_slice will be NULL if setdiskinfo() has not been called yet. 799 * setdiskinfo() is typically called whether the disk is present 800 * or not (e.g. CD), but the base disk device is created first 801 * and there may be a race. 802 */ 803 dp = dev->si_disk; 804 if (dp == NULL || dp->d_slice == NULL) 805 return (ENXIO); 806 error = 0; 807 808 /* 809 * Deal with open races 810 */ 811 get_mplock(); 812 while (dp->d_flags & DISKFLAG_LOCK) { 813 dp->d_flags |= DISKFLAG_WANTED; 814 error = tsleep(dp, PCATCH, "diskopen", hz); 815 if (error) { 816 rel_mplock(); 817 return (error); 818 } 819 } 820 dp->d_flags |= DISKFLAG_LOCK; 821 822 /* 823 * Open the underlying raw device. 824 */ 825 if (!dsisopen(dp->d_slice)) { 826 #if 0 827 if (!pdev->si_iosize_max) 828 pdev->si_iosize_max = dev->si_iosize_max; 829 #endif 830 error = dev_dopen(dp->d_rawdev, ap->a_oflags, 831 ap->a_devtype, ap->a_cred); 832 } 833 #if 0 834 /* 835 * Inherit properties from the underlying device now that it is 836 * open. 837 */ 838 dev_dclone(dev); 839 #endif 840 841 if (error) 842 goto out; 843 error = dsopen(dev, ap->a_devtype, dp->d_info.d_dsflags, 844 &dp->d_slice, &dp->d_info); 845 if (!dsisopen(dp->d_slice)) { 846 dev_dclose(dp->d_rawdev, ap->a_oflags, ap->a_devtype); 847 } 848 out: 849 dp->d_flags &= ~DISKFLAG_LOCK; 850 if (dp->d_flags & DISKFLAG_WANTED) { 851 dp->d_flags &= ~DISKFLAG_WANTED; 852 wakeup(dp); 853 } 854 rel_mplock(); 855 856 return(error); 857 } 858 859 /* 860 * Close a disk device or partition 861 */ 862 static 863 int 864 diskclose(struct dev_close_args *ap) 865 { 866 cdev_t dev = ap->a_head.a_dev; 867 struct disk *dp; 868 int error; 869 870 error = 0; 871 dp = dev->si_disk; 872 873 get_mplock(); 874 dsclose(dev, ap->a_devtype, dp->d_slice); 875 if (!dsisopen(dp->d_slice)) { 876 error = dev_dclose(dp->d_rawdev, ap->a_fflag, ap->a_devtype); 877 } 878 rel_mplock(); 879 return (error); 880 } 881 882 /* 883 * First execute the ioctl on the disk device, and if it isn't supported 884 * try running it on the backing device. 885 */ 886 static 887 int 888 diskioctl(struct dev_ioctl_args *ap) 889 { 890 cdev_t dev = ap->a_head.a_dev; 891 struct disk *dp; 892 int error; 893 u_int u; 894 895 dp = dev->si_disk; 896 if (dp == NULL) 897 return (ENXIO); 898 899 devfs_debug(DEVFS_DEBUG_DEBUG, 900 "diskioctl: cmd is: %lx (name: %s)\n", 901 ap->a_cmd, dev->si_name); 902 devfs_debug(DEVFS_DEBUG_DEBUG, 903 "diskioctl: &dp->d_slice is: %p, %p\n", 904 &dp->d_slice, dp->d_slice); 905 906 if (ap->a_cmd == DIOCGKERNELDUMP) { 907 u = *(u_int *)ap->a_data; 908 return disk_dumpconf(dev, u); 909 } 910 911 if (&dp->d_slice == NULL || dp->d_slice == NULL) { 912 error = ENOIOCTL; 913 } else { 914 get_mplock(); 915 error = dsioctl(dev, ap->a_cmd, ap->a_data, ap->a_fflag, 916 &dp->d_slice, &dp->d_info); 917 rel_mplock(); 918 } 919 920 if (error == ENOIOCTL) { 921 error = dev_dioctl(dp->d_rawdev, ap->a_cmd, ap->a_data, 922 ap->a_fflag, ap->a_cred, NULL); 923 } 924 return (error); 925 } 926 927 /* 928 * Execute strategy routine 929 */ 930 static 931 int 932 diskstrategy(struct dev_strategy_args *ap) 933 { 934 cdev_t dev = ap->a_head.a_dev; 935 struct bio *bio = ap->a_bio; 936 struct bio *nbio; 937 struct disk *dp; 938 939 dp = dev->si_disk; 940 941 if (dp == NULL) { 942 bio->bio_buf->b_error = ENXIO; 943 bio->bio_buf->b_flags |= B_ERROR; 944 biodone(bio); 945 return(0); 946 } 947 KKASSERT(dev->si_disk == dp); 948 949 /* 950 * The dscheck() function will also transform the slice relative 951 * block number i.e. bio->bio_offset into a block number that can be 952 * passed directly to the underlying raw device. If dscheck() 953 * returns NULL it will have handled the bio for us (e.g. EOF 954 * or error due to being beyond the device size). 955 */ 956 if ((nbio = dscheck(dev, bio, dp->d_slice)) != NULL) { 957 dsched_queue(dp, nbio); 958 } else { 959 biodone(bio); 960 } 961 return(0); 962 } 963 964 /* 965 * Return the partition size in ?blocks? 966 */ 967 static 968 int 969 diskpsize(struct dev_psize_args *ap) 970 { 971 cdev_t dev = ap->a_head.a_dev; 972 struct disk *dp; 973 974 dp = dev->si_disk; 975 if (dp == NULL) 976 return(ENODEV); 977 ap->a_result = dssize(dev, &dp->d_slice); 978 return(0); 979 } 980 981 /* 982 * When new device entries are instantiated, make sure they inherit our 983 * si_disk structure and block and iosize limits from the raw device. 984 * 985 * This routine is always called synchronously in the context of the 986 * client. 987 * 988 * XXX The various io and block size constraints are not always initialized 989 * properly by devices. 990 */ 991 static 992 int 993 diskclone(struct dev_clone_args *ap) 994 { 995 cdev_t dev = ap->a_head.a_dev; 996 struct disk *dp; 997 dp = dev->si_disk; 998 999 KKASSERT(dp != NULL); 1000 dev->si_disk = dp; 1001 dev->si_iosize_max = dp->d_rawdev->si_iosize_max; 1002 dev->si_bsize_phys = dp->d_rawdev->si_bsize_phys; 1003 dev->si_bsize_best = dp->d_rawdev->si_bsize_best; 1004 return(0); 1005 } 1006 1007 int 1008 diskdump(struct dev_dump_args *ap) 1009 { 1010 cdev_t dev = ap->a_head.a_dev; 1011 struct disk *dp = dev->si_disk; 1012 u_int64_t size, offset; 1013 int error; 1014 1015 error = disk_dumpcheck(dev, &size, &ap->a_blkno, &ap->a_secsize); 1016 /* XXX: this should probably go in disk_dumpcheck somehow */ 1017 if (ap->a_length != 0) { 1018 size *= DEV_BSIZE; 1019 offset = ap->a_blkno * DEV_BSIZE; 1020 if ((ap->a_offset < offset) || 1021 (ap->a_offset + ap->a_length - offset > size)) { 1022 kprintf("Attempt to write outside dump device boundaries.\n"); 1023 error = ENOSPC; 1024 } 1025 } 1026 1027 if (error == 0) { 1028 ap->a_head.a_dev = dp->d_rawdev; 1029 error = dev_doperate(&ap->a_head); 1030 } 1031 1032 return(error); 1033 } 1034 1035 1036 SYSCTL_INT(_debug_sizeof, OID_AUTO, diskslices, CTLFLAG_RD, 1037 0, sizeof(struct diskslices), "sizeof(struct diskslices)"); 1038 1039 SYSCTL_INT(_debug_sizeof, OID_AUTO, disk, CTLFLAG_RD, 1040 0, sizeof(struct disk), "sizeof(struct disk)"); 1041 1042 /* 1043 * Reorder interval for burst write allowance and minor write 1044 * allowance. 1045 * 1046 * We always want to trickle some writes in to make use of the 1047 * disk's zone cache. Bursting occurs on a longer interval and only 1048 * runningbufspace is well over the hirunningspace limit. 1049 */ 1050 int bioq_reorder_burst_interval = 60; /* should be multiple of minor */ 1051 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_interval, 1052 CTLFLAG_RW, &bioq_reorder_burst_interval, 0, ""); 1053 int bioq_reorder_minor_interval = 5; 1054 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_interval, 1055 CTLFLAG_RW, &bioq_reorder_minor_interval, 0, ""); 1056 1057 int bioq_reorder_burst_bytes = 3000000; 1058 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_burst_bytes, 1059 CTLFLAG_RW, &bioq_reorder_burst_bytes, 0, ""); 1060 int bioq_reorder_minor_bytes = 262144; 1061 SYSCTL_INT(_kern, OID_AUTO, bioq_reorder_minor_bytes, 1062 CTLFLAG_RW, &bioq_reorder_minor_bytes, 0, ""); 1063 1064 1065 /* 1066 * Order I/Os. Generally speaking this code is designed to make better 1067 * use of drive zone caches. A drive zone cache can typically track linear 1068 * reads or writes for around 16 zones simultaniously. 1069 * 1070 * Read prioritization issues: It is possible for hundreds of megabytes worth 1071 * of writes to be queued asynchronously. This creates a huge bottleneck 1072 * for reads which reduce read bandwidth to a trickle. 1073 * 1074 * To solve this problem we generally reorder reads before writes. 1075 * 1076 * However, a large number of random reads can also starve writes and 1077 * make poor use of the drive zone cache so we allow writes to trickle 1078 * in every N reads. 1079 */ 1080 void 1081 bioqdisksort(struct bio_queue_head *bioq, struct bio *bio) 1082 { 1083 /* 1084 * The BIO wants to be ordered. Adding to the tail also 1085 * causes transition to be set to NULL, forcing the ordering 1086 * of all prior I/O's. 1087 */ 1088 if (bio->bio_buf->b_flags & B_ORDERED) { 1089 bioq_insert_tail(bioq, bio); 1090 return; 1091 } 1092 1093 switch(bio->bio_buf->b_cmd) { 1094 case BUF_CMD_READ: 1095 if (bioq->transition) { 1096 /* 1097 * Insert before the first write. Bleedover writes 1098 * based on reorder intervals to prevent starvation. 1099 */ 1100 TAILQ_INSERT_BEFORE(bioq->transition, bio, bio_act); 1101 ++bioq->reorder; 1102 if (bioq->reorder % bioq_reorder_minor_interval == 0) { 1103 bioqwritereorder(bioq); 1104 if (bioq->reorder >= 1105 bioq_reorder_burst_interval) { 1106 bioq->reorder = 0; 1107 } 1108 } 1109 } else { 1110 /* 1111 * No writes queued (or ordering was forced), 1112 * insert at tail. 1113 */ 1114 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act); 1115 } 1116 break; 1117 case BUF_CMD_WRITE: 1118 /* 1119 * Writes are always appended. If no writes were previously 1120 * queued or an ordered tail insertion occured the transition 1121 * field will be NULL. 1122 */ 1123 TAILQ_INSERT_TAIL(&bioq->queue, bio, bio_act); 1124 if (bioq->transition == NULL) 1125 bioq->transition = bio; 1126 break; 1127 default: 1128 /* 1129 * All other request types are forced to be ordered. 1130 */ 1131 bioq_insert_tail(bioq, bio); 1132 break; 1133 } 1134 } 1135 1136 /* 1137 * Move the read-write transition point to prevent reads from 1138 * completely starving our writes. This brings a number of writes into 1139 * the fold every N reads. 1140 * 1141 * We bring a few linear writes into the fold on a minor interval 1142 * and we bring a non-linear burst of writes into the fold on a major 1143 * interval. Bursting only occurs if runningbufspace is really high 1144 * (typically from syncs, fsyncs, or HAMMER flushes). 1145 */ 1146 static 1147 void 1148 bioqwritereorder(struct bio_queue_head *bioq) 1149 { 1150 struct bio *bio; 1151 off_t next_offset; 1152 size_t left; 1153 size_t n; 1154 int check_off; 1155 1156 if (bioq->reorder < bioq_reorder_burst_interval || 1157 !buf_runningbufspace_severe()) { 1158 left = (size_t)bioq_reorder_minor_bytes; 1159 check_off = 1; 1160 } else { 1161 left = (size_t)bioq_reorder_burst_bytes; 1162 check_off = 0; 1163 } 1164 1165 next_offset = bioq->transition->bio_offset; 1166 while ((bio = bioq->transition) != NULL && 1167 (check_off == 0 || next_offset == bio->bio_offset) 1168 ) { 1169 n = bio->bio_buf->b_bcount; 1170 next_offset = bio->bio_offset + n; 1171 bioq->transition = TAILQ_NEXT(bio, bio_act); 1172 if (left < n) 1173 break; 1174 left -= n; 1175 } 1176 } 1177 1178 /* 1179 * Bounds checking against the media size, used for the raw partition. 1180 * secsize, mediasize and b_blkno must all be the same units. 1181 * Possibly this has to be DEV_BSIZE (512). 1182 */ 1183 int 1184 bounds_check_with_mediasize(struct bio *bio, int secsize, uint64_t mediasize) 1185 { 1186 struct buf *bp = bio->bio_buf; 1187 int64_t sz; 1188 1189 sz = howmany(bp->b_bcount, secsize); 1190 1191 if (bio->bio_offset/DEV_BSIZE + sz > mediasize) { 1192 sz = mediasize - bio->bio_offset/DEV_BSIZE; 1193 if (sz == 0) { 1194 /* If exactly at end of disk, return EOF. */ 1195 bp->b_resid = bp->b_bcount; 1196 return 0; 1197 } 1198 if (sz < 0) { 1199 /* If past end of disk, return EINVAL. */ 1200 bp->b_error = EINVAL; 1201 return 0; 1202 } 1203 /* Otherwise, truncate request. */ 1204 bp->b_bcount = sz * secsize; 1205 } 1206 1207 return 1; 1208 } 1209 1210 /* 1211 * Disk error is the preface to plaintive error messages 1212 * about failing disk transfers. It prints messages of the form 1213 1214 hp0g: hard error reading fsbn 12345 of 12344-12347 (hp0 bn %d cn %d tn %d sn %d) 1215 1216 * if the offset of the error in the transfer and a disk label 1217 * are both available. blkdone should be -1 if the position of the error 1218 * is unknown; the disklabel pointer may be null from drivers that have not 1219 * been converted to use them. The message is printed with kprintf 1220 * if pri is LOG_PRINTF, otherwise it uses log at the specified priority. 1221 * The message should be completed (with at least a newline) with kprintf 1222 * or log(-1, ...), respectively. There is no trailing space. 1223 */ 1224 void 1225 diskerr(struct bio *bio, cdev_t dev, const char *what, int pri, int donecnt) 1226 { 1227 struct buf *bp = bio->bio_buf; 1228 const char *term; 1229 1230 switch(bp->b_cmd) { 1231 case BUF_CMD_READ: 1232 term = "read"; 1233 break; 1234 case BUF_CMD_WRITE: 1235 term = "write"; 1236 break; 1237 default: 1238 term = "access"; 1239 break; 1240 } 1241 kprintf("%s: %s %sing ", dev->si_name, what, term); 1242 kprintf("offset %012llx for %d", 1243 (long long)bio->bio_offset, 1244 bp->b_bcount); 1245 1246 if (donecnt) 1247 kprintf(" (%d bytes completed)", donecnt); 1248 } 1249 1250 /* 1251 * Locate a disk device 1252 */ 1253 cdev_t 1254 disk_locate(const char *devname) 1255 { 1256 return devfs_find_device_by_name(devname); 1257 } 1258 1259 void 1260 disk_config(void *arg) 1261 { 1262 disk_msg_send_sync(DISK_SYNC, NULL, NULL); 1263 } 1264 1265 static void 1266 disk_init(void) 1267 { 1268 struct thread* td_core; 1269 1270 disk_msg_cache = objcache_create("disk-msg-cache", 0, 0, 1271 NULL, NULL, NULL, 1272 objcache_malloc_alloc, 1273 objcache_malloc_free, 1274 &disk_msg_malloc_args); 1275 1276 lwkt_token_init(&disklist_token, 1, "disks"); 1277 1278 /* 1279 * Initialize the reply-only port which acts as a message drain 1280 */ 1281 lwkt_initport_replyonly(&disk_dispose_port, disk_msg_autofree_reply); 1282 1283 lwkt_gettoken(&disklist_token); 1284 lwkt_create(disk_msg_core, /*args*/NULL, &td_core, NULL, 1285 0, 0, "disk_msg_core"); 1286 tsleep(td_core, 0, "diskcore", 0); 1287 lwkt_reltoken(&disklist_token); 1288 } 1289 1290 static void 1291 disk_uninit(void) 1292 { 1293 objcache_destroy(disk_msg_cache); 1294 } 1295 1296 /* 1297 * Clean out illegal characters in serial numbers. 1298 */ 1299 static void 1300 disk_cleanserial(char *serno) 1301 { 1302 char c; 1303 1304 while ((c = *serno) != 0) { 1305 if (c >= 'a' && c <= 'z') 1306 ; 1307 else if (c >= 'A' && c <= 'Z') 1308 ; 1309 else if (c >= '0' && c <= '9') 1310 ; 1311 else if (c == '-' || c == '@' || c == '+' || c == '.') 1312 ; 1313 else 1314 c = '_'; 1315 *serno++= c; 1316 } 1317 } 1318 1319 TUNABLE_INT("kern.disk_debug", &disk_debug_enable); 1320 SYSCTL_INT(_kern, OID_AUTO, disk_debug, CTLFLAG_RW, &disk_debug_enable, 1321 0, "Enable subr_disk debugging"); 1322 1323 SYSINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_FIRST, disk_init, NULL); 1324 SYSUNINIT(disk_register, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, disk_uninit, NULL); 1325