1/*- 2 * Copyright (c) 2002 Poul-Henning Kamp 3 * Copyright (c) 2002 Networks Associates Technology, Inc. 4 * All rights reserved. 5 * 6 * This software was developed for the FreeBSD Project by Poul-Henning Kamp 7 * and NAI Labs, the Security Research Division of Network Associates, Inc. 8 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 9 * DARPA CHATS research program. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. The names of the authors may not be used to endorse or promote 20 * products derived from this software without specific prior written 21 * permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 36#include <sys/cdefs.h> 37__FBSDID("$FreeBSD: stable/11/sys/geom/geom_disk.c 356589 2020-01-10 00:55:37Z mav $"); 38 39#include "opt_geom.h" 40 41#include <sys/param.h> 42#include <sys/systm.h> 43#include <sys/kernel.h> 44#include <sys/sysctl.h> 45#include <sys/bio.h> 46#include <sys/bus.h> 47#include <sys/ctype.h> 48#include <sys/fcntl.h> 49#include <sys/malloc.h> 50#include <sys/sbuf.h> 51#include <sys/devicestat.h> 52#include <machine/md_var.h> 53 54#include <sys/lock.h> 55#include <sys/mutex.h> 56#include <geom/geom.h> 57#include <geom/geom_disk.h> 58#include <geom/geom_int.h> 59 60#include <dev/led/led.h> 61 62#include <machine/bus.h> 63 64struct g_disk_softc { 65 struct mtx done_mtx; 66 struct disk *dp; 67 struct devstat *d_devstat; 68 struct sysctl_ctx_list sysctl_ctx; 69 struct sysctl_oid *sysctl_tree; 70 char led[64]; 71 uint32_t state; 72 struct mtx start_mtx; 73}; 74 75static g_access_t g_disk_access; 76static g_start_t g_disk_start; 77static g_ioctl_t g_disk_ioctl; 78static g_dumpconf_t g_disk_dumpconf; 79static g_provgone_t g_disk_providergone; 80 81static struct g_class g_disk_class = { 82 .name = G_DISK_CLASS_NAME, 83 .version = G_VERSION, 84 .start = g_disk_start, 85 .access = g_disk_access, 86 .ioctl = g_disk_ioctl, 87 .providergone = g_disk_providergone, 88 .dumpconf = g_disk_dumpconf, 89}; 90 91SYSCTL_DECL(_kern_geom); 92static SYSCTL_NODE(_kern_geom, OID_AUTO, disk, CTLFLAG_RW, 0, 93 "GEOM_DISK stuff"); 94 95DECLARE_GEOM_CLASS(g_disk_class, g_disk); 96 97static int 98g_disk_access(struct g_provider *pp, int r, int w, int e) 99{ 100 struct disk *dp; 101 struct g_disk_softc *sc; 102 int error; 103 104 g_trace(G_T_ACCESS, "g_disk_access(%s, %d, %d, %d)", 105 pp->name, r, w, e); 106 g_topology_assert(); 107 sc = pp->private; 108 if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) { 109 /* 110 * Allow decreasing access count even if disk is not 111 * available anymore. 112 */ 113 if (r <= 0 && w <= 0 && e <= 0) 114 return (0); 115 return (ENXIO); 116 } 117 r += pp->acr; 118 w += pp->acw; 119 e += pp->ace; 120 error = 0; 121 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 122 /* 123 * It would be better to defer this decision to d_open if 124 * it was able to take flags. 125 */ 126 if (w > 0 && (dp->d_flags & DISKFLAG_WRITE_PROTECT) != 0) 127 error = EROFS; 128 if (error == 0 && dp->d_open != NULL) 129 error = dp->d_open(dp); 130 if (bootverbose && error != 0) 131 printf("Opened disk %s -> %d\n", pp->name, error); 132 if (error != 0) 133 return (error); 134 pp->sectorsize = dp->d_sectorsize; 135 if (dp->d_maxsize == 0) { 136 printf("WARNING: Disk drive %s%d has no d_maxsize\n", 137 dp->d_name, dp->d_unit); 138 dp->d_maxsize = DFLTPHYS; 139 } 140 if (dp->d_delmaxsize == 0) { 141 if (bootverbose && dp->d_flags & DISKFLAG_CANDELETE) { 142 printf("WARNING: Disk drive %s%d has no " 143 "d_delmaxsize\n", dp->d_name, dp->d_unit); 144 } 145 dp->d_delmaxsize = dp->d_maxsize; 146 } 147 pp->stripeoffset = dp->d_stripeoffset; 148 pp->stripesize = dp->d_stripesize; 149 dp->d_flags |= DISKFLAG_OPEN; 150 /* 151 * Do not invoke resize event when initial size was zero. 152 * Some disks report its size only after first opening. 153 */ 154 if (pp->mediasize == 0) 155 pp->mediasize = dp->d_mediasize; 156 else 157 g_resize_provider(pp, dp->d_mediasize); 158 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 159 if (dp->d_close != NULL) { 160 error = dp->d_close(dp); 161 if (error != 0) 162 printf("Closed disk %s -> %d\n", 163 pp->name, error); 164 } 165 sc->state = G_STATE_ACTIVE; 166 if (sc->led[0] != 0) 167 led_set(sc->led, "0"); 168 dp->d_flags &= ~DISKFLAG_OPEN; 169 } 170 return (error); 171} 172 173static void 174g_disk_kerneldump(struct bio *bp, struct disk *dp) 175{ 176 struct g_kerneldump *gkd; 177 struct g_geom *gp; 178 179 gkd = (struct g_kerneldump*)bp->bio_data; 180 gp = bp->bio_to->geom; 181 g_trace(G_T_TOPOLOGY, "g_disk_kerneldump(%s, %jd, %jd)", 182 gp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); 183 if (dp->d_dump == NULL) { 184 g_io_deliver(bp, ENODEV); 185 return; 186 } 187 gkd->di.dumper = dp->d_dump; 188 gkd->di.priv = dp; 189 gkd->di.blocksize = dp->d_sectorsize; 190 gkd->di.maxiosize = dp->d_maxsize; 191 gkd->di.mediaoffset = gkd->offset; 192 if ((gkd->offset + gkd->length) > dp->d_mediasize) 193 gkd->length = dp->d_mediasize - gkd->offset; 194 gkd->di.mediasize = gkd->length; 195 g_io_deliver(bp, 0); 196} 197 198static void 199g_disk_setstate(struct bio *bp, struct g_disk_softc *sc) 200{ 201 const char *cmd; 202 203 memcpy(&sc->state, bp->bio_data, sizeof(sc->state)); 204 if (sc->led[0] != 0) { 205 switch (sc->state) { 206 case G_STATE_FAILED: 207 cmd = "1"; 208 break; 209 case G_STATE_REBUILD: 210 cmd = "f5"; 211 break; 212 case G_STATE_RESYNC: 213 cmd = "f1"; 214 break; 215 default: 216 cmd = "0"; 217 break; 218 } 219 led_set(sc->led, cmd); 220 } 221 g_io_deliver(bp, 0); 222} 223 224static void 225g_disk_done(struct bio *bp) 226{ 227 struct bintime now; 228 struct bio *bp2; 229 struct g_disk_softc *sc; 230 231 /* See "notes" for why we need a mutex here */ 232 sc = bp->bio_caller1; 233 bp2 = bp->bio_parent; 234 binuptime(&now); 235 mtx_lock(&sc->done_mtx); 236 if (bp2->bio_error == 0) 237 bp2->bio_error = bp->bio_error; 238 bp2->bio_completed += bp->bio_length - bp->bio_resid; 239 240 switch (bp->bio_cmd) { 241 case BIO_ZONE: 242 bcopy(&bp->bio_zone, &bp2->bio_zone, sizeof(bp->bio_zone)); 243 /*FALLTHROUGH*/ 244 case BIO_READ: 245 case BIO_WRITE: 246 case BIO_DELETE: 247 case BIO_FLUSH: 248 devstat_end_transaction_bio_bt(sc->d_devstat, bp, &now); 249 break; 250 default: 251 break; 252 } 253 bp2->bio_inbed++; 254 if (bp2->bio_children == bp2->bio_inbed) { 255 mtx_unlock(&sc->done_mtx); 256 bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed; 257 g_io_deliver(bp2, bp2->bio_error); 258 } else 259 mtx_unlock(&sc->done_mtx); 260 g_destroy_bio(bp); 261} 262 263static int 264g_disk_ioctl(struct g_provider *pp, u_long cmd, void * data, int fflag, struct thread *td) 265{ 266 struct disk *dp; 267 struct g_disk_softc *sc; 268 int error; 269 270 sc = pp->private; 271 dp = sc->dp; 272 273 if (dp->d_ioctl == NULL) 274 return (ENOIOCTL); 275 error = dp->d_ioctl(dp, cmd, data, fflag, td); 276 return (error); 277} 278 279static off_t 280g_disk_maxsize(struct disk *dp, struct bio *bp) 281{ 282 if (bp->bio_cmd == BIO_DELETE) 283 return (dp->d_delmaxsize); 284 return (dp->d_maxsize); 285} 286 287static int 288g_disk_maxsegs(struct disk *dp, struct bio *bp) 289{ 290 return ((g_disk_maxsize(dp, bp) / PAGE_SIZE) + 1); 291} 292 293static void 294g_disk_advance(struct disk *dp, struct bio *bp, off_t off) 295{ 296 297 bp->bio_offset += off; 298 bp->bio_length -= off; 299 300 if ((bp->bio_flags & BIO_VLIST) != 0) { 301 bus_dma_segment_t *seg, *end; 302 303 seg = (bus_dma_segment_t *)bp->bio_data; 304 end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n; 305 off += bp->bio_ma_offset; 306 while (off >= seg->ds_len) { 307 KASSERT((seg != end), 308 ("vlist request runs off the end")); 309 off -= seg->ds_len; 310 seg++; 311 } 312 bp->bio_ma_offset = off; 313 bp->bio_ma_n = end - seg; 314 bp->bio_data = (void *)seg; 315 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 316 bp->bio_ma += off / PAGE_SIZE; 317 bp->bio_ma_offset += off; 318 bp->bio_ma_offset %= PAGE_SIZE; 319 bp->bio_ma_n -= off / PAGE_SIZE; 320 } else { 321 bp->bio_data += off; 322 } 323} 324 325static void 326g_disk_seg_limit(bus_dma_segment_t *seg, off_t *poffset, 327 off_t *plength, int *ppages) 328{ 329 uintptr_t seg_page_base; 330 uintptr_t seg_page_end; 331 off_t offset; 332 off_t length; 333 int seg_pages; 334 335 offset = *poffset; 336 length = *plength; 337 338 if (length > seg->ds_len - offset) 339 length = seg->ds_len - offset; 340 341 seg_page_base = trunc_page(seg->ds_addr + offset); 342 seg_page_end = round_page(seg->ds_addr + offset + length); 343 seg_pages = (seg_page_end - seg_page_base) >> PAGE_SHIFT; 344 345 if (seg_pages > *ppages) { 346 seg_pages = *ppages; 347 length = (seg_page_base + (seg_pages << PAGE_SHIFT)) - 348 (seg->ds_addr + offset); 349 } 350 351 *poffset = 0; 352 *plength -= length; 353 *ppages -= seg_pages; 354} 355 356static off_t 357g_disk_vlist_limit(struct disk *dp, struct bio *bp, bus_dma_segment_t **pendseg) 358{ 359 bus_dma_segment_t *seg, *end; 360 off_t residual; 361 off_t offset; 362 int pages; 363 364 seg = (bus_dma_segment_t *)bp->bio_data; 365 end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n; 366 residual = bp->bio_length; 367 offset = bp->bio_ma_offset; 368 pages = g_disk_maxsegs(dp, bp); 369 while (residual != 0 && pages != 0) { 370 KASSERT((seg != end), 371 ("vlist limit runs off the end")); 372 g_disk_seg_limit(seg, &offset, &residual, &pages); 373 seg++; 374 } 375 if (pendseg != NULL) 376 *pendseg = seg; 377 return (residual); 378} 379 380static bool 381g_disk_limit(struct disk *dp, struct bio *bp) 382{ 383 bool limited = false; 384 off_t maxsz; 385 386 maxsz = g_disk_maxsize(dp, bp); 387 388 /* 389 * XXX: If we have a stripesize we should really use it here. 390 * Care should be taken in the delete case if this is done 391 * as deletes can be very sensitive to size given how they 392 * are processed. 393 */ 394 if (bp->bio_length > maxsz) { 395 bp->bio_length = maxsz; 396 limited = true; 397 } 398 399 if ((bp->bio_flags & BIO_VLIST) != 0) { 400 bus_dma_segment_t *firstseg, *endseg; 401 off_t residual; 402 403 firstseg = (bus_dma_segment_t*)bp->bio_data; 404 residual = g_disk_vlist_limit(dp, bp, &endseg); 405 if (residual != 0) { 406 bp->bio_ma_n = endseg - firstseg; 407 bp->bio_length -= residual; 408 limited = true; 409 } 410 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 411 bp->bio_ma_n = 412 howmany(bp->bio_ma_offset + bp->bio_length, PAGE_SIZE); 413 } 414 415 return (limited); 416} 417 418static void 419g_disk_start(struct bio *bp) 420{ 421 struct bio *bp2, *bp3; 422 struct disk *dp; 423 struct g_disk_softc *sc; 424 int error; 425 off_t off; 426 427 sc = bp->bio_to->private; 428 if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) { 429 g_io_deliver(bp, ENXIO); 430 return; 431 } 432 error = EJUSTRETURN; 433 switch(bp->bio_cmd) { 434 case BIO_DELETE: 435 if (!(dp->d_flags & DISKFLAG_CANDELETE)) { 436 error = EOPNOTSUPP; 437 break; 438 } 439 /* fall-through */ 440 case BIO_READ: 441 case BIO_WRITE: 442 KASSERT((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0 || 443 (bp->bio_flags & BIO_UNMAPPED) == 0, 444 ("unmapped bio not supported by disk %s", dp->d_name)); 445 off = 0; 446 bp3 = NULL; 447 bp2 = g_clone_bio(bp); 448 if (bp2 == NULL) { 449 error = ENOMEM; 450 break; 451 } 452 for (;;) { 453 if (g_disk_limit(dp, bp2)) { 454 off += bp2->bio_length; 455 456 /* 457 * To avoid a race, we need to grab the next bio 458 * before we schedule this one. See "notes". 459 */ 460 bp3 = g_clone_bio(bp); 461 if (bp3 == NULL) 462 bp->bio_error = ENOMEM; 463 } 464 bp2->bio_done = g_disk_done; 465 bp2->bio_caller1 = sc; 466 bp2->bio_pblkno = bp2->bio_offset / dp->d_sectorsize; 467 bp2->bio_bcount = bp2->bio_length; 468 bp2->bio_disk = dp; 469 mtx_lock(&sc->start_mtx); 470 devstat_start_transaction_bio(dp->d_devstat, bp2); 471 mtx_unlock(&sc->start_mtx); 472 dp->d_strategy(bp2); 473 474 if (bp3 == NULL) 475 break; 476 477 bp2 = bp3; 478 bp3 = NULL; 479 g_disk_advance(dp, bp2, off); 480 } 481 break; 482 case BIO_GETATTR: 483 /* Give the driver a chance to override */ 484 if (dp->d_getattr != NULL) { 485 if (bp->bio_disk == NULL) 486 bp->bio_disk = dp; 487 error = dp->d_getattr(bp); 488 if (error != -1) 489 break; 490 error = EJUSTRETURN; 491 } 492 if (g_handleattr_int(bp, "GEOM::candelete", 493 (dp->d_flags & DISKFLAG_CANDELETE) != 0)) 494 break; 495 else if (g_handleattr_int(bp, "GEOM::fwsectors", 496 dp->d_fwsectors)) 497 break; 498 else if (g_handleattr_int(bp, "GEOM::fwheads", dp->d_fwheads)) 499 break; 500 else if (g_handleattr_off_t(bp, "GEOM::frontstuff", 0)) 501 break; 502 else if (g_handleattr_str(bp, "GEOM::ident", dp->d_ident)) 503 break; 504 else if (g_handleattr_str(bp, "GEOM::descr", dp->d_descr)) 505 break; 506 else if (g_handleattr_uint16_t(bp, "GEOM::hba_vendor", 507 dp->d_hba_vendor)) 508 break; 509 else if (g_handleattr_uint16_t(bp, "GEOM::hba_device", 510 dp->d_hba_device)) 511 break; 512 else if (g_handleattr_uint16_t(bp, "GEOM::hba_subvendor", 513 dp->d_hba_subvendor)) 514 break; 515 else if (g_handleattr_uint16_t(bp, "GEOM::hba_subdevice", 516 dp->d_hba_subdevice)) 517 break; 518 else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) 519 g_disk_kerneldump(bp, dp); 520 else if (!strcmp(bp->bio_attribute, "GEOM::setstate")) 521 g_disk_setstate(bp, sc); 522 else if (g_handleattr_uint16_t(bp, "GEOM::rotation_rate", 523 dp->d_rotation_rate)) 524 break; 525 else 526 error = ENOIOCTL; 527 break; 528 case BIO_FLUSH: 529 g_trace(G_T_BIO, "g_disk_flushcache(%s)", 530 bp->bio_to->name); 531 if (!(dp->d_flags & DISKFLAG_CANFLUSHCACHE)) { 532 error = EOPNOTSUPP; 533 break; 534 } 535 /*FALLTHROUGH*/ 536 case BIO_ZONE: 537 if (bp->bio_cmd == BIO_ZONE) { 538 if (!(dp->d_flags & DISKFLAG_CANZONE)) { 539 error = EOPNOTSUPP; 540 break; 541 } 542 g_trace(G_T_BIO, "g_disk_zone(%s)", 543 bp->bio_to->name); 544 } 545 bp2 = g_clone_bio(bp); 546 if (bp2 == NULL) { 547 g_io_deliver(bp, ENOMEM); 548 return; 549 } 550 bp2->bio_done = g_disk_done; 551 bp2->bio_caller1 = sc; 552 bp2->bio_disk = dp; 553 mtx_lock(&sc->start_mtx); 554 devstat_start_transaction_bio(dp->d_devstat, bp2); 555 mtx_unlock(&sc->start_mtx); 556 dp->d_strategy(bp2); 557 break; 558 default: 559 error = EOPNOTSUPP; 560 break; 561 } 562 if (error != EJUSTRETURN) 563 g_io_deliver(bp, error); 564 return; 565} 566 567static void 568g_disk_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp) 569{ 570 struct bio *bp; 571 struct disk *dp; 572 struct g_disk_softc *sc; 573 char *buf; 574 int res = 0; 575 576 sc = gp->softc; 577 if (sc == NULL || (dp = sc->dp) == NULL) 578 return; 579 if (indent == NULL) { 580 sbuf_printf(sb, " hd %u", dp->d_fwheads); 581 sbuf_printf(sb, " sc %u", dp->d_fwsectors); 582 return; 583 } 584 if (pp != NULL) { 585 sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n", 586 indent, dp->d_fwheads); 587 sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n", 588 indent, dp->d_fwsectors); 589 590 /* 591 * "rotationrate" is a little complicated, because the value 592 * returned by the drive might not be the RPM; 0 and 1 are 593 * special cases, and there's also a valid range. 594 */ 595 sbuf_printf(sb, "%s<rotationrate>", indent); 596 if (dp->d_rotation_rate == DISK_RR_UNKNOWN) /* Old drives */ 597 sbuf_printf(sb, "unknown"); /* don't report RPM. */ 598 else if (dp->d_rotation_rate == DISK_RR_NON_ROTATING) 599 sbuf_printf(sb, "0"); 600 else if ((dp->d_rotation_rate >= DISK_RR_MIN) && 601 (dp->d_rotation_rate <= DISK_RR_MAX)) 602 sbuf_printf(sb, "%u", dp->d_rotation_rate); 603 else 604 sbuf_printf(sb, "invalid"); 605 sbuf_printf(sb, "</rotationrate>\n"); 606 if (dp->d_getattr != NULL) { 607 buf = g_malloc(DISK_IDENT_SIZE, M_WAITOK); 608 bp = g_alloc_bio(); 609 bp->bio_disk = dp; 610 bp->bio_attribute = "GEOM::ident"; 611 bp->bio_length = DISK_IDENT_SIZE; 612 bp->bio_data = buf; 613 res = dp->d_getattr(bp); 614 sbuf_printf(sb, "%s<ident>", indent); 615 g_conf_printf_escaped(sb, "%s", 616 res == 0 ? buf: dp->d_ident); 617 sbuf_printf(sb, "</ident>\n"); 618 bp->bio_attribute = "GEOM::lunid"; 619 bp->bio_length = DISK_IDENT_SIZE; 620 bp->bio_data = buf; 621 if (dp->d_getattr(bp) == 0) { 622 sbuf_printf(sb, "%s<lunid>", indent); 623 g_conf_printf_escaped(sb, "%s", buf); 624 sbuf_printf(sb, "</lunid>\n"); 625 } 626 bp->bio_attribute = "GEOM::lunname"; 627 bp->bio_length = DISK_IDENT_SIZE; 628 bp->bio_data = buf; 629 if (dp->d_getattr(bp) == 0) { 630 sbuf_printf(sb, "%s<lunname>", indent); 631 g_conf_printf_escaped(sb, "%s", buf); 632 sbuf_printf(sb, "</lunname>\n"); 633 } 634 g_destroy_bio(bp); 635 g_free(buf); 636 } else { 637 sbuf_printf(sb, "%s<ident>", indent); 638 g_conf_printf_escaped(sb, "%s", dp->d_ident); 639 sbuf_printf(sb, "</ident>\n"); 640 } 641 sbuf_printf(sb, "%s<descr>", indent); 642 g_conf_printf_escaped(sb, "%s", dp->d_descr); 643 sbuf_printf(sb, "</descr>\n"); 644 } 645} 646 647static void 648g_disk_resize(void *ptr, int flag) 649{ 650 struct disk *dp; 651 struct g_geom *gp; 652 struct g_provider *pp; 653 654 if (flag == EV_CANCEL) 655 return; 656 g_topology_assert(); 657 658 dp = ptr; 659 gp = dp->d_geom; 660 661 if (dp->d_destroyed || gp == NULL) 662 return; 663 664 LIST_FOREACH(pp, &gp->provider, provider) { 665 if (pp->sectorsize != 0 && 666 pp->sectorsize != dp->d_sectorsize) 667 g_wither_provider(pp, ENXIO); 668 else 669 g_resize_provider(pp, dp->d_mediasize); 670 } 671} 672 673static void 674g_disk_create(void *arg, int flag) 675{ 676 struct g_geom *gp; 677 struct g_provider *pp; 678 struct disk *dp; 679 struct g_disk_softc *sc; 680 char tmpstr[80]; 681 682 if (flag == EV_CANCEL) 683 return; 684 g_topology_assert(); 685 dp = arg; 686 687 mtx_pool_lock(mtxpool_sleep, dp); 688 dp->d_init_level = DISK_INIT_START; 689 690 /* 691 * If the disk has already gone away, we can just stop here and 692 * call the user's callback to tell him we've cleaned things up. 693 */ 694 if (dp->d_goneflag != 0) { 695 mtx_pool_unlock(mtxpool_sleep, dp); 696 if (dp->d_gone != NULL) 697 dp->d_gone(dp); 698 return; 699 } 700 mtx_pool_unlock(mtxpool_sleep, dp); 701 702 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO); 703 mtx_init(&sc->start_mtx, "g_disk_start", NULL, MTX_DEF); 704 mtx_init(&sc->done_mtx, "g_disk_done", NULL, MTX_DEF); 705 sc->dp = dp; 706 sc->d_devstat = dp->d_devstat; 707 gp = g_new_geomf(&g_disk_class, "%s%d", dp->d_name, dp->d_unit); 708 gp->softc = sc; 709 pp = g_new_providerf(gp, "%s", gp->name); 710 devstat_remove_entry(pp->stat); 711 pp->stat = NULL; 712 dp->d_devstat->id = pp; 713 pp->mediasize = dp->d_mediasize; 714 pp->sectorsize = dp->d_sectorsize; 715 pp->stripeoffset = dp->d_stripeoffset; 716 pp->stripesize = dp->d_stripesize; 717 if ((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0) 718 pp->flags |= G_PF_ACCEPT_UNMAPPED; 719 if ((dp->d_flags & DISKFLAG_DIRECT_COMPLETION) != 0) 720 pp->flags |= G_PF_DIRECT_SEND; 721 pp->flags |= G_PF_DIRECT_RECEIVE; 722 if (bootverbose) 723 printf("GEOM: new disk %s\n", gp->name); 724 sysctl_ctx_init(&sc->sysctl_ctx); 725 snprintf(tmpstr, sizeof(tmpstr), "GEOM disk %s", gp->name); 726 sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx, 727 SYSCTL_STATIC_CHILDREN(_kern_geom_disk), OID_AUTO, gp->name, 728 CTLFLAG_RD, 0, tmpstr); 729 if (sc->sysctl_tree != NULL) { 730 SYSCTL_ADD_STRING(&sc->sysctl_ctx, 731 SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "led", 732 CTLFLAG_RWTUN, sc->led, sizeof(sc->led), 733 "LED name"); 734 } 735 pp->private = sc; 736 dp->d_geom = gp; 737 g_error_provider(pp, 0); 738 739 mtx_pool_lock(mtxpool_sleep, dp); 740 dp->d_init_level = DISK_INIT_DONE; 741 742 /* 743 * If the disk has gone away at this stage, start the withering 744 * process for it. 745 */ 746 if (dp->d_goneflag != 0) { 747 mtx_pool_unlock(mtxpool_sleep, dp); 748 g_wither_provider(pp, ENXIO); 749 return; 750 } 751 mtx_pool_unlock(mtxpool_sleep, dp); 752 753} 754 755/* 756 * We get this callback after all of the consumers have gone away, and just 757 * before the provider is freed. If the disk driver provided a d_gone 758 * callback, let them know that it is okay to free resources -- they won't 759 * be getting any more accesses from GEOM. 760 */ 761static void 762g_disk_providergone(struct g_provider *pp) 763{ 764 struct disk *dp; 765 struct g_disk_softc *sc; 766 767 sc = (struct g_disk_softc *)pp->private; 768 dp = sc->dp; 769 if (dp != NULL && dp->d_gone != NULL) 770 dp->d_gone(dp); 771 if (sc->sysctl_tree != NULL) { 772 sysctl_ctx_free(&sc->sysctl_ctx); 773 sc->sysctl_tree = NULL; 774 } 775 if (sc->led[0] != 0) { 776 led_set(sc->led, "0"); 777 sc->led[0] = 0; 778 } 779 pp->private = NULL; 780 pp->geom->softc = NULL; 781 mtx_destroy(&sc->done_mtx); 782 mtx_destroy(&sc->start_mtx); 783 g_free(sc); 784} 785 786static void 787g_disk_destroy(void *ptr, int flag) 788{ 789 struct disk *dp; 790 struct g_geom *gp; 791 struct g_disk_softc *sc; 792 793 g_topology_assert(); 794 dp = ptr; 795 gp = dp->d_geom; 796 if (gp != NULL) { 797 sc = gp->softc; 798 if (sc != NULL) 799 sc->dp = NULL; 800 dp->d_geom = NULL; 801 g_wither_geom(gp, ENXIO); 802 } 803 804 g_free(dp); 805} 806 807/* 808 * We only allow printable characters in disk ident, 809 * the rest is converted to 'x<HH>'. 810 */ 811static void 812g_disk_ident_adjust(char *ident, size_t size) 813{ 814 char *p, tmp[4], newid[DISK_IDENT_SIZE]; 815 816 newid[0] = '\0'; 817 for (p = ident; *p != '\0'; p++) { 818 if (isprint(*p)) { 819 tmp[0] = *p; 820 tmp[1] = '\0'; 821 } else { 822 snprintf(tmp, sizeof(tmp), "x%02hhx", 823 *(unsigned char *)p); 824 } 825 if (strlcat(newid, tmp, sizeof(newid)) >= sizeof(newid)) 826 break; 827 } 828 bzero(ident, size); 829 strlcpy(ident, newid, size); 830} 831 832struct disk * 833disk_alloc(void) 834{ 835 836 return (g_malloc(sizeof(struct disk), M_WAITOK | M_ZERO)); 837} 838 839void 840disk_create(struct disk *dp, int version) 841{ 842 843 if (version != DISK_VERSION) { 844 printf("WARNING: Attempt to add disk %s%d %s", 845 dp->d_name, dp->d_unit, 846 " using incompatible ABI version of disk(9)\n"); 847 printf("WARNING: Ignoring disk %s%d\n", 848 dp->d_name, dp->d_unit); 849 return; 850 } 851 if (dp->d_flags & DISKFLAG_RESERVED) { 852 printf("WARNING: Attempt to add non-MPSAFE disk %s%d\n", 853 dp->d_name, dp->d_unit); 854 printf("WARNING: Ignoring disk %s%d\n", 855 dp->d_name, dp->d_unit); 856 return; 857 } 858 KASSERT(dp->d_strategy != NULL, ("disk_create need d_strategy")); 859 KASSERT(dp->d_name != NULL, ("disk_create need d_name")); 860 KASSERT(*dp->d_name != 0, ("disk_create need d_name")); 861 KASSERT(strlen(dp->d_name) < SPECNAMELEN - 4, ("disk name too long")); 862 if (dp->d_devstat == NULL) 863 dp->d_devstat = devstat_new_entry(dp->d_name, dp->d_unit, 864 dp->d_sectorsize, DEVSTAT_ALL_SUPPORTED, 865 DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX); 866 dp->d_geom = NULL; 867 868 dp->d_init_level = DISK_INIT_NONE; 869 870 g_disk_ident_adjust(dp->d_ident, sizeof(dp->d_ident)); 871 g_post_event(g_disk_create, dp, M_WAITOK, dp, NULL); 872} 873 874void 875disk_destroy(struct disk *dp) 876{ 877 878 g_cancel_event(dp); 879 dp->d_destroyed = 1; 880 if (dp->d_devstat != NULL) 881 devstat_remove_entry(dp->d_devstat); 882 g_post_event(g_disk_destroy, dp, M_WAITOK, NULL); 883} 884 885void 886disk_gone(struct disk *dp) 887{ 888 struct g_geom *gp; 889 struct g_provider *pp; 890 891 mtx_pool_lock(mtxpool_sleep, dp); 892 dp->d_goneflag = 1; 893 894 /* 895 * If we're still in the process of creating this disk (the 896 * g_disk_create() function is still queued, or is in 897 * progress), the init level will not yet be DISK_INIT_DONE. 898 * 899 * If that is the case, g_disk_create() will see d_goneflag 900 * and take care of cleaning things up. 901 * 902 * If the disk has already been created, we default to 903 * withering the provider as usual below. 904 * 905 * If the caller has not set a d_gone() callback, he will 906 * not be any worse off by returning here, because the geom 907 * has not been fully setup in any case. 908 */ 909 if (dp->d_init_level < DISK_INIT_DONE) { 910 mtx_pool_unlock(mtxpool_sleep, dp); 911 return; 912 } 913 mtx_pool_unlock(mtxpool_sleep, dp); 914 915 gp = dp->d_geom; 916 if (gp != NULL) { 917 pp = LIST_FIRST(&gp->provider); 918 if (pp != NULL) { 919 KASSERT(LIST_NEXT(pp, provider) == NULL, 920 ("geom %p has more than one provider", gp)); 921 g_wither_provider(pp, ENXIO); 922 } 923 } 924} 925 926void 927disk_attr_changed(struct disk *dp, const char *attr, int flag) 928{ 929 struct g_geom *gp; 930 struct g_provider *pp; 931 char devnamebuf[128]; 932 933 gp = dp->d_geom; 934 if (gp != NULL) 935 LIST_FOREACH(pp, &gp->provider, provider) 936 (void)g_attr_changed(pp, attr, flag); 937 snprintf(devnamebuf, sizeof(devnamebuf), "devname=%s%d", dp->d_name, 938 dp->d_unit); 939 devctl_notify("GEOM", "disk", attr, devnamebuf); 940} 941 942void 943disk_media_changed(struct disk *dp, int flag) 944{ 945 struct g_geom *gp; 946 struct g_provider *pp; 947 948 gp = dp->d_geom; 949 if (gp != NULL) { 950 pp = LIST_FIRST(&gp->provider); 951 if (pp != NULL) { 952 KASSERT(LIST_NEXT(pp, provider) == NULL, 953 ("geom %p has more than one provider", gp)); 954 g_media_changed(pp, flag); 955 } 956 } 957} 958 959void 960disk_media_gone(struct disk *dp, int flag) 961{ 962 struct g_geom *gp; 963 struct g_provider *pp; 964 965 gp = dp->d_geom; 966 if (gp != NULL) { 967 pp = LIST_FIRST(&gp->provider); 968 if (pp != NULL) { 969 KASSERT(LIST_NEXT(pp, provider) == NULL, 970 ("geom %p has more than one provider", gp)); 971 g_media_gone(pp, flag); 972 } 973 } 974} 975 976int 977disk_resize(struct disk *dp, int flag) 978{ 979 980 if (dp->d_destroyed || dp->d_geom == NULL) 981 return (0); 982 983 return (g_post_event(g_disk_resize, dp, flag, NULL)); 984} 985 986static void 987g_kern_disks(void *p, int flag __unused) 988{ 989 struct sbuf *sb; 990 struct g_geom *gp; 991 char *sp; 992 993 sb = p; 994 sp = ""; 995 g_topology_assert(); 996 LIST_FOREACH(gp, &g_disk_class.geom, geom) { 997 sbuf_printf(sb, "%s%s", sp, gp->name); 998 sp = " "; 999 } 1000 sbuf_finish(sb); 1001} 1002 1003static int 1004sysctl_disks(SYSCTL_HANDLER_ARGS) 1005{ 1006 int error; 1007 struct sbuf *sb; 1008 1009 sb = sbuf_new_auto(); 1010 g_waitfor_event(g_kern_disks, sb, M_WAITOK, NULL); 1011 error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1); 1012 sbuf_delete(sb); 1013 return error; 1014} 1015 1016SYSCTL_PROC(_kern, OID_AUTO, disks, 1017 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, 1018 sysctl_disks, "A", "names of available disks"); 1019