1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2002 Poul-Henning Kamp 5 * Copyright (c) 2002 Networks Associates Technology, Inc. 6 * All rights reserved. 7 * 8 * This software was developed for the FreeBSD Project by Poul-Henning Kamp 9 * and NAI Labs, the Security Research Division of Network Associates, Inc. 10 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 11 * DARPA CHATS research program. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. The names of the authors may not be used to endorse or promote 22 * products derived from this software without specific prior written 23 * permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 */ 37 38 #include <sys/cdefs.h> 39 __FBSDID("$FreeBSD$"); 40 41 #include "opt_geom.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/kernel.h> 46 #include <sys/sysctl.h> 47 #include <sys/bio.h> 48 #include <sys/bus.h> 49 #include <sys/ctype.h> 50 #include <sys/fcntl.h> 51 #include <sys/malloc.h> 52 #include <sys/sbuf.h> 53 #include <sys/devicestat.h> 54 #include <machine/md_var.h> 55 56 #include <sys/lock.h> 57 #include <sys/mutex.h> 58 #include <geom/geom.h> 59 #include <geom/geom_disk.h> 60 #include <geom/geom_int.h> 61 62 #include <dev/led/led.h> 63 64 #include <machine/bus.h> 65 66 struct g_disk_softc { 67 struct disk *dp; 68 struct sysctl_ctx_list sysctl_ctx; 69 struct sysctl_oid *sysctl_tree; 70 char led[64]; 71 uint32_t state; 72 struct mtx done_mtx; 73 }; 74 75 static g_access_t g_disk_access; 76 static g_start_t g_disk_start; 77 static g_ioctl_t g_disk_ioctl; 78 static g_dumpconf_t g_disk_dumpconf; 79 static g_provgone_t g_disk_providergone; 80 81 static int g_disk_sysctl_flags(SYSCTL_HANDLER_ARGS); 82 83 static struct g_class g_disk_class = { 84 .name = G_DISK_CLASS_NAME, 85 .version = G_VERSION, 86 .start = g_disk_start, 87 .access = g_disk_access, 88 .ioctl = g_disk_ioctl, 89 .providergone = g_disk_providergone, 90 .dumpconf = g_disk_dumpconf, 91 }; 92 93 SYSCTL_DECL(_kern_geom); 94 static SYSCTL_NODE(_kern_geom, OID_AUTO, disk, CTLFLAG_RW, 0, 95 "GEOM_DISK stuff"); 96 97 DECLARE_GEOM_CLASS(g_disk_class, g_disk); 98 99 static int 100 g_disk_access(struct g_provider *pp, int r, int w, int e) 101 { 102 struct disk *dp; 103 struct g_disk_softc *sc; 104 int error; 105 106 g_trace(G_T_ACCESS, "g_disk_access(%s, %d, %d, %d)", 107 pp->name, r, w, e); 108 g_topology_assert(); 109 sc = pp->private; 110 if ((dp = sc->dp) == NULL || dp->d_destroyed) { 111 /* 112 * Allow decreasing access count even if disk is not 113 * available anymore. 114 */ 115 if (r <= 0 && w <= 0 && e <= 0) 116 return (0); 117 return (ENXIO); 118 } 119 r += pp->acr; 120 w += pp->acw; 121 e += pp->ace; 122 error = 0; 123 if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) { 124 /* 125 * It would be better to defer this decision to d_open if 126 * it was able to take flags. 127 */ 128 if (w > 0 && (dp->d_flags & DISKFLAG_WRITE_PROTECT) != 0) 129 error = EROFS; 130 if (error == 0 && dp->d_open != NULL) 131 error = dp->d_open(dp); 132 if (bootverbose && error != 0) 133 printf("Opened disk %s -> %d\n", pp->name, error); 134 if (error != 0) 135 return (error); 136 pp->sectorsize = dp->d_sectorsize; 137 if (dp->d_maxsize == 0) { 138 printf("WARNING: Disk drive %s%d has no d_maxsize\n", 139 dp->d_name, dp->d_unit); 140 dp->d_maxsize = DFLTPHYS; 141 } 142 if (dp->d_delmaxsize == 0) { 143 if (bootverbose && dp->d_flags & DISKFLAG_CANDELETE) { 144 printf("WARNING: Disk drive %s%d has no " 145 "d_delmaxsize\n", dp->d_name, dp->d_unit); 146 } 147 dp->d_delmaxsize = dp->d_maxsize; 148 } 149 pp->stripeoffset = dp->d_stripeoffset; 150 pp->stripesize = dp->d_stripesize; 151 dp->d_flags |= DISKFLAG_OPEN; 152 /* 153 * Do not invoke resize event when initial size was zero. 154 * Some disks report its size only after first opening. 155 */ 156 if (pp->mediasize == 0) 157 pp->mediasize = dp->d_mediasize; 158 else 159 g_resize_provider(pp, dp->d_mediasize); 160 } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) { 161 if (dp->d_close != NULL) { 162 error = dp->d_close(dp); 163 if (error != 0) 164 printf("Closed disk %s -> %d\n", 165 pp->name, error); 166 } 167 sc->state = G_STATE_ACTIVE; 168 if (sc->led[0] != 0) 169 led_set(sc->led, "0"); 170 dp->d_flags &= ~DISKFLAG_OPEN; 171 } 172 return (error); 173 } 174 175 static void 176 g_disk_kerneldump(struct bio *bp, struct disk *dp) 177 { 178 struct g_kerneldump *gkd; 179 struct g_geom *gp; 180 181 gkd = (struct g_kerneldump*)bp->bio_data; 182 gp = bp->bio_to->geom; 183 g_trace(G_T_TOPOLOGY, "g_disk_kerneldump(%s, %jd, %jd)", 184 gp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); 185 if (dp->d_dump == NULL) { 186 g_io_deliver(bp, ENODEV); 187 return; 188 } 189 gkd->di.dumper = dp->d_dump; 190 gkd->di.priv = dp; 191 gkd->di.blocksize = dp->d_sectorsize; 192 gkd->di.maxiosize = dp->d_maxsize; 193 gkd->di.mediaoffset = gkd->offset; 194 if ((gkd->offset + gkd->length) > dp->d_mediasize) 195 gkd->length = dp->d_mediasize - gkd->offset; 196 gkd->di.mediasize = gkd->length; 197 g_io_deliver(bp, 0); 198 } 199 200 static void 201 g_disk_setstate(struct bio *bp, struct g_disk_softc *sc) 202 { 203 const char *cmd; 204 205 memcpy(&sc->state, bp->bio_data, sizeof(sc->state)); 206 if (sc->led[0] != 0) { 207 switch (sc->state) { 208 case G_STATE_FAILED: 209 cmd = "1"; 210 break; 211 case G_STATE_REBUILD: 212 cmd = "f5"; 213 break; 214 case G_STATE_RESYNC: 215 cmd = "f1"; 216 break; 217 default: 218 cmd = "0"; 219 break; 220 } 221 led_set(sc->led, cmd); 222 } 223 g_io_deliver(bp, 0); 224 } 225 226 static void 227 g_disk_done(struct bio *bp) 228 { 229 struct bintime now; 230 struct bio *bp2; 231 struct g_disk_softc *sc; 232 233 /* See "notes" for why we need a mutex here */ 234 /* XXX: will witness accept a mix of Giant/unGiant drivers here ? */ 235 bp2 = bp->bio_parent; 236 sc = bp2->bio_to->private; 237 bp->bio_completed = bp->bio_length - bp->bio_resid; 238 binuptime(&now); 239 mtx_lock(&sc->done_mtx); 240 if (bp2->bio_error == 0) 241 bp2->bio_error = bp->bio_error; 242 bp2->bio_completed += bp->bio_completed; 243 244 switch (bp->bio_cmd) { 245 case BIO_ZONE: 246 bcopy(&bp->bio_zone, &bp2->bio_zone, sizeof(bp->bio_zone)); 247 /*FALLTHROUGH*/ 248 case BIO_READ: 249 case BIO_WRITE: 250 case BIO_DELETE: 251 case BIO_FLUSH: 252 devstat_end_transaction_bio_bt(sc->dp->d_devstat, bp, &now); 253 break; 254 default: 255 break; 256 } 257 bp2->bio_inbed++; 258 if (bp2->bio_children == bp2->bio_inbed) { 259 mtx_unlock(&sc->done_mtx); 260 bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed; 261 g_io_deliver(bp2, bp2->bio_error); 262 } else 263 mtx_unlock(&sc->done_mtx); 264 g_destroy_bio(bp); 265 } 266 267 static int 268 g_disk_ioctl(struct g_provider *pp, u_long cmd, void * data, int fflag, struct thread *td) 269 { 270 struct disk *dp; 271 struct g_disk_softc *sc; 272 int error; 273 274 sc = pp->private; 275 dp = sc->dp; 276 KASSERT(dp != NULL && !dp->d_destroyed, 277 ("g_disk_ioctl(%lx) on destroyed disk %s", cmd, pp->name)); 278 279 if (dp->d_ioctl == NULL) 280 return (ENOIOCTL); 281 error = dp->d_ioctl(dp, cmd, data, fflag, td); 282 return (error); 283 } 284 285 static off_t 286 g_disk_maxsize(struct disk *dp, struct bio *bp) 287 { 288 if (bp->bio_cmd == BIO_DELETE) 289 return (dp->d_delmaxsize); 290 return (dp->d_maxsize); 291 } 292 293 static int 294 g_disk_maxsegs(struct disk *dp, struct bio *bp) 295 { 296 return ((g_disk_maxsize(dp, bp) / PAGE_SIZE) + 1); 297 } 298 299 static void 300 g_disk_advance(struct disk *dp, struct bio *bp, off_t off) 301 { 302 303 bp->bio_offset += off; 304 bp->bio_length -= off; 305 306 if ((bp->bio_flags & BIO_VLIST) != 0) { 307 bus_dma_segment_t *seg, *end; 308 309 seg = (bus_dma_segment_t *)bp->bio_data; 310 end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n; 311 off += bp->bio_ma_offset; 312 while (off >= seg->ds_len) { 313 KASSERT((seg != end), 314 ("vlist request runs off the end")); 315 off -= seg->ds_len; 316 seg++; 317 } 318 bp->bio_ma_offset = off; 319 bp->bio_ma_n = end - seg; 320 bp->bio_data = (void *)seg; 321 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 322 bp->bio_ma += off / PAGE_SIZE; 323 bp->bio_ma_offset += off; 324 bp->bio_ma_offset %= PAGE_SIZE; 325 bp->bio_ma_n -= off / PAGE_SIZE; 326 } else { 327 bp->bio_data += off; 328 } 329 } 330 331 static void 332 g_disk_seg_limit(bus_dma_segment_t *seg, off_t *poffset, 333 off_t *plength, int *ppages) 334 { 335 uintptr_t seg_page_base; 336 uintptr_t seg_page_end; 337 off_t offset; 338 off_t length; 339 int seg_pages; 340 341 offset = *poffset; 342 length = *plength; 343 344 if (length > seg->ds_len - offset) 345 length = seg->ds_len - offset; 346 347 seg_page_base = trunc_page(seg->ds_addr + offset); 348 seg_page_end = round_page(seg->ds_addr + offset + length); 349 seg_pages = (seg_page_end - seg_page_base) >> PAGE_SHIFT; 350 351 if (seg_pages > *ppages) { 352 seg_pages = *ppages; 353 length = (seg_page_base + (seg_pages << PAGE_SHIFT)) - 354 (seg->ds_addr + offset); 355 } 356 357 *poffset = 0; 358 *plength -= length; 359 *ppages -= seg_pages; 360 } 361 362 static off_t 363 g_disk_vlist_limit(struct disk *dp, struct bio *bp, bus_dma_segment_t **pendseg) 364 { 365 bus_dma_segment_t *seg, *end; 366 off_t residual; 367 off_t offset; 368 int pages; 369 370 seg = (bus_dma_segment_t *)bp->bio_data; 371 end = (bus_dma_segment_t *)bp->bio_data + bp->bio_ma_n; 372 residual = bp->bio_length; 373 offset = bp->bio_ma_offset; 374 pages = g_disk_maxsegs(dp, bp); 375 while (residual != 0 && pages != 0) { 376 KASSERT((seg != end), 377 ("vlist limit runs off the end")); 378 g_disk_seg_limit(seg, &offset, &residual, &pages); 379 seg++; 380 } 381 if (pendseg != NULL) 382 *pendseg = seg; 383 return (residual); 384 } 385 386 static bool 387 g_disk_limit(struct disk *dp, struct bio *bp) 388 { 389 bool limited = false; 390 off_t maxsz; 391 392 maxsz = g_disk_maxsize(dp, bp); 393 394 /* 395 * XXX: If we have a stripesize we should really use it here. 396 * Care should be taken in the delete case if this is done 397 * as deletes can be very sensitive to size given how they 398 * are processed. 399 */ 400 if (bp->bio_length > maxsz) { 401 bp->bio_length = maxsz; 402 limited = true; 403 } 404 405 if ((bp->bio_flags & BIO_VLIST) != 0) { 406 bus_dma_segment_t *firstseg, *endseg; 407 off_t residual; 408 409 firstseg = (bus_dma_segment_t*)bp->bio_data; 410 residual = g_disk_vlist_limit(dp, bp, &endseg); 411 if (residual != 0) { 412 bp->bio_ma_n = endseg - firstseg; 413 bp->bio_length -= residual; 414 limited = true; 415 } 416 } else if ((bp->bio_flags & BIO_UNMAPPED) != 0) { 417 bp->bio_ma_n = 418 howmany(bp->bio_ma_offset + bp->bio_length, PAGE_SIZE); 419 } 420 421 return (limited); 422 } 423 424 static void 425 g_disk_start(struct bio *bp) 426 { 427 struct bio *bp2, *bp3; 428 struct disk *dp; 429 struct g_disk_softc *sc; 430 int error; 431 off_t off; 432 433 biotrack(bp, __func__); 434 435 sc = bp->bio_to->private; 436 dp = sc->dp; 437 KASSERT(dp != NULL && !dp->d_destroyed, 438 ("g_disk_start(%p) on destroyed disk %s", bp, bp->bio_to->name)); 439 error = EJUSTRETURN; 440 switch(bp->bio_cmd) { 441 case BIO_DELETE: 442 if (!(dp->d_flags & DISKFLAG_CANDELETE)) { 443 error = EOPNOTSUPP; 444 break; 445 } 446 /* fall-through */ 447 case BIO_READ: 448 case BIO_WRITE: 449 KASSERT((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0 || 450 (bp->bio_flags & BIO_UNMAPPED) == 0, 451 ("unmapped bio not supported by disk %s", dp->d_name)); 452 off = 0; 453 bp3 = NULL; 454 bp2 = g_clone_bio(bp); 455 if (bp2 == NULL) { 456 error = ENOMEM; 457 break; 458 } 459 for (;;) { 460 if (g_disk_limit(dp, bp2)) { 461 off += bp2->bio_length; 462 463 /* 464 * To avoid a race, we need to grab the next bio 465 * before we schedule this one. See "notes". 466 */ 467 bp3 = g_clone_bio(bp); 468 if (bp3 == NULL) 469 bp->bio_error = ENOMEM; 470 } 471 bp2->bio_done = g_disk_done; 472 bp2->bio_pblkno = bp2->bio_offset / dp->d_sectorsize; 473 bp2->bio_bcount = bp2->bio_length; 474 bp2->bio_disk = dp; 475 devstat_start_transaction_bio(dp->d_devstat, bp2); 476 dp->d_strategy(bp2); 477 478 if (bp3 == NULL) 479 break; 480 481 bp2 = bp3; 482 bp3 = NULL; 483 g_disk_advance(dp, bp2, off); 484 } 485 break; 486 case BIO_GETATTR: 487 /* Give the driver a chance to override */ 488 if (dp->d_getattr != NULL) { 489 if (bp->bio_disk == NULL) 490 bp->bio_disk = dp; 491 error = dp->d_getattr(bp); 492 if (error != -1) 493 break; 494 error = EJUSTRETURN; 495 } 496 if (g_handleattr_int(bp, "GEOM::candelete", 497 (dp->d_flags & DISKFLAG_CANDELETE) != 0)) 498 break; 499 else if (g_handleattr_int(bp, "GEOM::fwsectors", 500 dp->d_fwsectors)) 501 break; 502 else if (g_handleattr_int(bp, "GEOM::fwheads", dp->d_fwheads)) 503 break; 504 else if (g_handleattr_off_t(bp, "GEOM::frontstuff", 0)) 505 break; 506 else if (g_handleattr_str(bp, "GEOM::ident", dp->d_ident)) 507 break; 508 else if (g_handleattr_str(bp, "GEOM::descr", dp->d_descr)) 509 break; 510 else if (g_handleattr_uint16_t(bp, "GEOM::hba_vendor", 511 dp->d_hba_vendor)) 512 break; 513 else if (g_handleattr_uint16_t(bp, "GEOM::hba_device", 514 dp->d_hba_device)) 515 break; 516 else if (g_handleattr_uint16_t(bp, "GEOM::hba_subvendor", 517 dp->d_hba_subvendor)) 518 break; 519 else if (g_handleattr_uint16_t(bp, "GEOM::hba_subdevice", 520 dp->d_hba_subdevice)) 521 break; 522 else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) 523 g_disk_kerneldump(bp, dp); 524 else if (!strcmp(bp->bio_attribute, "GEOM::setstate")) 525 g_disk_setstate(bp, sc); 526 else if (g_handleattr_uint16_t(bp, "GEOM::rotation_rate", 527 dp->d_rotation_rate)) 528 break; 529 else if (g_handleattr_str(bp, "GEOM::attachment", 530 dp->d_attachment)) 531 break; 532 else 533 error = ENOIOCTL; 534 break; 535 case BIO_FLUSH: 536 g_trace(G_T_BIO, "g_disk_flushcache(%s)", 537 bp->bio_to->name); 538 if (!(dp->d_flags & DISKFLAG_CANFLUSHCACHE)) { 539 error = EOPNOTSUPP; 540 break; 541 } 542 /*FALLTHROUGH*/ 543 case BIO_ZONE: 544 if (bp->bio_cmd == BIO_ZONE) { 545 if (!(dp->d_flags & DISKFLAG_CANZONE)) { 546 error = EOPNOTSUPP; 547 break; 548 } 549 g_trace(G_T_BIO, "g_disk_zone(%s)", 550 bp->bio_to->name); 551 } 552 bp2 = g_clone_bio(bp); 553 if (bp2 == NULL) { 554 g_io_deliver(bp, ENOMEM); 555 return; 556 } 557 bp2->bio_done = g_disk_done; 558 bp2->bio_disk = dp; 559 devstat_start_transaction_bio(dp->d_devstat, bp2); 560 dp->d_strategy(bp2); 561 break; 562 default: 563 error = EOPNOTSUPP; 564 break; 565 } 566 if (error != EJUSTRETURN) 567 g_io_deliver(bp, error); 568 return; 569 } 570 571 static void 572 g_disk_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp) 573 { 574 struct bio *bp; 575 struct disk *dp; 576 struct g_disk_softc *sc; 577 char *buf; 578 int res = 0; 579 580 sc = gp->softc; 581 if (sc == NULL || (dp = sc->dp) == NULL) 582 return; 583 if (indent == NULL) { 584 sbuf_printf(sb, " hd %u", dp->d_fwheads); 585 sbuf_printf(sb, " sc %u", dp->d_fwsectors); 586 return; 587 } 588 if (pp != NULL) { 589 sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n", 590 indent, dp->d_fwheads); 591 sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n", 592 indent, dp->d_fwsectors); 593 594 /* 595 * "rotationrate" is a little complicated, because the value 596 * returned by the drive might not be the RPM; 0 and 1 are 597 * special cases, and there's also a valid range. 598 */ 599 sbuf_printf(sb, "%s<rotationrate>", indent); 600 if (dp->d_rotation_rate == DISK_RR_UNKNOWN) /* Old drives */ 601 sbuf_cat(sb, "unknown"); /* don't report RPM. */ 602 else if (dp->d_rotation_rate == DISK_RR_NON_ROTATING) 603 sbuf_cat(sb, "0"); 604 else if ((dp->d_rotation_rate >= DISK_RR_MIN) && 605 (dp->d_rotation_rate <= DISK_RR_MAX)) 606 sbuf_printf(sb, "%u", dp->d_rotation_rate); 607 else 608 sbuf_cat(sb, "invalid"); 609 sbuf_cat(sb, "</rotationrate>\n"); 610 if (dp->d_getattr != NULL) { 611 buf = g_malloc(DISK_IDENT_SIZE, M_WAITOK); 612 bp = g_alloc_bio(); 613 bp->bio_disk = dp; 614 bp->bio_attribute = "GEOM::ident"; 615 bp->bio_length = DISK_IDENT_SIZE; 616 bp->bio_data = buf; 617 res = dp->d_getattr(bp); 618 sbuf_printf(sb, "%s<ident>", indent); 619 g_conf_cat_escaped(sb, res == 0 ? buf : dp->d_ident); 620 sbuf_cat(sb, "</ident>\n"); 621 bp->bio_attribute = "GEOM::lunid"; 622 bp->bio_length = DISK_IDENT_SIZE; 623 bp->bio_data = buf; 624 if (dp->d_getattr(bp) == 0) { 625 sbuf_printf(sb, "%s<lunid>", indent); 626 g_conf_cat_escaped(sb, buf); 627 sbuf_cat(sb, "</lunid>\n"); 628 } 629 bp->bio_attribute = "GEOM::lunname"; 630 bp->bio_length = DISK_IDENT_SIZE; 631 bp->bio_data = buf; 632 if (dp->d_getattr(bp) == 0) { 633 sbuf_printf(sb, "%s<lunname>", indent); 634 g_conf_cat_escaped(sb, buf); 635 sbuf_cat(sb, "</lunname>\n"); 636 } 637 g_destroy_bio(bp); 638 g_free(buf); 639 } else { 640 sbuf_printf(sb, "%s<ident>", indent); 641 g_conf_cat_escaped(sb, dp->d_ident); 642 sbuf_cat(sb, "</ident>\n"); 643 } 644 sbuf_printf(sb, "%s<descr>", indent); 645 g_conf_cat_escaped(sb, dp->d_descr); 646 sbuf_cat(sb, "</descr>\n"); 647 } 648 } 649 650 static void 651 g_disk_resize(void *ptr, int flag) 652 { 653 struct disk *dp; 654 struct g_geom *gp; 655 struct g_provider *pp; 656 657 if (flag == EV_CANCEL) 658 return; 659 g_topology_assert(); 660 661 dp = ptr; 662 gp = dp->d_geom; 663 664 if (dp->d_destroyed || gp == NULL) 665 return; 666 667 LIST_FOREACH(pp, &gp->provider, provider) { 668 if (pp->sectorsize != 0 && 669 pp->sectorsize != dp->d_sectorsize) 670 g_wither_provider(pp, ENXIO); 671 else 672 g_resize_provider(pp, dp->d_mediasize); 673 } 674 } 675 676 static void 677 g_disk_create(void *arg, int flag) 678 { 679 struct g_geom *gp; 680 struct g_provider *pp; 681 struct disk *dp; 682 struct g_disk_softc *sc; 683 struct disk_alias *dap; 684 char tmpstr[80]; 685 686 if (flag == EV_CANCEL) 687 return; 688 g_topology_assert(); 689 dp = arg; 690 691 mtx_pool_lock(mtxpool_sleep, dp); 692 dp->d_init_level = DISK_INIT_START; 693 694 /* 695 * If the disk has already gone away, we can just stop here and 696 * call the user's callback to tell him we've cleaned things up. 697 */ 698 if (dp->d_goneflag != 0) { 699 mtx_pool_unlock(mtxpool_sleep, dp); 700 if (dp->d_gone != NULL) 701 dp->d_gone(dp); 702 return; 703 } 704 mtx_pool_unlock(mtxpool_sleep, dp); 705 706 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO); 707 mtx_init(&sc->done_mtx, "g_disk_done", NULL, MTX_DEF); 708 sc->dp = dp; 709 gp = g_new_geomf(&g_disk_class, "%s%d", dp->d_name, dp->d_unit); 710 gp->softc = sc; 711 LIST_FOREACH(dap, &dp->d_aliases, da_next) { 712 snprintf(tmpstr, sizeof(tmpstr), "%s%d", dap->da_alias, dp->d_unit); 713 g_geom_add_alias(gp, tmpstr); 714 } 715 pp = g_new_providerf(gp, "%s", gp->name); 716 devstat_remove_entry(pp->stat); 717 pp->stat = NULL; 718 dp->d_devstat->id = pp; 719 pp->mediasize = dp->d_mediasize; 720 pp->sectorsize = dp->d_sectorsize; 721 pp->stripeoffset = dp->d_stripeoffset; 722 pp->stripesize = dp->d_stripesize; 723 if ((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0) 724 pp->flags |= G_PF_ACCEPT_UNMAPPED; 725 if ((dp->d_flags & DISKFLAG_DIRECT_COMPLETION) != 0) 726 pp->flags |= G_PF_DIRECT_SEND; 727 pp->flags |= G_PF_DIRECT_RECEIVE; 728 if (bootverbose) 729 printf("GEOM: new disk %s\n", gp->name); 730 sysctl_ctx_init(&sc->sysctl_ctx); 731 snprintf(tmpstr, sizeof(tmpstr), "GEOM disk %s", gp->name); 732 sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx, 733 SYSCTL_STATIC_CHILDREN(_kern_geom_disk), OID_AUTO, gp->name, 734 CTLFLAG_RD, 0, tmpstr); 735 if (sc->sysctl_tree != NULL) { 736 SYSCTL_ADD_STRING(&sc->sysctl_ctx, 737 SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "led", 738 CTLFLAG_RWTUN, sc->led, sizeof(sc->led), 739 "LED name"); 740 SYSCTL_ADD_PROC(&sc->sysctl_ctx, 741 SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "flags", 742 CTLTYPE_STRING | CTLFLAG_RD, dp, 0, g_disk_sysctl_flags, 743 "A", "Report disk flags"); 744 } 745 pp->private = sc; 746 dp->d_geom = gp; 747 g_error_provider(pp, 0); 748 749 mtx_pool_lock(mtxpool_sleep, dp); 750 dp->d_init_level = DISK_INIT_DONE; 751 752 /* 753 * If the disk has gone away at this stage, start the withering 754 * process for it. 755 */ 756 if (dp->d_goneflag != 0) { 757 mtx_pool_unlock(mtxpool_sleep, dp); 758 g_wither_provider(pp, ENXIO); 759 return; 760 } 761 mtx_pool_unlock(mtxpool_sleep, dp); 762 763 } 764 765 /* 766 * We get this callback after all of the consumers have gone away, and just 767 * before the provider is freed. If the disk driver provided a d_gone 768 * callback, let them know that it is okay to free resources -- they won't 769 * be getting any more accesses from GEOM. 770 */ 771 static void 772 g_disk_providergone(struct g_provider *pp) 773 { 774 struct disk *dp; 775 struct g_disk_softc *sc; 776 777 sc = (struct g_disk_softc *)pp->private; 778 dp = sc->dp; 779 if (dp != NULL && dp->d_gone != NULL) 780 dp->d_gone(dp); 781 if (sc->sysctl_tree != NULL) { 782 sysctl_ctx_free(&sc->sysctl_ctx); 783 sc->sysctl_tree = NULL; 784 } 785 if (sc->led[0] != 0) { 786 led_set(sc->led, "0"); 787 sc->led[0] = 0; 788 } 789 pp->private = NULL; 790 pp->geom->softc = NULL; 791 mtx_destroy(&sc->done_mtx); 792 g_free(sc); 793 } 794 795 static void 796 g_disk_destroy(void *ptr, int flag) 797 { 798 struct disk *dp; 799 struct g_geom *gp; 800 struct g_disk_softc *sc; 801 struct disk_alias *dap, *daptmp; 802 803 g_topology_assert(); 804 dp = ptr; 805 gp = dp->d_geom; 806 if (gp != NULL) { 807 sc = gp->softc; 808 if (sc != NULL) 809 sc->dp = NULL; 810 dp->d_geom = NULL; 811 g_wither_geom(gp, ENXIO); 812 } 813 LIST_FOREACH_SAFE(dap, &dp->d_aliases, da_next, daptmp) 814 g_free(dap); 815 816 g_free(dp); 817 } 818 819 /* 820 * We only allow printable characters in disk ident, 821 * the rest is converted to 'x<HH>'. 822 */ 823 static void 824 g_disk_ident_adjust(char *ident, size_t size) 825 { 826 char *p, tmp[4], newid[DISK_IDENT_SIZE]; 827 828 newid[0] = '\0'; 829 for (p = ident; *p != '\0'; p++) { 830 if (isprint(*p)) { 831 tmp[0] = *p; 832 tmp[1] = '\0'; 833 } else { 834 snprintf(tmp, sizeof(tmp), "x%02hhx", 835 *(unsigned char *)p); 836 } 837 if (strlcat(newid, tmp, sizeof(newid)) >= sizeof(newid)) 838 break; 839 } 840 bzero(ident, size); 841 strlcpy(ident, newid, size); 842 } 843 844 struct disk * 845 disk_alloc(void) 846 { 847 struct disk *dp; 848 849 dp = g_malloc(sizeof(struct disk), M_WAITOK | M_ZERO); 850 LIST_INIT(&dp->d_aliases); 851 return (dp); 852 } 853 854 void 855 disk_create(struct disk *dp, int version) 856 { 857 858 if (version != DISK_VERSION) { 859 printf("WARNING: Attempt to add disk %s%d %s", 860 dp->d_name, dp->d_unit, 861 " using incompatible ABI version of disk(9)\n"); 862 printf("WARNING: Ignoring disk %s%d\n", 863 dp->d_name, dp->d_unit); 864 return; 865 } 866 if (dp->d_flags & DISKFLAG_RESERVED) { 867 printf("WARNING: Attempt to add non-MPSAFE disk %s%d\n", 868 dp->d_name, dp->d_unit); 869 printf("WARNING: Ignoring disk %s%d\n", 870 dp->d_name, dp->d_unit); 871 return; 872 } 873 KASSERT(dp->d_strategy != NULL, ("disk_create need d_strategy")); 874 KASSERT(dp->d_name != NULL, ("disk_create need d_name")); 875 KASSERT(*dp->d_name != 0, ("disk_create need d_name")); 876 KASSERT(strlen(dp->d_name) < SPECNAMELEN - 4, ("disk name too long")); 877 if (dp->d_devstat == NULL) 878 dp->d_devstat = devstat_new_entry(dp->d_name, dp->d_unit, 879 dp->d_sectorsize, DEVSTAT_ALL_SUPPORTED, 880 DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX); 881 dp->d_geom = NULL; 882 883 dp->d_init_level = DISK_INIT_NONE; 884 885 g_disk_ident_adjust(dp->d_ident, sizeof(dp->d_ident)); 886 g_post_event(g_disk_create, dp, M_WAITOK, dp, NULL); 887 } 888 889 void 890 disk_destroy(struct disk *dp) 891 { 892 893 disk_gone(dp); 894 dp->d_destroyed = 1; 895 g_cancel_event(dp); 896 if (dp->d_devstat != NULL) 897 devstat_remove_entry(dp->d_devstat); 898 g_post_event(g_disk_destroy, dp, M_WAITOK, NULL); 899 } 900 901 void 902 disk_add_alias(struct disk *dp, const char *name) 903 { 904 struct disk_alias *dap; 905 906 dap = (struct disk_alias *)g_malloc( 907 sizeof(struct disk_alias) + strlen(name) + 1, M_WAITOK); 908 strcpy((char *)(dap + 1), name); 909 dap->da_alias = (const char *)(dap + 1); 910 LIST_INSERT_HEAD(&dp->d_aliases, dap, da_next); 911 } 912 913 void 914 disk_gone(struct disk *dp) 915 { 916 struct g_geom *gp; 917 struct g_provider *pp; 918 919 mtx_pool_lock(mtxpool_sleep, dp); 920 921 /* 922 * Second wither call makes no sense, plus we can not access the list 923 * of providers without topology lock after calling wither once. 924 */ 925 if (dp->d_goneflag != 0) { 926 mtx_pool_unlock(mtxpool_sleep, dp); 927 return; 928 } 929 930 dp->d_goneflag = 1; 931 932 /* 933 * If we're still in the process of creating this disk (the 934 * g_disk_create() function is still queued, or is in 935 * progress), the init level will not yet be DISK_INIT_DONE. 936 * 937 * If that is the case, g_disk_create() will see d_goneflag 938 * and take care of cleaning things up. 939 * 940 * If the disk has already been created, we default to 941 * withering the provider as usual below. 942 * 943 * If the caller has not set a d_gone() callback, he will 944 * not be any worse off by returning here, because the geom 945 * has not been fully setup in any case. 946 */ 947 if (dp->d_init_level < DISK_INIT_DONE) { 948 mtx_pool_unlock(mtxpool_sleep, dp); 949 return; 950 } 951 mtx_pool_unlock(mtxpool_sleep, dp); 952 953 gp = dp->d_geom; 954 pp = LIST_FIRST(&gp->provider); 955 if (pp != NULL) { 956 KASSERT(LIST_NEXT(pp, provider) == NULL, 957 ("geom %p has more than one provider", gp)); 958 g_wither_provider(pp, ENXIO); 959 } 960 } 961 962 void 963 disk_attr_changed(struct disk *dp, const char *attr, int flag) 964 { 965 struct g_geom *gp; 966 struct g_provider *pp; 967 char devnamebuf[128]; 968 969 gp = dp->d_geom; 970 if (gp != NULL) 971 LIST_FOREACH(pp, &gp->provider, provider) 972 (void)g_attr_changed(pp, attr, flag); 973 snprintf(devnamebuf, sizeof(devnamebuf), "devname=%s%d", dp->d_name, 974 dp->d_unit); 975 devctl_notify("GEOM", "disk", attr, devnamebuf); 976 } 977 978 void 979 disk_media_changed(struct disk *dp, int flag) 980 { 981 struct g_geom *gp; 982 struct g_provider *pp; 983 984 gp = dp->d_geom; 985 if (gp != NULL) { 986 pp = LIST_FIRST(&gp->provider); 987 if (pp != NULL) { 988 KASSERT(LIST_NEXT(pp, provider) == NULL, 989 ("geom %p has more than one provider", gp)); 990 g_media_changed(pp, flag); 991 } 992 } 993 } 994 995 void 996 disk_media_gone(struct disk *dp, int flag) 997 { 998 struct g_geom *gp; 999 struct g_provider *pp; 1000 1001 gp = dp->d_geom; 1002 if (gp != NULL) { 1003 pp = LIST_FIRST(&gp->provider); 1004 if (pp != NULL) { 1005 KASSERT(LIST_NEXT(pp, provider) == NULL, 1006 ("geom %p has more than one provider", gp)); 1007 g_media_gone(pp, flag); 1008 } 1009 } 1010 } 1011 1012 int 1013 disk_resize(struct disk *dp, int flag) 1014 { 1015 1016 if (dp->d_destroyed || dp->d_geom == NULL) 1017 return (0); 1018 1019 return (g_post_event(g_disk_resize, dp, flag, NULL)); 1020 } 1021 1022 static void 1023 g_kern_disks(void *p, int flag __unused) 1024 { 1025 struct sbuf *sb; 1026 struct g_geom *gp; 1027 char *sp; 1028 1029 sb = p; 1030 sp = ""; 1031 g_topology_assert(); 1032 LIST_FOREACH(gp, &g_disk_class.geom, geom) { 1033 sbuf_printf(sb, "%s%s", sp, gp->name); 1034 sp = " "; 1035 } 1036 sbuf_finish(sb); 1037 } 1038 1039 static int 1040 g_disk_sysctl_flags(SYSCTL_HANDLER_ARGS) 1041 { 1042 struct disk *dp; 1043 struct sbuf *sb; 1044 int error; 1045 1046 sb = sbuf_new_auto(); 1047 dp = (struct disk *)arg1; 1048 sbuf_printf(sb, "%b", dp->d_flags, 1049 "\20" 1050 "\2OPEN" 1051 "\3CANDELETE" 1052 "\4CANFLUSHCACHE" 1053 "\5UNMAPPEDBIO" 1054 "\6DIRECTCOMPLETION" 1055 "\10CANZONE" 1056 "\11WRITEPROTECT"); 1057 1058 sbuf_finish(sb); 1059 error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1); 1060 sbuf_delete(sb); 1061 return (error); 1062 } 1063 1064 static int 1065 sysctl_disks(SYSCTL_HANDLER_ARGS) 1066 { 1067 int error; 1068 struct sbuf *sb; 1069 1070 sb = sbuf_new_auto(); 1071 g_waitfor_event(g_kern_disks, sb, M_WAITOK, NULL); 1072 error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1); 1073 sbuf_delete(sb); 1074 return error; 1075 } 1076 1077 SYSCTL_PROC(_kern, OID_AUTO, disks, 1078 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, 1079 sysctl_disks, "A", "names of available disks"); 1080