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