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("$FreeBSD$"); 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/malloc.h> 42 #include <sys/kernel.h> 43 #include <sys/conf.h> 44 #include <sys/ctype.h> 45 #include <sys/bio.h> 46 #include <sys/bus.h> 47 #include <sys/lock.h> 48 #include <sys/mutex.h> 49 #include <sys/proc.h> 50 #include <sys/errno.h> 51 #include <sys/time.h> 52 #include <sys/disk.h> 53 #include <sys/fcntl.h> 54 #include <sys/limits.h> 55 #include <geom/geom.h> 56 #include <geom/geom_int.h> 57 #include <machine/stdarg.h> 58 59 struct g_dev_softc { 60 struct mtx sc_mtx; 61 struct cdev *sc_dev; 62 struct cdev *sc_alias; 63 int sc_open; 64 int sc_active; 65 }; 66 67 static d_open_t g_dev_open; 68 static d_close_t g_dev_close; 69 static d_strategy_t g_dev_strategy; 70 static d_ioctl_t g_dev_ioctl; 71 72 static struct cdevsw g_dev_cdevsw = { 73 .d_version = D_VERSION, 74 .d_open = g_dev_open, 75 .d_close = g_dev_close, 76 .d_read = physread, 77 .d_write = physwrite, 78 .d_ioctl = g_dev_ioctl, 79 .d_strategy = g_dev_strategy, 80 .d_name = "g_dev", 81 .d_flags = D_DISK | D_TRACKCLOSE | D_UNMAPPED_IO, 82 }; 83 84 static g_taste_t g_dev_taste; 85 static g_orphan_t g_dev_orphan; 86 static g_attrchanged_t g_dev_attrchanged; 87 88 static struct g_class g_dev_class = { 89 .name = "DEV", 90 .version = G_VERSION, 91 .taste = g_dev_taste, 92 .orphan = g_dev_orphan, 93 .attrchanged = g_dev_attrchanged 94 }; 95 96 static void 97 g_dev_destroy(void *arg, int flags __unused) 98 { 99 struct g_consumer *cp; 100 struct g_geom *gp; 101 struct g_dev_softc *sc; 102 103 g_topology_assert(); 104 cp = arg; 105 gp = cp->geom; 106 sc = cp->private; 107 g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name); 108 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 109 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 110 g_detach(cp); 111 g_destroy_consumer(cp); 112 g_destroy_geom(gp); 113 mtx_destroy(&sc->sc_mtx); 114 g_free(sc); 115 } 116 117 void 118 g_dev_print(void) 119 { 120 struct g_geom *gp; 121 char const *p = ""; 122 123 LIST_FOREACH(gp, &g_dev_class.geom, geom) { 124 printf("%s%s", p, gp->name); 125 p = " "; 126 } 127 printf("\n"); 128 } 129 130 static void 131 g_dev_attrchanged(struct g_consumer *cp, const char *attr) 132 { 133 struct g_dev_softc *sc; 134 struct cdev *dev; 135 char buf[SPECNAMELEN + 6]; 136 137 sc = cp->private; 138 if (strcmp(attr, "GEOM::media") == 0) { 139 dev = sc->sc_dev; 140 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name); 141 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK); 142 dev = sc->sc_alias; 143 if (dev != NULL) { 144 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name); 145 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, 146 M_WAITOK); 147 } 148 return; 149 } 150 151 if (strcmp(attr, "GEOM::physpath") != 0) 152 return; 153 154 if (g_access(cp, 1, 0, 0) == 0) { 155 char *physpath; 156 int error, physpath_len; 157 158 physpath_len = MAXPATHLEN; 159 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO); 160 error = 161 g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath); 162 g_access(cp, -1, 0, 0); 163 if (error == 0 && strlen(physpath) != 0) { 164 struct cdev *old_alias_dev; 165 struct cdev **alias_devp; 166 167 dev = sc->sc_dev; 168 old_alias_dev = sc->sc_alias; 169 alias_devp = (struct cdev **)&sc->sc_alias; 170 make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp, 171 dev, old_alias_dev, physpath); 172 } else if (sc->sc_alias) { 173 destroy_dev((struct cdev *)sc->sc_alias); 174 sc->sc_alias = NULL; 175 } 176 g_free(physpath); 177 } 178 } 179 180 struct g_provider * 181 g_dev_getprovider(struct cdev *dev) 182 { 183 struct g_consumer *cp; 184 185 g_topology_assert(); 186 if (dev == NULL) 187 return (NULL); 188 if (dev->si_devsw != &g_dev_cdevsw) 189 return (NULL); 190 cp = dev->si_drv2; 191 return (cp->provider); 192 } 193 194 static struct g_geom * 195 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused) 196 { 197 struct g_geom *gp; 198 struct g_consumer *cp; 199 struct g_dev_softc *sc; 200 int error, len; 201 struct cdev *dev, *adev; 202 char buf[64], *val; 203 204 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name); 205 g_topology_assert(); 206 gp = g_new_geomf(mp, "%s", pp->name); 207 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO); 208 mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF); 209 cp = g_new_consumer(gp); 210 cp->private = sc; 211 error = g_attach(cp, pp); 212 KASSERT(error == 0, 213 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error)); 214 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev, 215 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name); 216 if (error != 0) { 217 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n", 218 __func__, gp->name, error); 219 g_detach(cp); 220 g_destroy_consumer(cp); 221 g_destroy_geom(gp); 222 mtx_destroy(&sc->sc_mtx); 223 g_free(sc); 224 return (NULL); 225 } 226 sc->sc_dev = dev; 227 228 /* Search for device alias name and create it if found. */ 229 adev = NULL; 230 for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) { 231 snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name); 232 buf[14 + len] = 0; 233 val = getenv(buf); 234 if (val != NULL) { 235 snprintf(buf, sizeof(buf), "%s%s", 236 val, gp->name + len); 237 freeenv(val); 238 make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, 239 &adev, dev, "%s", buf); 240 break; 241 } 242 } 243 244 dev->si_iosize_max = MAXPHYS; 245 dev->si_drv2 = cp; 246 if (adev != NULL) { 247 adev->si_iosize_max = MAXPHYS; 248 adev->si_drv2 = cp; 249 } 250 251 g_dev_attrchanged(cp, "GEOM::physpath"); 252 253 return (gp); 254 } 255 256 static int 257 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td) 258 { 259 struct g_consumer *cp; 260 struct g_dev_softc *sc; 261 int error, r, w, e; 262 263 cp = dev->si_drv2; 264 if (cp == NULL) 265 return(ENXIO); /* g_dev_taste() not done yet */ 266 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", 267 cp->geom->name, flags, fmt, td); 268 269 r = flags & FREAD ? 1 : 0; 270 w = flags & FWRITE ? 1 : 0; 271 #ifdef notyet 272 e = flags & O_EXCL ? 1 : 0; 273 #else 274 e = 0; 275 #endif 276 if (w) { 277 /* 278 * When running in very secure mode, do not allow 279 * opens for writing of any disks. 280 */ 281 error = securelevel_ge(td->td_ucred, 2); 282 if (error) 283 return (error); 284 } 285 g_topology_lock(); 286 error = g_access(cp, r, w, e); 287 g_topology_unlock(); 288 if (error == 0) { 289 sc = cp->private; 290 mtx_lock(&sc->sc_mtx); 291 if (sc->sc_open == 0 && sc->sc_active != 0) 292 wakeup(&sc->sc_active); 293 sc->sc_open += r + w + e; 294 mtx_unlock(&sc->sc_mtx); 295 } 296 return(error); 297 } 298 299 static int 300 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td) 301 { 302 struct g_consumer *cp; 303 struct g_dev_softc *sc; 304 int error, r, w, e; 305 306 cp = dev->si_drv2; 307 if (cp == NULL) 308 return(ENXIO); 309 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)", 310 cp->geom->name, flags, fmt, td); 311 312 r = flags & FREAD ? -1 : 0; 313 w = flags & FWRITE ? -1 : 0; 314 #ifdef notyet 315 e = flags & O_EXCL ? -1 : 0; 316 #else 317 e = 0; 318 #endif 319 sc = cp->private; 320 mtx_lock(&sc->sc_mtx); 321 sc->sc_open += r + w + e; 322 while (sc->sc_open == 0 && sc->sc_active != 0) 323 msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0); 324 mtx_unlock(&sc->sc_mtx); 325 g_topology_lock(); 326 error = g_access(cp, r, w, e); 327 g_topology_unlock(); 328 return (error); 329 } 330 331 /* 332 * XXX: Until we have unmessed the ioctl situation, there is a race against 333 * XXX: a concurrent orphanization. We cannot close it by holding topology 334 * XXX: since that would prevent us from doing our job, and stalling events 335 * XXX: will break (actually: stall) the BSD disklabel hacks. 336 */ 337 static int 338 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 339 { 340 struct g_consumer *cp; 341 struct g_provider *pp; 342 struct g_kerneldump kd; 343 off_t offset, length, chunk; 344 int i, error; 345 u_int u; 346 347 cp = dev->si_drv2; 348 pp = cp->provider; 349 350 error = 0; 351 KASSERT(cp->acr || cp->acw, 352 ("Consumer with zero access count in g_dev_ioctl")); 353 354 i = IOCPARM_LEN(cmd); 355 switch (cmd) { 356 case DIOCGSECTORSIZE: 357 *(u_int *)data = cp->provider->sectorsize; 358 if (*(u_int *)data == 0) 359 error = ENOENT; 360 break; 361 case DIOCGMEDIASIZE: 362 *(off_t *)data = cp->provider->mediasize; 363 if (*(off_t *)data == 0) 364 error = ENOENT; 365 break; 366 case DIOCGFWSECTORS: 367 error = g_io_getattr("GEOM::fwsectors", cp, &i, data); 368 if (error == 0 && *(u_int *)data == 0) 369 error = ENOENT; 370 break; 371 case DIOCGFWHEADS: 372 error = g_io_getattr("GEOM::fwheads", cp, &i, data); 373 if (error == 0 && *(u_int *)data == 0) 374 error = ENOENT; 375 break; 376 case DIOCGFRONTSTUFF: 377 error = g_io_getattr("GEOM::frontstuff", cp, &i, data); 378 break; 379 case DIOCSKERNELDUMP: 380 u = *((u_int *)data); 381 if (!u) { 382 set_dumper(NULL, NULL); 383 error = 0; 384 break; 385 } 386 kd.offset = 0; 387 kd.length = OFF_MAX; 388 i = sizeof kd; 389 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd); 390 if (!error) { 391 error = set_dumper(&kd.di, devtoname(dev)); 392 if (!error) 393 dev->si_flags |= SI_DUMPDEV; 394 } 395 break; 396 case DIOCGFLUSH: 397 error = g_io_flush(cp); 398 break; 399 case DIOCGDELETE: 400 offset = ((off_t *)data)[0]; 401 length = ((off_t *)data)[1]; 402 if ((offset % cp->provider->sectorsize) != 0 || 403 (length % cp->provider->sectorsize) != 0 || length <= 0) { 404 printf("%s: offset=%jd length=%jd\n", __func__, offset, 405 length); 406 error = EINVAL; 407 break; 408 } 409 while (length > 0) { 410 chunk = length; 411 if (chunk > 65536 * cp->provider->sectorsize) 412 chunk = 65536 * cp->provider->sectorsize; 413 error = g_delete_data(cp, offset, chunk); 414 length -= chunk; 415 offset += chunk; 416 if (error) 417 break; 418 /* 419 * Since the request size is unbounded, the service 420 * time is likewise. We make this ioctl interruptible 421 * by checking for signals for each bio. 422 */ 423 if (SIGPENDING(td)) 424 break; 425 } 426 break; 427 case DIOCGIDENT: 428 error = g_io_getattr("GEOM::ident", cp, &i, data); 429 break; 430 case DIOCGPROVIDERNAME: 431 if (pp == NULL) 432 return (ENOENT); 433 strlcpy(data, pp->name, i); 434 break; 435 case DIOCGSTRIPESIZE: 436 *(off_t *)data = cp->provider->stripesize; 437 break; 438 case DIOCGSTRIPEOFFSET: 439 *(off_t *)data = cp->provider->stripeoffset; 440 break; 441 case DIOCGPHYSPATH: 442 error = g_io_getattr("GEOM::physpath", cp, &i, data); 443 if (error == 0 && *(char *)data == '\0') 444 error = ENOENT; 445 break; 446 default: 447 if (cp->provider->geom->ioctl != NULL) { 448 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td); 449 } else { 450 error = ENOIOCTL; 451 } 452 } 453 454 return (error); 455 } 456 457 static void 458 g_dev_done(struct bio *bp2) 459 { 460 struct g_consumer *cp; 461 struct g_dev_softc *sc; 462 struct bio *bp; 463 int destroy; 464 465 cp = bp2->bio_from; 466 sc = cp->private; 467 bp = bp2->bio_parent; 468 bp->bio_error = bp2->bio_error; 469 if (bp->bio_error != 0) { 470 g_trace(G_T_BIO, "g_dev_done(%p) had error %d", 471 bp2, bp->bio_error); 472 bp->bio_flags |= BIO_ERROR; 473 } else { 474 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", 475 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed); 476 } 477 bp->bio_resid = bp->bio_length - bp2->bio_completed; 478 bp->bio_completed = bp2->bio_completed; 479 g_destroy_bio(bp2); 480 destroy = 0; 481 mtx_lock(&sc->sc_mtx); 482 if ((--sc->sc_active) == 0) { 483 if (sc->sc_open == 0) 484 wakeup(&sc->sc_active); 485 if (sc->sc_dev == NULL) 486 destroy = 1; 487 } 488 mtx_unlock(&sc->sc_mtx); 489 if (destroy) 490 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL); 491 biodone(bp); 492 } 493 494 static void 495 g_dev_strategy(struct bio *bp) 496 { 497 struct g_consumer *cp; 498 struct bio *bp2; 499 struct cdev *dev; 500 struct g_dev_softc *sc; 501 502 KASSERT(bp->bio_cmd == BIO_READ || 503 bp->bio_cmd == BIO_WRITE || 504 bp->bio_cmd == BIO_DELETE || 505 bp->bio_cmd == BIO_FLUSH, 506 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); 507 dev = bp->bio_dev; 508 cp = dev->si_drv2; 509 sc = cp->private; 510 KASSERT(cp->acr || cp->acw, 511 ("Consumer with zero access count in g_dev_strategy")); 512 #ifdef INVARIANTS 513 if ((bp->bio_offset % cp->provider->sectorsize) != 0 || 514 (bp->bio_bcount % cp->provider->sectorsize) != 0) { 515 bp->bio_resid = bp->bio_bcount; 516 biofinish(bp, NULL, EINVAL); 517 return; 518 } 519 #endif 520 mtx_lock(&sc->sc_mtx); 521 KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy")); 522 sc->sc_active++; 523 mtx_unlock(&sc->sc_mtx); 524 525 for (;;) { 526 /* 527 * XXX: This is not an ideal solution, but I belive it to 528 * XXX: deadlock safe, all things considered. 529 */ 530 bp2 = g_clone_bio(bp); 531 if (bp2 != NULL) 532 break; 533 pause("gdstrat", hz / 10); 534 } 535 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); 536 bp2->bio_done = g_dev_done; 537 g_trace(G_T_BIO, 538 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", 539 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, 540 bp2->bio_data, bp2->bio_cmd); 541 g_io_request(bp2, cp); 542 KASSERT(cp->acr || cp->acw, 543 ("g_dev_strategy raced with g_dev_close and lost")); 544 545 } 546 547 /* 548 * g_dev_callback() 549 * 550 * Called by devfs when asynchronous device destruction is completed. 551 * - Mark that we have no attached device any more. 552 * - If there are no outstanding requests, schedule geom destruction. 553 * Otherwise destruction will be scheduled later by g_dev_done(). 554 */ 555 556 static void 557 g_dev_callback(void *arg) 558 { 559 struct g_consumer *cp; 560 struct g_dev_softc *sc; 561 int destroy; 562 563 cp = arg; 564 sc = cp->private; 565 g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name); 566 567 mtx_lock(&sc->sc_mtx); 568 sc->sc_dev = NULL; 569 sc->sc_alias = NULL; 570 destroy = (sc->sc_active == 0); 571 mtx_unlock(&sc->sc_mtx); 572 if (destroy) 573 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL); 574 } 575 576 /* 577 * g_dev_orphan() 578 * 579 * Called from below when the provider orphaned us. 580 * - Clear any dump settings. 581 * - Request asynchronous device destruction to prevent any more requests 582 * from coming in. The provider is already marked with an error, so 583 * anything which comes in in the interrim will be returned immediately. 584 */ 585 586 static void 587 g_dev_orphan(struct g_consumer *cp) 588 { 589 struct cdev *dev; 590 struct g_dev_softc *sc; 591 592 g_topology_assert(); 593 sc = cp->private; 594 dev = sc->sc_dev; 595 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name); 596 597 /* Reset any dump-area set on this device */ 598 if (dev->si_flags & SI_DUMPDEV) 599 set_dumper(NULL, NULL); 600 601 /* Destroy the struct cdev *so we get no more requests */ 602 destroy_dev_sched_cb(dev, g_dev_callback, cp); 603 } 604 605 DECLARE_GEOM_CLASS(g_dev_class, g_dev); 606