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 <sys/sysctl.h> 56 #include <geom/geom.h> 57 #include <geom/geom_int.h> 58 #include <machine/stdarg.h> 59 60 struct g_dev_softc { 61 struct mtx sc_mtx; 62 struct cdev *sc_dev; 63 struct cdev *sc_alias; 64 int sc_open; 65 int sc_active; 66 }; 67 68 static d_open_t g_dev_open; 69 static d_close_t g_dev_close; 70 static d_strategy_t g_dev_strategy; 71 static d_ioctl_t g_dev_ioctl; 72 73 static struct cdevsw g_dev_cdevsw = { 74 .d_version = D_VERSION, 75 .d_open = g_dev_open, 76 .d_close = g_dev_close, 77 .d_read = physread, 78 .d_write = physwrite, 79 .d_ioctl = g_dev_ioctl, 80 .d_strategy = g_dev_strategy, 81 .d_name = "g_dev", 82 .d_flags = D_DISK | D_TRACKCLOSE, 83 }; 84 85 static g_taste_t g_dev_taste; 86 static g_orphan_t g_dev_orphan; 87 static g_attrchanged_t g_dev_attrchanged; 88 89 static struct g_class g_dev_class = { 90 .name = "DEV", 91 .version = G_VERSION, 92 .taste = g_dev_taste, 93 .orphan = g_dev_orphan, 94 .attrchanged = g_dev_attrchanged 95 }; 96 97 /* 98 * We target 262144 (8 x 32768) sectors by default as this significantly 99 * increases the throughput on commonly used SSD's with a marginal 100 * increase in non-interruptible request latency. 101 */ 102 static uint64_t g_dev_del_max_sectors = 262144; 103 SYSCTL_DECL(_kern_geom); 104 SYSCTL_NODE(_kern_geom, OID_AUTO, dev, CTLFLAG_RW, 0, "GEOM_DEV stuff"); 105 SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW, 106 &g_dev_del_max_sectors, 0, "Maximum number of sectors in a single " 107 "delete request sent to the provider. Larger requests are chunked " 108 "so they can be interrupted. (0 = disable chunking)"); 109 110 static void 111 g_dev_destroy(void *arg, int flags __unused) 112 { 113 struct g_consumer *cp; 114 struct g_geom *gp; 115 struct g_dev_softc *sc; 116 117 g_topology_assert(); 118 cp = arg; 119 gp = cp->geom; 120 sc = cp->private; 121 g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name); 122 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 123 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 124 g_detach(cp); 125 g_destroy_consumer(cp); 126 g_destroy_geom(gp); 127 mtx_destroy(&sc->sc_mtx); 128 g_free(sc); 129 } 130 131 void 132 g_dev_print(void) 133 { 134 struct g_geom *gp; 135 char const *p = ""; 136 137 LIST_FOREACH(gp, &g_dev_class.geom, geom) { 138 printf("%s%s", p, gp->name); 139 p = " "; 140 } 141 printf("\n"); 142 } 143 144 static void 145 g_dev_attrchanged(struct g_consumer *cp, const char *attr) 146 { 147 struct g_dev_softc *sc; 148 struct cdev *dev; 149 char buf[SPECNAMELEN + 6]; 150 151 sc = cp->private; 152 if (strcmp(attr, "GEOM::media") == 0) { 153 dev = sc->sc_dev; 154 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name); 155 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK); 156 dev = sc->sc_alias; 157 if (dev != NULL) { 158 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name); 159 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, 160 M_WAITOK); 161 } 162 return; 163 } 164 165 if (strcmp(attr, "GEOM::physpath") != 0) 166 return; 167 168 if (g_access(cp, 1, 0, 0) == 0) { 169 char *physpath; 170 int error, physpath_len; 171 172 physpath_len = MAXPATHLEN; 173 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO); 174 error = 175 g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath); 176 g_access(cp, -1, 0, 0); 177 if (error == 0 && strlen(physpath) != 0) { 178 struct cdev *old_alias_dev; 179 struct cdev **alias_devp; 180 181 dev = sc->sc_dev; 182 old_alias_dev = sc->sc_alias; 183 alias_devp = (struct cdev **)&sc->sc_alias; 184 make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp, 185 dev, old_alias_dev, physpath); 186 } else if (sc->sc_alias) { 187 destroy_dev((struct cdev *)sc->sc_alias); 188 sc->sc_alias = NULL; 189 } 190 g_free(physpath); 191 } 192 } 193 194 struct g_provider * 195 g_dev_getprovider(struct cdev *dev) 196 { 197 struct g_consumer *cp; 198 199 g_topology_assert(); 200 if (dev == NULL) 201 return (NULL); 202 if (dev->si_devsw != &g_dev_cdevsw) 203 return (NULL); 204 cp = dev->si_drv2; 205 return (cp->provider); 206 } 207 208 static struct g_geom * 209 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused) 210 { 211 struct g_geom *gp; 212 struct g_consumer *cp; 213 struct g_dev_softc *sc; 214 int error, len; 215 struct cdev *dev, *adev; 216 char buf[64], *val; 217 218 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name); 219 g_topology_assert(); 220 gp = g_new_geomf(mp, "%s", pp->name); 221 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO); 222 mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF); 223 cp = g_new_consumer(gp); 224 cp->private = sc; 225 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; 226 error = g_attach(cp, pp); 227 KASSERT(error == 0, 228 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error)); 229 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev, 230 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name); 231 if (error != 0) { 232 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n", 233 __func__, gp->name, error); 234 g_detach(cp); 235 g_destroy_consumer(cp); 236 g_destroy_geom(gp); 237 mtx_destroy(&sc->sc_mtx); 238 g_free(sc); 239 return (NULL); 240 } 241 dev->si_flags |= SI_UNMAPPED; 242 sc->sc_dev = dev; 243 244 /* Search for device alias name and create it if found. */ 245 adev = NULL; 246 for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) { 247 snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name); 248 buf[14 + len] = 0; 249 val = getenv(buf); 250 if (val != NULL) { 251 snprintf(buf, sizeof(buf), "%s%s", 252 val, gp->name + len); 253 freeenv(val); 254 if ((make_dev_alias_p(MAKEDEV_CHECKNAME|MAKEDEV_WAITOK, 255 &adev, dev, "%s", buf)) != 0) 256 printf("Warning: unable to create device " 257 "alias %s\n", buf); 258 break; 259 } 260 } 261 262 dev->si_iosize_max = MAXPHYS; 263 dev->si_drv2 = cp; 264 if (adev != NULL) { 265 adev->si_iosize_max = MAXPHYS; 266 adev->si_drv2 = cp; 267 adev->si_flags |= SI_UNMAPPED; 268 } 269 270 g_dev_attrchanged(cp, "GEOM::physpath"); 271 272 return (gp); 273 } 274 275 static int 276 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td) 277 { 278 struct g_consumer *cp; 279 struct g_dev_softc *sc; 280 int error, r, w, e; 281 282 cp = dev->si_drv2; 283 if (cp == NULL) 284 return(ENXIO); /* g_dev_taste() not done yet */ 285 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", 286 cp->geom->name, flags, fmt, td); 287 288 r = flags & FREAD ? 1 : 0; 289 w = flags & FWRITE ? 1 : 0; 290 #ifdef notyet 291 e = flags & O_EXCL ? 1 : 0; 292 #else 293 e = 0; 294 #endif 295 if (w) { 296 /* 297 * When running in very secure mode, do not allow 298 * opens for writing of any disks. 299 */ 300 error = securelevel_ge(td->td_ucred, 2); 301 if (error) 302 return (error); 303 } 304 g_topology_lock(); 305 error = g_access(cp, r, w, e); 306 g_topology_unlock(); 307 if (error == 0) { 308 sc = cp->private; 309 mtx_lock(&sc->sc_mtx); 310 if (sc->sc_open == 0 && sc->sc_active != 0) 311 wakeup(&sc->sc_active); 312 sc->sc_open += r + w + e; 313 mtx_unlock(&sc->sc_mtx); 314 } 315 return(error); 316 } 317 318 static int 319 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td) 320 { 321 struct g_consumer *cp; 322 struct g_dev_softc *sc; 323 int error, r, w, e; 324 325 cp = dev->si_drv2; 326 if (cp == NULL) 327 return(ENXIO); 328 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)", 329 cp->geom->name, flags, fmt, td); 330 331 r = flags & FREAD ? -1 : 0; 332 w = flags & FWRITE ? -1 : 0; 333 #ifdef notyet 334 e = flags & O_EXCL ? -1 : 0; 335 #else 336 e = 0; 337 #endif 338 sc = cp->private; 339 mtx_lock(&sc->sc_mtx); 340 sc->sc_open += r + w + e; 341 while (sc->sc_open == 0 && sc->sc_active != 0) 342 msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0); 343 mtx_unlock(&sc->sc_mtx); 344 g_topology_lock(); 345 error = g_access(cp, r, w, e); 346 g_topology_unlock(); 347 return (error); 348 } 349 350 /* 351 * XXX: Until we have unmessed the ioctl situation, there is a race against 352 * XXX: a concurrent orphanization. We cannot close it by holding topology 353 * XXX: since that would prevent us from doing our job, and stalling events 354 * XXX: will break (actually: stall) the BSD disklabel hacks. 355 */ 356 static int 357 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 358 { 359 struct g_consumer *cp; 360 struct g_provider *pp; 361 struct g_kerneldump kd; 362 off_t offset, length, chunk; 363 int i, error; 364 u_int u; 365 366 cp = dev->si_drv2; 367 pp = cp->provider; 368 369 error = 0; 370 KASSERT(cp->acr || cp->acw, 371 ("Consumer with zero access count in g_dev_ioctl")); 372 373 i = IOCPARM_LEN(cmd); 374 switch (cmd) { 375 case DIOCGSECTORSIZE: 376 *(u_int *)data = cp->provider->sectorsize; 377 if (*(u_int *)data == 0) 378 error = ENOENT; 379 break; 380 case DIOCGMEDIASIZE: 381 *(off_t *)data = cp->provider->mediasize; 382 if (*(off_t *)data == 0) 383 error = ENOENT; 384 break; 385 case DIOCGFWSECTORS: 386 error = g_io_getattr("GEOM::fwsectors", cp, &i, data); 387 if (error == 0 && *(u_int *)data == 0) 388 error = ENOENT; 389 break; 390 case DIOCGFWHEADS: 391 error = g_io_getattr("GEOM::fwheads", cp, &i, data); 392 if (error == 0 && *(u_int *)data == 0) 393 error = ENOENT; 394 break; 395 case DIOCGFRONTSTUFF: 396 error = g_io_getattr("GEOM::frontstuff", cp, &i, data); 397 break; 398 case DIOCSKERNELDUMP: 399 u = *((u_int *)data); 400 if (!u) { 401 set_dumper(NULL, NULL); 402 error = 0; 403 break; 404 } 405 kd.offset = 0; 406 kd.length = OFF_MAX; 407 i = sizeof kd; 408 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd); 409 if (!error) { 410 error = set_dumper(&kd.di, devtoname(dev)); 411 if (!error) 412 dev->si_flags |= SI_DUMPDEV; 413 } 414 break; 415 case DIOCGFLUSH: 416 error = g_io_flush(cp); 417 break; 418 case DIOCGDELETE: 419 offset = ((off_t *)data)[0]; 420 length = ((off_t *)data)[1]; 421 if ((offset % cp->provider->sectorsize) != 0 || 422 (length % cp->provider->sectorsize) != 0 || length <= 0) { 423 printf("%s: offset=%jd length=%jd\n", __func__, offset, 424 length); 425 error = EINVAL; 426 break; 427 } 428 while (length > 0) { 429 chunk = length; 430 if (g_dev_del_max_sectors != 0 && chunk > 431 g_dev_del_max_sectors * cp->provider->sectorsize) { 432 chunk = g_dev_del_max_sectors * 433 cp->provider->sectorsize; 434 } 435 error = g_delete_data(cp, offset, chunk); 436 length -= chunk; 437 offset += chunk; 438 if (error) 439 break; 440 /* 441 * Since the request size can be large, the service 442 * time can be is likewise. We make this ioctl 443 * interruptible by checking for signals for each bio. 444 */ 445 if (SIGPENDING(td)) 446 break; 447 } 448 break; 449 case DIOCGIDENT: 450 error = g_io_getattr("GEOM::ident", cp, &i, data); 451 break; 452 case DIOCGPROVIDERNAME: 453 if (pp == NULL) 454 return (ENOENT); 455 strlcpy(data, pp->name, i); 456 break; 457 case DIOCGSTRIPESIZE: 458 *(off_t *)data = cp->provider->stripesize; 459 break; 460 case DIOCGSTRIPEOFFSET: 461 *(off_t *)data = cp->provider->stripeoffset; 462 break; 463 case DIOCGPHYSPATH: 464 error = g_io_getattr("GEOM::physpath", cp, &i, data); 465 if (error == 0 && *(char *)data == '\0') 466 error = ENOENT; 467 break; 468 default: 469 if (cp->provider->geom->ioctl != NULL) { 470 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td); 471 } else { 472 error = ENOIOCTL; 473 } 474 } 475 476 return (error); 477 } 478 479 static void 480 g_dev_done(struct bio *bp2) 481 { 482 struct g_consumer *cp; 483 struct g_dev_softc *sc; 484 struct bio *bp; 485 int destroy; 486 487 cp = bp2->bio_from; 488 sc = cp->private; 489 bp = bp2->bio_parent; 490 bp->bio_error = bp2->bio_error; 491 bp->bio_completed = bp2->bio_completed; 492 bp->bio_resid = bp->bio_length - bp2->bio_completed; 493 if (bp2->bio_error != 0) { 494 g_trace(G_T_BIO, "g_dev_done(%p) had error %d", 495 bp2, bp2->bio_error); 496 bp->bio_flags |= BIO_ERROR; 497 } else { 498 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", 499 bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed); 500 } 501 g_destroy_bio(bp2); 502 destroy = 0; 503 mtx_lock(&sc->sc_mtx); 504 if ((--sc->sc_active) == 0) { 505 if (sc->sc_open == 0) 506 wakeup(&sc->sc_active); 507 if (sc->sc_dev == NULL) 508 destroy = 1; 509 } 510 mtx_unlock(&sc->sc_mtx); 511 if (destroy) 512 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL); 513 biodone(bp); 514 } 515 516 static void 517 g_dev_strategy(struct bio *bp) 518 { 519 struct g_consumer *cp; 520 struct bio *bp2; 521 struct cdev *dev; 522 struct g_dev_softc *sc; 523 524 KASSERT(bp->bio_cmd == BIO_READ || 525 bp->bio_cmd == BIO_WRITE || 526 bp->bio_cmd == BIO_DELETE || 527 bp->bio_cmd == BIO_FLUSH, 528 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); 529 dev = bp->bio_dev; 530 cp = dev->si_drv2; 531 sc = cp->private; 532 KASSERT(cp->acr || cp->acw, 533 ("Consumer with zero access count in g_dev_strategy")); 534 #ifdef INVARIANTS 535 if ((bp->bio_offset % cp->provider->sectorsize) != 0 || 536 (bp->bio_bcount % cp->provider->sectorsize) != 0) { 537 bp->bio_resid = bp->bio_bcount; 538 biofinish(bp, NULL, EINVAL); 539 return; 540 } 541 #endif 542 mtx_lock(&sc->sc_mtx); 543 KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy")); 544 sc->sc_active++; 545 mtx_unlock(&sc->sc_mtx); 546 547 for (;;) { 548 /* 549 * XXX: This is not an ideal solution, but I belive it to 550 * XXX: deadlock safe, all things considered. 551 */ 552 bp2 = g_clone_bio(bp); 553 if (bp2 != NULL) 554 break; 555 pause("gdstrat", hz / 10); 556 } 557 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); 558 bp2->bio_done = g_dev_done; 559 g_trace(G_T_BIO, 560 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", 561 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, 562 bp2->bio_data, bp2->bio_cmd); 563 g_io_request(bp2, cp); 564 KASSERT(cp->acr || cp->acw, 565 ("g_dev_strategy raced with g_dev_close and lost")); 566 567 } 568 569 /* 570 * g_dev_callback() 571 * 572 * Called by devfs when asynchronous device destruction is completed. 573 * - Mark that we have no attached device any more. 574 * - If there are no outstanding requests, schedule geom destruction. 575 * Otherwise destruction will be scheduled later by g_dev_done(). 576 */ 577 578 static void 579 g_dev_callback(void *arg) 580 { 581 struct g_consumer *cp; 582 struct g_dev_softc *sc; 583 int destroy; 584 585 cp = arg; 586 sc = cp->private; 587 g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name); 588 589 mtx_lock(&sc->sc_mtx); 590 sc->sc_dev = NULL; 591 sc->sc_alias = NULL; 592 destroy = (sc->sc_active == 0); 593 mtx_unlock(&sc->sc_mtx); 594 if (destroy) 595 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL); 596 } 597 598 /* 599 * g_dev_orphan() 600 * 601 * Called from below when the provider orphaned us. 602 * - Clear any dump settings. 603 * - Request asynchronous device destruction to prevent any more requests 604 * from coming in. The provider is already marked with an error, so 605 * anything which comes in in the interrim will be returned immediately. 606 */ 607 608 static void 609 g_dev_orphan(struct g_consumer *cp) 610 { 611 struct cdev *dev; 612 struct g_dev_softc *sc; 613 614 g_topology_assert(); 615 sc = cp->private; 616 dev = sc->sc_dev; 617 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name); 618 619 /* Reset any dump-area set on this device */ 620 if (dev->si_flags & SI_DUMPDEV) 621 set_dumper(NULL, NULL); 622 623 /* Destroy the struct cdev *so we get no more requests */ 624 destroy_dev_sched_cb(dev, g_dev_callback, cp); 625 } 626 627 DECLARE_GEOM_CLASS(g_dev_class, g_dev); 628