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 /* 60 * Use the consumer private field to reference a physdev alias (if any). 61 */ 62 #define cp_alias_dev private 63 64 static d_open_t g_dev_open; 65 static d_close_t g_dev_close; 66 static d_strategy_t g_dev_strategy; 67 static d_ioctl_t g_dev_ioctl; 68 69 static struct cdevsw g_dev_cdevsw = { 70 .d_version = D_VERSION, 71 .d_open = g_dev_open, 72 .d_close = g_dev_close, 73 .d_read = physread, 74 .d_write = physwrite, 75 .d_ioctl = g_dev_ioctl, 76 .d_strategy = g_dev_strategy, 77 .d_name = "g_dev", 78 .d_flags = D_DISK | D_TRACKCLOSE, 79 }; 80 81 static g_taste_t g_dev_taste; 82 static g_orphan_t g_dev_orphan; 83 static g_attrchanged_t g_dev_attrchanged; 84 85 static struct g_class g_dev_class = { 86 .name = "DEV", 87 .version = G_VERSION, 88 .taste = g_dev_taste, 89 .orphan = g_dev_orphan, 90 .attrchanged = g_dev_attrchanged 91 }; 92 93 void 94 g_dev_print(void) 95 { 96 struct g_geom *gp; 97 char const *p = ""; 98 99 LIST_FOREACH(gp, &g_dev_class.geom, geom) { 100 printf("%s%s", p, gp->name); 101 p = " "; 102 } 103 printf("\n"); 104 } 105 106 static void 107 g_dev_attrchanged(struct g_consumer *cp, const char *attr) 108 { 109 struct cdev *dev; 110 char buf[SPECNAMELEN + 6]; 111 112 if (strcmp(attr, "GEOM::media") == 0) { 113 dev = cp->geom->softc; 114 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name); 115 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK); 116 dev = cp->cp_alias_dev; 117 if (dev != NULL) { 118 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name); 119 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, 120 M_WAITOK); 121 } 122 return; 123 } 124 125 if (strcmp(attr, "GEOM::physpath") != 0) 126 return; 127 128 if (g_access(cp, 1, 0, 0) == 0) { 129 char *physpath; 130 int error, physpath_len; 131 132 physpath_len = MAXPATHLEN; 133 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO); 134 error = 135 g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath); 136 g_access(cp, -1, 0, 0); 137 if (error == 0 && strlen(physpath) != 0) { 138 struct cdev *old_alias_dev; 139 struct cdev **alias_devp; 140 141 dev = cp->geom->softc; 142 old_alias_dev = cp->cp_alias_dev; 143 alias_devp = (struct cdev **)&cp->cp_alias_dev; 144 make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp, 145 dev, old_alias_dev, physpath); 146 } else if (cp->cp_alias_dev) { 147 destroy_dev((struct cdev *)cp->cp_alias_dev); 148 cp->cp_alias_dev = NULL; 149 } 150 g_free(physpath); 151 } 152 } 153 154 struct g_provider * 155 g_dev_getprovider(struct cdev *dev) 156 { 157 struct g_consumer *cp; 158 159 g_topology_assert(); 160 if (dev == NULL) 161 return (NULL); 162 if (dev->si_devsw != &g_dev_cdevsw) 163 return (NULL); 164 cp = dev->si_drv2; 165 return (cp->provider); 166 } 167 168 static struct g_geom * 169 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused) 170 { 171 struct g_geom *gp; 172 struct g_consumer *cp; 173 int error, len; 174 struct cdev *dev, *adev; 175 char buf[64], *val; 176 177 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name); 178 g_topology_assert(); 179 gp = g_new_geomf(mp, "%s", pp->name); 180 cp = g_new_consumer(gp); 181 error = g_attach(cp, pp); 182 KASSERT(error == 0, 183 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error)); 184 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev, 185 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name); 186 if (error != 0) { 187 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n", 188 __func__, gp->name, error); 189 g_detach(cp); 190 g_destroy_consumer(cp); 191 g_destroy_geom(gp); 192 return (NULL); 193 } 194 195 /* Search for device alias name and create it if found. */ 196 adev = NULL; 197 for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) { 198 snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name); 199 buf[14 + len] = 0; 200 val = getenv(buf); 201 if (val != NULL) { 202 snprintf(buf, sizeof(buf), "%s%s", 203 val, gp->name + len); 204 freeenv(val); 205 make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, 206 &adev, dev, "%s", buf); 207 break; 208 } 209 } 210 211 dev->si_iosize_max = MAXPHYS; 212 gp->softc = dev; 213 dev->si_drv1 = gp; 214 dev->si_drv2 = cp; 215 if (adev != NULL) { 216 adev->si_iosize_max = MAXPHYS; 217 adev->si_drv1 = gp; 218 adev->si_drv2 = cp; 219 } 220 221 g_dev_attrchanged(cp, "GEOM::physpath"); 222 223 return (gp); 224 } 225 226 static int 227 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td) 228 { 229 struct g_geom *gp; 230 struct g_consumer *cp; 231 int error, r, w, e; 232 233 gp = dev->si_drv1; 234 cp = dev->si_drv2; 235 if (gp == NULL || cp == NULL || gp->softc != dev) 236 return(ENXIO); /* g_dev_taste() not done yet */ 237 238 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", 239 gp->name, flags, fmt, td); 240 241 r = flags & FREAD ? 1 : 0; 242 w = flags & FWRITE ? 1 : 0; 243 #ifdef notyet 244 e = flags & O_EXCL ? 1 : 0; 245 #else 246 e = 0; 247 #endif 248 if (w) { 249 /* 250 * When running in very secure mode, do not allow 251 * opens for writing of any disks. 252 */ 253 error = securelevel_ge(td->td_ucred, 2); 254 if (error) 255 return (error); 256 } 257 g_topology_lock(); 258 if (dev->si_devsw == NULL) 259 error = ENXIO; /* We were orphaned */ 260 else 261 error = g_access(cp, r, w, e); 262 g_topology_unlock(); 263 return(error); 264 } 265 266 static int 267 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td) 268 { 269 struct g_geom *gp; 270 struct g_consumer *cp; 271 int error, r, w, e, i; 272 273 gp = dev->si_drv1; 274 cp = dev->si_drv2; 275 if (gp == NULL || cp == NULL) 276 return(ENXIO); 277 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)", 278 gp->name, flags, fmt, td); 279 r = flags & FREAD ? -1 : 0; 280 w = flags & FWRITE ? -1 : 0; 281 #ifdef notyet 282 e = flags & O_EXCL ? -1 : 0; 283 #else 284 e = 0; 285 #endif 286 g_topology_lock(); 287 if (dev->si_devsw == NULL) 288 error = ENXIO; /* We were orphaned */ 289 else 290 error = g_access(cp, r, w, e); 291 for (i = 0; i < 10 * hz;) { 292 if (cp->acr != 0 || cp->acw != 0) 293 break; 294 if (cp->nstart == cp->nend) 295 break; 296 pause("gdevwclose", hz / 10); 297 i += hz / 10; 298 } 299 if (cp->acr == 0 && cp->acw == 0 && cp->nstart != cp->nend) { 300 printf("WARNING: Final close of geom_dev(%s) %s %s\n", 301 gp->name, 302 "still has outstanding I/O after 10 seconds.", 303 "Completing close anyway, panic may happen later."); 304 } 305 g_topology_unlock(); 306 return (error); 307 } 308 309 /* 310 * XXX: Until we have unmessed the ioctl situation, there is a race against 311 * XXX: a concurrent orphanization. We cannot close it by holding topology 312 * XXX: since that would prevent us from doing our job, and stalling events 313 * XXX: will break (actually: stall) the BSD disklabel hacks. 314 */ 315 static int 316 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 317 { 318 struct g_geom *gp; 319 struct g_consumer *cp; 320 struct g_provider *pp; 321 struct g_kerneldump kd; 322 off_t offset, length, chunk; 323 int i, error; 324 u_int u; 325 326 gp = dev->si_drv1; 327 cp = dev->si_drv2; 328 pp = cp->provider; 329 330 error = 0; 331 KASSERT(cp->acr || cp->acw, 332 ("Consumer with zero access count in g_dev_ioctl")); 333 334 i = IOCPARM_LEN(cmd); 335 switch (cmd) { 336 case DIOCGSECTORSIZE: 337 *(u_int *)data = cp->provider->sectorsize; 338 if (*(u_int *)data == 0) 339 error = ENOENT; 340 break; 341 case DIOCGMEDIASIZE: 342 *(off_t *)data = cp->provider->mediasize; 343 if (*(off_t *)data == 0) 344 error = ENOENT; 345 break; 346 case DIOCGFWSECTORS: 347 error = g_io_getattr("GEOM::fwsectors", cp, &i, data); 348 if (error == 0 && *(u_int *)data == 0) 349 error = ENOENT; 350 break; 351 case DIOCGFWHEADS: 352 error = g_io_getattr("GEOM::fwheads", cp, &i, data); 353 if (error == 0 && *(u_int *)data == 0) 354 error = ENOENT; 355 break; 356 case DIOCGFRONTSTUFF: 357 error = g_io_getattr("GEOM::frontstuff", cp, &i, data); 358 break; 359 case DIOCSKERNELDUMP: 360 u = *((u_int *)data); 361 if (!u) { 362 set_dumper(NULL, NULL); 363 error = 0; 364 break; 365 } 366 kd.offset = 0; 367 kd.length = OFF_MAX; 368 i = sizeof kd; 369 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd); 370 if (!error) { 371 error = set_dumper(&kd.di, devtoname(dev)); 372 if (!error) 373 dev->si_flags |= SI_DUMPDEV; 374 } 375 break; 376 case DIOCGFLUSH: 377 error = g_io_flush(cp); 378 break; 379 case DIOCGDELETE: 380 offset = ((off_t *)data)[0]; 381 length = ((off_t *)data)[1]; 382 if ((offset % cp->provider->sectorsize) != 0 || 383 (length % cp->provider->sectorsize) != 0 || length <= 0) { 384 printf("%s: offset=%jd length=%jd\n", __func__, offset, 385 length); 386 error = EINVAL; 387 break; 388 } 389 while (length > 0) { 390 chunk = length; 391 if (chunk > 65536 * cp->provider->sectorsize) 392 chunk = 65536 * cp->provider->sectorsize; 393 error = g_delete_data(cp, offset, chunk); 394 length -= chunk; 395 offset += chunk; 396 if (error) 397 break; 398 /* 399 * Since the request size is unbounded, the service 400 * time is likewise. We make this ioctl interruptible 401 * by checking for signals for each bio. 402 */ 403 if (SIGPENDING(td)) 404 break; 405 } 406 break; 407 case DIOCGIDENT: 408 error = g_io_getattr("GEOM::ident", cp, &i, data); 409 break; 410 case DIOCGPROVIDERNAME: 411 if (pp == NULL) 412 return (ENOENT); 413 strlcpy(data, pp->name, i); 414 break; 415 case DIOCGSTRIPESIZE: 416 *(off_t *)data = cp->provider->stripesize; 417 break; 418 case DIOCGSTRIPEOFFSET: 419 *(off_t *)data = cp->provider->stripeoffset; 420 break; 421 case DIOCGPHYSPATH: 422 error = g_io_getattr("GEOM::physpath", cp, &i, data); 423 if (error == 0 && *(char *)data == '\0') 424 error = ENOENT; 425 break; 426 default: 427 if (cp->provider->geom->ioctl != NULL) { 428 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td); 429 } else { 430 error = ENOIOCTL; 431 } 432 } 433 434 return (error); 435 } 436 437 static void 438 g_dev_done(struct bio *bp2) 439 { 440 struct bio *bp; 441 442 bp = bp2->bio_parent; 443 bp->bio_error = bp2->bio_error; 444 if (bp->bio_error != 0) { 445 g_trace(G_T_BIO, "g_dev_done(%p) had error %d", 446 bp2, bp->bio_error); 447 bp->bio_flags |= BIO_ERROR; 448 } else { 449 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", 450 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed); 451 } 452 bp->bio_resid = bp->bio_length - bp2->bio_completed; 453 bp->bio_completed = bp2->bio_completed; 454 g_destroy_bio(bp2); 455 biodone(bp); 456 } 457 458 static void 459 g_dev_strategy(struct bio *bp) 460 { 461 struct g_consumer *cp; 462 struct bio *bp2; 463 struct cdev *dev; 464 465 KASSERT(bp->bio_cmd == BIO_READ || 466 bp->bio_cmd == BIO_WRITE || 467 bp->bio_cmd == BIO_DELETE, 468 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); 469 dev = bp->bio_dev; 470 cp = dev->si_drv2; 471 KASSERT(cp->acr || cp->acw, 472 ("Consumer with zero access count in g_dev_strategy")); 473 #ifdef INVARIANTS 474 if ((bp->bio_offset % cp->provider->sectorsize) != 0 || 475 (bp->bio_bcount % cp->provider->sectorsize) != 0) { 476 bp->bio_resid = bp->bio_bcount; 477 biofinish(bp, NULL, EINVAL); 478 return; 479 } 480 #endif 481 for (;;) { 482 /* 483 * XXX: This is not an ideal solution, but I belive it to 484 * XXX: deadlock safe, all things considered. 485 */ 486 bp2 = g_clone_bio(bp); 487 if (bp2 != NULL) 488 break; 489 pause("gdstrat", hz / 10); 490 } 491 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); 492 bp2->bio_done = g_dev_done; 493 g_trace(G_T_BIO, 494 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", 495 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, 496 bp2->bio_data, bp2->bio_cmd); 497 g_io_request(bp2, cp); 498 KASSERT(cp->acr || cp->acw, 499 ("g_dev_strategy raced with g_dev_close and lost")); 500 501 } 502 503 /* 504 * g_dev_orphan() 505 * 506 * Called from below when the provider orphaned us. 507 * - Clear any dump settings. 508 * - Destroy the struct cdev to prevent any more request from coming in. The 509 * provider is already marked with an error, so anything which comes in 510 * in the interrim will be returned immediately. 511 * - Wait for any outstanding I/O to finish. 512 * - Set our access counts to zero, whatever they were. 513 * - Detach and self-destruct. 514 */ 515 516 static void 517 g_dev_orphan(struct g_consumer *cp) 518 { 519 struct g_geom *gp; 520 struct cdev *dev; 521 522 g_topology_assert(); 523 gp = cp->geom; 524 dev = gp->softc; 525 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name); 526 527 /* Reset any dump-area set on this device */ 528 if (dev->si_flags & SI_DUMPDEV) 529 set_dumper(NULL, NULL); 530 531 /* Destroy the struct cdev *so we get no more requests */ 532 destroy_dev(dev); 533 534 /* Wait for the cows to come home */ 535 while (cp->nstart != cp->nend) 536 pause("gdevorphan", hz / 10); 537 538 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 539 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 540 541 g_detach(cp); 542 g_destroy_consumer(cp); 543 g_destroy_geom(gp); 544 } 545 546 DECLARE_GEOM_CLASS(g_dev_class, g_dev); 547