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/bio.h> 45 #include <sys/lock.h> 46 #include <sys/mutex.h> 47 #include <sys/proc.h> 48 #include <sys/errno.h> 49 #include <sys/time.h> 50 #include <sys/disk.h> 51 #include <sys/fcntl.h> 52 #include <sys/limits.h> 53 #include <geom/geom.h> 54 #include <geom/geom_int.h> 55 56 static d_open_t g_dev_open; 57 static d_close_t g_dev_close; 58 static d_strategy_t g_dev_strategy; 59 static d_ioctl_t g_dev_ioctl; 60 61 static struct cdevsw g_dev_cdevsw = { 62 .d_version = D_VERSION, 63 .d_open = g_dev_open, 64 .d_close = g_dev_close, 65 .d_read = physread, 66 .d_write = physwrite, 67 .d_ioctl = g_dev_ioctl, 68 .d_strategy = g_dev_strategy, 69 .d_name = "g_dev", 70 .d_flags = D_DISK | D_TRACKCLOSE, 71 }; 72 73 static g_taste_t g_dev_taste; 74 static g_orphan_t g_dev_orphan; 75 76 static struct g_class g_dev_class = { 77 .name = "DEV", 78 .version = G_VERSION, 79 .taste = g_dev_taste, 80 .orphan = g_dev_orphan, 81 }; 82 83 void 84 g_dev_print(void) 85 { 86 struct g_geom *gp; 87 char const *p = ""; 88 89 LIST_FOREACH(gp, &g_dev_class.geom, geom) { 90 printf("%s%s", p, gp->name); 91 p = " "; 92 } 93 printf("\n"); 94 } 95 96 struct g_provider * 97 g_dev_getprovider(struct cdev *dev) 98 { 99 struct g_consumer *cp; 100 101 g_topology_assert(); 102 if (dev == NULL) 103 return (NULL); 104 if (dev->si_devsw != &g_dev_cdevsw) 105 return (NULL); 106 cp = dev->si_drv2; 107 return (cp->provider); 108 } 109 110 111 static struct g_geom * 112 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused) 113 { 114 struct g_geom *gp; 115 struct g_consumer *cp; 116 int error, len; 117 struct cdev *dev, *adev; 118 char buf[64], *val; 119 120 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name); 121 g_topology_assert(); 122 LIST_FOREACH(cp, &pp->consumers, consumers) 123 if (cp->geom->class == mp) 124 return (NULL); 125 gp = g_new_geomf(mp, pp->name); 126 cp = g_new_consumer(gp); 127 error = g_attach(cp, pp); 128 KASSERT(error == 0, 129 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error)); 130 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev, 131 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name); 132 if (error != 0) { 133 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n", 134 __func__, gp->name, error); 135 g_detach(cp); 136 g_destroy_consumer(cp); 137 g_destroy_geom(gp); 138 return (NULL); 139 } 140 141 /* Search for device alias name and create it if found. */ 142 adev = NULL; 143 for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) { 144 snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name); 145 buf[14 + len] = 0; 146 val = getenv(buf); 147 if (val != NULL) { 148 snprintf(buf, sizeof(buf), "%s%s", 149 val, gp->name + len); 150 freeenv(val); 151 make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, 152 &adev, dev, "%s", buf); 153 break; 154 } 155 } 156 157 if (pp->flags & G_PF_CANDELETE) 158 dev->si_flags |= SI_CANDELETE; 159 dev->si_iosize_max = MAXPHYS; 160 gp->softc = dev; 161 dev->si_drv1 = gp; 162 dev->si_drv2 = cp; 163 if (adev != NULL) { 164 if (pp->flags & G_PF_CANDELETE) 165 adev->si_flags |= SI_CANDELETE; 166 adev->si_iosize_max = MAXPHYS; 167 adev->si_drv1 = gp; 168 adev->si_drv2 = cp; 169 } 170 return (gp); 171 } 172 173 static int 174 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td) 175 { 176 struct g_geom *gp; 177 struct g_consumer *cp; 178 int error, r, w, e; 179 180 gp = dev->si_drv1; 181 cp = dev->si_drv2; 182 if (gp == NULL || cp == NULL || gp->softc != dev) 183 return(ENXIO); /* g_dev_taste() not done yet */ 184 185 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", 186 gp->name, flags, fmt, td); 187 188 r = flags & FREAD ? 1 : 0; 189 w = flags & FWRITE ? 1 : 0; 190 #ifdef notyet 191 e = flags & O_EXCL ? 1 : 0; 192 #else 193 e = 0; 194 #endif 195 if (w) { 196 /* 197 * When running in very secure mode, do not allow 198 * opens for writing of any disks. 199 */ 200 error = securelevel_ge(td->td_ucred, 2); 201 if (error) 202 return (error); 203 } 204 g_topology_lock(); 205 if (dev->si_devsw == NULL) 206 error = ENXIO; /* We were orphaned */ 207 else 208 error = g_access(cp, r, w, e); 209 g_topology_unlock(); 210 return(error); 211 } 212 213 static int 214 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td) 215 { 216 struct g_geom *gp; 217 struct g_consumer *cp; 218 int error, r, w, e, i; 219 220 gp = dev->si_drv1; 221 cp = dev->si_drv2; 222 if (gp == NULL || cp == NULL) 223 return(ENXIO); 224 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)", 225 gp->name, flags, fmt, td); 226 r = flags & FREAD ? -1 : 0; 227 w = flags & FWRITE ? -1 : 0; 228 #ifdef notyet 229 e = flags & O_EXCL ? -1 : 0; 230 #else 231 e = 0; 232 #endif 233 g_topology_lock(); 234 if (dev->si_devsw == NULL) 235 error = ENXIO; /* We were orphaned */ 236 else 237 error = g_access(cp, r, w, e); 238 for (i = 0; i < 10 * hz;) { 239 if (cp->acr != 0 || cp->acw != 0) 240 break; 241 if (cp->nstart == cp->nend) 242 break; 243 pause("gdevwclose", hz / 10); 244 i += hz / 10; 245 } 246 if (cp->acr == 0 && cp->acw == 0 && cp->nstart != cp->nend) { 247 printf("WARNING: Final close of geom_dev(%s) %s %s\n", 248 gp->name, 249 "still has outstanding I/O after 10 seconds.", 250 "Completing close anyway, panic may happen later."); 251 } 252 g_topology_unlock(); 253 return (error); 254 } 255 256 /* 257 * XXX: Until we have unmessed the ioctl situation, there is a race against 258 * XXX: a concurrent orphanization. We cannot close it by holding topology 259 * XXX: since that would prevent us from doing our job, and stalling events 260 * XXX: will break (actually: stall) the BSD disklabel hacks. 261 */ 262 static int 263 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 264 { 265 struct g_geom *gp; 266 struct g_consumer *cp; 267 struct g_provider *pp; 268 struct g_kerneldump kd; 269 off_t offset, length, chunk; 270 int i, error; 271 u_int u; 272 273 gp = dev->si_drv1; 274 cp = dev->si_drv2; 275 pp = cp->provider; 276 277 error = 0; 278 KASSERT(cp->acr || cp->acw, 279 ("Consumer with zero access count in g_dev_ioctl")); 280 281 i = IOCPARM_LEN(cmd); 282 switch (cmd) { 283 case DIOCGSECTORSIZE: 284 *(u_int *)data = cp->provider->sectorsize; 285 if (*(u_int *)data == 0) 286 error = ENOENT; 287 break; 288 case DIOCGMEDIASIZE: 289 *(off_t *)data = cp->provider->mediasize; 290 if (*(off_t *)data == 0) 291 error = ENOENT; 292 break; 293 case DIOCGFWSECTORS: 294 error = g_io_getattr("GEOM::fwsectors", cp, &i, data); 295 if (error == 0 && *(u_int *)data == 0) 296 error = ENOENT; 297 break; 298 case DIOCGFWHEADS: 299 error = g_io_getattr("GEOM::fwheads", cp, &i, data); 300 if (error == 0 && *(u_int *)data == 0) 301 error = ENOENT; 302 break; 303 case DIOCGFRONTSTUFF: 304 error = g_io_getattr("GEOM::frontstuff", cp, &i, data); 305 break; 306 case DIOCSKERNELDUMP: 307 u = *((u_int *)data); 308 if (!u) { 309 set_dumper(NULL); 310 error = 0; 311 break; 312 } 313 kd.offset = 0; 314 kd.length = OFF_MAX; 315 i = sizeof kd; 316 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd); 317 if (!error) { 318 error = set_dumper(&kd.di); 319 if (!error) 320 dev->si_flags |= SI_DUMPDEV; 321 } 322 break; 323 case DIOCGFLUSH: 324 error = g_io_flush(cp); 325 break; 326 case DIOCGDELETE: 327 offset = ((off_t *)data)[0]; 328 length = ((off_t *)data)[1]; 329 if ((offset % cp->provider->sectorsize) != 0 || 330 (length % cp->provider->sectorsize) != 0 || length <= 0) { 331 printf("%s: offset=%jd length=%jd\n", __func__, offset, 332 length); 333 error = EINVAL; 334 break; 335 } 336 while (length > 0) { 337 chunk = length; 338 if (chunk > 65536 * cp->provider->sectorsize) 339 chunk = 65536 * cp->provider->sectorsize; 340 error = g_delete_data(cp, offset, chunk); 341 length -= chunk; 342 offset += chunk; 343 if (error) 344 break; 345 /* 346 * Since the request size is unbounded, the service 347 * time is likewise. We make this ioctl interruptible 348 * by checking for signals for each bio. 349 */ 350 if (SIGPENDING(td)) 351 break; 352 } 353 break; 354 case DIOCGIDENT: 355 error = g_io_getattr("GEOM::ident", cp, &i, data); 356 break; 357 case DIOCGPROVIDERNAME: 358 if (pp == NULL) 359 return (ENOENT); 360 strlcpy(data, pp->name, i); 361 break; 362 case DIOCGSTRIPESIZE: 363 *(off_t *)data = cp->provider->stripesize; 364 break; 365 case DIOCGSTRIPEOFFSET: 366 *(off_t *)data = cp->provider->stripeoffset; 367 break; 368 default: 369 if (cp->provider->geom->ioctl != NULL) { 370 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td); 371 } else { 372 error = ENOIOCTL; 373 } 374 } 375 376 return (error); 377 } 378 379 static void 380 g_dev_done(struct bio *bp2) 381 { 382 struct bio *bp; 383 384 bp = bp2->bio_parent; 385 bp->bio_error = bp2->bio_error; 386 if (bp->bio_error != 0) { 387 g_trace(G_T_BIO, "g_dev_done(%p) had error %d", 388 bp2, bp->bio_error); 389 bp->bio_flags |= BIO_ERROR; 390 } else { 391 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", 392 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed); 393 } 394 bp->bio_resid = bp->bio_length - bp2->bio_completed; 395 bp->bio_completed = bp2->bio_completed; 396 g_destroy_bio(bp2); 397 biodone(bp); 398 } 399 400 static void 401 g_dev_strategy(struct bio *bp) 402 { 403 struct g_consumer *cp; 404 struct bio *bp2; 405 struct cdev *dev; 406 407 KASSERT(bp->bio_cmd == BIO_READ || 408 bp->bio_cmd == BIO_WRITE || 409 bp->bio_cmd == BIO_DELETE, 410 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); 411 dev = bp->bio_dev; 412 cp = dev->si_drv2; 413 KASSERT(cp->acr || cp->acw, 414 ("Consumer with zero access count in g_dev_strategy")); 415 #ifdef INVARIANTS 416 if ((bp->bio_offset % cp->provider->sectorsize) != 0 || 417 (bp->bio_bcount % cp->provider->sectorsize) != 0) { 418 bp->bio_resid = bp->bio_bcount; 419 biofinish(bp, NULL, EINVAL); 420 return; 421 } 422 #endif 423 for (;;) { 424 /* 425 * XXX: This is not an ideal solution, but I belive it to 426 * XXX: deadlock safe, all things considered. 427 */ 428 bp2 = g_clone_bio(bp); 429 if (bp2 != NULL) 430 break; 431 pause("gdstrat", hz / 10); 432 } 433 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); 434 bp2->bio_done = g_dev_done; 435 g_trace(G_T_BIO, 436 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", 437 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, 438 bp2->bio_data, bp2->bio_cmd); 439 g_io_request(bp2, cp); 440 KASSERT(cp->acr || cp->acw, 441 ("g_dev_strategy raced with g_dev_close and lost")); 442 443 } 444 445 /* 446 * g_dev_orphan() 447 * 448 * Called from below when the provider orphaned us. 449 * - Clear any dump settings. 450 * - Destroy the struct cdev *to prevent any more request from coming in. The 451 * provider is already marked with an error, so anything which comes in 452 * in the interrim will be returned immediately. 453 * - Wait for any outstanding I/O to finish. 454 * - Set our access counts to zero, whatever they were. 455 * - Detach and self-destruct. 456 */ 457 458 static void 459 g_dev_orphan(struct g_consumer *cp) 460 { 461 struct g_geom *gp; 462 struct cdev *dev; 463 464 g_topology_assert(); 465 gp = cp->geom; 466 dev = gp->softc; 467 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name); 468 469 /* Reset any dump-area set on this device */ 470 if (dev->si_flags & SI_DUMPDEV) 471 set_dumper(NULL); 472 473 /* Destroy the struct cdev *so we get no more requests */ 474 destroy_dev(dev); 475 476 /* Wait for the cows to come home */ 477 while (cp->nstart != cp->nend) 478 pause("gdevorphan", hz / 10); 479 480 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 481 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 482 483 g_detach(cp); 484 g_destroy_consumer(cp); 485 g_destroy_geom(gp); 486 } 487 488 DECLARE_GEOM_CLASS(g_dev_class, g_dev); 489