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 char *alias; 117 int error; 118 struct cdev *dev; 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 dev = make_dev(&g_dev_cdevsw, 0, 131 UID_ROOT, GID_OPERATOR, 0640, gp->name); 132 if (pp->flags & G_PF_CANDELETE) 133 dev->si_flags |= SI_CANDELETE; 134 dev->si_iosize_max = MAXPHYS; 135 gp->softc = dev; 136 dev->si_drv1 = gp; 137 dev->si_drv2 = cp; 138 139 g_topology_unlock(); 140 141 alias = g_malloc(MAXPATHLEN, M_WAITOK | M_ZERO); 142 error = (pp->geom->ioctl == NULL) ? ENODEV : 143 pp->geom->ioctl(pp, DIOCGPROVIDERALIAS, alias, 0, curthread); 144 if (!error && alias[0] != '\0') 145 make_dev_alias(dev, "%s", alias); 146 g_free(alias); 147 148 g_topology_lock(); 149 return (gp); 150 } 151 152 static int 153 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td) 154 { 155 struct g_geom *gp; 156 struct g_consumer *cp; 157 int error, r, w, e; 158 159 gp = dev->si_drv1; 160 cp = dev->si_drv2; 161 if (gp == NULL || cp == NULL || gp->softc != dev) 162 return(ENXIO); /* g_dev_taste() not done yet */ 163 164 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", 165 gp->name, flags, fmt, td); 166 167 r = flags & FREAD ? 1 : 0; 168 w = flags & FWRITE ? 1 : 0; 169 #ifdef notyet 170 e = flags & O_EXCL ? 1 : 0; 171 #else 172 e = 0; 173 #endif 174 if (w) { 175 /* 176 * When running in very secure mode, do not allow 177 * opens for writing of any disks. 178 */ 179 error = securelevel_ge(td->td_ucred, 2); 180 if (error) 181 return (error); 182 } 183 g_topology_lock(); 184 if (dev->si_devsw == NULL) 185 error = ENXIO; /* We were orphaned */ 186 else 187 error = g_access(cp, r, w, e); 188 g_topology_unlock(); 189 return(error); 190 } 191 192 static int 193 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td) 194 { 195 struct g_geom *gp; 196 struct g_consumer *cp; 197 int error, r, w, e, i; 198 199 gp = dev->si_drv1; 200 cp = dev->si_drv2; 201 if (gp == NULL || cp == NULL) 202 return(ENXIO); 203 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)", 204 gp->name, flags, fmt, td); 205 r = flags & FREAD ? -1 : 0; 206 w = flags & FWRITE ? -1 : 0; 207 #ifdef notyet 208 e = flags & O_EXCL ? -1 : 0; 209 #else 210 e = 0; 211 #endif 212 g_topology_lock(); 213 if (dev->si_devsw == NULL) 214 error = ENXIO; /* We were orphaned */ 215 else 216 error = g_access(cp, r, w, e); 217 for (i = 0; i < 10 * hz;) { 218 if (cp->acr != 0 || cp->acw != 0) 219 break; 220 if (cp->nstart == cp->nend) 221 break; 222 pause("gdevwclose", hz / 10); 223 i += hz / 10; 224 } 225 if (cp->acr == 0 && cp->acw == 0 && cp->nstart != cp->nend) { 226 printf("WARNING: Final close of geom_dev(%s) %s %s\n", 227 gp->name, 228 "still has outstanding I/O after 10 seconds.", 229 "Completing close anyway, panic may happen later."); 230 } 231 g_topology_unlock(); 232 return (error); 233 } 234 235 /* 236 * XXX: Until we have unmessed the ioctl situation, there is a race against 237 * XXX: a concurrent orphanization. We cannot close it by holding topology 238 * XXX: since that would prevent us from doing our job, and stalling events 239 * XXX: will break (actually: stall) the BSD disklabel hacks. 240 */ 241 static int 242 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 243 { 244 struct g_geom *gp; 245 struct g_consumer *cp; 246 struct g_provider *pp; 247 struct g_kerneldump kd; 248 off_t offset, length, chunk; 249 int i, error; 250 u_int u; 251 252 gp = dev->si_drv1; 253 cp = dev->si_drv2; 254 pp = cp->provider; 255 256 error = 0; 257 KASSERT(cp->acr || cp->acw, 258 ("Consumer with zero access count in g_dev_ioctl")); 259 260 i = IOCPARM_LEN(cmd); 261 switch (cmd) { 262 case DIOCGSECTORSIZE: 263 *(u_int *)data = cp->provider->sectorsize; 264 if (*(u_int *)data == 0) 265 error = ENOENT; 266 break; 267 case DIOCGMEDIASIZE: 268 *(off_t *)data = cp->provider->mediasize; 269 if (*(off_t *)data == 0) 270 error = ENOENT; 271 break; 272 case DIOCGFWSECTORS: 273 error = g_io_getattr("GEOM::fwsectors", cp, &i, data); 274 if (error == 0 && *(u_int *)data == 0) 275 error = ENOENT; 276 break; 277 case DIOCGFWHEADS: 278 error = g_io_getattr("GEOM::fwheads", cp, &i, data); 279 if (error == 0 && *(u_int *)data == 0) 280 error = ENOENT; 281 break; 282 case DIOCGFRONTSTUFF: 283 error = g_io_getattr("GEOM::frontstuff", cp, &i, data); 284 break; 285 case DIOCSKERNELDUMP: 286 u = *((u_int *)data); 287 if (!u) { 288 set_dumper(NULL); 289 error = 0; 290 break; 291 } 292 kd.offset = 0; 293 kd.length = OFF_MAX; 294 i = sizeof kd; 295 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd); 296 if (!error) 297 dev->si_flags |= SI_DUMPDEV; 298 break; 299 case DIOCGFLUSH: 300 error = g_io_flush(cp); 301 break; 302 case DIOCGDELETE: 303 offset = ((off_t *)data)[0]; 304 length = ((off_t *)data)[1]; 305 if ((offset % cp->provider->sectorsize) != 0 || 306 (length % cp->provider->sectorsize) != 0 || length <= 0) { 307 printf("%s: offset=%jd length=%jd\n", __func__, offset, 308 length); 309 error = EINVAL; 310 break; 311 } 312 while (length > 0) { 313 chunk = length; 314 if (chunk > 1024 * cp->provider->sectorsize) 315 chunk = 1024 * cp->provider->sectorsize; 316 error = g_delete_data(cp, offset, chunk); 317 length -= chunk; 318 offset += chunk; 319 if (error) 320 break; 321 /* 322 * Since the request size is unbounded, the service 323 * time is likewise. We make this ioctl interruptible 324 * by checking for signals for each bio. 325 */ 326 if (SIGPENDING(td)) 327 break; 328 } 329 break; 330 case DIOCGIDENT: 331 error = g_io_getattr("GEOM::ident", cp, &i, data); 332 break; 333 case DIOCGPROVIDERNAME: 334 if (pp == NULL) 335 return (ENOENT); 336 strlcpy(data, pp->name, i); 337 break; 338 339 default: 340 if (cp->provider->geom->ioctl != NULL) { 341 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td); 342 } else { 343 error = ENOIOCTL; 344 } 345 } 346 347 return (error); 348 } 349 350 static void 351 g_dev_done(struct bio *bp2) 352 { 353 struct bio *bp; 354 355 bp = bp2->bio_parent; 356 bp->bio_error = bp2->bio_error; 357 if (bp->bio_error != 0) { 358 g_trace(G_T_BIO, "g_dev_done(%p) had error %d", 359 bp2, bp->bio_error); 360 bp->bio_flags |= BIO_ERROR; 361 } else { 362 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", 363 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed); 364 } 365 bp->bio_resid = bp->bio_length - bp2->bio_completed; 366 bp->bio_completed = bp2->bio_completed; 367 g_destroy_bio(bp2); 368 biodone(bp); 369 } 370 371 static void 372 g_dev_strategy(struct bio *bp) 373 { 374 struct g_consumer *cp; 375 struct bio *bp2; 376 struct cdev *dev; 377 378 KASSERT(bp->bio_cmd == BIO_READ || 379 bp->bio_cmd == BIO_WRITE || 380 bp->bio_cmd == BIO_DELETE, 381 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); 382 dev = bp->bio_dev; 383 cp = dev->si_drv2; 384 KASSERT(cp->acr || cp->acw, 385 ("Consumer with zero access count in g_dev_strategy")); 386 387 if ((bp->bio_offset % cp->provider->sectorsize) != 0 || 388 (bp->bio_bcount % cp->provider->sectorsize) != 0) { 389 bp->bio_resid = bp->bio_bcount; 390 biofinish(bp, NULL, EINVAL); 391 return; 392 } 393 394 for (;;) { 395 /* 396 * XXX: This is not an ideal solution, but I belive it to 397 * XXX: deadlock safe, all things considered. 398 */ 399 bp2 = g_clone_bio(bp); 400 if (bp2 != NULL) 401 break; 402 pause("gdstrat", hz / 10); 403 } 404 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); 405 bp2->bio_done = g_dev_done; 406 g_trace(G_T_BIO, 407 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", 408 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, 409 bp2->bio_data, bp2->bio_cmd); 410 g_io_request(bp2, cp); 411 KASSERT(cp->acr || cp->acw, 412 ("g_dev_strategy raced with g_dev_close and lost")); 413 414 } 415 416 /* 417 * g_dev_orphan() 418 * 419 * Called from below when the provider orphaned us. 420 * - Clear any dump settings. 421 * - Destroy the struct cdev *to prevent any more request from coming in. The 422 * provider is already marked with an error, so anything which comes in 423 * in the interrim will be returned immediately. 424 * - Wait for any outstanding I/O to finish. 425 * - Set our access counts to zero, whatever they were. 426 * - Detach and self-destruct. 427 */ 428 429 static void 430 g_dev_orphan(struct g_consumer *cp) 431 { 432 struct g_geom *gp; 433 struct cdev *dev; 434 435 g_topology_assert(); 436 gp = cp->geom; 437 dev = gp->softc; 438 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name); 439 440 /* Reset any dump-area set on this device */ 441 if (dev->si_flags & SI_DUMPDEV) 442 set_dumper(NULL); 443 444 /* Destroy the struct cdev *so we get no more requests */ 445 destroy_dev(dev); 446 447 /* Wait for the cows to come home */ 448 while (cp->nstart != cp->nend) 449 pause("gdevorphan", hz / 10); 450 451 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 452 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 453 454 g_detach(cp); 455 g_destroy_consumer(cp); 456 g_destroy_geom(gp); 457 } 458 459 DECLARE_GEOM_CLASS(g_dev_class, g_dev); 460