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 static g_init_t g_dev_init; 76 77 static struct g_class g_dev_class = { 78 .name = "DEV", 79 .version = G_VERSION, 80 .taste = g_dev_taste, 81 .orphan = g_dev_orphan, 82 .init = g_dev_init, 83 }; 84 85 static struct unrhdr *unithdr; /* Locked by topology */ 86 87 static void 88 g_dev_init(struct g_class *mp) 89 { 90 91 unithdr = new_unrhdr(0, INT_MAX, NULL); 92 } 93 94 void 95 g_dev_print(void) 96 { 97 struct g_geom *gp; 98 char const *p = ""; 99 100 LIST_FOREACH(gp, &g_dev_class.geom, geom) { 101 printf("%s%s", p, gp->name); 102 p = " "; 103 } 104 printf("\n"); 105 } 106 107 struct g_provider * 108 g_dev_getprovider(struct cdev *dev) 109 { 110 struct g_consumer *cp; 111 112 g_topology_assert(); 113 if (dev == NULL) 114 return (NULL); 115 if (dev->si_devsw != &g_dev_cdevsw) 116 return (NULL); 117 cp = dev->si_drv2; 118 return (cp->provider); 119 } 120 121 122 static struct g_geom * 123 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused) 124 { 125 struct g_geom *gp; 126 struct g_consumer *cp; 127 int error; 128 struct cdev *dev; 129 u_int unit; 130 131 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name); 132 g_topology_assert(); 133 LIST_FOREACH(cp, &pp->consumers, consumers) 134 if (cp->geom->class == mp) 135 return (NULL); 136 gp = g_new_geomf(mp, pp->name); 137 cp = g_new_consumer(gp); 138 error = g_attach(cp, pp); 139 KASSERT(error == 0, 140 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error)); 141 unit = alloc_unr(unithdr); 142 dev = make_dev(&g_dev_cdevsw, unit, 143 UID_ROOT, GID_OPERATOR, 0640, gp->name); 144 if (pp->flags & G_PF_CANDELETE) 145 dev->si_flags |= SI_CANDELETE; 146 dev->si_iosize_max = MAXPHYS; 147 gp->softc = dev; 148 dev->si_drv1 = gp; 149 dev->si_drv2 = cp; 150 return (gp); 151 } 152 153 static int 154 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td) 155 { 156 struct g_geom *gp; 157 struct g_consumer *cp; 158 int error, r, w, e; 159 160 gp = dev->si_drv1; 161 cp = dev->si_drv2; 162 if (gp == NULL || cp == NULL || gp->softc != dev) 163 return(ENXIO); /* g_dev_taste() not done yet */ 164 165 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", 166 gp->name, flags, fmt, td); 167 168 r = flags & FREAD ? 1 : 0; 169 w = flags & FWRITE ? 1 : 0; 170 #ifdef notyet 171 e = flags & O_EXCL ? 1 : 0; 172 #else 173 e = 0; 174 #endif 175 if (w) { 176 /* 177 * When running in very secure mode, do not allow 178 * opens for writing of any disks. 179 */ 180 error = securelevel_ge(td->td_ucred, 2); 181 if (error) 182 return (error); 183 } 184 g_topology_lock(); 185 if (dev->si_devsw == NULL) 186 error = ENXIO; /* We were orphaned */ 187 else 188 error = g_access(cp, r, w, e); 189 g_topology_unlock(); 190 return(error); 191 } 192 193 static int 194 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td) 195 { 196 struct g_geom *gp; 197 struct g_consumer *cp; 198 int error, r, w, e, i; 199 200 gp = dev->si_drv1; 201 cp = dev->si_drv2; 202 if (gp == NULL || cp == NULL) 203 return(ENXIO); 204 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)", 205 gp->name, flags, fmt, td); 206 r = flags & FREAD ? -1 : 0; 207 w = flags & FWRITE ? -1 : 0; 208 #ifdef notyet 209 e = flags & O_EXCL ? -1 : 0; 210 #else 211 e = 0; 212 #endif 213 g_topology_lock(); 214 if (dev->si_devsw == NULL) 215 error = ENXIO; /* We were orphaned */ 216 else 217 error = g_access(cp, r, w, e); 218 for (i = 0; i < 10 * hz;) { 219 if (cp->acr != 0 || cp->acw != 0) 220 break; 221 if (cp->nstart == cp->nend) 222 break; 223 pause("gdevwclose", hz / 10); 224 i += hz / 10; 225 } 226 if (cp->acr == 0 && cp->acw == 0 && cp->nstart != cp->nend) { 227 printf("WARNING: Final close of geom_dev(%s) %s %s\n", 228 gp->name, 229 "still has outstanding I/O after 10 seconds.", 230 "Completing close anyway, panic may happen later."); 231 } 232 g_topology_unlock(); 233 return (error); 234 } 235 236 /* 237 * XXX: Until we have unmessed the ioctl situation, there is a race against 238 * XXX: a concurrent orphanization. We cannot close it by holding topology 239 * XXX: since that would prevent us from doing our job, and stalling events 240 * XXX: will break (actually: stall) the BSD disklabel hacks. 241 */ 242 static int 243 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 244 { 245 struct g_geom *gp; 246 struct g_consumer *cp; 247 struct g_provider *pp; 248 struct g_kerneldump kd; 249 off_t offset, length, chunk; 250 int i, error; 251 u_int u; 252 253 gp = dev->si_drv1; 254 cp = dev->si_drv2; 255 pp = cp->provider; 256 257 error = 0; 258 KASSERT(cp->acr || cp->acw, 259 ("Consumer with zero access count in g_dev_ioctl")); 260 261 i = IOCPARM_LEN(cmd); 262 switch (cmd) { 263 case DIOCGSECTORSIZE: 264 *(u_int *)data = cp->provider->sectorsize; 265 if (*(u_int *)data == 0) 266 error = ENOENT; 267 break; 268 case DIOCGMEDIASIZE: 269 *(off_t *)data = cp->provider->mediasize; 270 if (*(off_t *)data == 0) 271 error = ENOENT; 272 break; 273 case DIOCGFWSECTORS: 274 error = g_io_getattr("GEOM::fwsectors", cp, &i, data); 275 if (error == 0 && *(u_int *)data == 0) 276 error = ENOENT; 277 break; 278 case DIOCGFWHEADS: 279 error = g_io_getattr("GEOM::fwheads", cp, &i, data); 280 if (error == 0 && *(u_int *)data == 0) 281 error = ENOENT; 282 break; 283 case DIOCGFRONTSTUFF: 284 error = g_io_getattr("GEOM::frontstuff", cp, &i, data); 285 break; 286 case DIOCSKERNELDUMP: 287 u = *((u_int *)data); 288 if (!u) { 289 set_dumper(NULL); 290 error = 0; 291 break; 292 } 293 kd.offset = 0; 294 kd.length = OFF_MAX; 295 i = sizeof kd; 296 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd); 297 if (!error) 298 dev->si_flags |= SI_DUMPDEV; 299 break; 300 case DIOCGFLUSH: 301 error = g_io_flush(cp); 302 break; 303 case DIOCGDELETE: 304 offset = ((off_t *)data)[0]; 305 length = ((off_t *)data)[1]; 306 if ((offset % cp->provider->sectorsize) != 0 || 307 (length % cp->provider->sectorsize) != 0 || length <= 0) { 308 printf("%s: offset=%jd length=%jd\n", __func__, offset, 309 length); 310 error = EINVAL; 311 break; 312 } 313 while (length > 0) { 314 chunk = length; 315 if (chunk > 1024 * cp->provider->sectorsize) 316 chunk = 1024 * cp->provider->sectorsize; 317 error = g_delete_data(cp, offset, chunk); 318 length -= chunk; 319 offset += chunk; 320 if (error) 321 break; 322 /* 323 * Since the request size is unbounded, the service 324 * time is likewise. We make this ioctl interruptible 325 * by checking for signals for each bio. 326 */ 327 if (SIGPENDING(td)) 328 break; 329 } 330 break; 331 case DIOCGIDENT: 332 error = g_io_getattr("GEOM::ident", cp, &i, data); 333 break; 334 case DIOCGPROVIDERNAME: 335 if (pp == NULL) 336 return (ENOENT); 337 strlcpy(data, pp->name, i); 338 break; 339 340 default: 341 if (cp->provider->geom->ioctl != NULL) { 342 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td); 343 } else { 344 error = ENOIOCTL; 345 } 346 } 347 348 return (error); 349 } 350 351 static void 352 g_dev_done(struct bio *bp2) 353 { 354 struct bio *bp; 355 356 bp = bp2->bio_parent; 357 bp->bio_error = bp2->bio_error; 358 if (bp->bio_error != 0) { 359 g_trace(G_T_BIO, "g_dev_done(%p) had error %d", 360 bp2, bp->bio_error); 361 bp->bio_flags |= BIO_ERROR; 362 } else { 363 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", 364 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed); 365 } 366 bp->bio_resid = bp->bio_length - bp2->bio_completed; 367 bp->bio_completed = bp2->bio_completed; 368 g_destroy_bio(bp2); 369 biodone(bp); 370 } 371 372 static void 373 g_dev_strategy(struct bio *bp) 374 { 375 struct g_consumer *cp; 376 struct bio *bp2; 377 struct cdev *dev; 378 379 KASSERT(bp->bio_cmd == BIO_READ || 380 bp->bio_cmd == BIO_WRITE || 381 bp->bio_cmd == BIO_DELETE, 382 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); 383 dev = bp->bio_dev; 384 cp = dev->si_drv2; 385 KASSERT(cp->acr || cp->acw, 386 ("Consumer with zero access count in g_dev_strategy")); 387 388 if ((bp->bio_offset % cp->provider->sectorsize) != 0 || 389 (bp->bio_bcount % cp->provider->sectorsize) != 0) { 390 bp->bio_resid = bp->bio_bcount; 391 biofinish(bp, NULL, EINVAL); 392 return; 393 } 394 395 for (;;) { 396 /* 397 * XXX: This is not an ideal solution, but I belive it to 398 * XXX: deadlock safe, all things considered. 399 */ 400 bp2 = g_clone_bio(bp); 401 if (bp2 != NULL) 402 break; 403 pause("gdstrat", hz / 10); 404 } 405 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); 406 bp2->bio_done = g_dev_done; 407 g_trace(G_T_BIO, 408 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", 409 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, 410 bp2->bio_data, bp2->bio_cmd); 411 g_io_request(bp2, cp); 412 KASSERT(cp->acr || cp->acw, 413 ("g_dev_strategy raced with g_dev_close and lost")); 414 415 } 416 417 /* 418 * g_dev_orphan() 419 * 420 * Called from below when the provider orphaned us. 421 * - Clear any dump settings. 422 * - Destroy the struct cdev *to prevent any more request from coming in. The 423 * provider is already marked with an error, so anything which comes in 424 * in the interrim will be returned immediately. 425 * - Wait for any outstanding I/O to finish. 426 * - Set our access counts to zero, whatever they were. 427 * - Detach and self-destruct. 428 */ 429 430 static void 431 g_dev_orphan(struct g_consumer *cp) 432 { 433 struct g_geom *gp; 434 struct cdev *dev; 435 u_int unit; 436 437 g_topology_assert(); 438 gp = cp->geom; 439 dev = gp->softc; 440 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name); 441 442 /* Reset any dump-area set on this device */ 443 if (dev->si_flags & SI_DUMPDEV) 444 set_dumper(NULL); 445 446 /* Destroy the struct cdev *so we get no more requests */ 447 unit = dev2unit(dev); 448 destroy_dev(dev); 449 free_unr(unithdr, unit); 450 451 /* Wait for the cows to come home */ 452 while (cp->nstart != cp->nend) 453 pause("gdevorphan", hz / 10); 454 455 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 456 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 457 458 g_detach(cp); 459 g_destroy_consumer(cp); 460 g_destroy_geom(gp); 461 } 462 463 DECLARE_GEOM_CLASS(g_dev_class, g_dev); 464