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/errno.h> 48 #include <sys/time.h> 49 #include <sys/disk.h> 50 #include <sys/fcntl.h> 51 #include <sys/limits.h> 52 #include <geom/geom.h> 53 #include <geom/geom_int.h> 54 55 static d_open_t g_dev_open; 56 static d_close_t g_dev_close; 57 static d_strategy_t g_dev_strategy; 58 static d_ioctl_t g_dev_ioctl; 59 60 static struct cdevsw g_dev_cdevsw = { 61 .d_open = g_dev_open, 62 .d_close = g_dev_close, 63 .d_read = physread, 64 .d_write = physwrite, 65 .d_ioctl = g_dev_ioctl, 66 .d_strategy = g_dev_strategy, 67 .d_name = "g_dev", 68 .d_maj = GEOM_MAJOR, 69 .d_flags = D_DISK | D_TRACKCLOSE, 70 }; 71 72 static g_taste_t g_dev_taste; 73 static g_orphan_t g_dev_orphan; 74 75 static struct g_class g_dev_class = { 76 .name = "DEV", 77 .taste = g_dev_taste, 78 }; 79 80 void 81 g_dev_print(void) 82 { 83 struct g_geom *gp; 84 char const *p = ""; 85 86 LIST_FOREACH(gp, &g_dev_class.geom, geom) { 87 printf("%s%s", p, gp->name); 88 p = " "; 89 } 90 printf("\n"); 91 } 92 93 /* 94 * XXX: This is disgusting and wrong in every way imaginable: The only reason 95 * XXX: we have a clone function is because of the root-mount hack we currently 96 * XXX: employ. An improvment would be to unregister this cloner once we know 97 * XXX: we no longer need it. Ideally, root-fs would be mounted through DEVFS 98 * XXX: eliminating the need for this hack. 99 */ 100 static void 101 g_dev_clone(void *arg __unused, char *name, int namelen __unused, dev_t *dev) 102 { 103 struct g_geom *gp; 104 105 if (*dev != NODEV) 106 return; 107 108 g_waitidle(); 109 110 /* g_topology_lock(); */ 111 LIST_FOREACH(gp, &g_dev_class.geom, geom) { 112 if (strcmp(gp->name, name)) 113 continue; 114 *dev = gp->softc; 115 g_trace(G_T_TOPOLOGY, "g_dev_clone(%s) = %p", name, *dev); 116 return; 117 } 118 /* g_topology_unlock(); */ 119 return; 120 } 121 122 static void 123 g_dev_register_cloner(void *foo __unused) 124 { 125 static int once; 126 127 /* XXX: why would this happen more than once ?? */ 128 if (!once) { 129 EVENTHANDLER_REGISTER(dev_clone, g_dev_clone, 0, 1000); 130 once++; 131 } 132 } 133 134 SYSINIT(geomdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,g_dev_register_cloner,NULL); 135 136 static struct g_geom * 137 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused) 138 { 139 struct g_geom *gp; 140 struct g_consumer *cp; 141 static int unit = GEOM_MINOR_PROVIDERS; 142 int error; 143 dev_t dev; 144 145 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name); 146 g_topology_assert(); 147 LIST_FOREACH(cp, &pp->consumers, consumers) 148 if (cp->geom->class == mp) 149 return (NULL); 150 gp = g_new_geomf(mp, pp->name); 151 gp->orphan = g_dev_orphan; 152 cp = g_new_consumer(gp); 153 error = g_attach(cp, pp); 154 KASSERT(error == 0, 155 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error)); 156 /* 157 * XXX: I'm not 100% sure we can call make_dev(9) without Giant 158 * yet. Once we can, we don't need to drop topology here either. 159 */ 160 g_topology_unlock(); 161 mtx_lock(&Giant); 162 dev = make_dev(&g_dev_cdevsw, unit2minor(unit++), 163 UID_ROOT, GID_OPERATOR, 0640, gp->name); 164 if (pp->flags & G_PF_CANDELETE) 165 dev->si_flags |= SI_CANDELETE; 166 mtx_unlock(&Giant); 167 g_topology_lock(); 168 dev->si_iosize_max = MAXPHYS; 169 dev->si_stripesize = pp->stripesize; 170 dev->si_stripeoffset = pp->stripeoffset; 171 gp->softc = dev; 172 dev->si_drv1 = gp; 173 dev->si_drv2 = cp; 174 return (gp); 175 } 176 177 static int 178 g_dev_open(dev_t dev, int flags, int fmt, struct thread *td) 179 { 180 struct g_geom *gp; 181 struct g_consumer *cp; 182 int error, r, w, e; 183 184 gp = dev->si_drv1; 185 cp = dev->si_drv2; 186 if (gp == NULL || cp == NULL || gp->softc != dev) 187 return(ENXIO); /* g_dev_taste() not done yet */ 188 189 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", 190 gp->name, flags, fmt, td); 191 r = flags & FREAD ? 1 : 0; 192 w = flags & FWRITE ? 1 : 0; 193 #ifdef notyet 194 e = flags & O_EXCL ? 1 : 0; 195 #else 196 e = 0; 197 #endif 198 DROP_GIANT(); 199 g_topology_lock(); 200 if (dev->si_devsw == NULL) 201 error = ENXIO; /* We were orphaned */ 202 else 203 error = g_access_rel(cp, r, w, e); 204 g_topology_unlock(); 205 PICKUP_GIANT(); 206 g_waitidle(); 207 if (!error) 208 dev->si_bsize_phys = cp->provider->sectorsize; 209 return(error); 210 } 211 212 static int 213 g_dev_close(dev_t dev, int flags, int fmt, struct thread *td) 214 { 215 struct g_geom *gp; 216 struct g_consumer *cp; 217 int error, r, w, e, i; 218 219 gp = dev->si_drv1; 220 cp = dev->si_drv2; 221 if (gp == NULL || cp == NULL) 222 return(ENXIO); 223 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)", 224 gp->name, flags, fmt, td); 225 r = flags & FREAD ? -1 : 0; 226 w = flags & FWRITE ? -1 : 0; 227 #ifdef notyet 228 e = flags & O_EXCL ? -1 : 0; 229 #else 230 e = 0; 231 #endif 232 DROP_GIANT(); 233 g_topology_lock(); 234 if (dev->si_devsw == NULL) 235 error = ENXIO; /* We were orphaned */ 236 else 237 error = g_access_rel(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 tsleep(&i, PRIBIO, "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", 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 PICKUP_GIANT(); 254 g_waitidle(); 255 return (error); 256 } 257 258 /* 259 * XXX: Until we have unmessed the ioctl situation, there is a race against 260 * XXX: a concurrent orphanization. We cannot close it by holding topology 261 * XXX: since that would prevent us from doing our job, and stalling events 262 * XXX: will break (actually: stall) the BSD disklabel hacks. 263 */ 264 static int 265 g_dev_ioctl(dev_t dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 266 { 267 struct g_geom *gp; 268 struct g_consumer *cp; 269 struct g_kerneldump kd; 270 int i, error; 271 u_int u; 272 struct g_ioctl *gio; 273 274 gp = dev->si_drv1; 275 cp = dev->si_drv2; 276 gio = NULL; 277 278 error = 0; 279 KASSERT(cp->acr || cp->acw, 280 ("Consumer with zero access count in g_dev_ioctl")); 281 DROP_GIANT(); 282 283 gio = NULL; 284 i = IOCPARM_LEN(cmd); 285 switch (cmd) { 286 case DIOCGSECTORSIZE: 287 *(u_int *)data = cp->provider->sectorsize; 288 if (*(u_int *)data == 0) 289 error = ENOENT; 290 break; 291 case DIOCGMEDIASIZE: 292 *(off_t *)data = cp->provider->mediasize; 293 if (*(off_t *)data == 0) 294 error = ENOENT; 295 break; 296 case DIOCGFWSECTORS: 297 error = g_io_getattr("GEOM::fwsectors", cp, &i, data); 298 if (error == 0 && *(u_int *)data == 0) 299 error = ENOENT; 300 break; 301 case DIOCGFWHEADS: 302 error = g_io_getattr("GEOM::fwheads", cp, &i, data); 303 if (error == 0 && *(u_int *)data == 0) 304 error = ENOENT; 305 break; 306 case DIOCGFRONTSTUFF: 307 error = g_io_getattr("GEOM::frontstuff", cp, &i, data); 308 break; 309 case DIOCSKERNELDUMP: 310 u = *((u_int *)data); 311 if (!u) { 312 set_dumper(NULL); 313 error = 0; 314 break; 315 } 316 kd.offset = 0; 317 kd.length = OFF_MAX; 318 i = sizeof kd; 319 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd); 320 if (!error) 321 dev->si_flags |= SI_DUMPDEV; 322 break; 323 324 default: 325 gio = g_malloc(sizeof *gio, M_WAITOK | M_ZERO); 326 gio->cmd = cmd; 327 gio->data = data; 328 gio->fflag = fflag; 329 gio->td = td; 330 i = sizeof *gio; 331 /* 332 * We always issue ioctls as getattr since the direction of data 333 * movement in ioctl is no indication of the ioctl being a "set" 334 * or "get" type ioctl or if such simplistic terms even apply 335 */ 336 error = g_io_getattr("GEOM::ioctl", cp, &i, gio); 337 break; 338 } 339 340 PICKUP_GIANT(); 341 if (error == EDIRIOCTL) { 342 KASSERT(gio != NULL, ("NULL gio but EDIRIOCTL")); 343 KASSERT(gio->func != NULL, ("NULL function but EDIRIOCTL")); 344 error = (gio->func)(gio->dev, cmd, data, fflag, td); 345 } 346 g_waitidle(); 347 if (gio != NULL && (error == EOPNOTSUPP || error == ENOIOCTL)) { 348 if (g_debugflags & G_T_TOPOLOGY) { 349 i = IOCGROUP(cmd); 350 printf("IOCTL(0x%lx) \"%s\"", cmd, gp->name); 351 if (i > ' ' && i <= '~') 352 printf(" '%c'", (int)IOCGROUP(cmd)); 353 else 354 printf(" 0x%lx", IOCGROUP(cmd)); 355 printf("/%ld ", cmd & 0xff); 356 if (cmd & IOC_IN) 357 printf("I"); 358 if (cmd & IOC_OUT) 359 printf("O"); 360 printf("(%ld) = ENOIOCTL\n", IOCPARM_LEN(cmd)); 361 } 362 error = ENOTTY; 363 } 364 if (gio != NULL) 365 g_free(gio); 366 return (error); 367 } 368 369 static void 370 g_dev_done(struct bio *bp2) 371 { 372 struct bio *bp; 373 374 bp = bp2->bio_parent; 375 bp->bio_error = bp2->bio_error; 376 if (bp->bio_error != 0) { 377 g_trace(G_T_BIO, "g_dev_done(%p) had error %d", 378 bp2, bp->bio_error); 379 bp->bio_flags |= BIO_ERROR; 380 } else { 381 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", 382 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed); 383 } 384 bp->bio_resid = bp->bio_bcount - bp2->bio_completed; 385 g_destroy_bio(bp2); 386 mtx_lock(&Giant); 387 biodone(bp); 388 mtx_unlock(&Giant); 389 } 390 391 static void 392 g_dev_strategy(struct bio *bp) 393 { 394 struct g_consumer *cp; 395 struct bio *bp2; 396 dev_t dev; 397 398 KASSERT(bp->bio_cmd == BIO_READ || 399 bp->bio_cmd == BIO_WRITE || 400 bp->bio_cmd == BIO_DELETE, 401 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); 402 dev = bp->bio_dev; 403 cp = dev->si_drv2; 404 KASSERT(cp->acr || cp->acw, 405 ("Consumer with zero access count in g_dev_strategy")); 406 407 bp2 = g_clone_bio(bp); 408 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); 409 bp2->bio_offset = (off_t)bp->bio_blkno << DEV_BSHIFT; 410 KASSERT(bp2->bio_offset >= 0, 411 ("Negative bio_offset (%jd) on bio %p", 412 (intmax_t)bp2->bio_offset, bp)); 413 bp2->bio_length = (off_t)bp->bio_bcount; 414 bp2->bio_done = g_dev_done; 415 g_trace(G_T_BIO, 416 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", 417 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, 418 bp2->bio_data, bp2->bio_cmd); 419 g_io_request(bp2, cp); 420 KASSERT(cp->acr || cp->acw, 421 ("g_dev_strategy raced with g_dev_close and lost")); 422 423 } 424 425 /* 426 * g_dev_orphan() 427 * 428 * Called from below when the provider orphaned us. 429 * - Clear any dump settings. 430 * - Destroy the dev_t to prevent any more request from coming in. The 431 * provider is already marked with an error, so anything which comes in 432 * in the interrim will be returned immediately. 433 * - Wait for any outstanding I/O to finish. 434 * - Set our access counts to zero, whatever they were. 435 * - Detach and self-destruct. 436 */ 437 438 static void 439 g_dev_orphan(struct g_consumer *cp) 440 { 441 struct g_geom *gp; 442 dev_t dev; 443 444 g_topology_assert(); 445 gp = cp->geom; 446 dev = gp->softc; 447 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name); 448 449 /* Reset any dump-area set on this device */ 450 if (dev->si_flags & SI_DUMPDEV) 451 set_dumper(NULL); 452 453 /* Destroy the dev_t so we get no more requests */ 454 destroy_dev(dev); 455 456 /* Wait for the cows to come home */ 457 while (cp->nstart != cp->nend) 458 msleep(&dev, NULL, PRIBIO, "gdevorphan", hz / 10); 459 460 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 461 g_access_rel(cp, -cp->acr, -cp->acw, -cp->ace); 462 463 g_detach(cp); 464 g_destroy_consumer(cp); 465 g_destroy_geom(gp); 466 } 467 468 DECLARE_GEOM_CLASS(g_dev_class, g_dev); 469