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 * $FreeBSD$ 36 */ 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/malloc.h> 41 #include <sys/kernel.h> 42 #include <sys/conf.h> 43 #include <sys/bio.h> 44 #include <sys/lock.h> 45 #include <sys/mutex.h> 46 #include <sys/errno.h> 47 #include <sys/time.h> 48 #include <sys/disk.h> 49 #include <sys/fcntl.h> 50 #include <sys/limits.h> 51 #include <geom/geom.h> 52 #include <geom/geom_int.h> 53 54 static d_open_t g_dev_open; 55 static d_close_t g_dev_close; 56 static d_strategy_t g_dev_strategy; 57 static d_ioctl_t g_dev_ioctl; 58 59 static struct cdevsw g_dev_cdevsw = { 60 .d_open = g_dev_open, 61 .d_close = g_dev_close, 62 .d_read = physread, 63 .d_write = physwrite, 64 .d_ioctl = g_dev_ioctl, 65 .d_strategy = g_dev_strategy, 66 .d_name = "g_dev", 67 .d_maj = GEOM_MAJOR, 68 .d_flags = D_DISK | D_TRACKCLOSE, 69 }; 70 71 static g_taste_t g_dev_taste; 72 static g_orphan_t g_dev_orphan; 73 74 static struct g_class g_dev_class = { 75 .name = "DEV", 76 .taste = g_dev_taste, 77 G_CLASS_INITIALIZER 78 }; 79 80 int 81 g_dev_print(void) 82 { 83 struct g_geom *gp; 84 85 if (LIST_EMPTY(&g_dev_class.geom)) 86 return (0); 87 printf("List of GEOM disk devices:\n "); 88 LIST_FOREACH(gp, &g_dev_class.geom, geom) 89 printf(" %s", gp->name); 90 printf("\n"); 91 return (1); 92 } 93 94 /* 95 * XXX: This is disgusting and wrong in every way imaginable: The only reason 96 * XXX: we have a clone function is because of the root-mount hack we currently 97 * XXX: employ. An improvment would be to unregister this cloner once we know 98 * XXX: we no longer need it. Ideally, root-fs would be mounted through DEVFS 99 * XXX: eliminating the need for this hack. 100 */ 101 static void 102 g_dev_clone(void *arg __unused, char *name, int namelen __unused, dev_t *dev) 103 { 104 struct g_geom *gp; 105 106 if (*dev != NODEV) 107 return; 108 109 g_waitidle(); 110 111 /* g_topology_lock(); */ 112 LIST_FOREACH(gp, &g_dev_class.geom, geom) { 113 if (strcmp(gp->name, name)) 114 continue; 115 *dev = gp->softc; 116 g_trace(G_T_TOPOLOGY, "g_dev_clone(%s) = %p", name, *dev); 117 return; 118 } 119 /* g_topology_unlock(); */ 120 return; 121 } 122 123 static void 124 g_dev_register_cloner(void *foo __unused) 125 { 126 static int once; 127 128 /* XXX: why would this happen more than once ?? */ 129 if (!once) { 130 EVENTHANDLER_REGISTER(dev_clone, g_dev_clone, 0, 1000); 131 once++; 132 } 133 } 134 135 SYSINIT(geomdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,g_dev_register_cloner,NULL); 136 137 static struct g_geom * 138 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused) 139 { 140 struct g_geom *gp; 141 struct g_consumer *cp; 142 static int unit = GEOM_MINOR_PROVIDERS; 143 int error; 144 dev_t dev; 145 146 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name); 147 g_topology_assert(); 148 LIST_FOREACH(cp, &pp->consumers, consumers) 149 if (cp->geom->class == mp) 150 return (NULL); 151 gp = g_new_geomf(mp, pp->name); 152 gp->orphan = g_dev_orphan; 153 cp = g_new_consumer(gp); 154 error = g_attach(cp, pp); 155 KASSERT(error == 0, 156 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error)); 157 /* 158 * XXX: I'm not 100% sure we can call make_dev(9) without Giant 159 * yet. Once we can, we don't need to drop topology here either. 160 */ 161 g_topology_unlock(); 162 mtx_lock(&Giant); 163 dev = make_dev(&g_dev_cdevsw, unit2minor(unit++), 164 UID_ROOT, GID_OPERATOR, 0640, gp->name); 165 if (pp->flags & G_PF_CANDELETE) 166 dev->si_flags |= SI_CANDELETE; 167 mtx_unlock(&Giant); 168 g_topology_lock(); 169 dev->si_iosize_max = MAXPHYS; 170 dev->si_stripesize = pp->stripesize; 171 dev->si_stripeoffset = pp->stripeoffset; 172 gp->softc = dev; 173 dev->si_drv1 = gp; 174 dev->si_drv2 = cp; 175 return (gp); 176 } 177 178 static int 179 g_dev_open(dev_t dev, int flags, int fmt, struct thread *td) 180 { 181 struct g_geom *gp; 182 struct g_consumer *cp; 183 int error, r, w, e; 184 185 gp = dev->si_drv1; 186 cp = dev->si_drv2; 187 if (gp == NULL || cp == NULL || gp->softc != dev) 188 return(ENXIO); /* g_dev_taste() not done yet */ 189 190 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", 191 gp->name, flags, fmt, td); 192 r = flags & FREAD ? 1 : 0; 193 w = flags & FWRITE ? 1 : 0; 194 #ifdef notyet 195 e = flags & O_EXCL ? 1 : 0; 196 #else 197 e = 0; 198 #endif 199 DROP_GIANT(); 200 g_topology_lock(); 201 if (dev->si_devsw == NULL) 202 error = ENXIO; /* We were orphaned */ 203 else 204 error = g_access_rel(cp, r, w, e); 205 g_topology_unlock(); 206 PICKUP_GIANT(); 207 g_waitidle(); 208 if (!error) 209 dev->si_bsize_phys = cp->provider->sectorsize; 210 return(error); 211 } 212 213 static int 214 g_dev_close(dev_t 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; 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 DROP_GIANT(); 234 g_topology_lock(); 235 if (dev->si_devsw == NULL) 236 error = ENXIO; /* We were orphaned */ 237 else 238 error = g_access_rel(cp, r, w, e); 239 KASSERT((cp->acr || cp->acw) || (cp->nstart == cp->nend), 240 ("final g_dev_close() with outstanding bios")); 241 g_topology_unlock(); 242 PICKUP_GIANT(); 243 g_waitidle(); 244 return (error); 245 } 246 247 /* 248 * XXX: Until we have unmessed the ioctl situation, there is a race against 249 * XXX: a concurrent orphanization. We cannot close it by holding topology 250 * XXX: since that would prevent us from doing our job, and stalling events 251 * XXX: will break (actually: stall) the BSD disklabel hacks. 252 */ 253 static int 254 g_dev_ioctl(dev_t dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 255 { 256 struct g_geom *gp, *gp2; 257 struct g_consumer *cp; 258 struct g_provider *pp2; 259 struct g_kerneldump kd; 260 int i, error; 261 u_int u; 262 struct g_ioctl *gio; 263 264 gp = dev->si_drv1; 265 cp = dev->si_drv2; 266 pp2 = cp->provider; 267 gp2 = pp2->geom; 268 gio = NULL; 269 270 error = 0; 271 KASSERT(cp->acr || cp->acw, 272 ("Consumer with zero access count in g_dev_ioctl")); 273 DROP_GIANT(); 274 275 gio = NULL; 276 i = IOCPARM_LEN(cmd); 277 switch (cmd) { 278 case DIOCGSECTORSIZE: 279 *(u_int *)data = cp->provider->sectorsize; 280 if (*(u_int *)data == 0) 281 error = ENOENT; 282 break; 283 case DIOCGMEDIASIZE: 284 *(off_t *)data = cp->provider->mediasize; 285 if (*(off_t *)data == 0) 286 error = ENOENT; 287 break; 288 case DIOCGFWSECTORS: 289 error = g_io_getattr("GEOM::fwsectors", cp, &i, data); 290 if (error == 0 && *(u_int *)data == 0) 291 error = ENOENT; 292 break; 293 case DIOCGFWHEADS: 294 error = g_io_getattr("GEOM::fwheads", cp, &i, data); 295 if (error == 0 && *(u_int *)data == 0) 296 error = ENOENT; 297 break; 298 case DIOCGFRONTSTUFF: 299 error = g_io_getattr("GEOM::frontstuff", cp, &i, data); 300 break; 301 case DIOCSKERNELDUMP: 302 u = *((u_int *)data); 303 if (!u) { 304 set_dumper(NULL); 305 error = 0; 306 break; 307 } 308 kd.offset = 0; 309 kd.length = OFF_MAX; 310 i = sizeof kd; 311 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd); 312 if (!error) 313 dev->si_flags |= SI_DUMPDEV; 314 break; 315 316 default: 317 gio = g_malloc(sizeof *gio, M_WAITOK | M_ZERO); 318 gio->cmd = cmd; 319 gio->data = data; 320 gio->fflag = fflag; 321 gio->td = td; 322 i = sizeof *gio; 323 /* 324 * We always issue ioctls as getattr since the direction of data 325 * movement in ioctl is no indication of the ioctl being a "set" 326 * or "get" type ioctl or if such simplistic terms even apply 327 */ 328 error = g_io_getattr("GEOM::ioctl", cp, &i, gio); 329 break; 330 } 331 332 PICKUP_GIANT(); 333 if (error == EDIRIOCTL) { 334 KASSERT(gio != NULL, ("NULL gio but EDIRIOCTL")); 335 KASSERT(gio->func != NULL, ("NULL function but EDIRIOCTL")); 336 error = (gio->func)(gio->dev, cmd, data, fflag, td); 337 } 338 g_waitidle(); 339 if (gio != NULL && (error == EOPNOTSUPP || error == ENOIOCTL)) { 340 if (g_debugflags & G_T_TOPOLOGY) { 341 i = IOCGROUP(cmd); 342 printf("IOCTL(0x%lx) \"%s\"", cmd, gp->name); 343 if (i > ' ' && i <= '~') 344 printf(" '%c'", (int)IOCGROUP(cmd)); 345 else 346 printf(" 0x%lx", IOCGROUP(cmd)); 347 printf("/%ld ", cmd & 0xff); 348 if (cmd & IOC_IN) 349 printf("I"); 350 if (cmd & IOC_OUT) 351 printf("O"); 352 printf("(%ld) = ENOIOCTL\n", IOCPARM_LEN(cmd)); 353 } 354 error = ENOTTY; 355 } 356 if (gio != NULL) 357 g_free(gio); 358 return (error); 359 } 360 361 static void 362 g_dev_done(struct bio *bp2) 363 { 364 struct bio *bp; 365 366 bp = bp2->bio_parent; 367 bp->bio_error = bp2->bio_error; 368 if (bp->bio_error != 0) { 369 g_trace(G_T_BIO, "g_dev_done(%p) had error %d", 370 bp2, bp->bio_error); 371 bp->bio_flags |= BIO_ERROR; 372 } else { 373 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", 374 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed); 375 } 376 bp->bio_resid = bp->bio_bcount - bp2->bio_completed; 377 g_destroy_bio(bp2); 378 mtx_lock(&Giant); 379 biodone(bp); 380 mtx_unlock(&Giant); 381 } 382 383 static void 384 g_dev_strategy(struct bio *bp) 385 { 386 struct g_geom *gp; 387 struct g_consumer *cp; 388 struct bio *bp2; 389 dev_t dev; 390 391 KASSERT(bp->bio_cmd == BIO_READ || 392 bp->bio_cmd == BIO_WRITE || 393 bp->bio_cmd == BIO_DELETE, 394 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); 395 dev = bp->bio_dev; 396 gp = dev->si_drv1; 397 cp = dev->si_drv2; 398 KASSERT(cp->acr || cp->acw, 399 ("Consumer with zero access count in g_dev_strategy")); 400 401 bp2 = g_clone_bio(bp); 402 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); 403 bp2->bio_offset = (off_t)bp->bio_blkno << DEV_BSHIFT; 404 KASSERT(bp2->bio_offset >= 0, 405 ("Negative bio_offset (%jd) on bio %p", 406 (intmax_t)bp2->bio_offset, bp)); 407 bp2->bio_length = (off_t)bp->bio_bcount; 408 bp2->bio_done = g_dev_done; 409 g_trace(G_T_BIO, 410 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", 411 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, 412 bp2->bio_data, bp2->bio_cmd); 413 g_io_request(bp2, cp); 414 KASSERT(cp->acr || cp->acw, 415 ("g_dev_strategy raced with g_dev_close and lost")); 416 417 } 418 419 /* 420 * g_dev_orphan() 421 * 422 * Called from below when the provider orphaned us. 423 * - Clear any dump settings. 424 * - Destroy the dev_t to prevent any more request from coming in. The 425 * provider is already marked with an error, so anything which comes in 426 * in the interrim will be returned immediately. 427 * - Wait for any outstanding I/O to finish. 428 * - Set our access counts to zero, whatever they were. 429 * - Detach and self-destruct. 430 */ 431 432 static void 433 g_dev_orphan(struct g_consumer *cp) 434 { 435 struct g_geom *gp; 436 dev_t dev; 437 438 g_topology_assert(); 439 gp = cp->geom; 440 dev = gp->softc; 441 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name); 442 443 /* Reset any dump-area set on this device */ 444 if (dev->si_flags & SI_DUMPDEV) 445 set_dumper(NULL); 446 447 /* Destroy the dev_t so we get no more requests */ 448 destroy_dev(dev); 449 450 /* Wait for the cows to come home */ 451 while (cp->nstart != cp->nend) 452 msleep(&dev, NULL, PRIBIO, "gdevorphan", hz / 10); 453 454 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 455 g_access_rel(cp, -cp->acr, -cp->acw, -cp->ace); 456 457 g_detach(cp); 458 g_destroy_consumer(cp); 459 g_destroy_geom(gp); 460 } 461 462 DECLARE_GEOM_CLASS(g_dev_class, g_dev); 463