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