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, minor2unit(MAXMINOR), 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 cp = NULL; 117 else 118 cp = dev->si_drv2; 119 return (cp->provider); 120 } 121 122 123 static struct g_geom * 124 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused) 125 { 126 struct g_geom *gp; 127 struct g_consumer *cp; 128 int error; 129 struct cdev *dev; 130 u_int unit; 131 132 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name); 133 g_topology_assert(); 134 LIST_FOREACH(cp, &pp->consumers, consumers) 135 if (cp->geom->class == mp) 136 return (NULL); 137 gp = g_new_geomf(mp, pp->name); 138 cp = g_new_consumer(gp); 139 error = g_attach(cp, pp); 140 KASSERT(error == 0, 141 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error)); 142 /* 143 * XXX: I'm not 100% sure we can call make_dev(9) without Giant 144 * yet. Once we can, we don't need to drop topology here either. 145 */ 146 unit = alloc_unr(unithdr); 147 g_topology_unlock(); 148 mtx_lock(&Giant); 149 dev = make_dev(&g_dev_cdevsw, unit2minor(unit), 150 UID_ROOT, GID_OPERATOR, 0640, gp->name); 151 if (pp->flags & G_PF_CANDELETE) 152 dev->si_flags |= SI_CANDELETE; 153 mtx_unlock(&Giant); 154 g_topology_lock(); 155 dev->si_iosize_max = MAXPHYS; 156 gp->softc = dev; 157 dev->si_drv1 = gp; 158 dev->si_drv2 = cp; 159 return (gp); 160 } 161 162 static int 163 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td) 164 { 165 struct g_geom *gp; 166 struct g_consumer *cp; 167 int error, r, w, e; 168 169 gp = dev->si_drv1; 170 cp = dev->si_drv2; 171 if (gp == NULL || cp == NULL || gp->softc != dev) 172 return(ENXIO); /* g_dev_taste() not done yet */ 173 174 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", 175 gp->name, flags, fmt, td); 176 177 r = flags & FREAD ? 1 : 0; 178 w = flags & FWRITE ? 1 : 0; 179 #ifdef notyet 180 e = flags & O_EXCL ? 1 : 0; 181 #else 182 e = 0; 183 #endif 184 if (w) { 185 /* 186 * When running in very secure mode, do not allow 187 * opens for writing of any disks. 188 */ 189 error = securelevel_ge(td->td_ucred, 2); 190 if (error) 191 return (error); 192 } 193 g_topology_lock(); 194 if (dev->si_devsw == NULL) 195 error = ENXIO; /* We were orphaned */ 196 else 197 error = g_access(cp, r, w, e); 198 g_topology_unlock(); 199 return(error); 200 } 201 202 static int 203 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td) 204 { 205 struct g_geom *gp; 206 struct g_consumer *cp; 207 int error, r, w, e, i; 208 209 gp = dev->si_drv1; 210 cp = dev->si_drv2; 211 if (gp == NULL || cp == NULL) 212 return(ENXIO); 213 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)", 214 gp->name, flags, fmt, td); 215 r = flags & FREAD ? -1 : 0; 216 w = flags & FWRITE ? -1 : 0; 217 #ifdef notyet 218 e = flags & O_EXCL ? -1 : 0; 219 #else 220 e = 0; 221 #endif 222 g_topology_lock(); 223 if (dev->si_devsw == NULL) 224 error = ENXIO; /* We were orphaned */ 225 else 226 error = g_access(cp, r, w, e); 227 for (i = 0; i < 10 * hz;) { 228 if (cp->acr != 0 || cp->acw != 0) 229 break; 230 if (cp->nstart == cp->nend) 231 break; 232 tsleep(&i, PRIBIO, "gdevwclose", hz / 10); 233 i += hz / 10; 234 } 235 if (cp->acr == 0 && cp->acw == 0 && cp->nstart != cp->nend) { 236 printf("WARNING: Final close of geom_dev(%s) %s %s\n", 237 gp->name, 238 "still has outstanding I/O after 10 seconds.", 239 "Completing close anyway, panic may happen later."); 240 } 241 g_topology_unlock(); 242 return (error); 243 } 244 245 /* 246 * XXX: Until we have unmessed the ioctl situation, there is a race against 247 * XXX: a concurrent orphanization. We cannot close it by holding topology 248 * XXX: since that would prevent us from doing our job, and stalling events 249 * XXX: will break (actually: stall) the BSD disklabel hacks. 250 */ 251 static int 252 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 253 { 254 struct g_geom *gp; 255 struct g_consumer *cp; 256 struct g_kerneldump kd; 257 int i, error; 258 u_int u; 259 260 gp = dev->si_drv1; 261 cp = dev->si_drv2; 262 263 error = 0; 264 KASSERT(cp->acr || cp->acw, 265 ("Consumer with zero access count in g_dev_ioctl")); 266 267 i = IOCPARM_LEN(cmd); 268 switch (cmd) { 269 case DIOCGSECTORSIZE: 270 *(u_int *)data = cp->provider->sectorsize; 271 if (*(u_int *)data == 0) 272 error = ENOENT; 273 break; 274 case DIOCGMEDIASIZE: 275 *(off_t *)data = cp->provider->mediasize; 276 if (*(off_t *)data == 0) 277 error = ENOENT; 278 break; 279 case DIOCGFWSECTORS: 280 error = g_io_getattr("GEOM::fwsectors", cp, &i, data); 281 if (error == 0 && *(u_int *)data == 0) 282 error = ENOENT; 283 break; 284 case DIOCGFWHEADS: 285 error = g_io_getattr("GEOM::fwheads", cp, &i, data); 286 if (error == 0 && *(u_int *)data == 0) 287 error = ENOENT; 288 break; 289 case DIOCGFRONTSTUFF: 290 error = g_io_getattr("GEOM::frontstuff", cp, &i, data); 291 break; 292 case DIOCSKERNELDUMP: 293 u = *((u_int *)data); 294 if (!u) { 295 set_dumper(NULL); 296 error = 0; 297 break; 298 } 299 kd.offset = 0; 300 kd.length = OFF_MAX; 301 i = sizeof kd; 302 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd); 303 if (!error) 304 dev->si_flags |= SI_DUMPDEV; 305 break; 306 307 default: 308 if (cp->provider->geom->ioctl != NULL) { 309 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td); 310 } else { 311 error = ENOIOCTL; 312 } 313 } 314 315 return (error); 316 } 317 318 static void 319 g_dev_done(struct bio *bp2) 320 { 321 struct bio *bp; 322 323 bp = bp2->bio_parent; 324 bp->bio_error = bp2->bio_error; 325 if (bp->bio_error != 0) { 326 g_trace(G_T_BIO, "g_dev_done(%p) had error %d", 327 bp2, bp->bio_error); 328 bp->bio_flags |= BIO_ERROR; 329 } else { 330 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", 331 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed); 332 } 333 bp->bio_resid = bp->bio_length - bp2->bio_completed; 334 bp->bio_completed = bp2->bio_completed; 335 g_destroy_bio(bp2); 336 biodone(bp); 337 } 338 339 static void 340 g_dev_strategy(struct bio *bp) 341 { 342 struct g_consumer *cp; 343 struct bio *bp2; 344 struct cdev *dev; 345 346 KASSERT(bp->bio_cmd == BIO_READ || 347 bp->bio_cmd == BIO_WRITE || 348 bp->bio_cmd == BIO_DELETE, 349 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); 350 dev = bp->bio_dev; 351 cp = dev->si_drv2; 352 KASSERT(cp->acr || cp->acw, 353 ("Consumer with zero access count in g_dev_strategy")); 354 355 if ((bp->bio_offset % cp->provider->sectorsize) != 0 || 356 (bp->bio_bcount % cp->provider->sectorsize) != 0) { 357 biofinish(bp, NULL, EINVAL); 358 return; 359 } 360 361 for (;;) { 362 /* 363 * XXX: This is not an ideal solution, but I belive it to 364 * XXX: deadlock safe, all things considered. 365 */ 366 bp2 = g_clone_bio(bp); 367 if (bp2 != NULL) 368 break; 369 tsleep(&bp, PRIBIO, "gdstrat", hz / 10); 370 } 371 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); 372 bp2->bio_done = g_dev_done; 373 g_trace(G_T_BIO, 374 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", 375 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, 376 bp2->bio_data, bp2->bio_cmd); 377 g_io_request(bp2, cp); 378 KASSERT(cp->acr || cp->acw, 379 ("g_dev_strategy raced with g_dev_close and lost")); 380 381 } 382 383 /* 384 * g_dev_orphan() 385 * 386 * Called from below when the provider orphaned us. 387 * - Clear any dump settings. 388 * - Destroy the struct cdev *to prevent any more request from coming in. The 389 * provider is already marked with an error, so anything which comes in 390 * in the interrim will be returned immediately. 391 * - Wait for any outstanding I/O to finish. 392 * - Set our access counts to zero, whatever they were. 393 * - Detach and self-destruct. 394 */ 395 396 static void 397 g_dev_orphan(struct g_consumer *cp) 398 { 399 struct g_geom *gp; 400 struct cdev *dev; 401 u_int unit; 402 403 g_topology_assert(); 404 gp = cp->geom; 405 dev = gp->softc; 406 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name); 407 408 /* Reset any dump-area set on this device */ 409 if (dev->si_flags & SI_DUMPDEV) 410 set_dumper(NULL); 411 412 /* Destroy the struct cdev *so we get no more requests */ 413 unit = dev2unit(dev); 414 destroy_dev(dev); 415 free_unr(unithdr, unit); 416 417 /* Wait for the cows to come home */ 418 while (cp->nstart != cp->nend) 419 msleep(&dev, NULL, PRIBIO, "gdevorphan", hz / 10); 420 421 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 422 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 423 424 g_detach(cp); 425 g_destroy_consumer(cp); 426 g_destroy_geom(gp); 427 } 428 429 DECLARE_GEOM_CLASS(g_dev_class, g_dev); 430