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_maj = GEOM_MAJOR, 71 .d_flags = D_DISK | D_TRACKCLOSE, 72 }; 73 74 static g_taste_t g_dev_taste; 75 static g_orphan_t g_dev_orphan; 76 77 static struct g_class g_dev_class = { 78 .name = "DEV", 79 .taste = g_dev_taste, 80 }; 81 82 void 83 g_dev_print(void) 84 { 85 struct g_geom *gp; 86 char const *p = ""; 87 88 LIST_FOREACH(gp, &g_dev_class.geom, geom) { 89 printf("%s%s", p, gp->name); 90 p = " "; 91 } 92 printf("\n"); 93 } 94 95 struct g_provider * 96 g_dev_getprovider(struct cdev *dev) 97 { 98 struct g_consumer *cp; 99 100 if (dev == NULL) 101 return (NULL); 102 if (devsw(dev) != &g_dev_cdevsw) 103 return (NULL); 104 cp = dev->si_drv2; 105 return (cp->provider); 106 } 107 108 109 static struct g_geom * 110 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused) 111 { 112 struct g_geom *gp; 113 struct g_consumer *cp; 114 static int unit = GEOM_MINOR_PROVIDERS; 115 int error; 116 struct cdev *dev; 117 118 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name); 119 g_topology_assert(); 120 LIST_FOREACH(cp, &pp->consumers, consumers) 121 if (cp->geom->class == mp) 122 return (NULL); 123 gp = g_new_geomf(mp, pp->name); 124 gp->orphan = g_dev_orphan; 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 /* 130 * XXX: I'm not 100% sure we can call make_dev(9) without Giant 131 * yet. Once we can, we don't need to drop topology here either. 132 */ 133 g_topology_unlock(); 134 mtx_lock(&Giant); 135 dev = make_dev(&g_dev_cdevsw, unit2minor(unit++), 136 UID_ROOT, GID_OPERATOR, 0640, gp->name); 137 if (pp->flags & G_PF_CANDELETE) 138 dev->si_flags |= SI_CANDELETE; 139 mtx_unlock(&Giant); 140 g_topology_lock(); 141 dev->si_iosize_max = MAXPHYS; 142 dev->si_stripesize = pp->stripesize; 143 dev->si_stripeoffset = pp->stripeoffset; 144 gp->softc = dev; 145 dev->si_drv1 = gp; 146 dev->si_drv2 = cp; 147 return (gp); 148 } 149 150 static int 151 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td) 152 { 153 struct g_geom *gp; 154 struct g_consumer *cp; 155 int error, r, w, e; 156 157 gp = dev->si_drv1; 158 cp = dev->si_drv2; 159 if (gp == NULL || cp == NULL || gp->softc != dev) 160 return(ENXIO); /* g_dev_taste() not done yet */ 161 162 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", 163 gp->name, flags, fmt, td); 164 165 r = flags & FREAD ? 1 : 0; 166 w = flags & FWRITE ? 1 : 0; 167 #ifdef notyet 168 e = flags & O_EXCL ? 1 : 0; 169 #else 170 e = 0; 171 #endif 172 if (w) { 173 /* 174 * When running in very secure mode, do not allow 175 * opens for writing of any disks. 176 */ 177 error = securelevel_ge(td->td_ucred, 2); 178 if (error) 179 return (error); 180 } 181 g_topology_lock(); 182 if (dev->si_devsw == NULL) 183 error = ENXIO; /* We were orphaned */ 184 else 185 error = g_access(cp, r, w, e); 186 g_topology_unlock(); 187 g_waitidle(); 188 if (!error) 189 dev->si_bsize_phys = cp->provider->sectorsize; 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 tsleep(&i, PRIBIO, "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 g_waitidle(); 234 return (error); 235 } 236 237 /* 238 * XXX: Until we have unmessed the ioctl situation, there is a race against 239 * XXX: a concurrent orphanization. We cannot close it by holding topology 240 * XXX: since that would prevent us from doing our job, and stalling events 241 * XXX: will break (actually: stall) the BSD disklabel hacks. 242 */ 243 static int 244 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 245 { 246 struct g_geom *gp; 247 struct g_consumer *cp; 248 struct g_kerneldump kd; 249 int i, error; 250 u_int u; 251 252 gp = dev->si_drv1; 253 cp = dev->si_drv2; 254 255 error = 0; 256 KASSERT(cp->acr || cp->acw, 257 ("Consumer with zero access count in g_dev_ioctl")); 258 259 i = IOCPARM_LEN(cmd); 260 switch (cmd) { 261 case DIOCGSECTORSIZE: 262 *(u_int *)data = cp->provider->sectorsize; 263 if (*(u_int *)data == 0) 264 error = ENOENT; 265 break; 266 case DIOCGMEDIASIZE: 267 *(off_t *)data = cp->provider->mediasize; 268 if (*(off_t *)data == 0) 269 error = ENOENT; 270 break; 271 case DIOCGFWSECTORS: 272 error = g_io_getattr("GEOM::fwsectors", cp, &i, data); 273 if (error == 0 && *(u_int *)data == 0) 274 error = ENOENT; 275 break; 276 case DIOCGFWHEADS: 277 error = g_io_getattr("GEOM::fwheads", cp, &i, data); 278 if (error == 0 && *(u_int *)data == 0) 279 error = ENOENT; 280 break; 281 case DIOCGFRONTSTUFF: 282 error = g_io_getattr("GEOM::frontstuff", cp, &i, data); 283 break; 284 case DIOCSKERNELDUMP: 285 u = *((u_int *)data); 286 if (!u) { 287 set_dumper(NULL); 288 error = 0; 289 break; 290 } 291 kd.offset = 0; 292 kd.length = OFF_MAX; 293 i = sizeof kd; 294 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd); 295 if (!error) 296 dev->si_flags |= SI_DUMPDEV; 297 break; 298 299 default: 300 if (cp->provider->geom->ioctl != NULL) { 301 error = cp->provider->geom->ioctl(cp->provider, cmd, data, td); 302 } else { 303 error = ENOIOCTL; 304 } 305 } 306 307 g_waitidle(); 308 return (error); 309 } 310 311 static void 312 g_dev_done(struct bio *bp2) 313 { 314 struct bio *bp; 315 316 bp = bp2->bio_parent; 317 bp->bio_error = bp2->bio_error; 318 if (bp->bio_error != 0) { 319 g_trace(G_T_BIO, "g_dev_done(%p) had error %d", 320 bp2, bp->bio_error); 321 bp->bio_flags |= BIO_ERROR; 322 } else { 323 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", 324 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed); 325 } 326 bp->bio_resid = bp->bio_bcount - bp2->bio_completed; 327 g_destroy_bio(bp2); 328 biodone(bp); 329 } 330 331 static void 332 g_dev_strategy(struct bio *bp) 333 { 334 struct g_consumer *cp; 335 struct bio *bp2; 336 struct cdev *dev; 337 338 KASSERT(bp->bio_cmd == BIO_READ || 339 bp->bio_cmd == BIO_WRITE || 340 bp->bio_cmd == BIO_DELETE, 341 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); 342 dev = bp->bio_dev; 343 cp = dev->si_drv2; 344 KASSERT(cp->acr || cp->acw, 345 ("Consumer with zero access count in g_dev_strategy")); 346 347 for (;;) { 348 /* 349 * XXX: This is not an ideal solution, but I belive it to 350 * XXX: deadlock safe, all things considered. 351 */ 352 bp2 = g_clone_bio(bp); 353 if (bp2 != NULL) 354 break; 355 tsleep(&bp, PRIBIO, "gdstrat", hz / 10); 356 } 357 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); 358 bp2->bio_length = (off_t)bp->bio_bcount; 359 bp2->bio_done = g_dev_done; 360 g_trace(G_T_BIO, 361 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", 362 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, 363 bp2->bio_data, bp2->bio_cmd); 364 g_io_request(bp2, cp); 365 KASSERT(cp->acr || cp->acw, 366 ("g_dev_strategy raced with g_dev_close and lost")); 367 368 } 369 370 /* 371 * g_dev_orphan() 372 * 373 * Called from below when the provider orphaned us. 374 * - Clear any dump settings. 375 * - Destroy the struct cdev *to prevent any more request from coming in. The 376 * provider is already marked with an error, so anything which comes in 377 * in the interrim will be returned immediately. 378 * - Wait for any outstanding I/O to finish. 379 * - Set our access counts to zero, whatever they were. 380 * - Detach and self-destruct. 381 */ 382 383 static void 384 g_dev_orphan(struct g_consumer *cp) 385 { 386 struct g_geom *gp; 387 struct cdev *dev; 388 389 g_topology_assert(); 390 gp = cp->geom; 391 dev = gp->softc; 392 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name); 393 394 /* Reset any dump-area set on this device */ 395 if (dev->si_flags & SI_DUMPDEV) 396 set_dumper(NULL); 397 398 /* Destroy the struct cdev *so we get no more requests */ 399 destroy_dev(dev); 400 401 /* Wait for the cows to come home */ 402 while (cp->nstart != cp->nend) 403 msleep(&dev, NULL, PRIBIO, "gdevorphan", hz / 10); 404 405 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 406 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 407 408 g_detach(cp); 409 g_destroy_consumer(cp); 410 g_destroy_geom(gp); 411 } 412 413 DECLARE_GEOM_CLASS(g_dev_class, g_dev); 414