geom_dev.c revision 135865
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: head/sys/geom/geom_dev.c 135865 2004-09-27 22:10:01Z pjd $"); 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 56static d_open_t g_dev_open; 57static d_close_t g_dev_close; 58static d_strategy_t g_dev_strategy; 59static d_ioctl_t g_dev_ioctl; 60 61static 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 74static g_taste_t g_dev_taste; 75static g_orphan_t g_dev_orphan; 76 77static struct g_class g_dev_class = { 78 .name = "DEV", 79 .version = G_VERSION, 80 .taste = g_dev_taste, 81 .orphan = g_dev_orphan, 82}; 83 84void 85g_dev_print(void) 86{ 87 struct g_geom *gp; 88 char const *p = ""; 89 90 LIST_FOREACH(gp, &g_dev_class.geom, geom) { 91 printf("%s%s", p, gp->name); 92 p = " "; 93 } 94 printf("\n"); 95} 96 97struct g_provider * 98g_dev_getprovider(struct cdev *dev) 99{ 100 struct g_consumer *cp; 101 102 g_topology_assert(); 103 if (dev == NULL) 104 return (NULL); 105 if (dev->si_devsw != &g_dev_cdevsw) 106 cp = NULL; 107 else 108 cp = dev->si_drv2; 109 return (cp->provider); 110} 111 112 113static struct g_geom * 114g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused) 115{ 116 struct g_geom *gp; 117 struct g_consumer *cp; 118 static int unit = GEOM_MINOR_PROVIDERS; 119 int error; 120 struct cdev *dev; 121 122 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name); 123 g_topology_assert(); 124 LIST_FOREACH(cp, &pp->consumers, consumers) 125 if (cp->geom->class == mp) 126 return (NULL); 127 gp = g_new_geomf(mp, pp->name); 128 cp = g_new_consumer(gp); 129 error = g_attach(cp, pp); 130 KASSERT(error == 0, 131 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error)); 132 /* 133 * XXX: I'm not 100% sure we can call make_dev(9) without Giant 134 * yet. Once we can, we don't need to drop topology here either. 135 */ 136 g_topology_unlock(); 137 mtx_lock(&Giant); 138 dev = make_dev(&g_dev_cdevsw, unit2minor(unit++), 139 UID_ROOT, GID_OPERATOR, 0640, gp->name); 140 if (pp->flags & G_PF_CANDELETE) 141 dev->si_flags |= SI_CANDELETE; 142 mtx_unlock(&Giant); 143 g_topology_lock(); 144 dev->si_iosize_max = MAXPHYS; 145 dev->si_stripesize = pp->stripesize; 146 dev->si_stripeoffset = pp->stripeoffset; 147 gp->softc = dev; 148 dev->si_drv1 = gp; 149 dev->si_drv2 = cp; 150 return (gp); 151} 152 153static int 154g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td) 155{ 156 struct g_geom *gp; 157 struct g_consumer *cp; 158 int error, r, w, e; 159 160 gp = dev->si_drv1; 161 cp = dev->si_drv2; 162 if (gp == NULL || cp == NULL || gp->softc != dev) 163 return(ENXIO); /* g_dev_taste() not done yet */ 164 165 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", 166 gp->name, flags, fmt, td); 167 168 r = flags & FREAD ? 1 : 0; 169 w = flags & FWRITE ? 1 : 0; 170#ifdef notyet 171 e = flags & O_EXCL ? 1 : 0; 172#else 173 e = 0; 174#endif 175 if (w) { 176 /* 177 * When running in very secure mode, do not allow 178 * opens for writing of any disks. 179 */ 180 error = securelevel_ge(td->td_ucred, 2); 181 if (error) 182 return (error); 183 } 184 g_topology_lock(); 185 if (dev->si_devsw == NULL) 186 error = ENXIO; /* We were orphaned */ 187 else 188 error = g_access(cp, r, w, e); 189 g_topology_unlock(); 190 g_waitidle(); 191 if (!error) 192 dev->si_bsize_phys = cp->provider->sectorsize; 193 return(error); 194} 195 196static int 197g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td) 198{ 199 struct g_geom *gp; 200 struct g_consumer *cp; 201 int error, r, w, e, i; 202 203 gp = dev->si_drv1; 204 cp = dev->si_drv2; 205 if (gp == NULL || cp == NULL) 206 return(ENXIO); 207 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)", 208 gp->name, flags, fmt, td); 209 r = flags & FREAD ? -1 : 0; 210 w = flags & FWRITE ? -1 : 0; 211#ifdef notyet 212 e = flags & O_EXCL ? -1 : 0; 213#else 214 e = 0; 215#endif 216 g_topology_lock(); 217 if (dev->si_devsw == NULL) 218 error = ENXIO; /* We were orphaned */ 219 else 220 error = g_access(cp, r, w, e); 221 for (i = 0; i < 10 * hz;) { 222 if (cp->acr != 0 || cp->acw != 0) 223 break; 224 if (cp->nstart == cp->nend) 225 break; 226 tsleep(&i, PRIBIO, "gdevwclose", hz / 10); 227 i += hz / 10; 228 } 229 if (cp->acr == 0 && cp->acw == 0 && cp->nstart != cp->nend) { 230 printf("WARNING: Final close of geom_dev(%s) %s %s\n", 231 gp->name, 232 "still has outstanding I/O after 10 seconds.", 233 "Completing close anyway, panic may happen later."); 234 } 235 g_topology_unlock(); 236 g_waitidle(); 237 return (error); 238} 239 240/* 241 * XXX: Until we have unmessed the ioctl situation, there is a race against 242 * XXX: a concurrent orphanization. We cannot close it by holding topology 243 * XXX: since that would prevent us from doing our job, and stalling events 244 * XXX: will break (actually: stall) the BSD disklabel hacks. 245 */ 246static int 247g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 248{ 249 struct g_geom *gp; 250 struct g_consumer *cp; 251 struct g_kerneldump kd; 252 int i, error; 253 u_int u; 254 255 gp = dev->si_drv1; 256 cp = dev->si_drv2; 257 258 error = 0; 259 KASSERT(cp->acr || cp->acw, 260 ("Consumer with zero access count in g_dev_ioctl")); 261 262 i = IOCPARM_LEN(cmd); 263 switch (cmd) { 264 case DIOCGSECTORSIZE: 265 *(u_int *)data = cp->provider->sectorsize; 266 if (*(u_int *)data == 0) 267 error = ENOENT; 268 break; 269 case DIOCGMEDIASIZE: 270 *(off_t *)data = cp->provider->mediasize; 271 if (*(off_t *)data == 0) 272 error = ENOENT; 273 break; 274 case DIOCGFWSECTORS: 275 error = g_io_getattr("GEOM::fwsectors", cp, &i, data); 276 if (error == 0 && *(u_int *)data == 0) 277 error = ENOENT; 278 break; 279 case DIOCGFWHEADS: 280 error = g_io_getattr("GEOM::fwheads", cp, &i, data); 281 if (error == 0 && *(u_int *)data == 0) 282 error = ENOENT; 283 break; 284 case DIOCGFRONTSTUFF: 285 error = g_io_getattr("GEOM::frontstuff", cp, &i, data); 286 break; 287 case DIOCSKERNELDUMP: 288 u = *((u_int *)data); 289 if (!u) { 290 set_dumper(NULL); 291 error = 0; 292 break; 293 } 294 kd.offset = 0; 295 kd.length = OFF_MAX; 296 i = sizeof kd; 297 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd); 298 if (!error) 299 dev->si_flags |= SI_DUMPDEV; 300 break; 301 302 default: 303 if (cp->provider->geom->ioctl != NULL) { 304 error = cp->provider->geom->ioctl(cp->provider, cmd, data, td); 305 } else { 306 error = ENOIOCTL; 307 } 308 } 309 310 g_waitidle(); 311 return (error); 312} 313 314static void 315g_dev_done(struct bio *bp2) 316{ 317 struct bio *bp; 318 319 bp = bp2->bio_parent; 320 bp->bio_error = bp2->bio_error; 321 if (bp->bio_error != 0) { 322 g_trace(G_T_BIO, "g_dev_done(%p) had error %d", 323 bp2, bp->bio_error); 324 bp->bio_flags |= BIO_ERROR; 325 } else { 326 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", 327 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed); 328 } 329 bp->bio_resid = bp->bio_bcount - bp2->bio_completed; 330 g_destroy_bio(bp2); 331 biodone(bp); 332} 333 334static void 335g_dev_strategy(struct bio *bp) 336{ 337 struct g_consumer *cp; 338 struct bio *bp2; 339 struct cdev *dev; 340 341 KASSERT(bp->bio_cmd == BIO_READ || 342 bp->bio_cmd == BIO_WRITE || 343 bp->bio_cmd == BIO_DELETE, 344 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); 345 dev = bp->bio_dev; 346 cp = dev->si_drv2; 347 KASSERT(cp->acr || cp->acw, 348 ("Consumer with zero access count in g_dev_strategy")); 349 350 if ((bp->bio_offset % cp->provider->sectorsize) != 0 || 351 (bp->bio_bcount % cp->provider->sectorsize) != 0) { 352 biofinish(bp, NULL, EINVAL); 353 return; 354 } 355 356 for (;;) { 357 /* 358 * XXX: This is not an ideal solution, but I belive it to 359 * XXX: deadlock safe, all things considered. 360 */ 361 bp2 = g_clone_bio(bp); 362 if (bp2 != NULL) 363 break; 364 tsleep(&bp, PRIBIO, "gdstrat", hz / 10); 365 } 366 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); 367 bp2->bio_length = (off_t)bp->bio_bcount; 368 bp2->bio_done = g_dev_done; 369 g_trace(G_T_BIO, 370 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", 371 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, 372 bp2->bio_data, bp2->bio_cmd); 373 g_io_request(bp2, cp); 374 KASSERT(cp->acr || cp->acw, 375 ("g_dev_strategy raced with g_dev_close and lost")); 376 377} 378 379/* 380 * g_dev_orphan() 381 * 382 * Called from below when the provider orphaned us. 383 * - Clear any dump settings. 384 * - Destroy the struct cdev *to prevent any more request from coming in. The 385 * provider is already marked with an error, so anything which comes in 386 * in the interrim will be returned immediately. 387 * - Wait for any outstanding I/O to finish. 388 * - Set our access counts to zero, whatever they were. 389 * - Detach and self-destruct. 390 */ 391 392static void 393g_dev_orphan(struct g_consumer *cp) 394{ 395 struct g_geom *gp; 396 struct cdev *dev; 397 398 g_topology_assert(); 399 gp = cp->geom; 400 dev = gp->softc; 401 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name); 402 403 /* Reset any dump-area set on this device */ 404 if (dev->si_flags & SI_DUMPDEV) 405 set_dumper(NULL); 406 407 /* Destroy the struct cdev *so we get no more requests */ 408 destroy_dev(dev); 409 410 /* Wait for the cows to come home */ 411 while (cp->nstart != cp->nend) 412 msleep(&dev, NULL, PRIBIO, "gdevorphan", hz / 10); 413 414 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 415 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 416 417 g_detach(cp); 418 g_destroy_consumer(cp); 419 g_destroy_geom(gp); 420} 421 422DECLARE_GEOM_CLASS(g_dev_class, g_dev); 423