geom_dev.c revision 137029
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 137029 2004-10-29 07:16:37Z phk $"); 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_flags = D_DISK | D_TRACKCLOSE, 71}; 72 73static g_taste_t g_dev_taste; 74static g_orphan_t g_dev_orphan; 75static g_init_t g_dev_init; 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 .init = g_dev_init, 83}; 84 85static struct unrhdr *unithdr; /* Locked by topology */ 86 87static void 88g_dev_init(struct g_class *mp) 89{ 90 91 /* XXX: should have a #define MAX_UNIT_MINOR */ 92 unithdr = new_unrhdr(0, 0xffffff); 93} 94 95void 96g_dev_print(void) 97{ 98 struct g_geom *gp; 99 char const *p = ""; 100 101 LIST_FOREACH(gp, &g_dev_class.geom, geom) { 102 printf("%s%s", p, gp->name); 103 p = " "; 104 } 105 printf("\n"); 106} 107 108struct g_provider * 109g_dev_getprovider(struct cdev *dev) 110{ 111 struct g_consumer *cp; 112 113 g_topology_assert(); 114 if (dev == NULL) 115 return (NULL); 116 if (dev->si_devsw != &g_dev_cdevsw) 117 cp = NULL; 118 else 119 cp = dev->si_drv2; 120 return (cp->provider); 121} 122 123 124static struct g_geom * 125g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused) 126{ 127 struct g_geom *gp; 128 struct g_consumer *cp; 129 int error; 130 struct cdev *dev; 131 u_int unit; 132 133 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name); 134 g_topology_assert(); 135 LIST_FOREACH(cp, &pp->consumers, consumers) 136 if (cp->geom->class == mp) 137 return (NULL); 138 gp = g_new_geomf(mp, pp->name); 139 cp = g_new_consumer(gp); 140 error = g_attach(cp, pp); 141 KASSERT(error == 0, 142 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error)); 143 /* 144 * XXX: I'm not 100% sure we can call make_dev(9) without Giant 145 * yet. Once we can, we don't need to drop topology here either. 146 */ 147 unit = alloc_unr(unithdr); 148 g_topology_unlock(); 149 mtx_lock(&Giant); 150 dev = make_dev(&g_dev_cdevsw, unit2minor(unit), 151 UID_ROOT, GID_OPERATOR, 0640, gp->name); 152 if (pp->flags & G_PF_CANDELETE) 153 dev->si_flags |= SI_CANDELETE; 154 mtx_unlock(&Giant); 155 g_topology_lock(); 156 dev->si_iosize_max = MAXPHYS; 157 gp->softc = dev; 158 dev->si_drv1 = gp; 159 dev->si_drv2 = cp; 160 return (gp); 161} 162 163static int 164g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td) 165{ 166 struct g_geom *gp; 167 struct g_consumer *cp; 168 int error, r, w, e; 169 170 gp = dev->si_drv1; 171 cp = dev->si_drv2; 172 if (gp == NULL || cp == NULL || gp->softc != dev) 173 return(ENXIO); /* g_dev_taste() not done yet */ 174 175 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", 176 gp->name, flags, fmt, td); 177 178 r = flags & FREAD ? 1 : 0; 179 w = flags & FWRITE ? 1 : 0; 180#ifdef notyet 181 e = flags & O_EXCL ? 1 : 0; 182#else 183 e = 0; 184#endif 185 if (w) { 186 /* 187 * When running in very secure mode, do not allow 188 * opens for writing of any disks. 189 */ 190 error = securelevel_ge(td->td_ucred, 2); 191 if (error) 192 return (error); 193 } 194 g_topology_lock(); 195 if (dev->si_devsw == NULL) 196 error = ENXIO; /* We were orphaned */ 197 else 198 error = g_access(cp, r, w, e); 199 g_topology_unlock(); 200 if (!error) 201 dev->si_bsize_phys = cp->provider->sectorsize; 202 return(error); 203} 204 205static int 206g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td) 207{ 208 struct g_geom *gp; 209 struct g_consumer *cp; 210 int error, r, w, e, i; 211 212 gp = dev->si_drv1; 213 cp = dev->si_drv2; 214 if (gp == NULL || cp == NULL) 215 return(ENXIO); 216 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)", 217 gp->name, flags, fmt, td); 218 r = flags & FREAD ? -1 : 0; 219 w = flags & FWRITE ? -1 : 0; 220#ifdef notyet 221 e = flags & O_EXCL ? -1 : 0; 222#else 223 e = 0; 224#endif 225 g_topology_lock(); 226 if (dev->si_devsw == NULL) 227 error = ENXIO; /* We were orphaned */ 228 else 229 error = g_access(cp, r, w, e); 230 for (i = 0; i < 10 * hz;) { 231 if (cp->acr != 0 || cp->acw != 0) 232 break; 233 if (cp->nstart == cp->nend) 234 break; 235 tsleep(&i, PRIBIO, "gdevwclose", hz / 10); 236 i += hz / 10; 237 } 238 if (cp->acr == 0 && cp->acw == 0 && cp->nstart != cp->nend) { 239 printf("WARNING: Final close of geom_dev(%s) %s %s\n", 240 gp->name, 241 "still has outstanding I/O after 10 seconds.", 242 "Completing close anyway, panic may happen later."); 243 } 244 g_topology_unlock(); 245 return (error); 246} 247 248/* 249 * XXX: Until we have unmessed the ioctl situation, there is a race against 250 * XXX: a concurrent orphanization. We cannot close it by holding topology 251 * XXX: since that would prevent us from doing our job, and stalling events 252 * XXX: will break (actually: stall) the BSD disklabel hacks. 253 */ 254static int 255g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 256{ 257 struct g_geom *gp; 258 struct g_consumer *cp; 259 struct g_kerneldump kd; 260 int i, error; 261 u_int u; 262 263 gp = dev->si_drv1; 264 cp = dev->si_drv2; 265 266 error = 0; 267 KASSERT(cp->acr || cp->acw, 268 ("Consumer with zero access count in g_dev_ioctl")); 269 270 i = IOCPARM_LEN(cmd); 271 switch (cmd) { 272 case DIOCGSECTORSIZE: 273 *(u_int *)data = cp->provider->sectorsize; 274 if (*(u_int *)data == 0) 275 error = ENOENT; 276 break; 277 case DIOCGMEDIASIZE: 278 *(off_t *)data = cp->provider->mediasize; 279 if (*(off_t *)data == 0) 280 error = ENOENT; 281 break; 282 case DIOCGFWSECTORS: 283 error = g_io_getattr("GEOM::fwsectors", cp, &i, data); 284 if (error == 0 && *(u_int *)data == 0) 285 error = ENOENT; 286 break; 287 case DIOCGFWHEADS: 288 error = g_io_getattr("GEOM::fwheads", cp, &i, data); 289 if (error == 0 && *(u_int *)data == 0) 290 error = ENOENT; 291 break; 292 case DIOCGFRONTSTUFF: 293 error = g_io_getattr("GEOM::frontstuff", cp, &i, data); 294 break; 295 case DIOCSKERNELDUMP: 296 u = *((u_int *)data); 297 if (!u) { 298 set_dumper(NULL); 299 error = 0; 300 break; 301 } 302 kd.offset = 0; 303 kd.length = OFF_MAX; 304 i = sizeof kd; 305 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd); 306 if (!error) 307 dev->si_flags |= SI_DUMPDEV; 308 break; 309 310 default: 311 if (cp->provider->geom->ioctl != NULL) { 312 error = cp->provider->geom->ioctl(cp->provider, cmd, data, td); 313 } else { 314 error = ENOIOCTL; 315 } 316 } 317 318 return (error); 319} 320 321static void 322g_dev_done(struct bio *bp2) 323{ 324 struct bio *bp; 325 326 bp = bp2->bio_parent; 327 bp->bio_error = bp2->bio_error; 328 if (bp->bio_error != 0) { 329 g_trace(G_T_BIO, "g_dev_done(%p) had error %d", 330 bp2, bp->bio_error); 331 bp->bio_flags |= BIO_ERROR; 332 } else { 333 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", 334 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed); 335 } 336 bp->bio_resid = bp->bio_length - bp2->bio_completed; 337 bp->bio_completed = bp2->bio_completed; 338 g_destroy_bio(bp2); 339 biodone(bp); 340} 341 342static void 343g_dev_strategy(struct bio *bp) 344{ 345 struct g_consumer *cp; 346 struct bio *bp2; 347 struct cdev *dev; 348 349 KASSERT(bp->bio_cmd == BIO_READ || 350 bp->bio_cmd == BIO_WRITE || 351 bp->bio_cmd == BIO_DELETE, 352 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); 353 dev = bp->bio_dev; 354 cp = dev->si_drv2; 355 KASSERT(cp->acr || cp->acw, 356 ("Consumer with zero access count in g_dev_strategy")); 357 358 if ((bp->bio_offset % cp->provider->sectorsize) != 0 || 359 (bp->bio_bcount % cp->provider->sectorsize) != 0) { 360 biofinish(bp, NULL, EINVAL); 361 return; 362 } 363 364 for (;;) { 365 /* 366 * XXX: This is not an ideal solution, but I belive it to 367 * XXX: deadlock safe, all things considered. 368 */ 369 bp2 = g_clone_bio(bp); 370 if (bp2 != NULL) 371 break; 372 tsleep(&bp, PRIBIO, "gdstrat", hz / 10); 373 } 374 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); 375 bp2->bio_done = g_dev_done; 376 g_trace(G_T_BIO, 377 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", 378 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, 379 bp2->bio_data, bp2->bio_cmd); 380 g_io_request(bp2, cp); 381 KASSERT(cp->acr || cp->acw, 382 ("g_dev_strategy raced with g_dev_close and lost")); 383 384} 385 386/* 387 * g_dev_orphan() 388 * 389 * Called from below when the provider orphaned us. 390 * - Clear any dump settings. 391 * - Destroy the struct cdev *to prevent any more request from coming in. The 392 * provider is already marked with an error, so anything which comes in 393 * in the interrim will be returned immediately. 394 * - Wait for any outstanding I/O to finish. 395 * - Set our access counts to zero, whatever they were. 396 * - Detach and self-destruct. 397 */ 398 399static void 400g_dev_orphan(struct g_consumer *cp) 401{ 402 struct g_geom *gp; 403 struct cdev *dev; 404 u_int unit; 405 406 g_topology_assert(); 407 gp = cp->geom; 408 dev = gp->softc; 409 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name); 410 411 /* Reset any dump-area set on this device */ 412 if (dev->si_flags & SI_DUMPDEV) 413 set_dumper(NULL); 414 415 /* Destroy the struct cdev *so we get no more requests */ 416 unit = dev2unit(dev); 417 destroy_dev(dev); 418 free_unr(unithdr, unit); 419 420 /* Wait for the cows to come home */ 421 while (cp->nstart != cp->nend) 422 msleep(&dev, NULL, PRIBIO, "gdevorphan", hz / 10); 423 424 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 425 g_access(cp, -cp->acr, -cp->acw, -cp->ace); 426 427 g_detach(cp); 428 g_destroy_consumer(cp); 429 g_destroy_geom(gp); 430} 431 432DECLARE_GEOM_CLASS(g_dev_class, g_dev); 433