geom_dev.c revision 112024
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 * $FreeBSD: head/sys/geom/geom_dev.c 112024 2003-03-09 09:28:45Z phk $ 36 */ 37 38#include <sys/param.h> 39#include <sys/stdint.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/errno.h> 48#include <sys/time.h> 49#include <sys/disk.h> 50#include <sys/fcntl.h> 51#include <geom/geom.h> 52#include <geom/geom_int.h> 53#include <geom/geom_stats.h> 54#include <machine/limits.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_open = g_dev_open, 63 .d_close = g_dev_close, 64 .d_read = physread, 65 .d_write = physwrite, 66 .d_ioctl = g_dev_ioctl, 67 .d_strategy = g_dev_strategy, 68 .d_name = "g_dev", 69 .d_maj = GEOM_MAJOR, 70 .d_flags = D_DISK | D_TRACKCLOSE, 71}; 72 73static g_taste_t g_dev_taste; 74static g_orphan_t g_dev_orphan; 75 76static struct g_class g_dev_class = { 77 "DEV", 78 g_dev_taste, 79 NULL, 80 G_CLASS_INITIALIZER 81}; 82 83int 84g_dev_print(void) 85{ 86 struct g_geom *gp; 87 88 if (LIST_EMPTY(&g_dev_class.geom)) 89 return (0); 90 printf("List of GEOM disk devices:\n "); 91 LIST_FOREACH(gp, &g_dev_class.geom, geom) 92 printf(" %s", gp->name); 93 printf("\n"); 94 return (1); 95} 96 97/* 98 * XXX: This is disgusting and wrong in every way imaginable: The only reason 99 * XXX: we have a clone function is because of the root-mount hack we currently 100 * XXX: employ. An improvment would be to unregister this cloner once we know 101 * XXX: we no longer need it. Ideally, root-fs would be mounted through DEVFS 102 * XXX: eliminating the need for this hack. 103 */ 104static void 105g_dev_clone(void *arg __unused, char *name, int namelen __unused, dev_t *dev) 106{ 107 struct g_geom *gp; 108 109 if (*dev != NODEV) 110 return; 111 112 g_waitidle(); 113 114 /* g_topology_lock(); */ 115 LIST_FOREACH(gp, &g_dev_class.geom, geom) { 116 if (strcmp(gp->name, name)) 117 continue; 118 *dev = gp->softc; 119 g_trace(G_T_TOPOLOGY, "g_dev_clone(%s) = %p", name, *dev); 120 return; 121 } 122 /* g_topology_unlock(); */ 123 return; 124} 125 126static void 127g_dev_register_cloner(void *foo __unused) 128{ 129 static int once; 130 131 /* XXX: why would this happen more than once ?? */ 132 if (!once) { 133 EVENTHANDLER_REGISTER(dev_clone, g_dev_clone, 0, 1000); 134 once++; 135 } 136} 137 138SYSINIT(geomdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,g_dev_register_cloner,NULL); 139 140static struct g_geom * 141g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused) 142{ 143 struct g_geom *gp; 144 struct g_consumer *cp; 145 static int unit = GEOM_MINOR_PROVIDERS; 146 int error; 147 dev_t dev; 148 149 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name); 150 g_topology_assert(); 151 LIST_FOREACH(cp, &pp->consumers, consumers) 152 if (cp->geom->class == mp) 153 return (NULL); 154 gp = g_new_geomf(mp, pp->name); 155 gp->orphan = g_dev_orphan; 156 cp = g_new_consumer(gp); 157 error = g_attach(cp, pp); 158 KASSERT(error == 0, 159 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error)); 160 /* 161 * XXX: I'm not 100% sure we can call make_dev(9) without Giant 162 * yet. Once we can, we don't need to drop topology here either. 163 */ 164 g_topology_unlock(); 165 mtx_lock(&Giant); 166 dev = make_dev(&g_dev_cdevsw, unit2minor(unit++), 167 UID_ROOT, GID_OPERATOR, 0640, gp->name); 168 if (pp->flags & G_PF_CANDELETE) 169 dev->si_flags |= SI_CANDELETE; 170 mtx_unlock(&Giant); 171 g_topology_lock(); 172 dev->si_iosize_max = MAXPHYS; 173 dev->si_stripesize = pp->stripesize; 174 dev->si_stripeoffset = pp->stripeoffset; 175 gp->softc = dev; 176 dev->si_drv1 = gp; 177 dev->si_drv2 = cp; 178 return (gp); 179} 180 181static int 182g_dev_open(dev_t dev, int flags, int fmt, struct thread *td) 183{ 184 struct g_geom *gp; 185 struct g_consumer *cp; 186 int error, r, w, e; 187 188 gp = dev->si_drv1; 189 cp = dev->si_drv2; 190 if (gp == NULL || cp == NULL) 191 return(ENXIO); 192 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", 193 gp->name, flags, fmt, td); 194 DROP_GIANT(); 195 g_topology_lock(); 196 r = flags & FREAD ? 1 : 0; 197 w = flags & FWRITE ? 1 : 0; 198#ifdef notyet 199 e = flags & O_EXCL ? 1 : 0; 200#else 201 e = 0; 202#endif 203 error = g_access_rel(cp, r, w, e); 204 g_topology_unlock(); 205 PICKUP_GIANT(); 206 g_waitidle(); 207 dev->si_bsize_phys = cp->provider->sectorsize; 208 return(error); 209} 210 211static int 212g_dev_close(dev_t dev, int flags, int fmt, struct thread *td) 213{ 214 struct g_geom *gp; 215 struct g_consumer *cp; 216 int error, r, w, e; 217 218 gp = dev->si_drv1; 219 cp = dev->si_drv2; 220 if (gp == NULL || cp == NULL) 221 return(ENXIO); 222 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)", 223 gp->name, flags, fmt, td); 224 DROP_GIANT(); 225 g_topology_lock(); 226 r = flags & FREAD ? -1 : 0; 227 w = flags & FWRITE ? -1 : 0; 228#ifdef notyet 229 e = flags & O_EXCL ? -1 : 0; 230#else 231 e = 0; 232#endif 233 error = g_access_rel(cp, r, w, e); 234 g_topology_unlock(); 235 PICKUP_GIANT(); 236 g_waitidle(); 237 return (error); 238} 239 240static int 241g_dev_ioctl(dev_t dev, u_long cmd, caddr_t data, int fflag, struct thread *td) 242{ 243 struct g_geom *gp, *gp2; 244 struct g_consumer *cp; 245 struct g_provider *pp2; 246 struct g_kerneldump kd; 247 int i, error; 248 u_int u; 249 struct g_ioctl *gio; 250 251 gp = dev->si_drv1; 252 cp = dev->si_drv2; 253 pp2 = cp->provider; 254 gp2 = pp2->geom; 255 gio = NULL; 256 257 error = 0; 258 DROP_GIANT(); 259 260 gio = NULL; 261 i = IOCPARM_LEN(cmd); 262 switch (cmd) { 263 case DIOCGSECTORSIZE: 264 *(u_int *)data = cp->provider->sectorsize; 265 if (*(u_int *)data == 0) 266 error = ENOENT; 267 break; 268 case DIOCGMEDIASIZE: 269 *(off_t *)data = cp->provider->mediasize; 270 if (*(off_t *)data == 0) 271 error = ENOENT; 272 break; 273 case DIOCGFWSECTORS: 274 error = g_io_getattr("GEOM::fwsectors", cp, &i, data); 275 if (error == 0 && *(u_int *)data == 0) 276 error = ENOENT; 277 break; 278 case DIOCGFWHEADS: 279 error = g_io_getattr("GEOM::fwheads", cp, &i, data); 280 if (error == 0 && *(u_int *)data == 0) 281 error = ENOENT; 282 break; 283 case DIOCGFRONTSTUFF: 284 error = g_io_getattr("GEOM::frontstuff", cp, &i, data); 285 break; 286 case DIOCSKERNELDUMP: 287 u = *((u_int *)data); 288 if (!u) { 289 set_dumper(NULL); 290 error = 0; 291 break; 292 } 293 kd.offset = 0; 294 kd.length = OFF_MAX; 295 i = sizeof kd; 296 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd); 297 if (!error) 298 dev->si_flags |= SI_DUMPDEV; 299 break; 300 301 default: 302 gio = g_malloc(sizeof *gio, M_WAITOK | M_ZERO); 303 gio->cmd = cmd; 304 gio->data = data; 305 gio->fflag = fflag; 306 gio->td = td; 307 i = sizeof *gio; 308 /* 309 * We always issue ioctls as getattr since the direction of data 310 * movement in ioctl is no indication of the ioctl being a "set" 311 * or "get" type ioctl or if such simplistic terms even apply 312 */ 313 error = g_io_getattr("GEOM::ioctl", cp, &i, gio); 314 break; 315 } 316 317 PICKUP_GIANT(); 318 if (error == EDIRIOCTL) { 319 KASSERT(gio != NULL, ("NULL gio but EDIRIOCTL")); 320 KASSERT(gio->func != NULL, ("NULL function but EDIRIOCTL")); 321 error = (gio->func)(gio->dev, cmd, data, fflag, td); 322 } 323 g_waitidle(); 324 if (gio != NULL && (error == EOPNOTSUPP || error == ENOIOCTL)) { 325 if (g_debugflags & G_T_TOPOLOGY) { 326 i = IOCGROUP(cmd); 327 printf("IOCTL(0x%lx) \"%s\"", cmd, gp->name); 328 if (i > ' ' && i <= '~') 329 printf(" '%c'", (int)IOCGROUP(cmd)); 330 else 331 printf(" 0x%lx", IOCGROUP(cmd)); 332 printf("/%ld ", cmd & 0xff); 333 if (cmd & IOC_IN) 334 printf("I"); 335 if (cmd & IOC_OUT) 336 printf("O"); 337 printf("(%ld) = ENOIOCTL\n", IOCPARM_LEN(cmd)); 338 } 339 error = ENOTTY; 340 } 341 if (gio != NULL) 342 g_free(gio); 343 return (error); 344} 345 346static void 347g_dev_done(struct bio *bp2) 348{ 349 struct bio *bp; 350 351 bp = bp2->bio_parent; 352 bp->bio_error = bp2->bio_error; 353 if (bp->bio_error != 0) { 354 g_trace(G_T_BIO, "g_dev_done(%p) had error %d", 355 bp2, bp->bio_error); 356 bp->bio_flags |= BIO_ERROR; 357 } else { 358 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", 359 bp2, bp, bp->bio_resid, (intmax_t)bp2->bio_completed); 360 } 361 bp->bio_resid = bp->bio_bcount - bp2->bio_completed; 362 g_destroy_bio(bp2); 363 mtx_lock(&Giant); 364 biodone(bp); 365 mtx_unlock(&Giant); 366} 367 368static void 369g_dev_strategy(struct bio *bp) 370{ 371 struct g_geom *gp; 372 struct g_consumer *cp; 373 struct bio *bp2; 374 dev_t dev; 375 376 KASSERT(bp->bio_cmd == BIO_READ || 377 bp->bio_cmd == BIO_WRITE || 378 bp->bio_cmd == BIO_DELETE, 379 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); 380 dev = bp->bio_dev; 381 gp = dev->si_drv1; 382 cp = dev->si_drv2; 383 bp2 = g_clone_bio(bp); 384 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); 385 bp2->bio_offset = (off_t)bp->bio_blkno << DEV_BSHIFT; 386 KASSERT(bp2->bio_offset >= 0, 387 ("Negative bio_offset (%jd) on bio %p", 388 (intmax_t)bp2->bio_offset, bp)); 389 bp2->bio_length = (off_t)bp->bio_bcount; 390 bp2->bio_done = g_dev_done; 391 g_trace(G_T_BIO, 392 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", 393 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, 394 bp2->bio_data, bp2->bio_cmd); 395 g_io_request(bp2, cp); 396} 397 398/* 399 * g_dev_orphan() 400 * 401 * Called from below when the provider orphaned us. 402 * - Clear any dump settings. 403 * - Destroy the dev_t to prevent any more request from coming in. The 404 * provider is already marked with an error, so anything which comes in 405 * in the interrim will be returned immediately. 406 * - Wait for any outstanding I/O to finish. 407 * - Set our access counts to zero, whatever they were. 408 * - Detach and self-destruct. 409 */ 410 411static void 412g_dev_orphan(struct g_consumer *cp) 413{ 414 struct g_geom *gp; 415 dev_t dev; 416 417 g_topology_assert(); 418 gp = cp->geom; 419 dev = gp->softc; 420 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, gp->name); 421 422 /* Reset any dump-area set on this device */ 423 if (dev->si_flags & SI_DUMPDEV) 424 set_dumper(NULL); 425 426 /* Destroy the dev_t so we get no more requests */ 427 destroy_dev(dev); 428 429 /* Wait for the cows to come home */ 430 while (cp->nstart != cp->nend) 431 msleep(&dev, NULL, PRIBIO, "gdevorphan", 1); 432 433 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 434 g_access_rel(cp, -cp->acr, -cp->acw, -cp->ace); 435 436 g_detach(cp); 437 g_destroy_consumer(cp); 438 g_destroy_geom(gp); 439} 440 441DECLARE_GEOM_CLASS(g_dev_class, g_dev); 442