sys/geom/geom_disk.c

   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 "opt_geom.h"
  40
  41 #include <sys/param.h>
  42 #include <sys/systm.h>
  43 #include <sys/kernel.h>
  44 #include <sys/sysctl.h>
  45 #include <sys/bio.h>
  46 #include <sys/ctype.h>
  47 #include <sys/fcntl.h>
  48 #include <sys/malloc.h>
  49 #include <sys/sbuf.h>
  50 #include <sys/sysctl.h>
  51 #include <sys/devicestat.h>
  52 #include <machine/md_var.h>
  53
  54 #include <sys/lock.h>
  55 #include <sys/mutex.h>
  56 #include <geom/geom.h>
  57 #include <geom/geom_disk.h>
  58 #include <geom/geom_int.h>
  59
  60 #include <dev/led/led.h>
  61
  62 struct g_disk_softc {
  63         struct mtx               done_mtx;
  64         struct disk             *dp;
  65         struct sysctl_ctx_list  sysctl_ctx;
  66         struct sysctl_oid       *sysctl_tree;
  67         char                    led[64];
  68         uint32_t                state;
  69         struct mtx               start_mtx;
  70 };
  71
  72 static g_access_t g_disk_access;
  73 static g_start_t g_disk_start;
  74 static g_ioctl_t g_disk_ioctl;
  75 static g_dumpconf_t g_disk_dumpconf;
  76 static g_provgone_t g_disk_providergone;
  77
  78 static struct g_class g_disk_class = {
  79         .name = G_DISK_CLASS_NAME,
  80         .version = G_VERSION,
  81         .start = g_disk_start,
  82         .access = g_disk_access,
  83         .ioctl = g_disk_ioctl,
  84         .providergone = g_disk_providergone,
  85         .dumpconf = g_disk_dumpconf,
  86 };
  87
  88 SYSCTL_DECL(_kern_geom);
  89 static SYSCTL_NODE(_kern_geom, OID_AUTO, disk, CTLFLAG_RW, 0,
  90     "GEOM_DISK stuff");
  91
  92 DECLARE_GEOM_CLASS(g_disk_class, g_disk);
  93
  94 static int
  95 g_disk_access(struct g_provider *pp, int r, int w, int e)
  96 {
  97         struct disk *dp;
  98         struct g_disk_softc *sc;
  99         int error;
 100
 101         g_trace(G_T_ACCESS, "g_disk_access(%s, %d, %d, %d)",
 102             pp->name, r, w, e);
 103         g_topology_assert();
 104         sc = pp->private;
 105         if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) {
 106                 /*
 107                  * Allow decreasing access count even if disk is not
 108                  * avaliable anymore.
 109                  */
 110                 if (r <= 0 && w <= 0 && e <= 0)
 111                         return (0);
 112                 return (ENXIO);
 113         }
 114         r += pp->acr;
 115         w += pp->acw;
 116         e += pp->ace;
 117         error = 0;
 118         if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
 119                 if (dp->d_open != NULL) {
 120                         error = dp->d_open(dp);
 121                         if (bootverbose && error != 0)
 122                                 printf("Opened disk %s -> %d\n",
 123                                     pp->name, error);
 124                         if (error != 0)
 125                                 return (error);
 126                 }
 127                 pp->mediasize = dp->d_mediasize;
 128                 pp->sectorsize = dp->d_sectorsize;
 129                 if (dp->d_maxsize == 0) {
 130                         printf("WARNING: Disk drive %s%d has no d_maxsize\n",
 131                             dp->d_name, dp->d_unit);
 132                         dp->d_maxsize = DFLTPHYS;
 133                 }
 134                 if (dp->d_delmaxsize == 0) {
 135                         if (bootverbose && dp->d_flags & DISKFLAG_CANDELETE) {
 136                                 printf("WARNING: Disk drive %s%d has no "
 137                                     "d_delmaxsize\n", dp->d_name, dp->d_unit);
 138                         }
 139                         dp->d_delmaxsize = dp->d_maxsize;
 140                 }
 141                 pp->stripeoffset = dp->d_stripeoffset;
 142                 pp->stripesize = dp->d_stripesize;
 143                 dp->d_flags |= DISKFLAG_OPEN;
 144         } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
 145                 if (dp->d_close != NULL) {
 146                         error = dp->d_close(dp);
 147                         if (error != 0)
 148                                 printf("Closed disk %s -> %d\n",
 149                                     pp->name, error);
 150                 }
 151                 sc->state = G_STATE_ACTIVE;
 152                 if (sc->led[0] != 0)
 153                         led_set(sc->led, "0");
 154                 dp->d_flags &= ~DISKFLAG_OPEN;
 155         }
 156         return (error);
 157 }
 158
 159 static void
 160 g_disk_kerneldump(struct bio *bp, struct disk *dp)
 161 {
 162         struct g_kerneldump *gkd;
 163         struct g_geom *gp;
 164
 165         gkd = (struct g_kerneldump*)bp->bio_data;
 166         gp = bp->bio_to->geom;
 167         g_trace(G_T_TOPOLOGY, "g_disk_kernedump(%s, %jd, %jd)",
 168                 gp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
 169         if (dp->d_dump == NULL) {
 170                 g_io_deliver(bp, ENODEV);
 171                 return;
 172         }
 173         gkd->di.dumper = dp->d_dump;
 174         gkd->di.priv = dp;
 175         gkd->di.blocksize = dp->d_sectorsize;
 176         gkd->di.maxiosize = dp->d_maxsize;
 177         gkd->di.mediaoffset = gkd->offset;
 178         if ((gkd->offset + gkd->length) > dp->d_mediasize)
 179                 gkd->length = dp->d_mediasize - gkd->offset;
 180         gkd->di.mediasize = gkd->length;
 181         g_io_deliver(bp, 0);
 182 }
 183
 184 static void
 185 g_disk_setstate(struct bio *bp, struct g_disk_softc *sc)
 186 {
 187         const char *cmd;
 188
 189         memcpy(&sc->state, bp->bio_data, sizeof(sc->state));
 190         if (sc->led[0] != 0) {
 191                 switch (sc->state) {
 192                 case G_STATE_FAILED:
 193                         cmd = "1";
 194                         break;
 195                 case G_STATE_REBUILD:
 196                         cmd = "f5";
 197                         break;
 198                 case G_STATE_RESYNC:
 199                         cmd = "f1";
 200                         break;
 201                 default:
 202                         cmd = "0";
 203                         break;
 204                 }
 205                 led_set(sc->led, cmd);
 206         }
 207         g_io_deliver(bp, 0);
 208 }
 209
 210 static void
 211 g_disk_done(struct bio *bp)
 212 {
 213         struct bintime now;
 214         struct bio *bp2;
 215         struct g_disk_softc *sc;
 216
 217         /* See "notes" for why we need a mutex here */
 218         /* XXX: will witness accept a mix of Giant/unGiant drivers here ? */
 219         bp2 = bp->bio_parent;
 220         sc = bp2->bio_to->private;
 221         bp->bio_completed = bp->bio_length - bp->bio_resid;
 222         binuptime(&now);
 223         mtx_lock(&sc->done_mtx);
 224         if (bp2->bio_error == 0)
 225                 bp2->bio_error = bp->bio_error;
 226         bp2->bio_completed += bp->bio_completed;
 227         if ((bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE)) != 0)
 228                 devstat_end_transaction_bio_bt(sc->dp->d_devstat, bp, &now);
 229         bp2->bio_inbed++;
 230         if (bp2->bio_children == bp2->bio_inbed) {
 231                 mtx_unlock(&sc->done_mtx);
 232                 bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed;
 233                 g_io_deliver(bp2, bp2->bio_error);
 234         } else
 235                 mtx_unlock(&sc->done_mtx);
 236         g_destroy_bio(bp);
 237 }
 238
 239 static void
 240 g_disk_done_single(struct bio *bp)
 241 {
 242         struct bintime now;
 243         struct g_disk_softc *sc;
 244
 245         bp->bio_completed = bp->bio_length - bp->bio_resid;
 246         bp->bio_done = (void *)bp->bio_to;
 247         bp->bio_to = LIST_FIRST(&bp->bio_disk->d_geom->provider);
 248         if ((bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE)) != 0) {
 249                 binuptime(&now);
 250                 sc = bp->bio_to->private;
 251                 mtx_lock(&sc->done_mtx);
 252                 devstat_end_transaction_bio_bt(sc->dp->d_devstat, bp, &now);
 253                 mtx_unlock(&sc->done_mtx);
 254         }
 255         g_io_deliver(bp, bp->bio_error);
 256 }
 257
 258 static int
 259 g_disk_ioctl(struct g_provider *pp, u_long cmd, void * data, int fflag, struct thread *td)
 260 {
 261         struct disk *dp;
 262         struct g_disk_softc *sc;
 263         int error;
 264
 265         sc = pp->private;
 266         dp = sc->dp;
 267
 268         if (dp->d_ioctl == NULL)
 269                 return (ENOIOCTL);
 270         error = dp->d_ioctl(dp, cmd, data, fflag, td);
 271         return (error);
 272 }
 273
 274 static void
 275 g_disk_start(struct bio *bp)
 276 {
 277         struct bio *bp2, *bp3;
 278         struct disk *dp;
 279         struct g_disk_softc *sc;
 280         int error;
 281         off_t d_maxsize, off;
 282
 283         sc = bp->bio_to->private;
 284         if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) {
 285                 g_io_deliver(bp, ENXIO);
 286                 return;
 287         }
 288         error = EJUSTRETURN;
 289         switch(bp->bio_cmd) {
 290         case BIO_DELETE:
 291                 if (!(dp->d_flags & DISKFLAG_CANDELETE)) {
 292                         error = EOPNOTSUPP;
 293                         break;
 294                 }
 295                 /* fall-through */
 296         case BIO_READ:
 297         case BIO_WRITE:
 298                 d_maxsize = (bp->bio_cmd == BIO_DELETE) ?
 299                     dp->d_delmaxsize : dp->d_maxsize;
 300                 if (bp->bio_length <= d_maxsize) {
 301                         bp->bio_disk = dp;
 302                         bp->bio_to = (void *)bp->bio_done;
 303                         bp->bio_done = g_disk_done_single;
 304                         bp->bio_pblkno = bp->bio_offset / dp->d_sectorsize;
 305                         bp->bio_bcount = bp->bio_length;
 306                         mtx_lock(&sc->start_mtx);
 307                         devstat_start_transaction_bio(dp->d_devstat, bp);
 308                         mtx_unlock(&sc->start_mtx);
 309                         dp->d_strategy(bp);
 310                         break;
 311                 }
 312                 off = 0;
 313                 bp3 = NULL;
 314                 bp2 = g_clone_bio(bp);
 315                 if (bp2 == NULL) {
 316                         error = ENOMEM;
 317                         break;
 318                 }
 319                 do {
 320                         bp2->bio_offset += off;
 321                         bp2->bio_length -= off;
 322                         if ((bp->bio_flags & BIO_UNMAPPED) == 0) {
 323                                 bp2->bio_data += off;
 324                         } else {
 325                                 KASSERT((dp->d_flags & DISKFLAG_UNMAPPED_BIO)
 326                                     != 0,
 327                                     ("unmapped bio not supported by disk %s",
 328                                     dp->d_name));
 329                                 bp2->bio_ma += off / PAGE_SIZE;
 330                                 bp2->bio_ma_offset += off;
 331                                 bp2->bio_ma_offset %= PAGE_SIZE;
 332                                 bp2->bio_ma_n -= off / PAGE_SIZE;
 333                         }
 334                         if (bp2->bio_length > d_maxsize) {
 335                                 /*
 336                                  * XXX: If we have a stripesize we should really
 337                                  * use it here. Care should be taken in the delete
 338                                  * case if this is done as deletes can be very
 339                                  * sensitive to size given how they are processed.
 340                                  */
 341                                 bp2->bio_length = d_maxsize;
 342                                 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
 343                                         bp2->bio_ma_n = howmany(
 344                                             bp2->bio_ma_offset +
 345                                             bp2->bio_length, PAGE_SIZE);
 346                                 }
 347                                 off += d_maxsize;
 348                                 /*
 349                                  * To avoid a race, we need to grab the next bio
 350                                  * before we schedule this one.  See "notes".
 351                                  */
 352                                 bp3 = g_clone_bio(bp);
 353                                 if (bp3 == NULL)
 354                                         bp->bio_error = ENOMEM;
 355                         }
 356                         bp2->bio_done = g_disk_done;
 357                         bp2->bio_pblkno = bp2->bio_offset / dp->d_sectorsize;
 358                         bp2->bio_bcount = bp2->bio_length;
 359                         bp2->bio_disk = dp;
 360                         mtx_lock(&sc->start_mtx);
 361                         devstat_start_transaction_bio(dp->d_devstat, bp2);
 362                         mtx_unlock(&sc->start_mtx);
 363                         dp->d_strategy(bp2);
 364                         bp2 = bp3;
 365                         bp3 = NULL;
 366                 } while (bp2 != NULL);
 367                 break;
 368         case BIO_GETATTR:
 369                 /* Give the driver a chance to override */
 370                 if (dp->d_getattr != NULL) {
 371                         if (bp->bio_disk == NULL)
 372                                 bp->bio_disk = dp;
 373                         error = dp->d_getattr(bp);
 374                         if (error != -1)
 375                                 break;
 376                         error = EJUSTRETURN;
 377                 }
 378                 if (g_handleattr_int(bp, "GEOM::candelete",
 379                     (dp->d_flags & DISKFLAG_CANDELETE) != 0))
 380                         break;
 381                 else if (g_handleattr_int(bp, "GEOM::fwsectors",
 382                     dp->d_fwsectors))
 383                         break;
 384                 else if (g_handleattr_int(bp, "GEOM::fwheads", dp->d_fwheads))
 385                         break;
 386                 else if (g_handleattr_off_t(bp, "GEOM::frontstuff", 0))
 387                         break;
 388                 else if (g_handleattr_str(bp, "GEOM::ident", dp->d_ident))
 389                         break;
 390                 else if (g_handleattr_uint16_t(bp, "GEOM::hba_vendor",
 391                     dp->d_hba_vendor))
 392                         break;
 393                 else if (g_handleattr_uint16_t(bp, "GEOM::hba_device",
 394                     dp->d_hba_device))
 395                         break;
 396                 else if (g_handleattr_uint16_t(bp, "GEOM::hba_subvendor",
 397                     dp->d_hba_subvendor))
 398                         break;
 399                 else if (g_handleattr_uint16_t(bp, "GEOM::hba_subdevice",
 400                     dp->d_hba_subdevice))
 401                         break;
 402                 else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
 403                         g_disk_kerneldump(bp, dp);
 404                 else if (!strcmp(bp->bio_attribute, "GEOM::setstate"))
 405                         g_disk_setstate(bp, sc);
 406                 else if (g_handleattr_uint16_t(bp, "GEOM::rotation_rate",
 407                     dp->d_rotation_rate))
 408                         break;
 409                 else
 410                         error = ENOIOCTL;
 411                 break;
 412         case BIO_FLUSH:
 413                 g_trace(G_T_BIO, "g_disk_flushcache(%s)",
 414                     bp->bio_to->name);
 415                 if (!(dp->d_flags & DISKFLAG_CANFLUSHCACHE)) {
 416                         error = EOPNOTSUPP;
 417                         break;
 418                 }
 419                 bp->bio_disk = dp;
 420                 bp->bio_to = (void *)bp->bio_done;
 421                 bp->bio_done = g_disk_done_single;
 422                 dp->d_strategy(bp);
 423                 break;
 424         default:
 425                 error = EOPNOTSUPP;
 426                 break;
 427         }
 428         if (error != EJUSTRETURN)
 429                 g_io_deliver(bp, error);
 430         return;
 431 }
 432
 433 static void
 434 g_disk_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp)
 435 {
 436         struct bio *bp;
 437         struct disk *dp;
 438         struct g_disk_softc *sc;
 439         char *buf;
 440         int res = 0;
 441
 442         sc = gp->softc;
 443         if (sc == NULL || (dp = sc->dp) == NULL)
 444                 return;
 445         if (indent == NULL) {
 446                 sbuf_printf(sb, " hd %u", dp->d_fwheads);
 447                 sbuf_printf(sb, " sc %u", dp->d_fwsectors);
 448                 return;
 449         }
 450         if (pp != NULL) {
 451                 sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n",
 452                     indent, dp->d_fwheads);
 453                 sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n",
 454                     indent, dp->d_fwsectors);
 455                 if (dp->d_getattr != NULL) {
 456                         buf = g_malloc(DISK_IDENT_SIZE, M_WAITOK);
 457                         bp = g_alloc_bio();
 458                         bp->bio_disk = dp;
 459                         bp->bio_attribute = "GEOM::ident";
 460                         bp->bio_length = DISK_IDENT_SIZE;
 461                         bp->bio_data = buf;
 462                         res = dp->d_getattr(bp);
 463                         sbuf_printf(sb, "%s<ident>", indent);
 464                         g_conf_printf_escaped(sb, "%s",
 465                             res == 0 ? buf: dp->d_ident);
 466                         sbuf_printf(sb, "</ident>\n");
 467                         bp->bio_attribute = "GEOM::lunid";
 468                         bp->bio_length = DISK_IDENT_SIZE;
 469                         bp->bio_data = buf;
 470                         if (dp->d_getattr(bp) == 0) {
 471                                 sbuf_printf(sb, "%s<lunid>", indent);
 472                                 g_conf_printf_escaped(sb, "%s", buf);
 473                                 sbuf_printf(sb, "</lunid>\n");
 474                         }
 475                         bp->bio_attribute = "GEOM::lunname";
 476                         bp->bio_length = DISK_IDENT_SIZE;
 477                         bp->bio_data = buf;
 478                         if (dp->d_getattr(bp) == 0) {
 479                                 sbuf_printf(sb, "%s<lunname>", indent);
 480                                 g_conf_printf_escaped(sb, "%s", buf);
 481                                 sbuf_printf(sb, "</lunname>\n");
 482                         }
 483                         g_destroy_bio(bp);
 484                         g_free(buf);
 485                 } else {
 486                         sbuf_printf(sb, "%s<ident>", indent);
 487                         g_conf_printf_escaped(sb, "%s", dp->d_ident);
 488                         sbuf_printf(sb, "</ident>\n");
 489                 }
 490                 sbuf_printf(sb, "%s<descr>", indent);
 491                 g_conf_printf_escaped(sb, "%s", dp->d_descr);
 492                 sbuf_printf(sb, "</descr>\n");
 493         }
 494 }
 495
 496 static void
 497 g_disk_resize(void *ptr, int flag)
 498 {
 499         struct disk *dp;
 500         struct g_geom *gp;
 501         struct g_provider *pp;
 502
 503         if (flag == EV_CANCEL)
 504                 return;
 505         g_topology_assert();
 506
 507         dp = ptr;
 508         gp = dp->d_geom;
 509
 510         if (dp->d_destroyed || gp == NULL)
 511                 return;
 512
 513         LIST_FOREACH(pp, &gp->provider, provider) {
 514                 if (pp->sectorsize != 0 &&
 515                     pp->sectorsize != dp->d_sectorsize)
 516                         g_wither_provider(pp, ENXIO);
 517                 else
 518                         g_resize_provider(pp, dp->d_mediasize);
 519         }
 520 }
 521
 522 static void
 523 g_disk_create(void *arg, int flag)
 524 {
 525         struct g_geom *gp;
 526         struct g_provider *pp;
 527         struct disk *dp;
 528         struct g_disk_softc *sc;
 529         char tmpstr[80];
 530
 531         if (flag == EV_CANCEL)
 532                 return;
 533         g_topology_assert();
 534         dp = arg;
 535         sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
 536         mtx_init(&sc->start_mtx, "g_disk_start", NULL, MTX_DEF);
 537         mtx_init(&sc->done_mtx, "g_disk_done", NULL, MTX_DEF);
 538         sc->dp = dp;
 539         gp = g_new_geomf(&g_disk_class, "%s%d", dp->d_name, dp->d_unit);
 540         gp->softc = sc;
 541         pp = g_new_providerf(gp, "%s", gp->name);
 542         devstat_remove_entry(pp->stat);
 543         pp->stat = NULL;
 544         dp->d_devstat->id = pp;
 545         pp->mediasize = dp->d_mediasize;
 546         pp->sectorsize = dp->d_sectorsize;
 547         pp->stripeoffset = dp->d_stripeoffset;
 548         pp->stripesize = dp->d_stripesize;
 549         if ((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0)
 550                 pp->flags |= G_PF_ACCEPT_UNMAPPED;
 551         if ((dp->d_flags & DISKFLAG_DIRECT_COMPLETION) != 0)
 552                 pp->flags |= G_PF_DIRECT_SEND;
 553         pp->flags |= G_PF_DIRECT_RECEIVE;
 554         if (bootverbose)
 555                 printf("GEOM: new disk %s\n", gp->name);
 556         sysctl_ctx_init(&sc->sysctl_ctx);
 557         snprintf(tmpstr, sizeof(tmpstr), "GEOM disk %s", gp->name);
 558         sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
 559                 SYSCTL_STATIC_CHILDREN(_kern_geom_disk), OID_AUTO, gp->name,
 560                 CTLFLAG_RD, 0, tmpstr);
 561         if (sc->sysctl_tree != NULL) {
 562                 snprintf(tmpstr, sizeof(tmpstr),
 563                     "kern.geom.disk.%s.led", gp->name);
 564                 TUNABLE_STR_FETCH(tmpstr, sc->led, sizeof(sc->led));
 565                 SYSCTL_ADD_STRING(&sc->sysctl_ctx,
 566                     SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "led",
 567                     CTLFLAG_RW | CTLFLAG_TUN, sc->led, sizeof(sc->led),
 568                     "LED name");
 569         }
 570         pp->private = sc;
 571         dp->d_geom = gp;
 572         g_error_provider(pp, 0);
 573 }
 574
 575 /*
 576  * We get this callback after all of the consumers have gone away, and just
 577  * before the provider is freed.  If the disk driver provided a d_gone
 578  * callback, let them know that it is okay to free resources -- they won't
 579  * be getting any more accesses from GEOM.
 580  */
 581 static void
 582 g_disk_providergone(struct g_provider *pp)
 583 {
 584         struct disk *dp;
 585         struct g_disk_softc *sc;
 586
 587         sc = (struct g_disk_softc *)pp->private;
 588         dp = sc->dp;
 589         if (dp != NULL && dp->d_gone != NULL)
 590                 dp->d_gone(dp);
 591         if (sc->sysctl_tree != NULL) {
 592                 sysctl_ctx_free(&sc->sysctl_ctx);
 593                 sc->sysctl_tree = NULL;
 594         }
 595         if (sc->led[0] != 0) {
 596                 led_set(sc->led, "0");
 597                 sc->led[0] = 0;
 598         }
 599         pp->private = NULL;
 600         pp->geom->softc = NULL;
 601         mtx_destroy(&sc->done_mtx);
 602         mtx_destroy(&sc->start_mtx);
 603         g_free(sc);
 604 }
 605
 606 static void
 607 g_disk_destroy(void *ptr, int flag)
 608 {
 609         struct disk *dp;
 610         struct g_geom *gp;
 611         struct g_disk_softc *sc;
 612
 613         g_topology_assert();
 614         dp = ptr;
 615         gp = dp->d_geom;
 616         if (gp != NULL) {
 617                 sc = gp->softc;
 618                 if (sc != NULL)
 619                         sc->dp = NULL;
 620                 dp->d_geom = NULL;
 621                 g_wither_geom(gp, ENXIO);
 622         }
 623         g_free(dp);
 624 }
 625
 626 /*
 627  * We only allow printable characters in disk ident,
 628  * the rest is converted to 'x<HH>'.
 629  */
 630 static void
 631 g_disk_ident_adjust(char *ident, size_t size)
 632 {
 633         char *p, tmp[4], newid[DISK_IDENT_SIZE];
 634
 635         newid[0] = '\0';
 636         for (p = ident; *p != '\0'; p++) {
 637                 if (isprint(*p)) {
 638                         tmp[0] = *p;
 639                         tmp[1] = '\0';
 640                 } else {
 641                         snprintf(tmp, sizeof(tmp), "x%02hhx",
 642                             *(unsigned char *)p);
 643                 }
 644                 if (strlcat(newid, tmp, sizeof(newid)) >= sizeof(newid))
 645                         break;
 646         }
 647         bzero(ident, size);
 648         strlcpy(ident, newid, size);
 649 }
 650
 651 struct disk *
 652 disk_alloc(void)
 653 {
 654
 655         return (g_malloc(sizeof(struct disk), M_WAITOK | M_ZERO));
 656 }
 657
 658 void
 659 disk_create(struct disk *dp, int version)
 660 {
 661
 662         if (version != DISK_VERSION) {
 663                 printf("WARNING: Attempt to add disk %s%d %s",
 664                     dp->d_name, dp->d_unit,
 665                     " using incompatible ABI version of disk(9)\n");
 666                 printf("WARNING: Ignoring disk %s%d\n",
 667                     dp->d_name, dp->d_unit);
 668                 return;
 669         }
 670         if (dp->d_flags & DISKFLAG_RESERVED) {
 671                 printf("WARNING: Attempt to add non-MPSAFE disk %s%d\n",
 672                     dp->d_name, dp->d_unit);
 673                 printf("WARNING: Ignoring disk %s%d\n",
 674                     dp->d_name, dp->d_unit);
 675                 return;
 676         }
 677         KASSERT(dp->d_strategy != NULL, ("disk_create need d_strategy"));
 678         KASSERT(dp->d_name != NULL, ("disk_create need d_name"));
 679         KASSERT(*dp->d_name != 0, ("disk_create need d_name"));
 680         KASSERT(strlen(dp->d_name) < SPECNAMELEN - 4, ("disk name too long"));
 681         if (dp->d_devstat == NULL)
 682                 dp->d_devstat = devstat_new_entry(dp->d_name, dp->d_unit,
 683                     dp->d_sectorsize, DEVSTAT_ALL_SUPPORTED,
 684                     DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
 685         dp->d_geom = NULL;
 686         g_disk_ident_adjust(dp->d_ident, sizeof(dp->d_ident));
 687         g_post_event(g_disk_create, dp, M_WAITOK, dp, NULL);
 688 }
 689
 690 void
 691 disk_destroy(struct disk *dp)
 692 {
 693
 694         g_cancel_event(dp);
 695         dp->d_destroyed = 1;
 696         if (dp->d_devstat != NULL)
 697                 devstat_remove_entry(dp->d_devstat);
 698         g_post_event(g_disk_destroy, dp, M_WAITOK, NULL);
 699 }
 700
 701 void
 702 disk_gone(struct disk *dp)
 703 {
 704         struct g_geom *gp;
 705         struct g_provider *pp;
 706
 707         gp = dp->d_geom;
 708         if (gp != NULL) {
 709                 pp = LIST_FIRST(&gp->provider);
 710                 if (pp != NULL) {
 711                         KASSERT(LIST_NEXT(pp, provider) == NULL,
 712                             ("geom %p has more than one provider", gp));
 713                         g_wither_provider(pp, ENXIO);
 714                 }
 715         }
 716 }
 717
 718 void
 719 disk_attr_changed(struct disk *dp, const char *attr, int flag)
 720 {
 721         struct g_geom *gp;
 722         struct g_provider *pp;
 723
 724         gp = dp->d_geom;
 725         if (gp != NULL)
 726                 LIST_FOREACH(pp, &gp->provider, provider)
 727                         (void)g_attr_changed(pp, attr, flag);
 728 }
 729
 730 void
 731 disk_media_changed(struct disk *dp, int flag)
 732 {
 733         struct g_geom *gp;
 734         struct g_provider *pp;
 735
 736         gp = dp->d_geom;
 737         if (gp != NULL) {
 738                 pp = LIST_FIRST(&gp->provider);
 739                 if (pp != NULL) {
 740                         KASSERT(LIST_NEXT(pp, provider) == NULL,
 741                             ("geom %p has more than one provider", gp));
 742                         g_media_changed(pp, flag);
 743                 }
 744         }
 745 }
 746
 747 void
 748 disk_media_gone(struct disk *dp, int flag)
 749 {
 750         struct g_geom *gp;
 751         struct g_provider *pp;
 752
 753         gp = dp->d_geom;
 754         if (gp != NULL) {
 755                 pp = LIST_FIRST(&gp->provider);
 756                 if (pp != NULL) {
 757                         KASSERT(LIST_NEXT(pp, provider) == NULL,
 758                             ("geom %p has more than one provider", gp));
 759                         g_media_gone(pp, flag);
 760                 }
 761         }
 762 }
 763
 764 int
 765 disk_resize(struct disk *dp, int flag)
 766 {
 767
 768         if (dp->d_destroyed || dp->d_geom == NULL)
 769                 return (0);
 770
 771         return (g_post_event(g_disk_resize, dp, flag, NULL));
 772 }
 773
 774 static void
 775 g_kern_disks(void *p, int flag __unused)
 776 {
 777         struct sbuf *sb;
 778         struct g_geom *gp;
 779         char *sp;
 780
 781         sb = p;
 782         sp = "";
 783         g_topology_assert();
 784         LIST_FOREACH(gp, &g_disk_class.geom, geom) {
 785                 sbuf_printf(sb, "%s%s", sp, gp->name);
 786                 sp = " ";
 787         }
 788         sbuf_finish(sb);
 789 }
 790
 791 static int
 792 sysctl_disks(SYSCTL_HANDLER_ARGS)
 793 {
 794         int error;
 795         struct sbuf *sb;
 796
 797         sb = sbuf_new_auto();
 798         g_waitfor_event(g_kern_disks, sb, M_WAITOK, NULL);
 799         error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
 800         sbuf_delete(sb);
 801         return error;
 802 }
 803
 804 SYSCTL_PROC(_kern, OID_AUTO, disks,
 805     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
 806     sysctl_disks, "A", "names of available disks");