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