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 void __inline
  94 g_disk_lock_giant(struct disk *dp)
  95 {
  96
  97         if (dp->d_flags & DISKFLAG_NEEDSGIANT)
  98                 mtx_lock(&Giant);
  99 }
 100
 101 static void __inline
 102 g_disk_unlock_giant(struct disk *dp)
 103 {
 104
 105         if (dp->d_flags & DISKFLAG_NEEDSGIANT)
 106                 mtx_unlock(&Giant);
 107 }
 108
 109 static int
 110 g_disk_access(struct g_provider *pp, int r, int w, int e)
 111 {
 112         struct disk *dp;
 113         struct g_disk_softc *sc;
 114         int error;
 115
 116         g_trace(G_T_ACCESS, "g_disk_access(%s, %d, %d, %d)",
 117             pp->name, r, w, e);
 118         g_topology_assert();
 119         sc = pp->private;
 120         if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) {
 121                 /*
 122                  * Allow decreasing access count even if disk is not
 123                  * avaliable anymore.
 124                  */
 125                 if (r <= 0 && w <= 0 && e <= 0)
 126                         return (0);
 127                 return (ENXIO);
 128         }
 129         r += pp->acr;
 130         w += pp->acw;
 131         e += pp->ace;
 132         error = 0;
 133         if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
 134                 if (dp->d_open != NULL) {
 135                         g_disk_lock_giant(dp);
 136                         error = dp->d_open(dp);
 137                         if (bootverbose && error != 0)
 138                                 printf("Opened disk %s -> %d\n",
 139                                     pp->name, error);
 140                         g_disk_unlock_giant(dp);
 141                         if (error != 0)
 142                                 return (error);
 143                 }
 144                 pp->mediasize = dp->d_mediasize;
 145                 pp->sectorsize = dp->d_sectorsize;
 146                 if (dp->d_maxsize == 0) {
 147                         printf("WARNING: Disk drive %s%d has no d_maxsize\n",
 148                             dp->d_name, dp->d_unit);
 149                         dp->d_maxsize = DFLTPHYS;
 150                 }
 151                 if (dp->d_delmaxsize == 0) {
 152                         if (bootverbose && dp->d_flags & DISKFLAG_CANDELETE) {
 153                                 printf("WARNING: Disk drive %s%d has no "
 154                                     "d_delmaxsize\n", dp->d_name, dp->d_unit);
 155                         }
 156                         dp->d_delmaxsize = dp->d_maxsize;
 157                 }
 158                 pp->stripeoffset = dp->d_stripeoffset;
 159                 pp->stripesize = dp->d_stripesize;
 160                 dp->d_flags |= DISKFLAG_OPEN;
 161         } else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
 162                 if (dp->d_close != NULL) {
 163                         g_disk_lock_giant(dp);
 164                         error = dp->d_close(dp);
 165                         if (error != 0)
 166                                 printf("Closed disk %s -> %d\n",
 167                                     pp->name, error);
 168                         g_disk_unlock_giant(dp);
 169                 }
 170                 sc->state = G_STATE_ACTIVE;
 171                 if (sc->led[0] != 0)
 172                         led_set(sc->led, "0");
 173                 dp->d_flags &= ~DISKFLAG_OPEN;
 174         }
 175         return (error);
 176 }
 177
 178 static void
 179 g_disk_kerneldump(struct bio *bp, struct disk *dp)
 180 {
 181         struct g_kerneldump *gkd;
 182         struct g_geom *gp;
 183
 184         gkd = (struct g_kerneldump*)bp->bio_data;
 185         gp = bp->bio_to->geom;
 186         g_trace(G_T_TOPOLOGY, "g_disk_kerneldump(%s, %jd, %jd)",
 187                 gp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
 188         if (dp->d_dump == NULL) {
 189                 g_io_deliver(bp, ENODEV);
 190                 return;
 191         }
 192         gkd->di.dumper = dp->d_dump;
 193         gkd->di.priv = dp;
 194         gkd->di.blocksize = dp->d_sectorsize;
 195         gkd->di.maxiosize = dp->d_maxsize;
 196         gkd->di.mediaoffset = gkd->offset;
 197         if ((gkd->offset + gkd->length) > dp->d_mediasize)
 198                 gkd->length = dp->d_mediasize - gkd->offset;
 199         gkd->di.mediasize = gkd->length;
 200         g_io_deliver(bp, 0);
 201 }
 202
 203 static void
 204 g_disk_setstate(struct bio *bp, struct g_disk_softc *sc)
 205 {
 206         const char *cmd;
 207
 208         memcpy(&sc->state, bp->bio_data, sizeof(sc->state));
 209         if (sc->led[0] != 0) {
 210                 switch (sc->state) {
 211                 case G_STATE_FAILED:
 212                         cmd = "1";
 213                         break;
 214                 case G_STATE_REBUILD:
 215                         cmd = "f5";
 216                         break;
 217                 case G_STATE_RESYNC:
 218                         cmd = "f1";
 219                         break;
 220                 default:
 221                         cmd = "0";
 222                         break;
 223                 }
 224                 led_set(sc->led, cmd);
 225         }
 226         g_io_deliver(bp, 0);
 227 }
 228
 229 static void
 230 g_disk_done(struct bio *bp)
 231 {
 232         struct bintime now;
 233         struct bio *bp2;
 234         struct g_disk_softc *sc;
 235
 236         /* See "notes" for why we need a mutex here */
 237         /* XXX: will witness accept a mix of Giant/unGiant drivers here ? */
 238         bp2 = bp->bio_parent;
 239         sc = bp2->bio_to->private;
 240         bp->bio_completed = bp->bio_length - bp->bio_resid;
 241         binuptime(&now);
 242         mtx_lock(&sc->done_mtx);
 243         if (bp2->bio_error == 0)
 244                 bp2->bio_error = bp->bio_error;
 245         bp2->bio_completed += bp->bio_completed;
 246         if ((bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE|BIO_FLUSH)) != 0)
 247                 devstat_end_transaction_bio_bt(sc->dp->d_devstat, bp, &now);
 248         bp2->bio_inbed++;
 249         if (bp2->bio_children == bp2->bio_inbed) {
 250                 mtx_unlock(&sc->done_mtx);
 251                 bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed;
 252                 g_io_deliver(bp2, bp2->bio_error);
 253         } else
 254                 mtx_unlock(&sc->done_mtx);
 255         g_destroy_bio(bp);
 256 }
 257
 258 static void
 259 g_disk_done_single(struct bio *bp)
 260 {
 261         struct bintime now;
 262         struct g_disk_softc *sc;
 263
 264         bp->bio_completed = bp->bio_length - bp->bio_resid;
 265         bp->bio_done = (void *)bp->bio_to;
 266         bp->bio_to = LIST_FIRST(&bp->bio_disk->d_geom->provider);
 267         if ((bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE|BIO_FLUSH)) != 0) {
 268                 binuptime(&now);
 269                 sc = bp->bio_to->private;
 270                 mtx_lock(&sc->done_mtx);
 271                 devstat_end_transaction_bio_bt(sc->dp->d_devstat, bp, &now);
 272                 mtx_unlock(&sc->done_mtx);
 273         }
 274         g_io_deliver(bp, bp->bio_error);
 275 }
 276
 277 static int
 278 g_disk_ioctl(struct g_provider *pp, u_long cmd, void * data, int fflag, struct thread *td)
 279 {
 280         struct disk *dp;
 281         struct g_disk_softc *sc;
 282         int error;
 283
 284         sc = pp->private;
 285         dp = sc->dp;
 286
 287         if (dp->d_ioctl == NULL)
 288                 return (ENOIOCTL);
 289         g_disk_lock_giant(dp);
 290         error = dp->d_ioctl(dp, cmd, data, fflag, td);
 291         g_disk_unlock_giant(dp);
 292         return (error);
 293 }
 294
 295 static void
 296 g_disk_start(struct bio *bp)
 297 {
 298         struct bio *bp2, *bp3;
 299         struct disk *dp;
 300         struct g_disk_softc *sc;
 301         int error;
 302         off_t d_maxsize, off;
 303
 304         sc = bp->bio_to->private;
 305         if (sc == NULL || (dp = sc->dp) == NULL || dp->d_destroyed) {
 306                 g_io_deliver(bp, ENXIO);
 307                 return;
 308         }
 309         error = EJUSTRETURN;
 310         switch(bp->bio_cmd) {
 311         case BIO_DELETE:
 312                 if (!(dp->d_flags & DISKFLAG_CANDELETE)) {
 313                         error = EOPNOTSUPP;
 314                         break;
 315                 }
 316                 /* fall-through */
 317         case BIO_READ:
 318         case BIO_WRITE:
 319                 d_maxsize = (bp->bio_cmd == BIO_DELETE) ?
 320                     dp->d_delmaxsize : dp->d_maxsize;
 321                 if (bp->bio_length <= d_maxsize) {
 322                         bp->bio_disk = dp;
 323                         bp->bio_to = (void *)bp->bio_done;
 324                         bp->bio_done = g_disk_done_single;
 325                         bp->bio_pblkno = bp->bio_offset / dp->d_sectorsize;
 326                         bp->bio_bcount = bp->bio_length;
 327                         mtx_lock(&sc->start_mtx);
 328                         devstat_start_transaction_bio(dp->d_devstat, bp);
 329                         mtx_unlock(&sc->start_mtx);
 330                         g_disk_lock_giant(dp);
 331                         dp->d_strategy(bp);
 332                         g_disk_unlock_giant(dp);
 333                         break;
 334                 }
 335                 off = 0;
 336                 bp3 = NULL;
 337                 bp2 = g_clone_bio(bp);
 338                 if (bp2 == NULL) {
 339                         error = ENOMEM;
 340                         break;
 341                 }
 342                 do {
 343                         bp2->bio_offset += off;
 344                         bp2->bio_length -= off;
 345                         if ((bp->bio_flags & BIO_UNMAPPED) == 0) {
 346                                 bp2->bio_data += off;
 347                         } else {
 348                                 KASSERT((dp->d_flags & DISKFLAG_UNMAPPED_BIO)
 349                                     != 0,
 350                                     ("unmapped bio not supported by disk %s",
 351                                     dp->d_name));
 352                                 bp2->bio_ma += off / PAGE_SIZE;
 353                                 bp2->bio_ma_offset += off;
 354                                 bp2->bio_ma_offset %= PAGE_SIZE;
 355                                 bp2->bio_ma_n -= off / PAGE_SIZE;
 356                         }
 357                         if (bp2->bio_length > d_maxsize) {
 358                                 /*
 359                                  * XXX: If we have a stripesize we should really
 360                                  * use it here. Care should be taken in the delete
 361                                  * case if this is done as deletes can be very
 362                                  * sensitive to size given how they are processed.
 363                                  */
 364                                 bp2->bio_length = d_maxsize;
 365                                 if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
 366                                         bp2->bio_ma_n = howmany(
 367                                             bp2->bio_ma_offset +
 368                                             bp2->bio_length, PAGE_SIZE);
 369                                 }
 370                                 off += d_maxsize;
 371                                 /*
 372                                  * To avoid a race, we need to grab the next bio
 373                                  * before we schedule this one.  See "notes".
 374                                  */
 375                                 bp3 = g_clone_bio(bp);
 376                                 if (bp3 == NULL)
 377                                         bp->bio_error = ENOMEM;
 378                         }
 379                         bp2->bio_done = g_disk_done;
 380                         bp2->bio_pblkno = bp2->bio_offset / dp->d_sectorsize;
 381                         bp2->bio_bcount = bp2->bio_length;
 382                         bp2->bio_disk = dp;
 383                         mtx_lock(&sc->start_mtx);
 384                         devstat_start_transaction_bio(dp->d_devstat, bp2);
 385                         mtx_unlock(&sc->start_mtx);
 386                         g_disk_lock_giant(dp);
 387                         dp->d_strategy(bp2);
 388                         g_disk_unlock_giant(dp);
 389                         bp2 = bp3;
 390                         bp3 = NULL;
 391                 } while (bp2 != NULL);
 392                 break;
 393         case BIO_GETATTR:
 394                 /* Give the driver a chance to override */
 395                 if (dp->d_getattr != NULL) {
 396                         if (bp->bio_disk == NULL)
 397                                 bp->bio_disk = dp;
 398                         error = dp->d_getattr(bp);
 399                         if (error != -1)
 400                                 break;
 401                         error = EJUSTRETURN;
 402                 }
 403                 if (g_handleattr_int(bp, "GEOM::candelete",
 404                     (dp->d_flags & DISKFLAG_CANDELETE) != 0))
 405                         break;
 406                 else if (g_handleattr_int(bp, "GEOM::fwsectors",
 407                     dp->d_fwsectors))
 408                         break;
 409                 else if (g_handleattr_int(bp, "GEOM::fwheads", dp->d_fwheads))
 410                         break;
 411                 else if (g_handleattr_off_t(bp, "GEOM::frontstuff", 0))
 412                         break;
 413                 else if (g_handleattr_str(bp, "GEOM::ident", dp->d_ident))
 414                         break;
 415                 else if (g_handleattr_uint16_t(bp, "GEOM::hba_vendor",
 416                     dp->d_hba_vendor))
 417                         break;
 418                 else if (g_handleattr_uint16_t(bp, "GEOM::hba_device",
 419                     dp->d_hba_device))
 420                         break;
 421                 else if (g_handleattr_uint16_t(bp, "GEOM::hba_subvendor",
 422                     dp->d_hba_subvendor))
 423                         break;
 424                 else if (g_handleattr_uint16_t(bp, "GEOM::hba_subdevice",
 425                     dp->d_hba_subdevice))
 426                         break;
 427                 else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
 428                         g_disk_kerneldump(bp, dp);
 429                 else if (!strcmp(bp->bio_attribute, "GEOM::setstate"))
 430                         g_disk_setstate(bp, sc);
 431                 else if (!strcmp(bp->bio_attribute, "GEOM::rotation_rate")) {
 432                         uint64_t v;
 433
 434                         if ((dp->d_flags & DISKFLAG_LACKS_ROTRATE) == 0)
 435                                 v = dp->d_rotation_rate;
 436                         else
 437                                 v = 0; /* rate unknown */
 438                         g_handleattr_uint16_t(bp, "GEOM::rotation_rate", v);
 439                         break;
 440                 } else
 441                         error = ENOIOCTL;
 442                 break;
 443         case BIO_FLUSH:
 444                 g_trace(G_T_BIO, "g_disk_flushcache(%s)",
 445                     bp->bio_to->name);
 446                 if (!(dp->d_flags & DISKFLAG_CANFLUSHCACHE)) {
 447                         error = EOPNOTSUPP;
 448                         break;
 449                 }
 450                 bp->bio_disk = dp;
 451                 bp->bio_to = (void *)bp->bio_done;
 452                 bp->bio_done = g_disk_done_single;
 453                 mtx_lock(&sc->start_mtx);
 454                 devstat_start_transaction_bio(dp->d_devstat, bp);
 455                 mtx_unlock(&sc->start_mtx);
 456                 g_disk_lock_giant(dp);
 457                 dp->d_strategy(bp);
 458                 g_disk_unlock_giant(dp);
 459                 break;
 460         default:
 461                 error = EOPNOTSUPP;
 462                 break;
 463         }
 464         if (error != EJUSTRETURN)
 465                 g_io_deliver(bp, error);
 466         return;
 467 }
 468
 469 static void
 470 g_disk_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp)
 471 {
 472         struct bio *bp;
 473         struct disk *dp;
 474         struct g_disk_softc *sc;
 475         char *buf;
 476         int res = 0;
 477
 478         sc = gp->softc;
 479         if (sc == NULL || (dp = sc->dp) == NULL)
 480                 return;
 481         if (indent == NULL) {
 482                 sbuf_printf(sb, " hd %u", dp->d_fwheads);
 483                 sbuf_printf(sb, " sc %u", dp->d_fwsectors);
 484                 return;
 485         }
 486         if (pp != NULL) {
 487                 sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n",
 488                     indent, dp->d_fwheads);
 489                 sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n",
 490                     indent, dp->d_fwsectors);
 491                 if (dp->d_getattr != NULL) {
 492                         buf = g_malloc(DISK_IDENT_SIZE, M_WAITOK);
 493                         bp = g_alloc_bio();
 494                         bp->bio_disk = dp;
 495                         bp->bio_attribute = "GEOM::ident";
 496                         bp->bio_length = DISK_IDENT_SIZE;
 497                         bp->bio_data = buf;
 498                         res = dp->d_getattr(bp);
 499                         sbuf_printf(sb, "%s<ident>", indent);
 500                         g_conf_printf_escaped(sb, "%s",
 501                             res == 0 ? buf: dp->d_ident);
 502                         sbuf_printf(sb, "</ident>\n");
 503                         bp->bio_attribute = "GEOM::lunid";
 504                         bp->bio_length = DISK_IDENT_SIZE;
 505                         bp->bio_data = buf;
 506                         if (dp->d_getattr(bp) == 0) {
 507                                 sbuf_printf(sb, "%s<lunid>", indent);
 508                                 g_conf_printf_escaped(sb, "%s", buf);
 509                                 sbuf_printf(sb, "</lunid>\n");
 510                         }
 511                         bp->bio_attribute = "GEOM::lunname";
 512                         bp->bio_length = DISK_IDENT_SIZE;
 513                         bp->bio_data = buf;
 514                         if (dp->d_getattr(bp) == 0) {
 515                                 sbuf_printf(sb, "%s<lunname>", indent);
 516                                 g_conf_printf_escaped(sb, "%s", buf);
 517                                 sbuf_printf(sb, "</lunname>\n");
 518                         }
 519                         g_destroy_bio(bp);
 520                         g_free(buf);
 521                 } else {
 522                         sbuf_printf(sb, "%s<ident>", indent);
 523                         g_conf_printf_escaped(sb, "%s", dp->d_ident);
 524                         sbuf_printf(sb, "</ident>\n");
 525                 }
 526                 sbuf_printf(sb, "%s<descr>", indent);
 527                 g_conf_printf_escaped(sb, "%s", dp->d_descr);
 528                 sbuf_printf(sb, "</descr>\n");
 529         }
 530 }
 531
 532 static void
 533 g_disk_resize(void *ptr, int flag)
 534 {
 535         struct disk *dp;
 536         struct g_geom *gp;
 537         struct g_provider *pp;
 538
 539         if (flag == EV_CANCEL)
 540                 return;
 541         g_topology_assert();
 542
 543         dp = ptr;
 544         gp = dp->d_geom;
 545
 546         if (dp->d_destroyed || gp == NULL)
 547                 return;
 548
 549         LIST_FOREACH(pp, &gp->provider, provider) {
 550                 if (pp->sectorsize != 0 &&
 551                     pp->sectorsize != dp->d_sectorsize)
 552                         g_wither_provider(pp, ENXIO);
 553                 else
 554                         g_resize_provider(pp, dp->d_mediasize);
 555         }
 556 }
 557
 558 static void
 559 g_disk_create(void *arg, int flag)
 560 {
 561         struct g_geom *gp;
 562         struct g_provider *pp;
 563         struct disk *dp;
 564         struct g_disk_softc *sc;
 565         char tmpstr[80];
 566
 567         if (flag == EV_CANCEL)
 568                 return;
 569         g_topology_assert();
 570         dp = arg;
 571         sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
 572         mtx_init(&sc->start_mtx, "g_disk_start", NULL, MTX_DEF);
 573         mtx_init(&sc->done_mtx, "g_disk_done", NULL, MTX_DEF);
 574         sc->dp = dp;
 575         gp = g_new_geomf(&g_disk_class, "%s%d", dp->d_name, dp->d_unit);
 576         gp->softc = sc;
 577         pp = g_new_providerf(gp, "%s", gp->name);
 578         devstat_remove_entry(pp->stat);
 579         pp->stat = NULL;
 580         dp->d_devstat->id = pp;
 581         pp->mediasize = dp->d_mediasize;
 582         pp->sectorsize = dp->d_sectorsize;
 583         pp->stripeoffset = dp->d_stripeoffset;
 584         pp->stripesize = dp->d_stripesize;
 585         if ((dp->d_flags & DISKFLAG_UNMAPPED_BIO) != 0)
 586                 pp->flags |= G_PF_ACCEPT_UNMAPPED;
 587         if ((dp->d_flags & DISKFLAG_DIRECT_COMPLETION) != 0)
 588                 pp->flags |= G_PF_DIRECT_SEND;
 589         pp->flags |= G_PF_DIRECT_RECEIVE;
 590         if (bootverbose)
 591                 printf("GEOM: new disk %s\n", gp->name);
 592         sysctl_ctx_init(&sc->sysctl_ctx);
 593         snprintf(tmpstr, sizeof(tmpstr), "GEOM disk %s", gp->name);
 594         sc->sysctl_tree = SYSCTL_ADD_NODE(&sc->sysctl_ctx,
 595                 SYSCTL_STATIC_CHILDREN(_kern_geom_disk), OID_AUTO, gp->name,
 596                 CTLFLAG_RD, 0, tmpstr);
 597         if (sc->sysctl_tree != NULL) {
 598                 snprintf(tmpstr, sizeof(tmpstr),
 599                     "kern.geom.disk.%s.led", gp->name);
 600                 TUNABLE_STR_FETCH(tmpstr, sc->led, sizeof(sc->led));
 601                 SYSCTL_ADD_STRING(&sc->sysctl_ctx,
 602                     SYSCTL_CHILDREN(sc->sysctl_tree), OID_AUTO, "led",
 603                     CTLFLAG_RW | CTLFLAG_TUN, sc->led, sizeof(sc->led),
 604                     "LED name");
 605         }
 606         pp->private = sc;
 607         dp->d_geom = gp;
 608         g_error_provider(pp, 0);
 609 }
 610
 611 /*
 612  * We get this callback after all of the consumers have gone away, and just
 613  * before the provider is freed.  If the disk driver provided a d_gone
 614  * callback, let them know that it is okay to free resources -- they won't
 615  * be getting any more accesses from GEOM.
 616  */
 617 static void
 618 g_disk_providergone(struct g_provider *pp)
 619 {
 620         struct disk *dp;
 621         struct g_disk_softc *sc;
 622
 623         sc = (struct g_disk_softc *)pp->private;
 624         dp = sc->dp;
 625         if (dp != NULL && dp->d_gone != NULL)
 626                 dp->d_gone(dp);
 627         if (sc->sysctl_tree != NULL) {
 628                 sysctl_ctx_free(&sc->sysctl_ctx);
 629                 sc->sysctl_tree = NULL;
 630         }
 631         if (sc->led[0] != 0) {
 632                 led_set(sc->led, "0");
 633                 sc->led[0] = 0;
 634         }
 635         pp->private = NULL;
 636         pp->geom->softc = NULL;
 637         mtx_destroy(&sc->done_mtx);
 638         mtx_destroy(&sc->start_mtx);
 639         g_free(sc);
 640 }
 641
 642 static void
 643 g_disk_destroy(void *ptr, int flag)
 644 {
 645         struct disk *dp;
 646         struct g_geom *gp;
 647         struct g_disk_softc *sc;
 648
 649         g_topology_assert();
 650         dp = ptr;
 651         gp = dp->d_geom;
 652         if (gp != NULL) {
 653                 sc = gp->softc;
 654                 if (sc != NULL)
 655                         sc->dp = NULL;
 656                 dp->d_geom = NULL;
 657                 g_wither_geom(gp, ENXIO);
 658         }
 659         g_free(dp);
 660 }
 661
 662 /*
 663  * We only allow printable characters in disk ident,
 664  * the rest is converted to 'x<HH>'.
 665  */
 666 static void
 667 g_disk_ident_adjust(char *ident, size_t size)
 668 {
 669         char *p, tmp[4], newid[DISK_IDENT_SIZE];
 670
 671         newid[0] = '\0';
 672         for (p = ident; *p != '\0'; p++) {
 673                 if (isprint(*p)) {
 674                         tmp[0] = *p;
 675                         tmp[1] = '\0';
 676                 } else {
 677                         snprintf(tmp, sizeof(tmp), "x%02hhx",
 678                             *(unsigned char *)p);
 679                 }
 680                 if (strlcat(newid, tmp, sizeof(newid)) >= sizeof(newid))
 681                         break;
 682         }
 683         bzero(ident, size);
 684         strlcpy(ident, newid, size);
 685 }
 686
 687 struct disk *
 688 disk_alloc(void)
 689 {
 690
 691         return (g_malloc(sizeof(struct disk), M_WAITOK | M_ZERO));
 692 }
 693
 694 void
 695 disk_create(struct disk *dp, int version)
 696 {
 697
 698         if (version != DISK_VERSION) {
 699                 printf("WARNING: Attempt to add disk %s%d %s",
 700                     dp->d_name, dp->d_unit,
 701                     " using incompatible ABI version of disk(9)\n");
 702                 printf("WARNING: Ignoring disk %s%d\n",
 703                     dp->d_name, dp->d_unit);
 704                 return;
 705         }
 706         if (version < DISK_VERSION_04)
 707                 dp->d_flags |= DISKFLAG_LACKS_ROTRATE;
 708         KASSERT(dp->d_strategy != NULL, ("disk_create need d_strategy"));
 709         KASSERT(dp->d_name != NULL, ("disk_create need d_name"));
 710         KASSERT(*dp->d_name != 0, ("disk_create need d_name"));
 711         KASSERT(strlen(dp->d_name) < SPECNAMELEN - 4, ("disk name too long"));
 712         if (dp->d_devstat == NULL)
 713                 dp->d_devstat = devstat_new_entry(dp->d_name, dp->d_unit,
 714                     dp->d_sectorsize, DEVSTAT_ALL_SUPPORTED,
 715                     DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
 716         dp->d_geom = NULL;
 717         g_disk_ident_adjust(dp->d_ident, sizeof(dp->d_ident));
 718         g_post_event(g_disk_create, dp, M_WAITOK, dp, NULL);
 719 }
 720
 721 void
 722 disk_destroy(struct disk *dp)
 723 {
 724
 725         g_cancel_event(dp);
 726         dp->d_destroyed = 1;
 727         if (dp->d_devstat != NULL)
 728                 devstat_remove_entry(dp->d_devstat);
 729         g_post_event(g_disk_destroy, dp, M_WAITOK, NULL);
 730 }
 731
 732 void
 733 disk_gone(struct disk *dp)
 734 {
 735         struct g_geom *gp;
 736         struct g_provider *pp;
 737
 738         gp = dp->d_geom;
 739         if (gp != NULL) {
 740                 pp = LIST_FIRST(&gp->provider);
 741                 if (pp != NULL) {
 742                         KASSERT(LIST_NEXT(pp, provider) == NULL,
 743                             ("geom %p has more than one provider", gp));
 744                         g_wither_provider(pp, ENXIO);
 745                 }
 746         }
 747 }
 748
 749 void
 750 disk_attr_changed(struct disk *dp, const char *attr, int flag)
 751 {
 752         struct g_geom *gp;
 753         struct g_provider *pp;
 754
 755         gp = dp->d_geom;
 756         if (gp != NULL)
 757                 LIST_FOREACH(pp, &gp->provider, provider)
 758                         (void)g_attr_changed(pp, attr, flag);
 759 }
 760
 761 void
 762 disk_media_changed(struct disk *dp, int flag)
 763 {
 764         struct g_geom *gp;
 765         struct g_provider *pp;
 766
 767         gp = dp->d_geom;
 768         if (gp != NULL) {
 769                 pp = LIST_FIRST(&gp->provider);
 770                 if (pp != NULL) {
 771                         KASSERT(LIST_NEXT(pp, provider) == NULL,
 772                             ("geom %p has more than one provider", gp));
 773                         g_media_changed(pp, flag);
 774                 }
 775         }
 776 }
 777
 778 void
 779 disk_media_gone(struct disk *dp, int flag)
 780 {
 781         struct g_geom *gp;
 782         struct g_provider *pp;
 783
 784         gp = dp->d_geom;
 785         if (gp != NULL) {
 786                 pp = LIST_FIRST(&gp->provider);
 787                 if (pp != NULL) {
 788                         KASSERT(LIST_NEXT(pp, provider) == NULL,
 789                             ("geom %p has more than one provider", gp));
 790                         g_media_gone(pp, flag);
 791                 }
 792         }
 793 }
 794
 795 int
 796 disk_resize(struct disk *dp, int flag)
 797 {
 798
 799         if (dp->d_destroyed || dp->d_geom == NULL)
 800                 return (0);
 801
 802         return (g_post_event(g_disk_resize, dp, flag, NULL));
 803 }
 804
 805 static void
 806 g_kern_disks(void *p, int flag __unused)
 807 {
 808         struct sbuf *sb;
 809         struct g_geom *gp;
 810         char *sp;
 811
 812         sb = p;
 813         sp = "";
 814         g_topology_assert();
 815         LIST_FOREACH(gp, &g_disk_class.geom, geom) {
 816                 sbuf_printf(sb, "%s%s", sp, gp->name);
 817                 sp = " ";
 818         }
 819         sbuf_finish(sb);
 820 }
 821
 822 static int
 823 sysctl_disks(SYSCTL_HANDLER_ARGS)
 824 {
 825         int error;
 826         struct sbuf *sb;
 827
 828         sb = sbuf_new_auto();
 829         g_waitfor_event(g_kern_disks, sb, M_WAITOK, NULL);
 830         error = SYSCTL_OUT(req, sbuf_data(sb), sbuf_len(sb) + 1);
 831         sbuf_delete(sb);
 832         return error;
 833 }
 834
 835 SYSCTL_PROC(_kern, OID_AUTO, disks,
 836     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0,
 837     sysctl_disks, "A", "names of available disks");