1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright 2015, Joyent, Inc. All rights reserved.
  26  */
  27 
  28 /*
  29  * zoneadmd manages zones; one zoneadmd process is launched for each
  30  * non-global zone on the system.  This daemon juggles four jobs:
  31  *
  32  * - Implement setup and teardown of the zone "virtual platform": mount and
  33  *   unmount filesystems; create and destroy network interfaces; communicate
  34  *   with devfsadmd to lay out devices for the zone; instantiate the zone
  35  *   console device; configure process runtime attributes such as resource
  36  *   controls, pool bindings, fine-grained privileges.
  37  *
  38  * - Launch the zone's init(1M) process.
  39  *
  40  * - Implement a door server; clients (like zoneadm) connect to the door
  41  *   server and request zone state changes.  The kernel is also a client of
  42  *   this door server.  A request to halt or reboot the zone which originates
  43  *   *inside* the zone results in a door upcall from the kernel into zoneadmd.
  44  *
  45  *   One minor problem is that messages emitted by zoneadmd need to be passed
  46  *   back to the zoneadm process making the request.  These messages need to
  47  *   be rendered in the client's locale; so, this is passed in as part of the
  48  *   request.  The exception is the kernel upcall to zoneadmd, in which case
  49  *   messages are syslog'd.
  50  *
  51  *   To make all of this work, the Makefile adds -a to xgettext to extract *all*
  52  *   strings, and an exclusion file (zoneadmd.xcl) is used to exclude those
  53  *   strings which do not need to be translated.
  54  *
  55  * - Act as a console server for zlogin -C processes; see comments in zcons.c
  56  *   for more information about the zone console architecture.
  57  *
  58  * DESIGN NOTES
  59  *
  60  * Restart:
  61  *   A chief design constraint of zoneadmd is that it should be restartable in
  62  *   the case that the administrator kills it off, or it suffers a fatal error,
  63  *   without the running zone being impacted; this is akin to being able to
  64  *   reboot the service processor of a server without affecting the OS instance.
  65  */
  66 
  67 #include <sys/param.h>
  68 #include <sys/mman.h>
  69 #include <sys/types.h>
  70 #include <sys/stat.h>
  71 #include <sys/sysmacros.h>
  72 #include <sys/time.h>
  73 
  74 #include <bsm/adt.h>
  75 #include <bsm/adt_event.h>
  76 
  77 #include <alloca.h>
  78 #include <assert.h>
  79 #include <errno.h>
  80 #include <door.h>
  81 #include <fcntl.h>
  82 #include <locale.h>
  83 #include <signal.h>
  84 #include <stdarg.h>
  85 #include <stdio.h>
  86 #include <stdlib.h>
  87 #include <string.h>
  88 #include <strings.h>
  89 #include <synch.h>
  90 #include <syslog.h>
  91 #include <thread.h>
  92 #include <unistd.h>
  93 #include <wait.h>
  94 #include <limits.h>
  95 #include <zone.h>
  96 #include <libbrand.h>
  97 #include <sys/brand.h>
  98 #include <libcontract.h>
  99 #include <libcontract_priv.h>
 100 #include <sys/brand.h>
 101 #include <sys/contract/process.h>
 102 #include <sys/ctfs.h>
 103 #include <libdladm.h>
 104 #include <sys/dls_mgmt.h>
 105 #include <libscf.h>
 106 
 107 #include <libzonecfg.h>
 108 #include <zonestat_impl.h>
 109 #include "zoneadmd.h"
 110 
 111 static char *progname;
 112 char *zone_name;        /* zone which we are managing */
 113 zone_dochandle_t snap_hndl;     /* handle for snapshot created when ready */
 114 char zonepath[MAXNAMELEN];
 115 char pool_name[MAXNAMELEN];
 116 char default_brand[MAXNAMELEN];
 117 char brand_name[MAXNAMELEN];
 118 boolean_t zone_isnative;
 119 boolean_t zone_iscluster;
 120 boolean_t zone_islabeled;
 121 boolean_t shutdown_in_progress;
 122 static zoneid_t zone_id;
 123 dladm_handle_t dld_handle = NULL;
 124 
 125 static char pre_statechg_hook[2 * MAXPATHLEN];
 126 static char post_statechg_hook[2 * MAXPATHLEN];
 127 char query_hook[2 * MAXPATHLEN];
 128 
 129 zlog_t logsys;
 130 
 131 mutex_t lock = DEFAULTMUTEX;    /* to serialize stuff */
 132 mutex_t msglock = DEFAULTMUTEX; /* for calling setlocale() */
 133 
 134 static sema_t scratch_sem;      /* for scratch zones */
 135 
 136 static char     zone_door_path[MAXPATHLEN];
 137 static int      zone_door = -1;
 138 
 139 boolean_t in_death_throes = B_FALSE;    /* daemon is dying */
 140 boolean_t bringup_failure_recovery = B_FALSE; /* ignore certain failures */
 141 
 142 #if !defined(TEXT_DOMAIN)               /* should be defined by cc -D */
 143 #define TEXT_DOMAIN     "SYS_TEST"      /* Use this only if it wasn't */
 144 #endif
 145 
 146 #define DEFAULT_LOCALE  "C"
 147 
 148 static const char *
 149 z_cmd_name(zone_cmd_t zcmd)
 150 {
 151         /* This list needs to match the enum in sys/zone.h */
 152         static const char *zcmdstr[] = {
 153                 "ready", "boot", "forceboot", "reboot", "halt",
 154                 "note_uninstalling", "mount", "forcemount", "unmount",
 155                 "shutdown"
 156         };
 157 
 158         if (zcmd >= sizeof (zcmdstr) / sizeof (*zcmdstr))
 159                 return ("unknown");
 160         else
 161                 return (zcmdstr[(int)zcmd]);
 162 }
 163 
 164 static char *
 165 get_execbasename(char *execfullname)
 166 {
 167         char *last_slash, *execbasename;
 168 
 169         /* guard against '/' at end of command invocation */
 170         for (;;) {
 171                 last_slash = strrchr(execfullname, '/');
 172                 if (last_slash == NULL) {
 173                         execbasename = execfullname;
 174                         break;
 175                 } else {
 176                         execbasename = last_slash + 1;
 177                         if (*execbasename == '\0') {
 178                                 *last_slash = '\0';
 179                                 continue;
 180                         }
 181                         break;
 182                 }
 183         }
 184         return (execbasename);
 185 }
 186 
 187 static void
 188 usage(void)
 189 {
 190         (void) fprintf(stderr, gettext("Usage: %s -z zonename\n"), progname);
 191         (void) fprintf(stderr,
 192             gettext("\tNote: %s should not be run directly.\n"), progname);
 193         exit(2);
 194 }
 195 
 196 /* ARGSUSED */
 197 static void
 198 sigchld(int sig)
 199 {
 200 }
 201 
 202 char *
 203 localize_msg(char *locale, const char *msg)
 204 {
 205         char *out;
 206 
 207         (void) mutex_lock(&msglock);
 208         (void) setlocale(LC_MESSAGES, locale);
 209         out = gettext(msg);
 210         (void) setlocale(LC_MESSAGES, DEFAULT_LOCALE);
 211         (void) mutex_unlock(&msglock);
 212         return (out);
 213 }
 214 
 215 /* PRINTFLIKE3 */
 216 void
 217 zerror(zlog_t *zlogp, boolean_t use_strerror, const char *fmt, ...)
 218 {
 219         va_list alist;
 220         char buf[MAXPATHLEN * 2]; /* enough space for err msg with a path */
 221         char *bp;
 222         int saved_errno = errno;
 223 
 224         if (zlogp == NULL)
 225                 return;
 226         if (zlogp == &logsys)
 227                 (void) snprintf(buf, sizeof (buf), "[zone '%s'] ",
 228                     zone_name);
 229         else
 230                 buf[0] = '\0';
 231         bp = &(buf[strlen(buf)]);
 232 
 233         /*
 234          * In theory, the locale pointer should be set to either "C" or a
 235          * char array, so it should never be NULL
 236          */
 237         assert(zlogp->locale != NULL);
 238         /* Locale is per process, but we are multi-threaded... */
 239         fmt = localize_msg(zlogp->locale, fmt);
 240 
 241         va_start(alist, fmt);
 242         (void) vsnprintf(bp, sizeof (buf) - (bp - buf), fmt, alist);
 243         va_end(alist);
 244         bp = &(buf[strlen(buf)]);
 245         if (use_strerror)
 246                 (void) snprintf(bp, sizeof (buf) - (bp - buf), ": %s",
 247                     strerror(saved_errno));
 248         if (zlogp == &logsys) {
 249                 (void) syslog(LOG_ERR, "%s", buf);
 250         } else if (zlogp->logfile != NULL) {
 251                 (void) fprintf(zlogp->logfile, "%s\n", buf);
 252         } else {
 253                 size_t buflen;
 254                 size_t copylen;
 255 
 256                 buflen = snprintf(zlogp->log, zlogp->loglen, "%s\n", buf);
 257                 copylen = MIN(buflen, zlogp->loglen);
 258                 zlogp->log += copylen;
 259                 zlogp->loglen -= copylen;
 260         }
 261 }
 262 
 263 /*
 264  * Since Solaris boot arguments are getopt(3c) compatible (see kernel(1m)), we
 265  * put the arguments into an argv style array, use getopt to process them,
 266  * and put the resultant argument string back into outargs. Non-Solaris brands
 267  * may support alternate forms of boot arguments so we must handle that as well.
 268  *
 269  * During the filtering, we pull out any arguments which are truly "boot"
 270  * arguments, leaving only those which are to be passed intact to the
 271  * progenitor process.  The one we support at the moment is -i, which
 272  * indicates to the kernel which program should be launched as 'init'.
 273  *
 274  * Except for Z_OK, all other return values are treated as fatal.
 275  */
 276 static int
 277 filter_bootargs(zlog_t *zlogp, const char *inargs, char *outargs,
 278     char *init_file)
 279 {
 280         int argc = 0, argc_save;
 281         int i;
 282         int err;
 283         char *arg, *lasts, **argv = NULL, **argv_save;
 284         char zonecfg_args[BOOTARGS_MAX];
 285         char scratchargs[BOOTARGS_MAX], *sargs;
 286         char c;
 287 
 288         bzero(outargs, BOOTARGS_MAX);
 289 
 290         /*
 291          * If the user didn't specify transient boot arguments, check
 292          * to see if there were any specified in the zone configuration,
 293          * and use them if applicable.
 294          */
 295         if (inargs == NULL || inargs[0] == '\0')  {
 296                 zone_dochandle_t handle;
 297                 if ((handle = zonecfg_init_handle()) == NULL) {
 298                         zerror(zlogp, B_TRUE,
 299                             "getting zone configuration handle");
 300                         return (Z_BAD_HANDLE);
 301                 }
 302                 err = zonecfg_get_snapshot_handle(zone_name, handle);
 303                 if (err != Z_OK) {
 304                         zerror(zlogp, B_FALSE,
 305                             "invalid configuration snapshot");
 306                         zonecfg_fini_handle(handle);
 307                         return (Z_BAD_HANDLE);
 308                 }
 309 
 310                 bzero(zonecfg_args, sizeof (zonecfg_args));
 311                 (void) zonecfg_get_bootargs(handle, zonecfg_args,
 312                     sizeof (zonecfg_args));
 313                 inargs = zonecfg_args;
 314                 zonecfg_fini_handle(handle);
 315         }
 316 
 317         if (strlen(inargs) >= BOOTARGS_MAX) {
 318                 zerror(zlogp, B_FALSE, "boot argument string too long");
 319                 return (Z_INVAL);
 320         }
 321 
 322         (void) strlcpy(scratchargs, inargs, sizeof (scratchargs));
 323         sargs = scratchargs;
 324         while ((arg = strtok_r(sargs, " \t", &lasts)) != NULL) {
 325                 sargs = NULL;
 326                 argc++;
 327         }
 328 
 329         if ((argv = calloc(argc + 1, sizeof (char *))) == NULL) {
 330                 zerror(zlogp, B_FALSE, "memory allocation failed");
 331                 return (Z_NOMEM);
 332         }
 333 
 334         argv_save = argv;
 335         argc_save = argc;
 336 
 337         (void) strlcpy(scratchargs, inargs, sizeof (scratchargs));
 338         sargs = scratchargs;
 339         i = 0;
 340         while ((arg = strtok_r(sargs, " \t", &lasts)) != NULL) {
 341                 sargs = NULL;
 342                 if ((argv[i] = strdup(arg)) == NULL) {
 343                         err = Z_NOMEM;
 344                         zerror(zlogp, B_FALSE, "memory allocation failed");
 345                         goto done;
 346                 }
 347                 i++;
 348         }
 349 
 350         /*
 351          * We preserve compatibility with the illumos system boot behavior,
 352          * which allows:
 353          *
 354          *      # reboot kernel/unix -s -m verbose
 355          *
 356          * In this example, kernel/unix tells the booter what file to boot. The
 357          * original intent of this was that we didn't want reboot in a zone to
 358          * be gratuitously different, so we would silently ignore the boot
 359          * file, if necessary. However, this usage is archaic and has never
 360          * been common, since it is impossible to boot a zone onto a different
 361          * kernel. Ignoring the first argument breaks for non-native brands
 362          * which pass boot arguments in a different style. e.g.
 363          *      systemd.log_level=debug
 364          * Thus, for backward compatibility we only ignore the first argument
 365          * if it appears to be in the illumos form and attempting to specify a
 366          * kernel.
 367          */
 368         if (argv[0] == NULL)
 369                 goto done;
 370 
 371         assert(argv[0][0] != ' ');
 372         assert(argv[0][0] != '\t');
 373 
 374         if (strncmp(argv[0], "kernel/", 7) == 0) {
 375                 argv = &argv[1];
 376                 argc--;
 377         }
 378 
 379         optind = 0;
 380         opterr = 0;
 381         err = Z_OK;
 382         while ((c = getopt(argc, argv, "fi:m:s")) != -1) {
 383                 switch (c) {
 384                 case 'i':
 385                         /*
 386                          * -i is handled by the runtime and is not passed
 387                          * along to userland
 388                          */
 389                         (void) strlcpy(init_file, optarg, MAXPATHLEN);
 390                         break;
 391                 case 'f':
 392                         /* This has already been processed by zoneadm */
 393                         break;
 394                 case 'm':
 395                 case 's':
 396                         /* These pass through unmolested */
 397                         (void) snprintf(outargs, BOOTARGS_MAX,
 398                             "%s -%c %s ", outargs, c, optarg ? optarg : "");
 399                         break;
 400                 case '?':
 401                         /*
 402                          * If a brand has its own init, we need to pass along
 403                          * whatever the user provides. We use the entire
 404                          * unknown string here so that we correctly handle
 405                          * unknown long options (e.g. --debug).
 406                          */
 407                         (void) snprintf(outargs, BOOTARGS_MAX,
 408                             "%s %s", outargs, argv[optind - 1]);
 409                         break;
 410                 }
 411         }
 412 
 413         /*
 414          * We need to pass along everything else since we don't know what
 415          * the brand's init is expecting. For example, an argument list like:
 416          *   --confdir /foo --debug
 417          * will cause the getopt parsing to stop at '/foo' but we need to pass
 418          * that on, along with the '--debug'. This does mean that we require
 419          * any of our known options (-ifms) to preceed the brand-specific ones.
 420          */
 421         while (optind < argc) {
 422                 (void) snprintf(outargs, BOOTARGS_MAX, "%s %s", outargs,
 423                     argv[optind]);
 424                 optind++;
 425         }
 426 
 427 done:
 428         for (i = 0; i < argc_save; i++) {
 429                 if (argv_save[i] != NULL)
 430                         free(argv_save[i]);
 431         }
 432         free(argv_save);
 433         return (err);
 434 }
 435 
 436 
 437 static int
 438 mkzonedir(zlog_t *zlogp)
 439 {
 440         struct stat st;
 441         /*
 442          * We must create and lock everyone but root out of ZONES_TMPDIR
 443          * since anyone can open any UNIX domain socket, regardless of
 444          * its file system permissions.  Sigh...
 445          */
 446         if (mkdir(ZONES_TMPDIR, S_IRWXU) < 0 && errno != EEXIST) {
 447                 zerror(zlogp, B_TRUE, "could not mkdir '%s'", ZONES_TMPDIR);
 448                 return (-1);
 449         }
 450         /* paranoia */
 451         if ((stat(ZONES_TMPDIR, &st) < 0) || !S_ISDIR(st.st_mode)) {
 452                 zerror(zlogp, B_TRUE, "'%s' is not a directory", ZONES_TMPDIR);
 453                 return (-1);
 454         }
 455         (void) chmod(ZONES_TMPDIR, S_IRWXU);
 456         return (0);
 457 }
 458 
 459 /*
 460  * Run the brand's pre-state change callback, if it exists.
 461  */
 462 static int
 463 brand_prestatechg(zlog_t *zlogp, int state, int cmd)
 464 {
 465         char cmdbuf[2 * MAXPATHLEN];
 466         const char *altroot;
 467 
 468         if (pre_statechg_hook[0] == '\0')
 469                 return (0);
 470 
 471         altroot = zonecfg_get_root();
 472         if (snprintf(cmdbuf, sizeof (cmdbuf), "%s %d %d %s", pre_statechg_hook,
 473             state, cmd, altroot) > sizeof (cmdbuf))
 474                 return (-1);
 475 
 476         if (do_subproc(zlogp, cmdbuf, NULL) != 0)
 477                 return (-1);
 478 
 479         return (0);
 480 }
 481 
 482 /*
 483  * Run the brand's post-state change callback, if it exists.
 484  */
 485 static int
 486 brand_poststatechg(zlog_t *zlogp, int state, int cmd)
 487 {
 488         char cmdbuf[2 * MAXPATHLEN];
 489         const char *altroot;
 490 
 491         if (post_statechg_hook[0] == '\0')
 492                 return (0);
 493 
 494         altroot = zonecfg_get_root();
 495         if (snprintf(cmdbuf, sizeof (cmdbuf), "%s %d %d %s", post_statechg_hook,
 496             state, cmd, altroot) > sizeof (cmdbuf))
 497                 return (-1);
 498 
 499         if (do_subproc(zlogp, cmdbuf, NULL) != 0)
 500                 return (-1);
 501 
 502         return (0);
 503 }
 504 
 505 /*
 506  * Notify zonestatd of the new zone.  If zonestatd is not running, this
 507  * will do nothing.
 508  */
 509 static void
 510 notify_zonestatd(zoneid_t zoneid)
 511 {
 512         int cmd[2];
 513         int fd;
 514         door_arg_t params;
 515 
 516         fd = open(ZS_DOOR_PATH, O_RDONLY);
 517         if (fd < 0)
 518                 return;
 519 
 520         cmd[0] = ZSD_CMD_NEW_ZONE;
 521         cmd[1] = zoneid;
 522         params.data_ptr = (char *)&cmd;
 523         params.data_size = sizeof (cmd);
 524         params.desc_ptr = NULL;
 525         params.desc_num = 0;
 526         params.rbuf = NULL;
 527         params.rsize = NULL;
 528         (void) door_call(fd, &params);
 529         (void) close(fd);
 530 }
 531 
 532 /*
 533  * Bring a zone up to the pre-boot "ready" stage.  The mount_cmd argument is
 534  * 'true' if this is being invoked as part of the processing for the "mount"
 535  * subcommand.
 536  */
 537 static int
 538 zone_ready(zlog_t *zlogp, zone_mnt_t mount_cmd, int zstate)
 539 {
 540         int err;
 541 
 542         if (brand_prestatechg(zlogp, zstate, Z_READY) != 0)
 543                 return (-1);
 544 
 545         if ((err = zonecfg_create_snapshot(zone_name)) != Z_OK) {
 546                 zerror(zlogp, B_FALSE, "unable to create snapshot: %s",
 547                     zonecfg_strerror(err));
 548                 goto bad;
 549         }
 550 
 551         if ((zone_id = vplat_create(zlogp, mount_cmd)) == -1) {
 552                 if ((err = zonecfg_destroy_snapshot(zone_name)) != Z_OK)
 553                         zerror(zlogp, B_FALSE, "destroying snapshot: %s",
 554                             zonecfg_strerror(err));
 555                 goto bad;
 556         }
 557         if (vplat_bringup(zlogp, mount_cmd, zone_id) != 0) {
 558                 bringup_failure_recovery = B_TRUE;
 559                 (void) vplat_teardown(NULL, (mount_cmd != Z_MNT_BOOT), B_FALSE);
 560                 if ((err = zonecfg_destroy_snapshot(zone_name)) != Z_OK)
 561                         zerror(zlogp, B_FALSE, "destroying snapshot: %s",
 562                             zonecfg_strerror(err));
 563                 goto bad;
 564         }
 565 
 566         if (brand_poststatechg(zlogp, zstate, Z_READY) != 0)
 567                 goto bad;
 568 
 569         return (0);
 570 
 571 bad:
 572         /*
 573          * If something goes wrong, we up the zones's state to the target
 574          * state, READY, and then invoke the hook as if we're halting.
 575          */
 576         (void) brand_poststatechg(zlogp, ZONE_STATE_READY, Z_HALT);
 577         return (-1);
 578 }
 579 
 580 int
 581 init_template(void)
 582 {
 583         int fd;
 584         int err = 0;
 585 
 586         fd = open64(CTFS_ROOT "/process/template", O_RDWR);
 587         if (fd == -1)
 588                 return (-1);
 589 
 590         /*
 591          * For now, zoneadmd doesn't do anything with the contract.
 592          * Deliver no events, don't inherit, and allow it to be orphaned.
 593          */
 594         err |= ct_tmpl_set_critical(fd, 0);
 595         err |= ct_tmpl_set_informative(fd, 0);
 596         err |= ct_pr_tmpl_set_fatal(fd, CT_PR_EV_HWERR);
 597         err |= ct_pr_tmpl_set_param(fd, CT_PR_PGRPONLY | CT_PR_REGENT);
 598         if (err || ct_tmpl_activate(fd)) {
 599                 (void) close(fd);
 600                 return (-1);
 601         }
 602 
 603         return (fd);
 604 }
 605 
 606 typedef struct fs_callback {
 607         zlog_t          *zlogp;
 608         zoneid_t        zoneid;
 609         boolean_t       mount_cmd;
 610 } fs_callback_t;
 611 
 612 static int
 613 mount_early_fs(void *data, const char *spec, const char *dir,
 614     const char *fstype, const char *opt)
 615 {
 616         zlog_t *zlogp = ((fs_callback_t *)data)->zlogp;
 617         zoneid_t zoneid = ((fs_callback_t *)data)->zoneid;
 618         boolean_t mount_cmd = ((fs_callback_t *)data)->mount_cmd;
 619         char rootpath[MAXPATHLEN];
 620         pid_t child;
 621         int child_status;
 622         int tmpl_fd;
 623         int rv;
 624         ctid_t ct;
 625 
 626         /* determine the zone rootpath */
 627         if (mount_cmd) {
 628                 char luroot[MAXPATHLEN];
 629 
 630                 (void) snprintf(luroot, sizeof (luroot), "%s/lu", zonepath);
 631                 resolve_lofs(zlogp, luroot, sizeof (luroot));
 632                 (void) strlcpy(rootpath, luroot, sizeof (rootpath));
 633         } else {
 634                 if (zone_get_rootpath(zone_name,
 635                     rootpath, sizeof (rootpath)) != Z_OK) {
 636                         zerror(zlogp, B_FALSE, "unable to determine zone root");
 637                         return (-1);
 638                 }
 639         }
 640 
 641         if ((rv = valid_mount_path(zlogp, rootpath, spec, dir, fstype)) < 0) {
 642                 zerror(zlogp, B_FALSE, "%s%s is not a valid mount point",
 643                     rootpath, dir);
 644                 return (-1);
 645         } else if (rv > 0) {
 646                 /* The mount point path doesn't exist, create it now. */
 647                 if (make_one_dir(zlogp, rootpath, dir,
 648                     DEFAULT_DIR_MODE, DEFAULT_DIR_USER,
 649                     DEFAULT_DIR_GROUP) != 0) {
 650                         zerror(zlogp, B_FALSE, "failed to create mount point");
 651                         return (-1);
 652                 }
 653 
 654                 /*
 655                  * Now this might seem weird, but we need to invoke
 656                  * valid_mount_path() again.  Why?  Because it checks
 657                  * to make sure that the mount point path is canonical,
 658                  * which it can only do if the path exists, so now that
 659                  * we've created the path we have to verify it again.
 660                  */
 661                 if ((rv = valid_mount_path(zlogp, rootpath, spec, dir,
 662                     fstype)) < 0) {
 663                         zerror(zlogp, B_FALSE,
 664                             "%s%s is not a valid mount point", rootpath, dir);
 665                         return (-1);
 666                 }
 667         }
 668 
 669         if ((tmpl_fd = init_template()) == -1) {
 670                 zerror(zlogp, B_TRUE, "failed to create contract");
 671                 return (-1);
 672         }
 673 
 674         if ((child = fork()) == -1) {
 675                 (void) ct_tmpl_clear(tmpl_fd);
 676                 (void) close(tmpl_fd);
 677                 zerror(zlogp, B_TRUE, "failed to fork");
 678                 return (-1);
 679 
 680         } else if (child == 0) {        /* child */
 681                 char opt_buf[MAX_MNTOPT_STR];
 682                 int optlen = 0;
 683                 int mflag = MS_DATA;
 684 
 685                 (void) ct_tmpl_clear(tmpl_fd);
 686                 /*
 687                  * Even though there are no procs running in the zone, we
 688                  * do this for paranoia's sake.
 689                  */
 690                 (void) closefrom(0);
 691 
 692                 if (zone_enter(zoneid) == -1) {
 693                         _exit(errno);
 694                 }
 695                 if (opt != NULL) {
 696                         /*
 697                          * The mount() system call is incredibly annoying.
 698                          * If options are specified, we need to copy them
 699                          * into a temporary buffer since the mount() system
 700                          * call will overwrite the options string.  It will
 701                          * also fail if the new option string it wants to
 702                          * write is bigger than the one we passed in, so
 703                          * you must pass in a buffer of the maximum possible
 704                          * option string length.  sigh.
 705                          */
 706                         (void) strlcpy(opt_buf, opt, sizeof (opt_buf));
 707                         opt = opt_buf;
 708                         optlen = MAX_MNTOPT_STR;
 709                         mflag = MS_OPTIONSTR;
 710                 }
 711                 if (mount(spec, dir, mflag, fstype, NULL, 0, opt, optlen) != 0)
 712                         _exit(errno);
 713                 _exit(0);
 714         }
 715 
 716         /* parent */
 717         if (contract_latest(&ct) == -1)
 718                 ct = -1;
 719         (void) ct_tmpl_clear(tmpl_fd);
 720         (void) close(tmpl_fd);
 721         if (waitpid(child, &child_status, 0) != child) {
 722                 /* unexpected: we must have been signalled */
 723                 (void) contract_abandon_id(ct);
 724                 return (-1);
 725         }
 726         (void) contract_abandon_id(ct);
 727         if (WEXITSTATUS(child_status) != 0) {
 728                 errno = WEXITSTATUS(child_status);
 729                 zerror(zlogp, B_TRUE, "mount of %s failed", dir);
 730                 return (-1);
 731         }
 732 
 733         return (0);
 734 }
 735 
 736 /*
 737  * If retstr is not NULL, the output of the subproc is returned in the str,
 738  * otherwise it is output using zerror().  Any memory allocated for retstr
 739  * should be freed by the caller.
 740  */
 741 int
 742 do_subproc(zlog_t *zlogp, char *cmdbuf, char **retstr)
 743 {
 744         char buf[1024];         /* arbitrary large amount */
 745         char *inbuf;
 746         FILE *file;
 747         int status;
 748         int rd_cnt;
 749 
 750         if (retstr != NULL) {
 751                 if ((*retstr = malloc(1024)) == NULL) {
 752                         zerror(zlogp, B_FALSE, "out of memory");
 753                         return (-1);
 754                 }
 755                 inbuf = *retstr;
 756                 rd_cnt = 0;
 757         } else {
 758                 inbuf = buf;
 759         }
 760 
 761         file = popen(cmdbuf, "r");
 762         if (file == NULL) {
 763                 zerror(zlogp, B_TRUE, "could not launch: %s", cmdbuf);
 764                 return (-1);
 765         }
 766 
 767         while (fgets(inbuf, 1024, file) != NULL) {
 768                 if (retstr == NULL) {
 769                         if (zlogp != &logsys)
 770                                 zerror(zlogp, B_FALSE, "%s", inbuf);
 771                 } else {
 772                         char *p;
 773 
 774                         rd_cnt += 1024 - 1;
 775                         if ((p = realloc(*retstr, rd_cnt + 1024)) == NULL) {
 776                                 zerror(zlogp, B_FALSE, "out of memory");
 777                                 (void) pclose(file);
 778                                 return (-1);
 779                         }
 780 
 781                         *retstr = p;
 782                         inbuf = *retstr + rd_cnt;
 783                 }
 784         }
 785         status = pclose(file);
 786 
 787         if (WIFSIGNALED(status)) {
 788                 zerror(zlogp, B_FALSE, "%s unexpectedly terminated due to "
 789                     "signal %d", cmdbuf, WTERMSIG(status));
 790                 return (-1);
 791         }
 792         assert(WIFEXITED(status));
 793         if (WEXITSTATUS(status) == ZEXIT_EXEC) {
 794                 zerror(zlogp, B_FALSE, "failed to exec %s", cmdbuf);
 795                 return (-1);
 796         }
 797         return (WEXITSTATUS(status));
 798 }
 799 
 800 #if 0 /* XXX KEBE SAYS not yet */
 801 /*
 802  * Get the path for this zone's init(1M) (or equivalent) process. First look
 803  * for a zone-specific init-name attr, then get it from the brand.
 804  */
 805 static int
 806 get_initname(brand_handle_t bh, char *initname, int len)
 807 {
 808         struct zone_attrtab a;
 809 
 810         bzero(&a, sizeof (a));
 811         (void) strlcpy(a.zone_attr_name, "init-name",
 812             sizeof (a.zone_attr_name));
 813 
 814         if (zonecfg_lookup_attr(snap_hndl, &a) == Z_OK) {
 815                 (void) strlcpy(initname, a.zone_attr_value, len);
 816                 return (0);
 817         }
 818 
 819         return (brand_get_initname(bh, initname, len));
 820 }
 821 
 822 /*
 823  * Get the restart-init flag for this zone's init(1M) (or equivalent) process.
 824  * First look for a zone-specific restart-init attr, then get it from the brand.
 825  */
 826 static boolean_t
 827 restartinit(brand_handle_t bh)
 828 {
 829         struct zone_attrtab a;
 830 
 831         bzero(&a, sizeof (a));
 832         (void) strlcpy(a.zone_attr_name, "restart-init",
 833             sizeof (a.zone_attr_name));
 834 
 835         if (zonecfg_lookup_attr(snap_hndl, &a) == Z_OK) {
 836                 if (strcmp(a.zone_attr_value, "false") == 0)
 837                         return (B_FALSE);
 838                 return (B_TRUE);
 839         }
 840 
 841         return (brand_restartinit(bh));
 842 }
 843 #endif /* XXX KEBE */
 844 
 845 /*
 846  * Get the app-svc-dependent flag for this zone's init process. This is a
 847  * zone-specific attr which controls the type of contract we create for the
 848  * zone's init. When true, the contract will include CT_PR_EV_EXIT in the fatal
 849  * set, so that when any service which is in the same contract exits, the init
 850  * application will be terminated.
 851  */
 852 static boolean_t
 853 is_app_svc_dep(brand_handle_t bh)
 854 {
 855         struct zone_attrtab a;
 856 
 857         bzero(&a, sizeof (a));
 858         (void) strlcpy(a.zone_attr_name, "app-svc-dependent",
 859             sizeof (a.zone_attr_name));
 860 
 861         if (zonecfg_lookup_attr(snap_hndl, &a) == Z_OK &&
 862             strcmp(a.zone_attr_value, "true") == 0) {
 863                 return (B_TRUE);
 864         }
 865 
 866         return (B_FALSE);
 867 }
 868 
 869 static int
 870 zone_bootup(zlog_t *zlogp, const char *bootargs, int zstate)
 871 {
 872         zoneid_t zoneid;
 873         struct stat st;
 874         char rpath[MAXPATHLEN], initpath[MAXPATHLEN], init_file[MAXPATHLEN];
 875         char nbootargs[BOOTARGS_MAX];
 876         char cmdbuf[MAXPATHLEN];
 877         fs_callback_t cb;
 878         brand_handle_t bh;
 879         zone_iptype_t iptype;
 880         dladm_status_t status;
 881         char errmsg[DLADM_STRSIZE];
 882         int err;
 883         boolean_t restart_init;
 884         boolean_t app_svc_dep;
 885 
 886         if (brand_prestatechg(zlogp, zstate, Z_BOOT) != 0)
 887                 return (-1);
 888 
 889         if ((zoneid = getzoneidbyname(zone_name)) == -1) {
 890                 zerror(zlogp, B_TRUE, "unable to get zoneid");
 891                 goto bad;
 892         }
 893 
 894         cb.zlogp = zlogp;
 895         cb.zoneid = zoneid;
 896         cb.mount_cmd = B_FALSE;
 897 
 898         /* Get a handle to the brand info for this zone */
 899         if ((bh = brand_open(brand_name)) == NULL) {
 900                 zerror(zlogp, B_FALSE, "unable to determine zone brand");
 901                 goto bad;
 902         }
 903 
 904         /*
 905          * Get the list of filesystems to mount from the brand
 906          * configuration.  These mounts are done via a thread that will
 907          * enter the zone, so they are done from within the context of the
 908          * zone.
 909          */
 910         if (brand_platform_iter_mounts(bh, mount_early_fs, &cb) != 0) {
 911                 zerror(zlogp, B_FALSE, "unable to mount filesystems");
 912                 brand_close(bh);
 913                 goto bad;
 914         }
 915 
 916         /*
 917          * Get the brand's boot callback if it exists.
 918          */
 919         (void) strcpy(cmdbuf, EXEC_PREFIX);
 920         if (brand_get_boot(bh, zone_name, zonepath, cmdbuf + EXEC_LEN,
 921             sizeof (cmdbuf) - EXEC_LEN) != 0) {
 922                 zerror(zlogp, B_FALSE,
 923                     "unable to determine branded zone's boot callback");
 924                 brand_close(bh);
 925                 goto bad;
 926         }
 927 
 928         /* Get the path for this zone's init(1M) (or equivalent) process.  */
 929         if (brand_get_initname(bh, init_file, MAXPATHLEN) != 0) {
 930                 zerror(zlogp, B_FALSE,
 931                     "unable to determine zone's init(1M) location");
 932                 brand_close(bh);
 933                 goto bad;
 934         }
 935 
 936         /* See if this zone's brand should restart init if it dies. */
 937         restart_init = brand_restartinit(bh);
 938 
 939         /*
 940          * See if we need to setup contract dependencies between the zone's
 941          * primary application and any of its services.
 942          */
 943         app_svc_dep = is_app_svc_dep(bh);
 944 
 945         brand_close(bh);
 946 
 947         err = filter_bootargs(zlogp, bootargs, nbootargs, init_file);
 948         if (err != Z_OK)
 949                 goto bad;
 950 
 951         assert(init_file[0] != '\0');
 952 
 953         /*
 954          * Try to anticipate possible problems: If possible, make sure init is
 955          * executable.
 956          */
 957         if (zone_get_rootpath(zone_name, rpath, sizeof (rpath)) != Z_OK) {
 958                 zerror(zlogp, B_FALSE, "unable to determine zone root");
 959                 goto bad;
 960         }
 961 
 962         (void) snprintf(initpath, sizeof (initpath), "%s%s", rpath, init_file);
 963 
 964         if (lstat(initpath, &st) == -1) {
 965                 zerror(zlogp, B_TRUE, "could not stat %s", initpath);
 966                 goto bad;
 967         }
 968 
 969         if ((st.st_mode & S_IFMT) == S_IFLNK) {
 970                 /* symlink, we'll have to wait and resolve when we boot */
 971         } else if ((st.st_mode & S_IXUSR) == 0) {
 972                 zerror(zlogp, B_FALSE, "%s is not executable", initpath);
 973                 goto bad;
 974         }
 975 
 976         /*
 977          * Exclusive stack zones interact with the dlmgmtd running in the
 978          * global zone.  dladm_zone_boot() tells dlmgmtd that this zone is
 979          * booting, and loads its datalinks from the zone's datalink
 980          * configuration file.
 981          */
 982         if (vplat_get_iptype(zlogp, &iptype) == 0 && iptype == ZS_EXCLUSIVE) {
 983                 status = dladm_zone_boot(dld_handle, zoneid);
 984                 if (status != DLADM_STATUS_OK) {
 985                         zerror(zlogp, B_FALSE, "unable to load zone datalinks: "
 986                             " %s", dladm_status2str(status, errmsg));
 987                         goto bad;
 988                 }
 989         }
 990 
 991         /*
 992          * If there is a brand 'boot' callback, execute it now to give the
 993          * brand one last chance to do any additional setup before the zone
 994          * is booted.
 995          */
 996         if ((strlen(cmdbuf) > EXEC_LEN) &&
 997             (do_subproc(zlogp, cmdbuf, NULL) != Z_OK)) {
 998                 zerror(zlogp, B_FALSE, "%s failed", cmdbuf);
 999                 goto bad;
1000         }
1001 
1002         if (zone_setattr(zoneid, ZONE_ATTR_INITNAME, init_file, 0) == -1) {
1003                 zerror(zlogp, B_TRUE, "could not set zone boot file");
1004                 goto bad;
1005         }
1006 
1007         if (zone_setattr(zoneid, ZONE_ATTR_BOOTARGS, nbootargs, 0) == -1) {
1008                 zerror(zlogp, B_TRUE, "could not set zone boot arguments");
1009                 goto bad;
1010         }
1011 
1012         if (!restart_init && zone_setattr(zoneid, ZONE_ATTR_INITNORESTART,
1013             NULL, 0) == -1) {
1014                 zerror(zlogp, B_TRUE, "could not set zone init-no-restart");
1015                 goto bad;
1016         }
1017 
1018         if (app_svc_dep && zone_setattr(zoneid, ZONE_ATTR_APP_SVC_CT,
1019             (void *)B_TRUE, sizeof (boolean_t)) == -1) {
1020                 zerror(zlogp, B_TRUE, "could not set zone app-die");
1021                 goto bad;
1022         }
1023 
1024         /*
1025          * Inform zonestatd of a new zone so that it can install a door for
1026          * the zone to contact it.
1027          */
1028         notify_zonestatd(zone_id);
1029 
1030         if (zone_boot(zoneid) == -1) {
1031                 zerror(zlogp, B_TRUE, "unable to boot zone");
1032                 goto bad;
1033         }
1034 
1035         if (brand_poststatechg(zlogp, zstate, Z_BOOT) != 0)
1036                 goto bad;
1037 
1038         /* Startup a thread to perform zfd logging/tty svc for the zone. */
1039         create_log_thread(zlogp, zone_id);
1040 
1041         /* Startup a thread to perform memory capping for the zone. */
1042         create_mcap_thread(zlogp, zone_id);
1043 
1044         return (0);
1045 
1046 bad:
1047         /*
1048          * If something goes wrong, we up the zones's state to the target
1049          * state, RUNNING, and then invoke the hook as if we're halting.
1050          */
1051         (void) brand_poststatechg(zlogp, ZONE_STATE_RUNNING, Z_HALT);
1052 
1053         return (-1);
1054 }
1055 
1056 static int
1057 zone_halt(zlog_t *zlogp, boolean_t unmount_cmd, boolean_t rebooting, int zstate)
1058 {
1059         int err;
1060 
1061         if (brand_prestatechg(zlogp, zstate, Z_HALT) != 0)
1062                 return (-1);
1063 
1064         /* Shutting down, stop the memcap thread */
1065         destroy_mcap_thread();
1066 
1067         if (vplat_teardown(zlogp, unmount_cmd, rebooting) != 0) {
1068                 if (!bringup_failure_recovery)
1069                         zerror(zlogp, B_FALSE, "unable to destroy zone");
1070                 destroy_log_thread();
1071                 return (-1);
1072         }
1073 
1074         /* Shut down is done, stop the log thread */
1075         destroy_log_thread();
1076 
1077         if (brand_poststatechg(zlogp, zstate, Z_HALT) != 0)
1078                 return (-1);
1079 
1080         if ((err = zonecfg_destroy_snapshot(zone_name)) != Z_OK)
1081                 zerror(zlogp, B_FALSE, "destroying snapshot: %s",
1082                     zonecfg_strerror(err));
1083 
1084         return (0);
1085 }
1086 
1087 static int
1088 zone_graceful_shutdown(zlog_t *zlogp)
1089 {
1090         zoneid_t zoneid;
1091         pid_t child;
1092         char cmdbuf[MAXPATHLEN];
1093         brand_handle_t bh = NULL;
1094         ctid_t ct;
1095         int tmpl_fd;
1096         int child_status;
1097 
1098         if (shutdown_in_progress) {
1099                 zerror(zlogp, B_FALSE, "shutdown already in progress");
1100                 return (-1);
1101         }
1102 
1103         if ((zoneid = getzoneidbyname(zone_name)) == -1) {
1104                 zerror(zlogp, B_TRUE, "unable to get zoneid");
1105                 return (-1);
1106         }
1107 
1108         /* Get a handle to the brand info for this zone */
1109         if ((bh = brand_open(brand_name)) == NULL) {
1110                 zerror(zlogp, B_FALSE, "unable to determine zone brand");
1111                 return (-1);
1112         }
1113 
1114         /*
1115          * If there is a brand 'shutdown' callback, execute it now to give the
1116          * brand a chance to cleanup any custom configuration.
1117          */
1118         (void) strcpy(cmdbuf, EXEC_PREFIX);
1119         if (brand_get_shutdown(bh, zone_name, zonepath, cmdbuf + EXEC_LEN,
1120             sizeof (cmdbuf) - EXEC_LEN) != 0 || strlen(cmdbuf) <= EXEC_LEN) {
1121                 (void) strcat(cmdbuf, SHUTDOWN_DEFAULT);
1122         }
1123         brand_close(bh);
1124 
1125         if ((tmpl_fd = init_template()) == -1) {
1126                 zerror(zlogp, B_TRUE, "failed to create contract");
1127                 return (-1);
1128         }
1129 
1130         if ((child = fork()) == -1) {
1131                 (void) ct_tmpl_clear(tmpl_fd);
1132                 (void) close(tmpl_fd);
1133                 zerror(zlogp, B_TRUE, "failed to fork");
1134                 return (-1);
1135         } else if (child == 0) {
1136                 (void) ct_tmpl_clear(tmpl_fd);
1137                 if (zone_enter(zoneid) == -1) {
1138                         _exit(errno);
1139                 }
1140                 _exit(execl("/bin/sh", "sh", "-c", cmdbuf, (char *)NULL));
1141         }
1142 
1143         if (contract_latest(&ct) == -1)
1144                 ct = -1;
1145         (void) ct_tmpl_clear(tmpl_fd);
1146         (void) close(tmpl_fd);
1147 
1148         if (waitpid(child, &child_status, 0) != child) {
1149                 /* unexpected: we must have been signalled */
1150                 (void) contract_abandon_id(ct);
1151                 return (-1);
1152         }
1153 
1154         (void) contract_abandon_id(ct);
1155         if (WEXITSTATUS(child_status) != 0) {
1156                 errno = WEXITSTATUS(child_status);
1157                 zerror(zlogp, B_FALSE, "unable to shutdown zone");
1158                 return (-1);
1159         }
1160 
1161         shutdown_in_progress = B_TRUE;
1162 
1163         return (0);
1164 }
1165 
1166 static int
1167 zone_wait_shutdown(zlog_t *zlogp)
1168 {
1169         zone_state_t zstate;
1170         uint64_t *tm = NULL;
1171         scf_simple_prop_t *prop = NULL;
1172         int timeout;
1173         int tries;
1174         int rc = -1;
1175 
1176         /* Get default stop timeout from SMF framework */
1177         timeout = SHUTDOWN_WAIT;
1178         if ((prop = scf_simple_prop_get(NULL, SHUTDOWN_FMRI, "stop",
1179             SCF_PROPERTY_TIMEOUT)) != NULL) {
1180                 if ((tm = scf_simple_prop_next_count(prop)) != NULL) {
1181                         if (tm != 0)
1182                                 timeout = *tm;
1183                 }
1184                 scf_simple_prop_free(prop);
1185         }
1186 
1187         /* allow time for zone to shutdown cleanly */
1188         for (tries = 0; tries < timeout; tries ++) {
1189                 (void) sleep(1);
1190                 if (zone_get_state(zone_name, &zstate) == Z_OK &&
1191                     zstate == ZONE_STATE_INSTALLED) {
1192                         rc = 0;
1193                         break;
1194                 }
1195         }
1196 
1197         if (rc != 0)
1198                 zerror(zlogp, B_FALSE, "unable to shutdown zone");
1199 
1200         shutdown_in_progress = B_FALSE;
1201 
1202         return (rc);
1203 }
1204 
1205 
1206 
1207 /*
1208  * Generate AUE_zone_state for a command that boots a zone.
1209  */
1210 static void
1211 audit_put_record(zlog_t *zlogp, ucred_t *uc, int return_val,
1212     char *new_state)
1213 {
1214         adt_session_data_t      *ah;
1215         adt_event_data_t        *event;
1216         int                     pass_fail, fail_reason;
1217 
1218         if (!adt_audit_enabled())
1219                 return;
1220 
1221         if (return_val == 0) {
1222                 pass_fail = ADT_SUCCESS;
1223                 fail_reason = ADT_SUCCESS;
1224         } else {
1225                 pass_fail = ADT_FAILURE;
1226                 fail_reason = ADT_FAIL_VALUE_PROGRAM;
1227         }
1228 
1229         if (adt_start_session(&ah, NULL, 0)) {
1230                 zerror(zlogp, B_TRUE, gettext("audit failure."));
1231                 return;
1232         }
1233         if (adt_set_from_ucred(ah, uc, ADT_NEW)) {
1234                 zerror(zlogp, B_TRUE, gettext("audit failure."));
1235                 (void) adt_end_session(ah);
1236                 return;
1237         }
1238 
1239         event = adt_alloc_event(ah, ADT_zone_state);
1240         if (event == NULL) {
1241                 zerror(zlogp, B_TRUE, gettext("audit failure."));
1242                 (void) adt_end_session(ah);
1243                 return;
1244         }
1245         event->adt_zone_state.zonename = zone_name;
1246         event->adt_zone_state.new_state = new_state;
1247 
1248         if (adt_put_event(event, pass_fail, fail_reason))
1249                 zerror(zlogp, B_TRUE, gettext("audit failure."));
1250 
1251         adt_free_event(event);
1252 
1253         (void) adt_end_session(ah);
1254 }
1255 
1256 /*
1257  * Log the exit time and status of the zone's init process into
1258  * {zonepath}/lastexited. If the zone shutdown normally, the exit status will
1259  * be -1, otherwise it will be the exit status as described in wait.3c.
1260  * If the zone is configured to restart init, then nothing will be logged if
1261  * init exits unexpectedly (the kernel will never upcall in this case).
1262  */
1263 static void
1264 log_init_exit(int status)
1265 {
1266         char p[MAXPATHLEN];
1267         char buf[128];
1268         struct timeval t;
1269         int fd;
1270 
1271         if (snprintf(p, sizeof (p), "%s/lastexited", zonepath) > sizeof (p))
1272                 return;
1273         if (gettimeofday(&t, NULL) != 0)
1274                 return;
1275         if (snprintf(buf, sizeof (buf), "%ld.%ld %d\n", t.tv_sec, t.tv_usec,
1276             status) > sizeof (buf))
1277                 return;
1278         if ((fd = open(p, O_WRONLY | O_CREAT | O_TRUNC, 0644)) < 0)
1279                 return;
1280 
1281         (void) write(fd, buf, strlen(buf));
1282 
1283         (void) close(fd);
1284 }
1285 
1286 /*
1287  * The main routine for the door server that deals with zone state transitions.
1288  */
1289 /* ARGSUSED */
1290 static void
1291 server(void *cookie, char *args, size_t alen, door_desc_t *dp,
1292     uint_t n_desc)
1293 {
1294         ucred_t *uc = NULL;
1295         const priv_set_t *eset;
1296 
1297         zone_state_t zstate;
1298         zone_cmd_t cmd;
1299         boolean_t debug;
1300         int init_status;
1301         zone_cmd_arg_t *zargp;
1302 
1303         boolean_t kernelcall;
1304 
1305         int rval = -1;
1306         uint64_t uniqid;
1307         zoneid_t zoneid = -1;
1308         zlog_t zlog;
1309         zlog_t *zlogp;
1310         zone_cmd_rval_t *rvalp;
1311         size_t rlen = getpagesize(); /* conservative */
1312         fs_callback_t cb;
1313         brand_handle_t bh;
1314         boolean_t wait_shut = B_FALSE;
1315 
1316         /* LINTED E_BAD_PTR_CAST_ALIGN */
1317         zargp = (zone_cmd_arg_t *)args;
1318 
1319         /*
1320          * When we get the door unref message, we've fdetach'd the door, and
1321          * it is time for us to shut down zoneadmd.
1322          */
1323         if (zargp == DOOR_UNREF_DATA) {
1324                 /*
1325                  * See comment at end of main() for info on the last rites.
1326                  */
1327                 exit(0);
1328         }
1329 
1330         if (zargp == NULL) {
1331                 (void) door_return(NULL, 0, 0, 0);
1332         }
1333 
1334         rvalp = alloca(rlen);
1335         bzero(rvalp, rlen);
1336         zlog.logfile = NULL;
1337         zlog.buflen = zlog.loglen = rlen - sizeof (zone_cmd_rval_t) + 1;
1338         zlog.buf = rvalp->errbuf;
1339         zlog.log = zlog.buf;
1340         /* defer initialization of zlog.locale until after credential check */
1341         zlogp = &zlog;
1342 
1343         if (alen != sizeof (zone_cmd_arg_t)) {
1344                 /*
1345                  * This really shouldn't be happening.
1346                  */
1347                 zerror(&logsys, B_FALSE, "argument size (%d bytes) "
1348                     "unexpected (expected %d bytes)", alen,
1349                     sizeof (zone_cmd_arg_t));
1350                 goto out;
1351         }
1352         cmd = zargp->cmd;
1353         debug = zargp->debug;
1354         init_status = zargp->status;
1355 
1356         if (door_ucred(&uc) != 0) {
1357                 zerror(&logsys, B_TRUE, "door_ucred");
1358                 goto out;
1359         }
1360         eset = ucred_getprivset(uc, PRIV_EFFECTIVE);
1361         if (ucred_getzoneid(uc) != GLOBAL_ZONEID ||
1362             (eset != NULL ? !priv_ismember(eset, PRIV_SYS_CONFIG) :
1363             ucred_geteuid(uc) != 0)) {
1364                 zerror(&logsys, B_FALSE, "insufficient privileges");
1365                 goto out;
1366         }
1367 
1368         kernelcall = ucred_getpid(uc) == 0;
1369 
1370         /*
1371          * This is safe because we only use a zlog_t throughout the
1372          * duration of a door call; i.e., by the time the pointer
1373          * might become invalid, the door call would be over.
1374          */
1375         zlog.locale = kernelcall ? DEFAULT_LOCALE : zargp->locale;
1376 
1377         (void) mutex_lock(&lock);
1378 
1379         /*
1380          * Once we start to really die off, we don't want more connections.
1381          */
1382         if (in_death_throes) {
1383                 (void) mutex_unlock(&lock);
1384                 ucred_free(uc);
1385                 (void) door_return(NULL, 0, 0, 0);
1386                 thr_exit(NULL);
1387         }
1388 
1389         /*
1390          * Check for validity of command.
1391          */
1392         if (cmd != Z_READY && cmd != Z_BOOT && cmd != Z_FORCEBOOT &&
1393             cmd != Z_REBOOT && cmd != Z_SHUTDOWN && cmd != Z_HALT &&
1394             cmd != Z_NOTE_UNINSTALLING && cmd != Z_MOUNT &&
1395             cmd != Z_FORCEMOUNT && cmd != Z_UNMOUNT) {
1396                 zerror(&logsys, B_FALSE, "invalid command %d", (int)cmd);
1397                 goto out;
1398         }
1399 
1400         if (kernelcall && (cmd != Z_HALT && cmd != Z_REBOOT)) {
1401                 /*
1402                  * Can't happen
1403                  */
1404                 zerror(&logsys, B_FALSE, "received unexpected kernel upcall %d",
1405                     cmd);
1406                 goto out;
1407         }
1408         /*
1409          * We ignore the possibility of someone calling zone_create(2)
1410          * explicitly; all requests must come through zoneadmd.
1411          */
1412         if (zone_get_state(zone_name, &zstate) != Z_OK) {
1413                 /*
1414                  * Something terribly wrong happened
1415                  */
1416                 zerror(&logsys, B_FALSE, "unable to determine state of zone");
1417                 goto out;
1418         }
1419 
1420         if (kernelcall) {
1421                 /*
1422                  * Kernel-initiated requests may lose their validity if the
1423                  * zone_t the kernel was referring to has gone away.
1424                  */
1425                 if ((zoneid = getzoneidbyname(zone_name)) == -1 ||
1426                     zone_getattr(zoneid, ZONE_ATTR_UNIQID, &uniqid,
1427                     sizeof (uniqid)) == -1 || uniqid != zargp->uniqid) {
1428                         /*
1429                          * We're not talking about the same zone. The request
1430                          * must have arrived too late.  Return error.
1431                          */
1432                         rval = -1;
1433                         goto out;
1434                 }
1435                 zlogp = &logsys;    /* Log errors to syslog */
1436         }
1437 
1438         /*
1439          * If we are being asked to forcibly mount or boot a zone, we
1440          * pretend that an INCOMPLETE zone is actually INSTALLED.
1441          */
1442         if (zstate == ZONE_STATE_INCOMPLETE &&
1443             (cmd == Z_FORCEBOOT || cmd == Z_FORCEMOUNT))
1444                 zstate = ZONE_STATE_INSTALLED;
1445 
1446         switch (zstate) {
1447         case ZONE_STATE_CONFIGURED:
1448         case ZONE_STATE_INCOMPLETE:
1449                 /*
1450                  * Not our area of expertise; we just print a nice message
1451                  * and die off.
1452                  */
1453                 zerror(zlogp, B_FALSE,
1454                     "%s operation is invalid for zones in state '%s'",
1455                     z_cmd_name(cmd), zone_state_str(zstate));
1456                 break;
1457 
1458         case ZONE_STATE_INSTALLED:
1459                 switch (cmd) {
1460                 case Z_READY:
1461                         rval = zone_ready(zlogp, Z_MNT_BOOT, zstate);
1462                         if (rval == 0)
1463                                 eventstream_write(Z_EVT_ZONE_READIED);
1464                         break;
1465                 case Z_BOOT:
1466                 case Z_FORCEBOOT:
1467                         eventstream_write(Z_EVT_ZONE_BOOTING);
1468                         if ((rval = zone_ready(zlogp, Z_MNT_BOOT, zstate))
1469                             == 0) {
1470                                 rval = zone_bootup(zlogp, zargp->bootbuf,
1471                                     zstate);
1472                         }
1473                         audit_put_record(zlogp, uc, rval, "boot");
1474                         if (rval != 0) {
1475                                 bringup_failure_recovery = B_TRUE;
1476                                 (void) zone_halt(zlogp, B_FALSE, B_FALSE,
1477                                     zstate);
1478                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1479                         }
1480                         break;
1481                 case Z_SHUTDOWN:
1482                 case Z_HALT:
1483                         if (kernelcall) /* Invalid; can't happen */
1484                                 abort();
1485                         /*
1486                          * We could have two clients racing to halt this
1487                          * zone; the second client loses, but his request
1488                          * doesn't fail, since the zone is now in the desired
1489                          * state.
1490                          */
1491                         zerror(zlogp, B_FALSE, "zone is already halted");
1492                         rval = 0;
1493                         break;
1494                 case Z_REBOOT:
1495                         if (kernelcall) /* Invalid; can't happen */
1496                                 abort();
1497                         zerror(zlogp, B_FALSE, "%s operation is invalid "
1498                             "for zones in state '%s'", z_cmd_name(cmd),
1499                             zone_state_str(zstate));
1500                         rval = -1;
1501                         break;
1502                 case Z_NOTE_UNINSTALLING:
1503                         if (kernelcall) /* Invalid; can't happen */
1504                                 abort();
1505                         /*
1506                          * Tell the console to print out a message about this.
1507                          * Once it does, we will be in_death_throes.
1508                          */
1509                         eventstream_write(Z_EVT_ZONE_UNINSTALLING);
1510                         break;
1511                 case Z_MOUNT:
1512                 case Z_FORCEMOUNT:
1513                         if (kernelcall) /* Invalid; can't happen */
1514                                 abort();
1515                         if (!zone_isnative && !zone_iscluster &&
1516                             !zone_islabeled) {
1517                                 /*
1518                                  * -U mounts the zone without lofs mounting
1519                                  * zone file systems back into the scratch
1520                                  * zone.  This is required when mounting
1521                                  * non-native branded zones.
1522                                  */
1523                                 (void) strlcpy(zargp->bootbuf, "-U",
1524                                     BOOTARGS_MAX);
1525                         }
1526 
1527                         rval = zone_ready(zlogp,
1528                             strcmp(zargp->bootbuf, "-U") == 0 ?
1529                             Z_MNT_UPDATE : Z_MNT_SCRATCH, zstate);
1530                         if (rval != 0)
1531                                 break;
1532 
1533                         eventstream_write(Z_EVT_ZONE_READIED);
1534 
1535                         /*
1536                          * Get a handle to the default brand info.
1537                          * We must always use the default brand file system
1538                          * list when mounting the zone.
1539                          */
1540                         if ((bh = brand_open(default_brand)) == NULL) {
1541                                 rval = -1;
1542                                 break;
1543                         }
1544 
1545                         /*
1546                          * Get the list of filesystems to mount from
1547                          * the brand configuration.  These mounts are done
1548                          * via a thread that will enter the zone, so they
1549                          * are done from within the context of the zone.
1550                          */
1551                         cb.zlogp = zlogp;
1552                         cb.zoneid = zone_id;
1553                         cb.mount_cmd = B_TRUE;
1554                         rval = brand_platform_iter_mounts(bh,
1555                             mount_early_fs, &cb);
1556 
1557                         brand_close(bh);
1558 
1559                         /*
1560                          * Ordinarily, /dev/fd would be mounted inside the zone
1561                          * by svc:/system/filesystem/usr:default, but since
1562                          * we're not booting the zone, we need to do this
1563                          * manually.
1564                          */
1565                         if (rval == 0)
1566                                 rval = mount_early_fs(&cb,
1567                                     "fd", "/dev/fd", "fd", NULL);
1568                         break;
1569                 case Z_UNMOUNT:
1570                         if (kernelcall) /* Invalid; can't happen */
1571                                 abort();
1572                         zerror(zlogp, B_FALSE, "zone is already unmounted");
1573                         rval = 0;
1574                         break;
1575                 }
1576                 break;
1577 
1578         case ZONE_STATE_READY:
1579                 switch (cmd) {
1580                 case Z_READY:
1581                         /*
1582                          * We could have two clients racing to ready this
1583                          * zone; the second client loses, but his request
1584                          * doesn't fail, since the zone is now in the desired
1585                          * state.
1586                          */
1587                         zerror(zlogp, B_FALSE, "zone is already ready");
1588                         rval = 0;
1589                         break;
1590                 case Z_BOOT:
1591                         (void) strlcpy(boot_args, zargp->bootbuf,
1592                             sizeof (boot_args));
1593                         eventstream_write(Z_EVT_ZONE_BOOTING);
1594                         rval = zone_bootup(zlogp, zargp->bootbuf, zstate);
1595                         audit_put_record(zlogp, uc, rval, "boot");
1596                         if (rval != 0) {
1597                                 bringup_failure_recovery = B_TRUE;
1598                                 (void) zone_halt(zlogp, B_FALSE, B_TRUE,
1599                                     zstate);
1600                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1601                         }
1602                         boot_args[0] = '\0';
1603                         break;
1604                 case Z_HALT:
1605                         if (kernelcall) /* Invalid; can't happen */
1606                                 abort();
1607                         if ((rval = zone_halt(zlogp, B_FALSE, B_FALSE, zstate))
1608                             != 0)
1609                                 break;
1610                         eventstream_write(Z_EVT_ZONE_HALTED);
1611                         break;
1612                 case Z_SHUTDOWN:
1613                 case Z_REBOOT:
1614                 case Z_NOTE_UNINSTALLING:
1615                 case Z_MOUNT:
1616                 case Z_UNMOUNT:
1617                         if (kernelcall) /* Invalid; can't happen */
1618                                 abort();
1619                         zerror(zlogp, B_FALSE, "%s operation is invalid "
1620                             "for zones in state '%s'", z_cmd_name(cmd),
1621                             zone_state_str(zstate));
1622                         rval = -1;
1623                         break;
1624                 }
1625                 break;
1626 
1627         case ZONE_STATE_MOUNTED:
1628                 switch (cmd) {
1629                 case Z_UNMOUNT:
1630                         if (kernelcall) /* Invalid; can't happen */
1631                                 abort();
1632                         rval = zone_halt(zlogp, B_TRUE, B_FALSE, zstate);
1633                         if (rval == 0) {
1634                                 eventstream_write(Z_EVT_ZONE_HALTED);
1635                                 (void) sema_post(&scratch_sem);
1636                         }
1637                         break;
1638                 default:
1639                         if (kernelcall) /* Invalid; can't happen */
1640                                 abort();
1641                         zerror(zlogp, B_FALSE, "%s operation is invalid "
1642                             "for zones in state '%s'", z_cmd_name(cmd),
1643                             zone_state_str(zstate));
1644                         rval = -1;
1645                         break;
1646                 }
1647                 break;
1648 
1649         case ZONE_STATE_RUNNING:
1650         case ZONE_STATE_SHUTTING_DOWN:
1651         case ZONE_STATE_DOWN:
1652                 switch (cmd) {
1653                 case Z_READY:
1654                         if ((rval = zone_halt(zlogp, B_FALSE, B_TRUE, zstate))
1655                             != 0)
1656                                 break;
1657                         if ((rval = zone_ready(zlogp, Z_MNT_BOOT, zstate)) == 0)
1658                                 eventstream_write(Z_EVT_ZONE_READIED);
1659                         else
1660                                 eventstream_write(Z_EVT_ZONE_HALTED);
1661                         break;
1662                 case Z_BOOT:
1663                         /*
1664                          * We could have two clients racing to boot this
1665                          * zone; the second client loses, but his request
1666                          * doesn't fail, since the zone is now in the desired
1667                          * state.
1668                          */
1669                         zerror(zlogp, B_FALSE, "zone is already booted");
1670                         rval = 0;
1671                         break;
1672                 case Z_HALT:
1673                         if (kernelcall) {
1674                                 log_init_exit(init_status);
1675                         } else {
1676                                 log_init_exit(-1);
1677                         }
1678                         if ((rval = zone_halt(zlogp, B_FALSE, B_FALSE, zstate))
1679                             != 0)
1680                                 break;
1681                         eventstream_write(Z_EVT_ZONE_HALTED);
1682                         break;
1683                 case Z_REBOOT:
1684                         (void) strlcpy(boot_args, zargp->bootbuf,
1685                             sizeof (boot_args));
1686                         eventstream_write(Z_EVT_ZONE_REBOOTING);
1687                         if ((rval = zone_halt(zlogp, B_FALSE, B_TRUE, zstate))
1688                             != 0) {
1689                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1690                                 boot_args[0] = '\0';
1691                                 break;
1692                         }
1693                         if ((rval = zone_ready(zlogp, Z_MNT_BOOT, zstate))
1694                             != 0) {
1695                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1696                                 boot_args[0] = '\0';
1697                                 break;
1698                         }
1699                         rval = zone_bootup(zlogp, zargp->bootbuf, zstate);
1700                         audit_put_record(zlogp, uc, rval, "reboot");
1701                         if (rval != 0) {
1702                                 (void) zone_halt(zlogp, B_FALSE, B_TRUE,
1703                                     zstate);
1704                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1705                         }
1706                         boot_args[0] = '\0';
1707                         break;
1708                 case Z_SHUTDOWN:
1709                         if ((rval = zone_graceful_shutdown(zlogp)) == 0) {
1710                                 wait_shut = B_TRUE;
1711                         }
1712                         break;
1713                 case Z_NOTE_UNINSTALLING:
1714                 case Z_MOUNT:
1715                 case Z_UNMOUNT:
1716                         zerror(zlogp, B_FALSE, "%s operation is invalid "
1717                             "for zones in state '%s'", z_cmd_name(cmd),
1718                             zone_state_str(zstate));
1719                         rval = -1;
1720                         break;
1721                 }
1722                 break;
1723         default:
1724                 abort();
1725         }
1726 
1727         /*
1728          * Because the state of the zone may have changed, we make sure
1729          * to wake the console poller, which is in charge of initiating
1730          * the shutdown procedure as necessary.
1731          */
1732         eventstream_write(Z_EVT_NULL);
1733 
1734 out:
1735         (void) mutex_unlock(&lock);
1736 
1737         /* Wait for the Z_SHUTDOWN commands to complete */
1738         if (wait_shut)
1739                 rval = zone_wait_shutdown(zlogp);
1740 
1741         if (kernelcall) {
1742                 rvalp = NULL;
1743                 rlen = 0;
1744         } else {
1745                 rvalp->rval = rval;
1746         }
1747         if (uc != NULL)
1748                 ucred_free(uc);
1749         (void) door_return((char *)rvalp, rlen, NULL, 0);
1750         thr_exit(NULL);
1751 }
1752 
1753 static int
1754 setup_door(zlog_t *zlogp)
1755 {
1756         if ((zone_door = door_create(server, NULL,
1757             DOOR_UNREF | DOOR_REFUSE_DESC | DOOR_NO_CANCEL)) < 0) {
1758                 zerror(zlogp, B_TRUE, "%s failed", "door_create");
1759                 return (-1);
1760         }
1761         (void) fdetach(zone_door_path);
1762 
1763         if (fattach(zone_door, zone_door_path) != 0) {
1764                 zerror(zlogp, B_TRUE, "fattach to %s failed", zone_door_path);
1765                 (void) door_revoke(zone_door);
1766                 (void) fdetach(zone_door_path);
1767                 zone_door = -1;
1768                 return (-1);
1769         }
1770         return (0);
1771 }
1772 
1773 /*
1774  * zoneadm(1m) will start zoneadmd if it thinks it isn't running; this
1775  * is where zoneadmd itself will check to see that another instance of
1776  * zoneadmd isn't already controlling this zone.
1777  *
1778  * The idea here is that we want to open the path to which we will
1779  * attach our door, lock it, and then make sure that no-one has beat us
1780  * to fattach(3c)ing onto it.
1781  *
1782  * fattach(3c) is really a mount, so there are actually two possible
1783  * vnodes we could be dealing with.  Our strategy is as follows:
1784  *
1785  * - If the file we opened is a regular file (common case):
1786  *      There is no fattach(3c)ed door, so we have a chance of becoming
1787  *      the managing zoneadmd. We attempt to lock the file: if it is
1788  *      already locked, that means someone else raced us here, so we
1789  *      lose and give up.  zoneadm(1m) will try to contact the zoneadmd
1790  *      that beat us to it.
1791  *
1792  * - If the file we opened is a namefs file:
1793  *      This means there is already an established door fattach(3c)'ed
1794  *      to the rendezvous path.  We've lost the race, so we give up.
1795  *      Note that in this case we also try to grab the file lock, and
1796  *      will succeed in acquiring it since the vnode locked by the
1797  *      "winning" zoneadmd was a regular one, and the one we locked was
1798  *      the fattach(3c)'ed door node.  At any rate, no harm is done, and
1799  *      we just return to zoneadm(1m) which knows to retry.
1800  */
1801 static int
1802 make_daemon_exclusive(zlog_t *zlogp)
1803 {
1804         int doorfd = -1;
1805         int err, ret = -1;
1806         struct stat st;
1807         struct flock flock;
1808         zone_state_t zstate;
1809 
1810 top:
1811         if ((err = zone_get_state(zone_name, &zstate)) != Z_OK) {
1812                 zerror(zlogp, B_FALSE, "failed to get zone state: %s",
1813                     zonecfg_strerror(err));
1814                 goto out;
1815         }
1816         if ((doorfd = open(zone_door_path, O_CREAT|O_RDWR,
1817             S_IREAD|S_IWRITE)) < 0) {
1818                 zerror(zlogp, B_TRUE, "failed to open %s", zone_door_path);
1819                 goto out;
1820         }
1821         if (fstat(doorfd, &st) < 0) {
1822                 zerror(zlogp, B_TRUE, "failed to stat %s", zone_door_path);
1823                 goto out;
1824         }
1825         /*
1826          * Lock the file to synchronize with other zoneadmd
1827          */
1828         flock.l_type = F_WRLCK;
1829         flock.l_whence = SEEK_SET;
1830         flock.l_start = (off_t)0;
1831         flock.l_len = (off_t)0;
1832         if (fcntl(doorfd, F_SETLK, &flock) < 0) {
1833                 /*
1834                  * Someone else raced us here and grabbed the lock file
1835                  * first.  A warning here is inappropriate since nothing
1836                  * went wrong.
1837                  */
1838                 goto out;
1839         }
1840 
1841         if (strcmp(st.st_fstype, "namefs") == 0) {
1842                 struct door_info info;
1843 
1844                 /*
1845                  * There is already something fattach()'ed to this file.
1846                  * Lets see what the door is up to.
1847                  */
1848                 if (door_info(doorfd, &info) == 0 && info.di_target != -1) {
1849                         /*
1850                          * Another zoneadmd process seems to be in
1851                          * control of the situation and we don't need to
1852                          * be here.  A warning here is inappropriate
1853                          * since nothing went wrong.
1854                          *
1855                          * If the door has been revoked, the zoneadmd
1856                          * process currently managing the zone is going
1857                          * away.  We'll return control to zoneadm(1m)
1858                          * which will try again (by which time zoneadmd
1859                          * will hopefully have exited).
1860                          */
1861                         goto out;
1862                 }
1863 
1864                 /*
1865                  * If we got this far, there's a fattach(3c)'ed door
1866                  * that belongs to a process that has exited, which can
1867                  * happen if the previous zoneadmd died unexpectedly.
1868                  *
1869                  * Let user know that something is amiss, but that we can
1870                  * recover; if the zone is in the installed state, then don't
1871                  * message, since having a running zoneadmd isn't really
1872                  * expected/needed.  We want to keep occurences of this message
1873                  * limited to times when zoneadmd is picking back up from a
1874                  * zoneadmd that died while the zone was in some non-trivial
1875                  * state.
1876                  */
1877                 if (zstate > ZONE_STATE_INSTALLED) {
1878                         static zoneid_t zid;
1879 
1880                         zerror(zlogp, B_FALSE,
1881                             "zone '%s': WARNING: zone is in state '%s', but "
1882                             "zoneadmd does not appear to be available; "
1883                             "restarted zoneadmd to recover.",
1884                             zone_name, zone_state_str(zstate));
1885 
1886                         /*
1887                          * Startup a thread to perform the zfd logging/tty svc
1888                          * and a thread to perform memory capping for the
1889                          * zone. zlogp won't be valid for much longer so use
1890                          * logsys.
1891                          */
1892                         if ((zid = getzoneidbyname(zone_name)) != -1) {
1893                                 create_log_thread(&logsys, zid);
1894                                 create_mcap_thread(&logsys, zid);
1895                         }
1896 
1897                         /* recover the global configuration snapshot */
1898                         if (snap_hndl == NULL) {
1899                                 if ((snap_hndl = zonecfg_init_handle())
1900                                     == NULL ||
1901                                     zonecfg_create_snapshot(zone_name)
1902                                     != Z_OK ||
1903                                     zonecfg_get_snapshot_handle(zone_name,
1904                                     snap_hndl) != Z_OK) {
1905                                         zerror(zlogp, B_FALSE, "recovering "
1906                                             "zone configuration handle");
1907                                         goto out;
1908                                 }
1909                         }
1910                 }
1911 
1912                 (void) fdetach(zone_door_path);
1913                 (void) close(doorfd);
1914                 goto top;
1915         }
1916         ret = 0;
1917 out:
1918         (void) close(doorfd);
1919         return (ret);
1920 }
1921 
1922 /*
1923  * Setup the brand's pre and post state change callbacks, as well as the
1924  * query callback, if any of these exist.
1925  */
1926 static int
1927 brand_callback_init(brand_handle_t bh, char *zone_name)
1928 {
1929         (void) strlcpy(pre_statechg_hook, EXEC_PREFIX,
1930             sizeof (pre_statechg_hook));
1931 
1932         if (brand_get_prestatechange(bh, zone_name, zonepath,
1933             pre_statechg_hook + EXEC_LEN,
1934             sizeof (pre_statechg_hook) - EXEC_LEN) != 0)
1935                 return (-1);
1936 
1937         if (strlen(pre_statechg_hook) <= EXEC_LEN)
1938                 pre_statechg_hook[0] = '\0';
1939 
1940         (void) strlcpy(post_statechg_hook, EXEC_PREFIX,
1941             sizeof (post_statechg_hook));
1942 
1943         if (brand_get_poststatechange(bh, zone_name, zonepath,
1944             post_statechg_hook + EXEC_LEN,
1945             sizeof (post_statechg_hook) - EXEC_LEN) != 0)
1946                 return (-1);
1947 
1948         if (strlen(post_statechg_hook) <= EXEC_LEN)
1949                 post_statechg_hook[0] = '\0';
1950 
1951         (void) strlcpy(query_hook, EXEC_PREFIX,
1952             sizeof (query_hook));
1953 
1954         if (brand_get_query(bh, zone_name, zonepath, query_hook + EXEC_LEN,
1955             sizeof (query_hook) - EXEC_LEN) != 0)
1956                 return (-1);
1957 
1958         if (strlen(query_hook) <= EXEC_LEN)
1959                 query_hook[0] = '\0';
1960 
1961         return (0);
1962 }
1963 
1964 int
1965 main(int argc, char *argv[])
1966 {
1967         int opt;
1968         zoneid_t zid;
1969         priv_set_t *privset;
1970         zone_state_t zstate;
1971         char parents_locale[MAXPATHLEN];
1972         brand_handle_t bh;
1973         int err;
1974 
1975         pid_t pid;
1976         sigset_t blockset;
1977         sigset_t block_cld;
1978 
1979         struct {
1980                 sema_t sem;
1981                 int status;
1982                 zlog_t log;
1983         } *shstate;
1984         size_t shstatelen = getpagesize();
1985 
1986         zlog_t errlog;
1987         zlog_t *zlogp;
1988 
1989         int ctfd;
1990 
1991         progname = get_execbasename(argv[0]);
1992 
1993         /*
1994          * Make sure stderr is unbuffered
1995          */
1996         (void) setbuffer(stderr, NULL, 0);
1997 
1998         /*
1999          * Get out of the way of mounted filesystems, since we will daemonize
2000          * soon.
2001          */
2002         (void) chdir("/");
2003 
2004         /*
2005          * Use the default system umask per PSARC 1998/110 rather than
2006          * anything that may have been set by the caller.
2007          */
2008         (void) umask(CMASK);
2009 
2010         /*
2011          * Initially we want to use our parent's locale.
2012          */
2013         (void) setlocale(LC_ALL, "");
2014         (void) textdomain(TEXT_DOMAIN);
2015         (void) strlcpy(parents_locale, setlocale(LC_MESSAGES, NULL),
2016             sizeof (parents_locale));
2017 
2018         /*
2019          * This zlog_t is used for writing to stderr
2020          */
2021         errlog.logfile = stderr;
2022         errlog.buflen = errlog.loglen = 0;
2023         errlog.buf = errlog.log = NULL;
2024         errlog.locale = parents_locale;
2025 
2026         /*
2027          * We start off writing to stderr until we're ready to daemonize.
2028          */
2029         zlogp = &errlog;
2030 
2031         /*
2032          * Process options.
2033          */
2034         while ((opt = getopt(argc, argv, "R:z:")) != EOF) {
2035                 switch (opt) {
2036                 case 'R':
2037                         zonecfg_set_root(optarg);
2038                         break;
2039                 case 'z':
2040                         zone_name = optarg;
2041                         break;
2042                 default:
2043                         usage();
2044                 }
2045         }
2046 
2047         if (zone_name == NULL)
2048                 usage();
2049 
2050         /*
2051          * Because usage() prints directly to stderr, it has gettext()
2052          * wrapping, which depends on the locale.  But since zerror() calls
2053          * localize() which tweaks the locale, it is not safe to call zerror()
2054          * until after the last call to usage().  Fortunately, the last call
2055          * to usage() is just above and the first call to zerror() is just
2056          * below.  Don't mess this up.
2057          */
2058         if (strcmp(zone_name, GLOBAL_ZONENAME) == 0) {
2059                 zerror(zlogp, B_FALSE, "cannot manage the %s zone",
2060                     GLOBAL_ZONENAME);
2061                 return (1);
2062         }
2063 
2064         if (zone_get_id(zone_name, &zid) != 0) {
2065                 zerror(zlogp, B_FALSE, "could not manage %s: %s", zone_name,
2066                     zonecfg_strerror(Z_NO_ZONE));
2067                 return (1);
2068         }
2069 
2070         if ((err = zone_get_state(zone_name, &zstate)) != Z_OK) {
2071                 zerror(zlogp, B_FALSE, "failed to get zone state: %s",
2072                     zonecfg_strerror(err));
2073                 return (1);
2074         }
2075         if (zstate < ZONE_STATE_INCOMPLETE) {
2076                 zerror(zlogp, B_FALSE,
2077                     "cannot manage a zone which is in state '%s'",
2078                     zone_state_str(zstate));
2079                 return (1);
2080         }
2081 
2082         if (zone_get_zonepath(zone_name, zonepath, sizeof (zonepath)) != Z_OK) {
2083                 zerror(zlogp, B_FALSE, "unable to determine zone path");
2084                 return (-1);
2085         }
2086 
2087         if (zonecfg_default_brand(default_brand,
2088             sizeof (default_brand)) != Z_OK) {
2089                 zerror(zlogp, B_FALSE, "unable to determine default brand");
2090                 return (1);
2091         }
2092 
2093         /* Get a handle to the brand info for this zone */
2094         if (zone_get_brand(zone_name, brand_name, sizeof (brand_name))
2095             != Z_OK) {
2096                 zerror(zlogp, B_FALSE, "unable to determine zone brand");
2097                 return (1);
2098         }
2099         zone_isnative = (strcmp(brand_name, NATIVE_BRAND_NAME) == 0);
2100         zone_islabeled = (strcmp(brand_name, LABELED_BRAND_NAME) == 0);
2101 
2102         /*
2103          * In the alternate root environment, the only supported
2104          * operations are mount and unmount.  In this case, just treat
2105          * the zone as native if it is cluster.  Cluster zones can be
2106          * native for the purpose of LU or upgrade, and the cluster
2107          * brand may not exist in the miniroot (such as in net install
2108          * upgrade).
2109          */
2110         if (strcmp(brand_name, CLUSTER_BRAND_NAME) == 0) {
2111                 zone_iscluster = B_TRUE;
2112                 if (zonecfg_in_alt_root()) {
2113                         (void) strlcpy(brand_name, default_brand,
2114                             sizeof (brand_name));
2115                 }
2116         } else {
2117                 zone_iscluster = B_FALSE;
2118         }
2119 
2120         if ((bh = brand_open(brand_name)) == NULL) {
2121                 zerror(zlogp, B_FALSE, "unable to open zone brand");
2122                 return (1);
2123         }
2124 
2125         /* Get state change brand hooks. */
2126         if (brand_callback_init(bh, zone_name) == -1) {
2127                 zerror(zlogp, B_TRUE,
2128                     "failed to initialize brand state change hooks");
2129                 brand_close(bh);
2130                 return (1);
2131         }
2132 
2133         brand_close(bh);
2134 
2135         /*
2136          * Check that we have all privileges.  It would be nice to pare
2137          * this down, but this is at least a first cut.
2138          */
2139         if ((privset = priv_allocset()) == NULL) {
2140                 zerror(zlogp, B_TRUE, "%s failed", "priv_allocset");
2141                 return (1);
2142         }
2143 
2144         if (getppriv(PRIV_EFFECTIVE, privset) != 0) {
2145                 zerror(zlogp, B_TRUE, "%s failed", "getppriv");
2146                 priv_freeset(privset);
2147                 return (1);
2148         }
2149 
2150         if (priv_isfullset(privset) == B_FALSE) {
2151                 zerror(zlogp, B_FALSE, "You lack sufficient privilege to "
2152                     "run this command (all privs required)");
2153                 priv_freeset(privset);
2154                 return (1);
2155         }
2156         priv_freeset(privset);
2157 
2158         if (mkzonedir(zlogp) != 0)
2159                 return (1);
2160 
2161         /*
2162          * Pre-fork: setup shared state
2163          */
2164         if ((shstate = (void *)mmap(NULL, shstatelen,
2165             PROT_READ|PROT_WRITE, MAP_SHARED|MAP_ANON, -1, (off_t)0)) ==
2166             MAP_FAILED) {
2167                 zerror(zlogp, B_TRUE, "%s failed", "mmap");
2168                 return (1);
2169         }
2170         if (sema_init(&shstate->sem, 0, USYNC_PROCESS, NULL) != 0) {
2171                 zerror(zlogp, B_TRUE, "%s failed", "sema_init()");
2172                 (void) munmap((char *)shstate, shstatelen);
2173                 return (1);
2174         }
2175         shstate->log.logfile = NULL;
2176         shstate->log.buflen = shstatelen - sizeof (*shstate);
2177         shstate->log.loglen = shstate->log.buflen;
2178         shstate->log.buf = (char *)shstate + sizeof (*shstate);
2179         shstate->log.log = shstate->log.buf;
2180         shstate->log.locale = parents_locale;
2181         shstate->status = -1;
2182 
2183         /*
2184          * We need a SIGCHLD handler so the sema_wait() below will wake
2185          * up if the child dies without doing a sema_post().
2186          */
2187         (void) sigset(SIGCHLD, sigchld);
2188         /*
2189          * We must mask SIGCHLD until after we've coped with the fork
2190          * sufficiently to deal with it; otherwise we can race and
2191          * receive the signal before pid has been initialized
2192          * (yes, this really happens).
2193          */
2194         (void) sigemptyset(&block_cld);
2195         (void) sigaddset(&block_cld, SIGCHLD);
2196         (void) sigprocmask(SIG_BLOCK, &block_cld, NULL);
2197 
2198         /*
2199          * The parent only needs stderr after the fork, so close other fd's
2200          * that we inherited from zoneadm so that the parent doesn't have those
2201          * open while waiting. The child will close the rest after the fork.
2202          */
2203         closefrom(3);
2204 
2205         if ((ctfd = init_template()) == -1) {
2206                 zerror(zlogp, B_TRUE, "failed to create contract");
2207                 return (1);
2208         }
2209 
2210         /*
2211          * Do not let another thread localize a message while we are forking.
2212          */
2213         (void) mutex_lock(&msglock);
2214         pid = fork();
2215         (void) mutex_unlock(&msglock);
2216 
2217         /*
2218          * In all cases (parent, child, and in the event of an error) we
2219          * don't want to cause creation of contracts on subsequent fork()s.
2220          */
2221         (void) ct_tmpl_clear(ctfd);
2222         (void) close(ctfd);
2223 
2224         if (pid == -1) {
2225                 zerror(zlogp, B_TRUE, "could not fork");
2226                 return (1);
2227 
2228         } else if (pid > 0) { /* parent */
2229                 (void) sigprocmask(SIG_UNBLOCK, &block_cld, NULL);
2230                 /*
2231                  * This marks a window of vulnerability in which we receive
2232                  * the SIGCLD before falling into sema_wait (normally we would
2233                  * get woken up from sema_wait with EINTR upon receipt of
2234                  * SIGCLD).  So we may need to use some other scheme like
2235                  * sema_posting in the sigcld handler.
2236                  * blech
2237                  */
2238                 (void) sema_wait(&shstate->sem);
2239                 (void) sema_destroy(&shstate->sem);
2240                 if (shstate->status != 0)
2241                         (void) waitpid(pid, NULL, WNOHANG);
2242                 /*
2243                  * It's ok if we die with SIGPIPE.  It's not like we could have
2244                  * done anything about it.
2245                  */
2246                 (void) fprintf(stderr, "%s", shstate->log.buf);
2247                 _exit(shstate->status == 0 ? 0 : 1);
2248         }
2249 
2250         /*
2251          * The child charges on.
2252          */
2253         (void) sigset(SIGCHLD, SIG_DFL);
2254         (void) sigprocmask(SIG_UNBLOCK, &block_cld, NULL);
2255 
2256         /*
2257          * SIGPIPE can be delivered if we write to a socket for which the
2258          * peer endpoint is gone.  That can lead to too-early termination
2259          * of zoneadmd, and that's not good eats.
2260          */
2261         (void) sigset(SIGPIPE, SIG_IGN);
2262         /*
2263          * Stop using stderr
2264          */
2265         zlogp = &shstate->log;
2266 
2267         /*
2268          * We don't need stdout/stderr from now on.
2269          */
2270         closefrom(0);
2271 
2272         /*
2273          * Initialize the syslog zlog_t.  This needs to be done after
2274          * the call to closefrom().
2275          */
2276         logsys.buf = logsys.log = NULL;
2277         logsys.buflen = logsys.loglen = 0;
2278         logsys.logfile = NULL;
2279         logsys.locale = DEFAULT_LOCALE;
2280 
2281         openlog("zoneadmd", LOG_PID, LOG_DAEMON);
2282 
2283         /*
2284          * The eventstream is used to publish state changes in the zone
2285          * from the door threads to the console I/O poller.
2286          */
2287         if (eventstream_init() == -1) {
2288                 zerror(zlogp, B_TRUE, "unable to create eventstream");
2289                 goto child_out;
2290         }
2291 
2292         (void) snprintf(zone_door_path, sizeof (zone_door_path),
2293             "%s" ZONE_DOOR_PATH, zonecfg_get_root(), zone_name);
2294 
2295         /*
2296          * See if another zoneadmd is running for this zone.  If not, then we
2297          * can now modify system state.
2298          */
2299         if (make_daemon_exclusive(zlogp) == -1)
2300                 goto child_out;
2301 
2302 
2303         /*
2304          * Create/join a new session; we need to be careful of what we do with
2305          * the console from now on so we don't end up being the session leader
2306          * for the terminal we're going to be handing out.
2307          */
2308         (void) setsid();
2309 
2310         /*
2311          * This thread shouldn't be receiving any signals; in particular,
2312          * SIGCHLD should be received by the thread doing the fork().
2313          */
2314         (void) sigfillset(&blockset);
2315         (void) thr_sigsetmask(SIG_BLOCK, &blockset, NULL);
2316 
2317         /*
2318          * Setup the console device and get ready to serve the console;
2319          * once this has completed, we're ready to let console clients
2320          * make an attempt to connect (they will block until
2321          * serve_console_sock() below gets called, and any pending
2322          * connection is accept()ed).
2323          */
2324         if (!zonecfg_in_alt_root() && init_console(zlogp) < 0)
2325                 goto child_out;
2326 
2327         /*
2328          * Take the lock now, so that when the door server gets going, we
2329          * are guaranteed that it won't take a request until we are sure
2330          * that everything is completely set up.  See the child_out: label
2331          * below to see why this matters.
2332          */
2333         (void) mutex_lock(&lock);
2334 
2335         /* Init semaphore for scratch zones. */
2336         if (sema_init(&scratch_sem, 0, USYNC_THREAD, NULL) == -1) {
2337                 zerror(zlogp, B_TRUE,
2338                     "failed to initialize semaphore for scratch zone");
2339                 goto child_out;
2340         }
2341 
2342         /* open the dladm handle */
2343         if (dladm_open(&dld_handle) != DLADM_STATUS_OK) {
2344                 zerror(zlogp, B_FALSE, "failed to open dladm handle");
2345                 goto child_out;
2346         }
2347 
2348         /*
2349          * Note: door setup must occur *after* the console is setup.
2350          * This is so that as zlogin tests the door to see if zoneadmd
2351          * is ready yet, we know that the console will get serviced
2352          * once door_info() indicates that the door is "up".
2353          */
2354         if (setup_door(zlogp) == -1)
2355                 goto child_out;
2356 
2357         /*
2358          * Things seem OK so far; tell the parent process that we're done
2359          * with setup tasks.  This will cause the parent to exit, signalling
2360          * to zoneadm, zlogin, or whatever forked it that we are ready to
2361          * service requests.
2362          */
2363         shstate->status = 0;
2364         (void) sema_post(&shstate->sem);
2365         (void) munmap((char *)shstate, shstatelen);
2366         shstate = NULL;
2367 
2368         (void) mutex_unlock(&lock);
2369 
2370         /*
2371          * zlogp is now invalid, so reset it to the syslog logger.
2372          */
2373         zlogp = &logsys;
2374 
2375         /*
2376          * Now that we are free of any parents, switch to the default locale.
2377          */
2378         (void) setlocale(LC_ALL, DEFAULT_LOCALE);
2379 
2380         /*
2381          * At this point the setup portion of main() is basically done, so
2382          * we reuse this thread to manage the zone console.  When
2383          * serve_console() has returned, we are past the point of no return
2384          * in the life of this zoneadmd.
2385          */
2386         if (zonecfg_in_alt_root()) {
2387                 /*
2388                  * This is just awful, but mounted scratch zones don't (and
2389                  * can't) have consoles.  We just wait for unmount instead.
2390                  */
2391                 while (sema_wait(&scratch_sem) == EINTR)
2392                         ;
2393         } else {
2394                 serve_console(zlogp);
2395                 assert(in_death_throes);
2396         }
2397 
2398         /*
2399          * This is the next-to-last part of the exit interlock.  Upon calling
2400          * fdetach(), the door will go unreferenced; once any
2401          * outstanding requests (like the door thread doing Z_HALT) are
2402          * done, the door will get an UNREF notification; when it handles
2403          * the UNREF, the door server will cause the exit.  It's possible
2404          * that fdetach() can fail because the file is in use, in which
2405          * case we'll retry the operation.
2406          */
2407         assert(!MUTEX_HELD(&lock));
2408         for (;;) {
2409                 if ((fdetach(zone_door_path) == 0) || (errno != EBUSY))
2410                         break;
2411                 yield();
2412         }
2413 
2414         for (;;)
2415                 (void) pause();
2416 
2417 child_out:
2418         assert(pid == 0);
2419         if (shstate != NULL) {
2420                 shstate->status = -1;
2421                 (void) sema_post(&shstate->sem);
2422                 (void) munmap((char *)shstate, shstatelen);
2423         }
2424 
2425         /*
2426          * This might trigger an unref notification, but if so,
2427          * we are still holding the lock, so our call to exit will
2428          * ultimately win the race and will publish the right exit
2429          * code.
2430          */
2431         if (zone_door != -1) {
2432                 assert(MUTEX_HELD(&lock));
2433                 (void) door_revoke(zone_door);
2434                 (void) fdetach(zone_door_path);
2435         }
2436 
2437         if (dld_handle != NULL)
2438                 dladm_close(dld_handle);
2439 
2440         return (1); /* return from main() forcibly exits an MT process */
2441 }