Print this page
OS-5330 zoneadm mounting an lx or joyent branded zone fails
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Approved by: Jerry Jelinek <jerry.jelinek@joyent.com>
(NOTE: Manual port, because of divergence from SmartOS.)
OS-3831 lxbrand /proc/cmdline should reflect zone boot arguments
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Reviewed by: Patrick Mooney <patrick.mooney@joyent.com>
Approved by: Jerry Jelinek <jerry.jelinek@joyent.com>
Remove most KEBE comments and accompanying unused code or variables/fields.
Merge cleanup from previous six commits
OS-200 need a better mechanism for storing persistent zone_did
OS-2564 zone boot failed: could not start zoneadmd
OS-1763 mount of /etc/svc/volatile failed: Device busy
OS-511 make zonecfg device resource extensible, like the net resource
OS-224 add more zonecfg net properties
Reduce lint
Add zfd.c to zoneadmd's Makefile, a bit more not-yet ifdef-out.
zoneadmd mismerge (we don't support debug yet)
OS-4932 zoneadm boot args not passed to lx init
Reviewed by: Patrick Mooney <patrick.mooney@joyent.com>
OS-4781 would like to be able to add CT_PR_EV_EXIT to fatal event set of current contract
OS-4253 lxbrand ubuntu 15.04 won't boot because /sbin/init is a symlink
OS-3524 in order to support interaction with docker containers, need to be able to connect to stdio for init from GZ
OS-3525 in order to support 'docker logs' need to be able to get stdio from zone to log file
OS-3429 Expose zone's init exit status
OS-3342 dlmgmtd needs to be mindful of lock ordering
OS-2608 dlmgmtd needs to record zone identifiers
OS-3492 zone_free asserts to its destruction when dlmgmtd has fallen
OS-3494 zoneadmd tears down networking too soon when boot fails
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
OS-3077 restarted zoneadmd uses invalid zlogp
OS-3075 zone long boot args aren't passed through
OS-11 rcapd behaves poorly when under extreme load


   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  */
  26 
  27 /*
  28  * zoneadmd manages zones; one zoneadmd process is launched for each
  29  * non-global zone on the system.  This daemon juggles four jobs:
  30  *
  31  * - Implement setup and teardown of the zone "virtual platform": mount and
  32  *   unmount filesystems; create and destroy network interfaces; communicate
  33  *   with devfsadmd to lay out devices for the zone; instantiate the zone
  34  *   console device; configure process runtime attributes such as resource
  35  *   controls, pool bindings, fine-grained privileges.
  36  *
  37  * - Launch the zone's init(1M) process.
  38  *
  39  * - Implement a door server; clients (like zoneadm) connect to the door
  40  *   server and request zone state changes.  The kernel is also a client of
  41  *   this door server.  A request to halt or reboot the zone which originates
  42  *   *inside* the zone results in a door upcall from the kernel into zoneadmd.
  43  *
  44  *   One minor problem is that messages emitted by zoneadmd need to be passed


  51  *   strings, and an exclusion file (zoneadmd.xcl) is used to exclude those
  52  *   strings which do not need to be translated.
  53  *
  54  * - Act as a console server for zlogin -C processes; see comments in zcons.c
  55  *   for more information about the zone console architecture.
  56  *
  57  * DESIGN NOTES
  58  *
  59  * Restart:
  60  *   A chief design constraint of zoneadmd is that it should be restartable in
  61  *   the case that the administrator kills it off, or it suffers a fatal error,
  62  *   without the running zone being impacted; this is akin to being able to
  63  *   reboot the service processor of a server without affecting the OS instance.
  64  */
  65 
  66 #include <sys/param.h>
  67 #include <sys/mman.h>
  68 #include <sys/types.h>
  69 #include <sys/stat.h>
  70 #include <sys/sysmacros.h>

  71 
  72 #include <bsm/adt.h>
  73 #include <bsm/adt_event.h>
  74 
  75 #include <alloca.h>
  76 #include <assert.h>
  77 #include <errno.h>
  78 #include <door.h>
  79 #include <fcntl.h>
  80 #include <locale.h>
  81 #include <signal.h>
  82 #include <stdarg.h>
  83 #include <stdio.h>
  84 #include <stdlib.h>
  85 #include <string.h>
  86 #include <strings.h>
  87 #include <synch.h>
  88 #include <syslog.h>
  89 #include <thread.h>
  90 #include <unistd.h>
  91 #include <wait.h>
  92 #include <limits.h>
  93 #include <zone.h>
  94 #include <libbrand.h>
  95 #include <sys/brand.h>
  96 #include <libcontract.h>
  97 #include <libcontract_priv.h>
  98 #include <sys/brand.h>
  99 #include <sys/contract/process.h>
 100 #include <sys/ctfs.h>
 101 #include <libdladm.h>
 102 #include <sys/dls_mgmt.h>
 103 #include <libscf.h>
 104 
 105 #include <libzonecfg.h>
 106 #include <zonestat_impl.h>
 107 #include "zoneadmd.h"
 108 
 109 static char *progname;
 110 char *zone_name;        /* zone which we are managing */


 111 char pool_name[MAXNAMELEN];
 112 char default_brand[MAXNAMELEN];
 113 char brand_name[MAXNAMELEN];
 114 boolean_t zone_isnative;
 115 boolean_t zone_iscluster;
 116 boolean_t zone_islabeled;
 117 boolean_t shutdown_in_progress;
 118 static zoneid_t zone_id;
 119 dladm_handle_t dld_handle = NULL;
 120 
 121 static char pre_statechg_hook[2 * MAXPATHLEN];
 122 static char post_statechg_hook[2 * MAXPATHLEN];
 123 char query_hook[2 * MAXPATHLEN];
 124 
 125 zlog_t logsys;
 126 
 127 mutex_t lock = DEFAULTMUTEX;    /* to serialize stuff */
 128 mutex_t msglock = DEFAULTMUTEX; /* for calling setlocale() */
 129 
 130 static sema_t scratch_sem;      /* for scratch zones */
 131 
 132 static char     zone_door_path[MAXPATHLEN];
 133 static int      zone_door = -1;
 134 
 135 boolean_t in_death_throes = B_FALSE;    /* daemon is dying */
 136 boolean_t bringup_failure_recovery = B_FALSE; /* ignore certain failures */
 137 
 138 #if !defined(TEXT_DOMAIN)               /* should be defined by cc -D */
 139 #define TEXT_DOMAIN     "SYS_TEST"      /* Use this only if it wasn't */
 140 #endif
 141 
 142 #define DEFAULT_LOCALE  "C"
 143 



 144 static const char *
 145 z_cmd_name(zone_cmd_t zcmd)
 146 {
 147         /* This list needs to match the enum in sys/zone.h */
 148         static const char *zcmdstr[] = {
 149                 "ready", "boot", "forceboot", "reboot", "halt",
 150                 "note_uninstalling", "mount", "forcemount", "unmount",
 151                 "shutdown"
 152         };
 153 
 154         if (zcmd >= sizeof (zcmdstr) / sizeof (*zcmdstr))
 155                 return ("unknown");
 156         else
 157                 return (zcmdstr[(int)zcmd]);
 158 }
 159 
 160 static char *
 161 get_execbasename(char *execfullname)
 162 {
 163         char *last_slash, *execbasename;


 240         bp = &(buf[strlen(buf)]);
 241         if (use_strerror)
 242                 (void) snprintf(bp, sizeof (buf) - (bp - buf), ": %s",
 243                     strerror(saved_errno));
 244         if (zlogp == &logsys) {
 245                 (void) syslog(LOG_ERR, "%s", buf);
 246         } else if (zlogp->logfile != NULL) {
 247                 (void) fprintf(zlogp->logfile, "%s\n", buf);
 248         } else {
 249                 size_t buflen;
 250                 size_t copylen;
 251 
 252                 buflen = snprintf(zlogp->log, zlogp->loglen, "%s\n", buf);
 253                 copylen = MIN(buflen, zlogp->loglen);
 254                 zlogp->log += copylen;
 255                 zlogp->loglen -= copylen;
 256         }
 257 }
 258 
 259 /*
 260  * Emit a warning for any boot arguments which are unrecognized.  Since
 261  * Solaris boot arguments are getopt(3c) compatible (see kernel(1m)), we










 262  * put the arguments into an argv style array, use getopt to process them,
 263  * and put the resultant argument string back into outargs.

 264  *
 265  * During the filtering, we pull out any arguments which are truly "boot"
 266  * arguments, leaving only those which are to be passed intact to the
 267  * progenitor process.  The one we support at the moment is -i, which
 268  * indicates to the kernel which program should be launched as 'init'.
 269  *
 270  * A return of Z_INVAL indicates specifically that the arguments are
 271  * not valid; this is a non-fatal error.  Except for Z_OK, all other return
 272  * values are treated as fatal.
 273  */
 274 static int
 275 filter_bootargs(zlog_t *zlogp, const char *inargs, char *outargs,
 276     char *init_file, char *badarg)
 277 {
 278         int argc = 0, argc_save;
 279         int i;
 280         int err;
 281         char *arg, *lasts, **argv = NULL, **argv_save;
 282         char zonecfg_args[BOOTARGS_MAX];
 283         char scratchargs[BOOTARGS_MAX], *sargs;

 284         char c;
 285 
 286         bzero(outargs, BOOTARGS_MAX);
 287         bzero(badarg, BOOTARGS_MAX);
 288 
 289         /*
 290          * If the user didn't specify transient boot arguments, check
 291          * to see if there were any specified in the zone configuration,
 292          * and use them if applicable.
 293          */
 294         if (inargs == NULL || inargs[0] == '\0')  {
 295                 zone_dochandle_t handle;
 296                 if ((handle = zonecfg_init_handle()) == NULL) {
 297                         zerror(zlogp, B_TRUE,
 298                             "getting zone configuration handle");
 299                         return (Z_BAD_HANDLE);
 300                 }
 301                 err = zonecfg_get_snapshot_handle(zone_name, handle);
 302                 if (err != Z_OK) {
 303                         zerror(zlogp, B_FALSE,
 304                             "invalid configuration snapshot");
 305                         zonecfg_fini_handle(handle);
 306                         return (Z_BAD_HANDLE);
 307                 }


 330                 return (Z_NOMEM);
 331         }
 332 
 333         argv_save = argv;
 334         argc_save = argc;
 335 
 336         (void) strlcpy(scratchargs, inargs, sizeof (scratchargs));
 337         sargs = scratchargs;
 338         i = 0;
 339         while ((arg = strtok_r(sargs, " \t", &lasts)) != NULL) {
 340                 sargs = NULL;
 341                 if ((argv[i] = strdup(arg)) == NULL) {
 342                         err = Z_NOMEM;
 343                         zerror(zlogp, B_FALSE, "memory allocation failed");
 344                         goto done;
 345                 }
 346                 i++;
 347         }
 348 
 349         /*
 350          * We preserve compatibility with the Solaris system boot behavior,
 351          * which allows:
 352          *
 353          *      # reboot kernel/unix -s -m verbose
 354          *
 355          * In this example, kernel/unix tells the booter what file to
 356          * boot.  We don't want reboot in a zone to be gratuitously different,
 357          * so we silently ignore the boot file, if necessary.








 358          */
 359         if (argv[0] == NULL)
 360                 goto done;
 361 
 362         assert(argv[0][0] != ' ');
 363         assert(argv[0][0] != '\t');
 364 
 365         if (argv[0][0] != '-' && argv[0][0] != '\0') {
 366                 argv = &argv[1];
 367                 argc--;
 368         }
 369 
 370         optind = 0;
 371         opterr = 0;
 372         err = Z_OK;
 373         while ((c = getopt(argc, argv, "fi:m:s")) != -1) {
 374                 switch (c) {
 375                 case 'i':
 376                         /*
 377                          * -i is handled by the runtime and is not passed
 378                          * along to userland
 379                          */
 380                         (void) strlcpy(init_file, optarg, MAXPATHLEN);
 381                         break;
 382                 case 'f':
 383                         /* This has already been processed by zoneadm */
 384                         break;
 385                 case 'm':
 386                 case 's':
 387                         /* These pass through unmolested */
 388                         (void) snprintf(outargs, BOOTARGS_MAX,
 389                             "%s -%c %s ", outargs, c, optarg ? optarg : "");



 390                         break;
 391                 case '?':
 392                         /*
 393                          * We warn about unknown arguments but pass them
 394                          * along anyway-- if someone wants to develop their
 395                          * own init replacement, they can pass it whatever
 396                          * args they want.
 397                          */
 398                         err = Z_INVAL;
 399                         (void) snprintf(outargs, BOOTARGS_MAX,
 400                             "%s -%c", outargs, optopt);
 401                         (void) snprintf(badarg, BOOTARGS_MAX,
 402                             "%s -%c", badarg, optopt);
 403                         break;
 404                 }
 405         }
 406 
 407         /*
 408          * For Solaris Zones we warn about and discard non-option arguments.
 409          * Hence 'boot foo bar baz gub' --> 'boot'.  However, to be similar
 410          * to the kernel, we concat up all the other remaining boot args.
 411          * and warn on them as a group.


 412          */
 413         if (optind < argc) {
 414                 err = Z_INVAL;
 415                 while (optind < argc) {
 416                         (void) snprintf(badarg, BOOTARGS_MAX, "%s%s%s",
 417                             badarg, strlen(badarg) > 0 ? " " : "",
 418                             argv[optind]);
 419                         optind++;
 420                 }
 421                 zerror(zlogp, B_FALSE, "WARNING: Unused or invalid boot "
 422                     "arguments `%s'.", badarg);
 423         }
 424 
 425 done:
 426         for (i = 0; i < argc_save; i++) {
 427                 if (argv_save[i] != NULL)
 428                         free(argv_save[i]);
 429         }
 430         free(argv_save);
 431         return (err);
 432 }
 433 
 434 
 435 static int
 436 mkzonedir(zlog_t *zlogp)
 437 {
 438         struct stat st;
 439         /*
 440          * We must create and lock everyone but root out of ZONES_TMPDIR
 441          * since anyone can open any UNIX domain socket, regardless of
 442          * its file system permissions.  Sigh...
 443          */


 520         params.data_ptr = (char *)&cmd;
 521         params.data_size = sizeof (cmd);
 522         params.desc_ptr = NULL;
 523         params.desc_num = 0;
 524         params.rbuf = NULL;
 525         params.rsize = NULL;
 526         (void) door_call(fd, &params);
 527         (void) close(fd);
 528 }
 529 
 530 /*
 531  * Bring a zone up to the pre-boot "ready" stage.  The mount_cmd argument is
 532  * 'true' if this is being invoked as part of the processing for the "mount"
 533  * subcommand.
 534  */
 535 static int
 536 zone_ready(zlog_t *zlogp, zone_mnt_t mount_cmd, int zstate)
 537 {
 538         int err;
 539 
 540         if (brand_prestatechg(zlogp, zstate, Z_READY) != 0)

 541                 return (-1);
 542 
 543         if ((err = zonecfg_create_snapshot(zone_name)) != Z_OK) {
 544                 zerror(zlogp, B_FALSE, "unable to create snapshot: %s",
 545                     zonecfg_strerror(err));
 546                 goto bad;
 547         }
 548 
 549         if ((zone_id = vplat_create(zlogp, mount_cmd)) == -1) {
 550                 if ((err = zonecfg_destroy_snapshot(zone_name)) != Z_OK)
 551                         zerror(zlogp, B_FALSE, "destroying snapshot: %s",
 552                             zonecfg_strerror(err));
 553                 goto bad;
 554         }
 555         if (vplat_bringup(zlogp, mount_cmd, zone_id) != 0) {
 556                 bringup_failure_recovery = B_TRUE;
 557                 (void) vplat_teardown(NULL, (mount_cmd != Z_MNT_BOOT), B_FALSE);
 558                 if ((err = zonecfg_destroy_snapshot(zone_name)) != Z_OK)
 559                         zerror(zlogp, B_FALSE, "destroying snapshot: %s",
 560                             zonecfg_strerror(err));
 561                 goto bad;
 562         }
 563 
 564         if (brand_poststatechg(zlogp, zstate, Z_READY) != 0)

 565                 goto bad;
 566 
 567         return (0);
 568 
 569 bad:
 570         /*
 571          * If something goes wrong, we up the zones's state to the target
 572          * state, READY, and then invoke the hook as if we're halting.
 573          */

 574         (void) brand_poststatechg(zlogp, ZONE_STATE_READY, Z_HALT);
 575         return (-1);
 576 }
 577 
 578 int
 579 init_template(void)
 580 {
 581         int fd;
 582         int err = 0;
 583 
 584         fd = open64(CTFS_ROOT "/process/template", O_RDWR);
 585         if (fd == -1)
 586                 return (-1);
 587 
 588         /*
 589          * For now, zoneadmd doesn't do anything with the contract.
 590          * Deliver no events, don't inherit, and allow it to be orphaned.
 591          */
 592         err |= ct_tmpl_set_critical(fd, 0);
 593         err |= ct_tmpl_set_informative(fd, 0);


 606         zoneid_t        zoneid;
 607         boolean_t       mount_cmd;
 608 } fs_callback_t;
 609 
 610 static int
 611 mount_early_fs(void *data, const char *spec, const char *dir,
 612     const char *fstype, const char *opt)
 613 {
 614         zlog_t *zlogp = ((fs_callback_t *)data)->zlogp;
 615         zoneid_t zoneid = ((fs_callback_t *)data)->zoneid;
 616         boolean_t mount_cmd = ((fs_callback_t *)data)->mount_cmd;
 617         char rootpath[MAXPATHLEN];
 618         pid_t child;
 619         int child_status;
 620         int tmpl_fd;
 621         int rv;
 622         ctid_t ct;
 623 
 624         /* determine the zone rootpath */
 625         if (mount_cmd) {
 626                 char zonepath[MAXPATHLEN];
 627                 char luroot[MAXPATHLEN];
 628 
 629                 if (zone_get_zonepath(zone_name,
 630                     zonepath, sizeof (zonepath)) != Z_OK) {
 631                         zerror(zlogp, B_FALSE, "unable to determine zone path");
 632                         return (-1);
 633                 }
 634 
 635                 (void) snprintf(luroot, sizeof (luroot), "%s/lu", zonepath);
 636                 resolve_lofs(zlogp, luroot, sizeof (luroot));
 637                 (void) strlcpy(rootpath, luroot, sizeof (rootpath));
 638         } else {
 639                 if (zone_get_rootpath(zone_name,
 640                     rootpath, sizeof (rootpath)) != Z_OK) {
 641                         zerror(zlogp, B_FALSE, "unable to determine zone root");
 642                         return (-1);
 643                 }
 644         }
 645 
 646         if ((rv = valid_mount_path(zlogp, rootpath, spec, dir, fstype)) < 0) {
 647                 zerror(zlogp, B_FALSE, "%s%s is not a valid mount point",
 648                     rootpath, dir);
 649                 return (-1);
 650         } else if (rv > 0) {
 651                 /* The mount point path doesn't exist, create it now. */
 652                 if (make_one_dir(zlogp, rootpath, dir,
 653                     DEFAULT_DIR_MODE, DEFAULT_DIR_USER,
 654                     DEFAULT_DIR_GROUP) != 0) {


 669                             "%s%s is not a valid mount point", rootpath, dir);
 670                         return (-1);
 671                 }
 672         }
 673 
 674         if ((tmpl_fd = init_template()) == -1) {
 675                 zerror(zlogp, B_TRUE, "failed to create contract");
 676                 return (-1);
 677         }
 678 
 679         if ((child = fork()) == -1) {
 680                 (void) ct_tmpl_clear(tmpl_fd);
 681                 (void) close(tmpl_fd);
 682                 zerror(zlogp, B_TRUE, "failed to fork");
 683                 return (-1);
 684 
 685         } else if (child == 0) {        /* child */
 686                 char opt_buf[MAX_MNTOPT_STR];
 687                 int optlen = 0;
 688                 int mflag = MS_DATA;


 689 
 690                 (void) ct_tmpl_clear(tmpl_fd);
 691                 /*
 692                  * Even though there are no procs running in the zone, we
 693                  * do this for paranoia's sake.
 694                  */
 695                 (void) closefrom(0);
 696 
 697                 if (zone_enter(zoneid) == -1) {
 698                         _exit(errno);
 699                 }
 700                 if (opt != NULL) {
 701                         /*
 702                          * The mount() system call is incredibly annoying.
 703                          * If options are specified, we need to copy them
 704                          * into a temporary buffer since the mount() system
 705                          * call will overwrite the options string.  It will
 706                          * also fail if the new option string it wants to
 707                          * write is bigger than the one we passed in, so
 708                          * you must pass in a buffer of the maximum possible
 709                          * option string length.  sigh.
 710                          */
 711                         (void) strlcpy(opt_buf, opt, sizeof (opt_buf));
 712                         opt = opt_buf;
 713                         optlen = MAX_MNTOPT_STR;
 714                         mflag = MS_OPTIONSTR;
 715                 }
 716                 if (mount(spec, dir, mflag, fstype, NULL, 0, opt, optlen) != 0)
 717                         _exit(errno);
 718                 _exit(0);















 719         }


 720 
 721         /* parent */
 722         if (contract_latest(&ct) == -1)
 723                 ct = -1;
 724         (void) ct_tmpl_clear(tmpl_fd);
 725         (void) close(tmpl_fd);
 726         if (waitpid(child, &child_status, 0) != child) {
 727                 /* unexpected: we must have been signalled */
 728                 (void) contract_abandon_id(ct);
 729                 return (-1);
 730         }
 731         (void) contract_abandon_id(ct);
 732         if (WEXITSTATUS(child_status) != 0) {
 733                 errno = WEXITSTATUS(child_status);
 734                 zerror(zlogp, B_TRUE, "mount of %s failed", dir);
 735                 return (-1);
 736         }
 737 
 738         return (0);
 739 }
 740 
 741 /*













































































































 742  * If retstr is not NULL, the output of the subproc is returned in the str,
 743  * otherwise it is output using zerror().  Any memory allocated for retstr
 744  * should be freed by the caller.
 745  */
 746 int
 747 do_subproc(zlog_t *zlogp, char *cmdbuf, char **retstr)
 748 {
 749         char buf[1024];         /* arbitrary large amount */
 750         char *inbuf;
 751         FILE *file;
 752         int status;
 753         int rd_cnt;
 754 
 755         if (retstr != NULL) {
 756                 if ((*retstr = malloc(1024)) == NULL) {
 757                         zerror(zlogp, B_FALSE, "out of memory");
 758                         return (-1);
 759                 }
 760                 inbuf = *retstr;
 761                 rd_cnt = 0;
 762         } else {
 763                 inbuf = buf;
 764         }
 765 





 766         file = popen(cmdbuf, "r");
 767         if (file == NULL) {
 768                 zerror(zlogp, B_TRUE, "could not launch: %s", cmdbuf);
 769                 return (-1);
 770         }
 771 
 772         while (fgets(inbuf, 1024, file) != NULL) {
 773                 if (retstr == NULL) {
 774                         if (zlogp != &logsys)
 775                                 zerror(zlogp, B_FALSE, "%s", inbuf);
 776                 } else {
 777                         char *p;
 778 
 779                         rd_cnt += 1024 - 1;
 780                         if ((p = realloc(*retstr, rd_cnt + 1024)) == NULL) {
 781                                 zerror(zlogp, B_FALSE, "out of memory");
 782                                 (void) pclose(file);
 783                                 return (-1);
 784                         }
 785 
 786                         *retstr = p;
 787                         inbuf = *retstr + rd_cnt;
 788                 }
 789         }
 790         status = pclose(file);
 791 
 792         if (WIFSIGNALED(status)) {
 793                 zerror(zlogp, B_FALSE, "%s unexpectedly terminated due to "
 794                     "signal %d", cmdbuf, WTERMSIG(status));
 795                 return (-1);
 796         }
 797         assert(WIFEXITED(status));
 798         if (WEXITSTATUS(status) == ZEXIT_EXEC) {
 799                 zerror(zlogp, B_FALSE, "failed to exec %s", cmdbuf);
 800                 return (-1);
 801         }
 802         return (WEXITSTATUS(status));
 803 }
 804 


























 805 static int
 806 zone_bootup(zlog_t *zlogp, const char *bootargs, int zstate)
 807 {
 808         zoneid_t zoneid;
 809         struct stat st;
 810         char zpath[MAXPATHLEN], initpath[MAXPATHLEN], init_file[MAXPATHLEN];
 811         char nbootargs[BOOTARGS_MAX];
 812         char cmdbuf[MAXPATHLEN];
 813         fs_callback_t cb;
 814         brand_handle_t bh;
 815         zone_iptype_t iptype;
 816         boolean_t links_loaded = B_FALSE;
 817         dladm_status_t status;
 818         char errmsg[DLADM_STRSIZE];
 819         int err;
 820         boolean_t restart_init;

 821 
 822         if (brand_prestatechg(zlogp, zstate, Z_BOOT) != 0)
 823                 return (-1);
 824 
 825         if ((zoneid = getzoneidbyname(zone_name)) == -1) {
 826                 zerror(zlogp, B_TRUE, "unable to get zoneid");
 827                 goto bad;
 828         }
 829 
 830         cb.zlogp = zlogp;
 831         cb.zoneid = zoneid;
 832         cb.mount_cmd = B_FALSE;
 833 
 834         /* Get a handle to the brand info for this zone */
 835         if ((bh = brand_open(brand_name)) == NULL) {
 836                 zerror(zlogp, B_FALSE, "unable to determine zone brand");
 837                 goto bad;
 838         }
 839 
 840         /*
 841          * Get the list of filesystems to mount from the brand
 842          * configuration.  These mounts are done via a thread that will
 843          * enter the zone, so they are done from within the context of the
 844          * zone.
 845          */
 846         if (brand_platform_iter_mounts(bh, mount_early_fs, &cb) != 0) {
 847                 zerror(zlogp, B_FALSE, "unable to mount filesystems");
 848                 brand_close(bh);
 849                 goto bad;
 850         }
 851 
 852         /*
 853          * Get the brand's boot callback if it exists.
 854          */
 855         if (zone_get_zonepath(zone_name, zpath, sizeof (zpath)) != Z_OK) {
 856                 zerror(zlogp, B_FALSE, "unable to determine zone path");
 857                 brand_close(bh);
 858                 goto bad;
 859         }
 860         (void) strcpy(cmdbuf, EXEC_PREFIX);
 861         if (brand_get_boot(bh, zone_name, zpath, cmdbuf + EXEC_LEN,
 862             sizeof (cmdbuf) - EXEC_LEN) != 0) {
 863                 zerror(zlogp, B_FALSE,
 864                     "unable to determine branded zone's boot callback");
 865                 brand_close(bh);
 866                 goto bad;
 867         }
 868 
 869         /* Get the path for this zone's init(1M) (or equivalent) process.  */
 870         if (brand_get_initname(bh, init_file, MAXPATHLEN) != 0) {
 871                 zerror(zlogp, B_FALSE,
 872                     "unable to determine zone's init(1M) location");
 873                 brand_close(bh);
 874                 goto bad;
 875         }
 876 
 877         /* See if this zone's brand should restart init if it dies. */
 878         restart_init = brand_restartinit(bh);
 879 






 880         brand_close(bh);
 881 
 882         err = filter_bootargs(zlogp, bootargs, nbootargs, init_file,
 883             bad_boot_arg);
 884         if (err == Z_INVAL)
 885                 eventstream_write(Z_EVT_ZONE_BADARGS);
 886         else if (err != Z_OK)
 887                 goto bad;
 888 
 889         assert(init_file[0] != '\0');
 890 
 891         /* Try to anticipate possible problems: Make sure init is executable. */
 892         if (zone_get_rootpath(zone_name, zpath, sizeof (zpath)) != Z_OK) {



 893                 zerror(zlogp, B_FALSE, "unable to determine zone root");
 894                 goto bad;
 895         }
 896 
 897         (void) snprintf(initpath, sizeof (initpath), "%s%s", zpath, init_file);
 898 
 899         if (stat(initpath, &st) == -1) {
 900                 zerror(zlogp, B_TRUE, "could not stat %s", initpath);
 901                 goto bad;
 902         }
 903 
 904         if ((st.st_mode & S_IXUSR) == 0) {




 905                 zerror(zlogp, B_FALSE, "%s is not executable", initpath);
 906                 goto bad;
 907         }
 908 
 909         /*
 910          * Exclusive stack zones interact with the dlmgmtd running in the
 911          * global zone.  dladm_zone_boot() tells dlmgmtd that this zone is
 912          * booting, and loads its datalinks from the zone's datalink
 913          * configuration file.
 914          */
 915         if (vplat_get_iptype(zlogp, &iptype) == 0 && iptype == ZS_EXCLUSIVE) {
 916                 status = dladm_zone_boot(dld_handle, zoneid);
 917                 if (status != DLADM_STATUS_OK) {
 918                         zerror(zlogp, B_FALSE, "unable to load zone datalinks: "
 919                             " %s", dladm_status2str(status, errmsg));
 920                         goto bad;
 921                 }
 922                 links_loaded = B_TRUE;
 923         }
 924 
 925         /*
 926          * If there is a brand 'boot' callback, execute it now to give the
 927          * brand one last chance to do any additional setup before the zone
 928          * is booted.
 929          */
 930         if ((strlen(cmdbuf) > EXEC_LEN) &&
 931             (do_subproc(zlogp, cmdbuf, NULL) != Z_OK)) {
 932                 zerror(zlogp, B_FALSE, "%s failed", cmdbuf);
 933                 goto bad;
 934         }
 935 
 936         if (zone_setattr(zoneid, ZONE_ATTR_INITNAME, init_file, 0) == -1) {
 937                 zerror(zlogp, B_TRUE, "could not set zone boot file");
 938                 goto bad;
 939         }
 940 
 941         if (zone_setattr(zoneid, ZONE_ATTR_BOOTARGS, nbootargs, 0) == -1) {
 942                 zerror(zlogp, B_TRUE, "could not set zone boot arguments");
 943                 goto bad;
 944         }
 945 
 946         if (!restart_init && zone_setattr(zoneid, ZONE_ATTR_INITNORESTART,
 947             NULL, 0) == -1) {
 948                 zerror(zlogp, B_TRUE, "could not set zone init-no-restart");
 949                 goto bad;
 950         }
 951 






 952         /*
 953          * Inform zonestatd of a new zone so that it can install a door for
 954          * the zone to contact it.
 955          */
 956         notify_zonestatd(zone_id);
 957 
 958         if (zone_boot(zoneid) == -1) {
 959                 zerror(zlogp, B_TRUE, "unable to boot zone");
 960                 goto bad;
 961         }
 962 
 963         if (brand_poststatechg(zlogp, zstate, Z_BOOT) != 0)
 964                 goto bad;
 965 






 966         return (0);
 967 
 968 bad:
 969         /*
 970          * If something goes wrong, we up the zones's state to the target
 971          * state, RUNNING, and then invoke the hook as if we're halting.
 972          */
 973         (void) brand_poststatechg(zlogp, ZONE_STATE_RUNNING, Z_HALT);
 974         if (links_loaded)
 975                 (void) dladm_zone_halt(dld_handle, zoneid);
 976         return (-1);
 977 }
 978 
 979 static int
 980 zone_halt(zlog_t *zlogp, boolean_t unmount_cmd, boolean_t rebooting, int zstate)
 981 {
 982         int err;
 983 
 984         if (brand_prestatechg(zlogp, zstate, Z_HALT) != 0)

 985                 return (-1);
 986 



 987         if (vplat_teardown(zlogp, unmount_cmd, rebooting) != 0) {
 988                 if (!bringup_failure_recovery)
 989                         zerror(zlogp, B_FALSE, "unable to destroy zone");

 990                 return (-1);
 991         }
 992 







 993         if ((err = zonecfg_destroy_snapshot(zone_name)) != Z_OK)
 994                 zerror(zlogp, B_FALSE, "destroying snapshot: %s",
 995                     zonecfg_strerror(err));
 996 
 997         if (brand_poststatechg(zlogp, zstate, Z_HALT) != 0)
 998                 return (-1);
 999 
1000         return (0);
1001 }
1002 
1003 static int
1004 zone_graceful_shutdown(zlog_t *zlogp)
1005 {
1006         zoneid_t zoneid;
1007         pid_t child;
1008         char cmdbuf[MAXPATHLEN];
1009         brand_handle_t bh = NULL;
1010         char zpath[MAXPATHLEN];
1011         ctid_t ct;
1012         int tmpl_fd;
1013         int child_status;
1014 
1015         if (shutdown_in_progress) {
1016                 zerror(zlogp, B_FALSE, "shutdown already in progress");
1017                 return (-1);
1018         }
1019 
1020         if ((zoneid = getzoneidbyname(zone_name)) == -1) {
1021                 zerror(zlogp, B_TRUE, "unable to get zoneid");
1022                 return (-1);
1023         }
1024 
1025         /* Get a handle to the brand info for this zone */
1026         if ((bh = brand_open(brand_name)) == NULL) {
1027                 zerror(zlogp, B_FALSE, "unable to determine zone brand");
1028                 return (-1);
1029         }
1030 
1031         if (zone_get_zonepath(zone_name, zpath, sizeof (zpath)) != Z_OK) {
1032                 zerror(zlogp, B_FALSE, "unable to determine zone path");
1033                 brand_close(bh);
1034                 return (-1);
1035         }
1036 
1037         /*
1038          * If there is a brand 'shutdown' callback, execute it now to give the
1039          * brand a chance to cleanup any custom configuration.
1040          */
1041         (void) strcpy(cmdbuf, EXEC_PREFIX);
1042         if (brand_get_shutdown(bh, zone_name, zpath, cmdbuf + EXEC_LEN,
1043             sizeof (cmdbuf) - EXEC_LEN) != 0 || strlen(cmdbuf) <= EXEC_LEN) {
1044                 (void) strcat(cmdbuf, SHUTDOWN_DEFAULT);
1045         }
1046         brand_close(bh);
1047 
1048         if ((tmpl_fd = init_template()) == -1) {
1049                 zerror(zlogp, B_TRUE, "failed to create contract");
1050                 return (-1);
1051         }
1052 
1053         if ((child = fork()) == -1) {
1054                 (void) ct_tmpl_clear(tmpl_fd);
1055                 (void) close(tmpl_fd);
1056                 zerror(zlogp, B_TRUE, "failed to fork");
1057                 return (-1);
1058         } else if (child == 0) {
1059                 (void) ct_tmpl_clear(tmpl_fd);
1060                 if (zone_enter(zoneid) == -1) {
1061                         _exit(errno);
1062                 }


1160         }
1161 
1162         event = adt_alloc_event(ah, ADT_zone_state);
1163         if (event == NULL) {
1164                 zerror(zlogp, B_TRUE, gettext("audit failure."));
1165                 (void) adt_end_session(ah);
1166                 return;
1167         }
1168         event->adt_zone_state.zonename = zone_name;
1169         event->adt_zone_state.new_state = new_state;
1170 
1171         if (adt_put_event(event, pass_fail, fail_reason))
1172                 zerror(zlogp, B_TRUE, gettext("audit failure."));
1173 
1174         adt_free_event(event);
1175 
1176         (void) adt_end_session(ah);
1177 }
1178 
1179 /*






























1180  * The main routine for the door server that deals with zone state transitions.
1181  */
1182 /* ARGSUSED */
1183 static void
1184 server(void *cookie, char *args, size_t alen, door_desc_t *dp,
1185     uint_t n_desc)
1186 {
1187         ucred_t *uc = NULL;
1188         const priv_set_t *eset;
1189 
1190         zone_state_t zstate;
1191         zone_cmd_t cmd;

1192         zone_cmd_arg_t *zargp;
1193 
1194         boolean_t kernelcall;
1195 
1196         int rval = -1;
1197         uint64_t uniqid;
1198         zoneid_t zoneid = -1;
1199         zlog_t zlog;
1200         zlog_t *zlogp;
1201         zone_cmd_rval_t *rvalp;
1202         size_t rlen = getpagesize(); /* conservative */
1203         fs_callback_t cb;
1204         brand_handle_t bh;
1205         boolean_t wait_shut = B_FALSE;
1206 
1207         /* LINTED E_BAD_PTR_CAST_ALIGN */
1208         zargp = (zone_cmd_arg_t *)args;
1209 
1210         /*
1211          * When we get the door unref message, we've fdetach'd the door, and


1224 
1225         rvalp = alloca(rlen);
1226         bzero(rvalp, rlen);
1227         zlog.logfile = NULL;
1228         zlog.buflen = zlog.loglen = rlen - sizeof (zone_cmd_rval_t) + 1;
1229         zlog.buf = rvalp->errbuf;
1230         zlog.log = zlog.buf;
1231         /* defer initialization of zlog.locale until after credential check */
1232         zlogp = &zlog;
1233 
1234         if (alen != sizeof (zone_cmd_arg_t)) {
1235                 /*
1236                  * This really shouldn't be happening.
1237                  */
1238                 zerror(&logsys, B_FALSE, "argument size (%d bytes) "
1239                     "unexpected (expected %d bytes)", alen,
1240                     sizeof (zone_cmd_arg_t));
1241                 goto out;
1242         }
1243         cmd = zargp->cmd;

1244 
1245         if (door_ucred(&uc) != 0) {
1246                 zerror(&logsys, B_TRUE, "door_ucred");
1247                 goto out;
1248         }
1249         eset = ucred_getprivset(uc, PRIV_EFFECTIVE);
1250         if (ucred_getzoneid(uc) != GLOBAL_ZONEID ||
1251             (eset != NULL ? !priv_ismember(eset, PRIV_SYS_CONFIG) :
1252             ucred_geteuid(uc) != 0)) {
1253                 zerror(&logsys, B_FALSE, "insufficient privileges");
1254                 goto out;
1255         }
1256 
1257         kernelcall = ucred_getpid(uc) == 0;
1258 
1259         /*
1260          * This is safe because we only use a zlog_t throughout the
1261          * duration of a door call; i.e., by the time the pointer
1262          * might become invalid, the door call would be over.
1263          */


1333                 zstate = ZONE_STATE_INSTALLED;
1334 
1335         switch (zstate) {
1336         case ZONE_STATE_CONFIGURED:
1337         case ZONE_STATE_INCOMPLETE:
1338                 /*
1339                  * Not our area of expertise; we just print a nice message
1340                  * and die off.
1341                  */
1342                 zerror(zlogp, B_FALSE,
1343                     "%s operation is invalid for zones in state '%s'",
1344                     z_cmd_name(cmd), zone_state_str(zstate));
1345                 break;
1346 
1347         case ZONE_STATE_INSTALLED:
1348                 switch (cmd) {
1349                 case Z_READY:
1350                         rval = zone_ready(zlogp, Z_MNT_BOOT, zstate);
1351                         if (rval == 0)
1352                                 eventstream_write(Z_EVT_ZONE_READIED);

1353                         break;
1354                 case Z_BOOT:
1355                 case Z_FORCEBOOT:
1356                         eventstream_write(Z_EVT_ZONE_BOOTING);
1357                         if ((rval = zone_ready(zlogp, Z_MNT_BOOT, zstate))
1358                             == 0) {
1359                                 rval = zone_bootup(zlogp, zargp->bootbuf,
1360                                     zstate);
1361                         }
1362                         audit_put_record(zlogp, uc, rval, "boot");

1363                         if (rval != 0) {
1364                                 bringup_failure_recovery = B_TRUE;
1365                                 (void) zone_halt(zlogp, B_FALSE, B_FALSE,
1366                                     zstate);
1367                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1368                         }
1369                         break;
1370                 case Z_SHUTDOWN:
1371                 case Z_HALT:
1372                         if (kernelcall) /* Invalid; can't happen */
1373                                 abort();
1374                         /*
1375                          * We could have two clients racing to halt this
1376                          * zone; the second client loses, but his request
1377                          * doesn't fail, since the zone is now in the desired
1378                          * state.
1379                          */
1380                         zerror(zlogp, B_FALSE, "zone is already halted");
1381                         rval = 0;
1382                         break;


1465                 break;
1466 
1467         case ZONE_STATE_READY:
1468                 switch (cmd) {
1469                 case Z_READY:
1470                         /*
1471                          * We could have two clients racing to ready this
1472                          * zone; the second client loses, but his request
1473                          * doesn't fail, since the zone is now in the desired
1474                          * state.
1475                          */
1476                         zerror(zlogp, B_FALSE, "zone is already ready");
1477                         rval = 0;
1478                         break;
1479                 case Z_BOOT:
1480                         (void) strlcpy(boot_args, zargp->bootbuf,
1481                             sizeof (boot_args));
1482                         eventstream_write(Z_EVT_ZONE_BOOTING);
1483                         rval = zone_bootup(zlogp, zargp->bootbuf, zstate);
1484                         audit_put_record(zlogp, uc, rval, "boot");

1485                         if (rval != 0) {
1486                                 bringup_failure_recovery = B_TRUE;
1487                                 (void) zone_halt(zlogp, B_FALSE, B_TRUE,
1488                                     zstate);
1489                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1490                         }
1491                         boot_args[0] = '\0';
1492                         break;
1493                 case Z_HALT:
1494                         if (kernelcall) /* Invalid; can't happen */
1495                                 abort();
1496                         if ((rval = zone_halt(zlogp, B_FALSE, B_FALSE, zstate))
1497                             != 0)
1498                                 break;

1499                         eventstream_write(Z_EVT_ZONE_HALTED);
1500                         break;
1501                 case Z_SHUTDOWN:
1502                 case Z_REBOOT:
1503                 case Z_NOTE_UNINSTALLING:
1504                 case Z_MOUNT:
1505                 case Z_UNMOUNT:
1506                         if (kernelcall) /* Invalid; can't happen */
1507                                 abort();
1508                         zerror(zlogp, B_FALSE, "%s operation is invalid "
1509                             "for zones in state '%s'", z_cmd_name(cmd),
1510                             zone_state_str(zstate));
1511                         rval = -1;
1512                         break;
1513                 }
1514                 break;
1515 
1516         case ZONE_STATE_MOUNTED:
1517                 switch (cmd) {
1518                 case Z_UNMOUNT:


1526                         break;
1527                 default:
1528                         if (kernelcall) /* Invalid; can't happen */
1529                                 abort();
1530                         zerror(zlogp, B_FALSE, "%s operation is invalid "
1531                             "for zones in state '%s'", z_cmd_name(cmd),
1532                             zone_state_str(zstate));
1533                         rval = -1;
1534                         break;
1535                 }
1536                 break;
1537 
1538         case ZONE_STATE_RUNNING:
1539         case ZONE_STATE_SHUTTING_DOWN:
1540         case ZONE_STATE_DOWN:
1541                 switch (cmd) {
1542                 case Z_READY:
1543                         if ((rval = zone_halt(zlogp, B_FALSE, B_TRUE, zstate))
1544                             != 0)
1545                                 break;

1546                         if ((rval = zone_ready(zlogp, Z_MNT_BOOT, zstate)) == 0)
1547                                 eventstream_write(Z_EVT_ZONE_READIED);
1548                         else
1549                                 eventstream_write(Z_EVT_ZONE_HALTED);
1550                         break;
1551                 case Z_BOOT:
1552                         /*
1553                          * We could have two clients racing to boot this
1554                          * zone; the second client loses, but his request
1555                          * doesn't fail, since the zone is now in the desired
1556                          * state.
1557                          */
1558                         zerror(zlogp, B_FALSE, "zone is already booted");
1559                         rval = 0;
1560                         break;
1561                 case Z_HALT:





1562                         if ((rval = zone_halt(zlogp, B_FALSE, B_FALSE, zstate))
1563                             != 0)
1564                                 break;
1565                         eventstream_write(Z_EVT_ZONE_HALTED);

1566                         break;
1567                 case Z_REBOOT:
1568                         (void) strlcpy(boot_args, zargp->bootbuf,
1569                             sizeof (boot_args));
1570                         eventstream_write(Z_EVT_ZONE_REBOOTING);
1571                         if ((rval = zone_halt(zlogp, B_FALSE, B_TRUE, zstate))
1572                             != 0) {
1573                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1574                                 boot_args[0] = '\0';
1575                                 break;
1576                         }
1577                         if ((rval = zone_ready(zlogp, Z_MNT_BOOT, zstate))
1578                             != 0) {

1579                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1580                                 boot_args[0] = '\0';
1581                                 break;
1582                         }
1583                         rval = zone_bootup(zlogp, zargp->bootbuf, zstate);
1584                         audit_put_record(zlogp, uc, rval, "reboot");
1585                         if (rval != 0) {
1586                                 (void) zone_halt(zlogp, B_FALSE, B_TRUE,
1587                                     zstate);
1588                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1589                         }
1590                         boot_args[0] = '\0';
1591                         break;
1592                 case Z_SHUTDOWN:
1593                         if ((rval = zone_graceful_shutdown(zlogp)) == 0) {
1594                                 wait_shut = B_TRUE;
1595                         }
1596                         break;
1597                 case Z_NOTE_UNINSTALLING:
1598                 case Z_MOUNT:


1742                          * which will try again (by which time zoneadmd
1743                          * will hopefully have exited).
1744                          */
1745                         goto out;
1746                 }
1747 
1748                 /*
1749                  * If we got this far, there's a fattach(3c)'ed door
1750                  * that belongs to a process that has exited, which can
1751                  * happen if the previous zoneadmd died unexpectedly.
1752                  *
1753                  * Let user know that something is amiss, but that we can
1754                  * recover; if the zone is in the installed state, then don't
1755                  * message, since having a running zoneadmd isn't really
1756                  * expected/needed.  We want to keep occurences of this message
1757                  * limited to times when zoneadmd is picking back up from a
1758                  * zoneadmd that died while the zone was in some non-trivial
1759                  * state.
1760                  */
1761                 if (zstate > ZONE_STATE_INSTALLED) {


1762                         zerror(zlogp, B_FALSE,
1763                             "zone '%s': WARNING: zone is in state '%s', but "
1764                             "zoneadmd does not appear to be available; "
1765                             "restarted zoneadmd to recover.",
1766                             zone_name, zone_state_str(zstate));










1767                 }
1768 















1769                 (void) fdetach(zone_door_path);
1770                 (void) close(doorfd);
1771                 goto top;
1772         }
1773         ret = 0;
1774 out:
1775         (void) close(doorfd);
1776         return (ret);
1777 }
1778 
1779 /*
1780  * Setup the brand's pre and post state change callbacks, as well as the
1781  * query callback, if any of these exist.
1782  */
1783 static int
1784 brand_callback_init(brand_handle_t bh, char *zone_name)
1785 {
1786         char zpath[MAXPATHLEN];
1787 
1788         if (zone_get_zonepath(zone_name, zpath, sizeof (zpath)) != Z_OK)
1789                 return (-1);
1790 
1791         (void) strlcpy(pre_statechg_hook, EXEC_PREFIX,
1792             sizeof (pre_statechg_hook));
1793 
1794         if (brand_get_prestatechange(bh, zone_name, zpath,
1795             pre_statechg_hook + EXEC_LEN,
1796             sizeof (pre_statechg_hook) - EXEC_LEN) != 0)
1797                 return (-1);
1798 
1799         if (strlen(pre_statechg_hook) <= EXEC_LEN)
1800                 pre_statechg_hook[0] = '\0';
1801 
1802         (void) strlcpy(post_statechg_hook, EXEC_PREFIX,
1803             sizeof (post_statechg_hook));
1804 
1805         if (brand_get_poststatechange(bh, zone_name, zpath,
1806             post_statechg_hook + EXEC_LEN,
1807             sizeof (post_statechg_hook) - EXEC_LEN) != 0)
1808                 return (-1);
1809 
1810         if (strlen(post_statechg_hook) <= EXEC_LEN)
1811                 post_statechg_hook[0] = '\0';
1812 
1813         (void) strlcpy(query_hook, EXEC_PREFIX,
1814             sizeof (query_hook));
1815 
1816         if (brand_get_query(bh, zone_name, zpath, query_hook + EXEC_LEN,
1817             sizeof (query_hook) - EXEC_LEN) != 0)
1818                 return (-1);
1819 
1820         if (strlen(query_hook) <= EXEC_LEN)
1821                 query_hook[0] = '\0';
1822 
1823         return (0);
1824 }
1825 
1826 int
1827 main(int argc, char *argv[])
1828 {
1829         int opt;
1830         zoneid_t zid;
1831         priv_set_t *privset;
1832         zone_state_t zstate;
1833         char parents_locale[MAXPATHLEN];
1834         brand_handle_t bh;
1835         int err;
1836 


1924         }
1925 
1926         if (zone_get_id(zone_name, &zid) != 0) {
1927                 zerror(zlogp, B_FALSE, "could not manage %s: %s", zone_name,
1928                     zonecfg_strerror(Z_NO_ZONE));
1929                 return (1);
1930         }
1931 
1932         if ((err = zone_get_state(zone_name, &zstate)) != Z_OK) {
1933                 zerror(zlogp, B_FALSE, "failed to get zone state: %s",
1934                     zonecfg_strerror(err));
1935                 return (1);
1936         }
1937         if (zstate < ZONE_STATE_INCOMPLETE) {
1938                 zerror(zlogp, B_FALSE,
1939                     "cannot manage a zone which is in state '%s'",
1940                     zone_state_str(zstate));
1941                 return (1);
1942         }
1943 





1944         if (zonecfg_default_brand(default_brand,
1945             sizeof (default_brand)) != Z_OK) {
1946                 zerror(zlogp, B_FALSE, "unable to determine default brand");
1947                 return (1);
1948         }
1949 
1950         /* Get a handle to the brand info for this zone */
1951         if (zone_get_brand(zone_name, brand_name, sizeof (brand_name))
1952             != Z_OK) {
1953                 zerror(zlogp, B_FALSE, "unable to determine zone brand");
1954                 return (1);
1955         }
1956         zone_isnative = (strcmp(brand_name, NATIVE_BRAND_NAME) == 0);
1957         zone_islabeled = (strcmp(brand_name, LABELED_BRAND_NAME) == 0);
1958 
1959         /*
1960          * In the alternate root environment, the only supported
1961          * operations are mount and unmount.  In this case, just treat
1962          * the zone as native if it is cluster.  Cluster zones can be
1963          * native for the purpose of LU or upgrade, and the cluster




   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 2016 Joyent, Inc.
  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


  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 #define RSRC_NET        "net"
 149 #define RSRC_DEV        "device"
 150 
 151 static const char *
 152 z_cmd_name(zone_cmd_t zcmd)
 153 {
 154         /* This list needs to match the enum in sys/zone.h */
 155         static const char *zcmdstr[] = {
 156                 "ready", "boot", "forceboot", "reboot", "halt",
 157                 "note_uninstalling", "mount", "forcemount", "unmount",
 158                 "shutdown"
 159         };
 160 
 161         if (zcmd >= sizeof (zcmdstr) / sizeof (*zcmdstr))
 162                 return ("unknown");
 163         else
 164                 return (zcmdstr[(int)zcmd]);
 165 }
 166 
 167 static char *
 168 get_execbasename(char *execfullname)
 169 {
 170         char *last_slash, *execbasename;


 247         bp = &(buf[strlen(buf)]);
 248         if (use_strerror)
 249                 (void) snprintf(bp, sizeof (buf) - (bp - buf), ": %s",
 250                     strerror(saved_errno));
 251         if (zlogp == &logsys) {
 252                 (void) syslog(LOG_ERR, "%s", buf);
 253         } else if (zlogp->logfile != NULL) {
 254                 (void) fprintf(zlogp->logfile, "%s\n", buf);
 255         } else {
 256                 size_t buflen;
 257                 size_t copylen;
 258 
 259                 buflen = snprintf(zlogp->log, zlogp->loglen, "%s\n", buf);
 260                 copylen = MIN(buflen, zlogp->loglen);
 261                 zlogp->log += copylen;
 262                 zlogp->loglen -= copylen;
 263         }
 264 }
 265 
 266 /*
 267  * Append src to dest, modifying dest in the process. Prefix src with
 268  * a space character if dest is a non-empty string.
 269  */
 270 static void
 271 strnappend(char *dest, size_t n, const char *src)
 272 {
 273         (void) snprintf(dest, n, "%s%s%s", dest,
 274             dest[0] == '\0' ? "" : " ", src);
 275 }
 276 
 277 /*
 278  * Since illumos boot arguments are getopt(3c) compatible (see kernel(1m)), we
 279  * put the arguments into an argv style array, use getopt to process them,
 280  * and put the resultant argument string back into outargs. Non-native brands
 281  * may support alternate forms of boot arguments so we must handle that as well.
 282  *
 283  * During the filtering, we pull out any arguments which are truly "boot"
 284  * arguments, leaving only those which are to be passed intact to the
 285  * progenitor process.  The one we support at the moment is -i, which
 286  * indicates to the kernel which program should be launched as 'init'.
 287  *
 288  * Except for Z_OK, all other return values are treated as fatal.


 289  */
 290 static int
 291 filter_bootargs(zlog_t *zlogp, const char *inargs, char *outargs,
 292     char *init_file)
 293 {
 294         int argc = 0, argc_save;
 295         int i;
 296         int err;
 297         char *arg, *lasts, **argv = NULL, **argv_save;
 298         char zonecfg_args[BOOTARGS_MAX];
 299         char scratchargs[BOOTARGS_MAX], *sargs;
 300         char scratchopt[3];
 301         char c;
 302 
 303         bzero(outargs, BOOTARGS_MAX);

 304 
 305         /*
 306          * If the user didn't specify transient boot arguments, check
 307          * to see if there were any specified in the zone configuration,
 308          * and use them if applicable.
 309          */
 310         if (inargs == NULL || inargs[0] == '\0')  {
 311                 zone_dochandle_t handle;
 312                 if ((handle = zonecfg_init_handle()) == NULL) {
 313                         zerror(zlogp, B_TRUE,
 314                             "getting zone configuration handle");
 315                         return (Z_BAD_HANDLE);
 316                 }
 317                 err = zonecfg_get_snapshot_handle(zone_name, handle);
 318                 if (err != Z_OK) {
 319                         zerror(zlogp, B_FALSE,
 320                             "invalid configuration snapshot");
 321                         zonecfg_fini_handle(handle);
 322                         return (Z_BAD_HANDLE);
 323                 }


 346                 return (Z_NOMEM);
 347         }
 348 
 349         argv_save = argv;
 350         argc_save = argc;
 351 
 352         (void) strlcpy(scratchargs, inargs, sizeof (scratchargs));
 353         sargs = scratchargs;
 354         i = 0;
 355         while ((arg = strtok_r(sargs, " \t", &lasts)) != NULL) {
 356                 sargs = NULL;
 357                 if ((argv[i] = strdup(arg)) == NULL) {
 358                         err = Z_NOMEM;
 359                         zerror(zlogp, B_FALSE, "memory allocation failed");
 360                         goto done;
 361                 }
 362                 i++;
 363         }
 364 
 365         /*
 366          * We preserve compatibility with the illumos system boot behavior,
 367          * which allows:
 368          *
 369          *      # reboot kernel/unix -s -m verbose
 370          *
 371          * In this example, kernel/unix tells the booter what file to boot. The
 372          * original intent of this was that we didn't want reboot in a zone to
 373          * be gratuitously different, so we would silently ignore the boot
 374          * file, if necessary. However, this usage is archaic and has never
 375          * been common, since it is impossible to boot a zone onto a different
 376          * kernel. Ignoring the first argument breaks for non-native brands
 377          * which pass boot arguments in a different style. e.g.
 378          *      systemd.log_level=debug
 379          * Thus, for backward compatibility we only ignore the first argument
 380          * if it appears to be in the illumos form and attempting to specify a
 381          * kernel.
 382          */
 383         if (argv[0] == NULL)
 384                 goto done;
 385 
 386         assert(argv[0][0] != ' ');
 387         assert(argv[0][0] != '\t');
 388 
 389         if (strncmp(argv[0], "kernel/", 7) == 0) {
 390                 argv = &argv[1];
 391                 argc--;
 392         }
 393 
 394         optind = 0;
 395         opterr = 0;
 396         err = Z_OK;
 397         while ((c = getopt(argc, argv, "fi:m:s")) != -1) {
 398                 switch (c) {
 399                 case 'i':
 400                         /*
 401                          * -i is handled by the runtime and is not passed
 402                          * along to userland
 403                          */
 404                         (void) strlcpy(init_file, optarg, MAXPATHLEN);
 405                         break;
 406                 case 'f':
 407                         /* This has already been processed by zoneadm */
 408                         break;
 409                 case 'm':
 410                 case 's':
 411                         /* These pass through unmolested */
 412                         (void) snprintf(scratchopt, sizeof (scratchopt),
 413                             "-%c", c);
 414                         strnappend(outargs, BOOTARGS_MAX, scratchopt);
 415                         if (optarg != NULL)
 416                                 strnappend(outargs, BOOTARGS_MAX, optarg);
 417                         break;
 418                 case '?':
 419                         /*
 420                          * If a brand has its own init, we need to pass along
 421                          * whatever the user provides. We use the entire
 422                          * unknown string here so that we correctly handle
 423                          * unknown long options (e.g. --debug).
 424                          */
 425                         strnappend(outargs, BOOTARGS_MAX, argv[optind - 1]);




 426                         break;
 427                 }
 428         }
 429 
 430         /*
 431          * We need to pass along everything else since we don't know what
 432          * the brand's init is expecting. For example, an argument list like:
 433          *   --confdir /foo --debug
 434          * will cause the getopt parsing to stop at '/foo' but we need to pass
 435          * that on, along with the '--debug'. This does mean that we require
 436          * any of our known options (-ifms) to preceed the brand-specific ones.
 437          */


 438         while (optind < argc) {
 439                 strnappend(outargs, BOOTARGS_MAX, argv[optind]);


 440                 optind++;
 441         }



 442 
 443 done:
 444         for (i = 0; i < argc_save; i++) {
 445                 if (argv_save[i] != NULL)
 446                         free(argv_save[i]);
 447         }
 448         free(argv_save);
 449         return (err);
 450 }
 451 
 452 
 453 static int
 454 mkzonedir(zlog_t *zlogp)
 455 {
 456         struct stat st;
 457         /*
 458          * We must create and lock everyone but root out of ZONES_TMPDIR
 459          * since anyone can open any UNIX domain socket, regardless of
 460          * its file system permissions.  Sigh...
 461          */


 538         params.data_ptr = (char *)&cmd;
 539         params.data_size = sizeof (cmd);
 540         params.desc_ptr = NULL;
 541         params.desc_num = 0;
 542         params.rbuf = NULL;
 543         params.rsize = NULL;
 544         (void) door_call(fd, &params);
 545         (void) close(fd);
 546 }
 547 
 548 /*
 549  * Bring a zone up to the pre-boot "ready" stage.  The mount_cmd argument is
 550  * 'true' if this is being invoked as part of the processing for the "mount"
 551  * subcommand.
 552  */
 553 static int
 554 zone_ready(zlog_t *zlogp, zone_mnt_t mount_cmd, int zstate)
 555 {
 556         int err;
 557 
 558         if (!ALT_MOUNT(mount_cmd) &&
 559             brand_prestatechg(zlogp, zstate, Z_READY) != 0)
 560                 return (-1);
 561 
 562         if ((err = zonecfg_create_snapshot(zone_name)) != Z_OK) {
 563                 zerror(zlogp, B_FALSE, "unable to create snapshot: %s",
 564                     zonecfg_strerror(err));
 565                 goto bad;
 566         }
 567 
 568         if ((zone_id = vplat_create(zlogp, mount_cmd)) == -1) {
 569                 if ((err = zonecfg_destroy_snapshot(zone_name)) != Z_OK)
 570                         zerror(zlogp, B_FALSE, "destroying snapshot: %s",
 571                             zonecfg_strerror(err));
 572                 goto bad;
 573         }
 574         if (vplat_bringup(zlogp, mount_cmd, zone_id) != 0) {
 575                 bringup_failure_recovery = B_TRUE;
 576                 (void) vplat_teardown(NULL, (mount_cmd != Z_MNT_BOOT), B_FALSE);
 577                 if ((err = zonecfg_destroy_snapshot(zone_name)) != Z_OK)
 578                         zerror(zlogp, B_FALSE, "destroying snapshot: %s",
 579                             zonecfg_strerror(err));
 580                 goto bad;
 581         }
 582 
 583         if (!ALT_MOUNT(mount_cmd) &&
 584             brand_poststatechg(zlogp, zstate, Z_READY) != 0)
 585                 goto bad;
 586 
 587         return (0);
 588 
 589 bad:
 590         /*
 591          * If something goes wrong, we up the zones's state to the target
 592          * state, READY, and then invoke the hook as if we're halting.
 593          */
 594         if (!ALT_MOUNT(mount_cmd))
 595                 (void) brand_poststatechg(zlogp, ZONE_STATE_READY, Z_HALT);
 596         return (-1);
 597 }
 598 
 599 int
 600 init_template(void)
 601 {
 602         int fd;
 603         int err = 0;
 604 
 605         fd = open64(CTFS_ROOT "/process/template", O_RDWR);
 606         if (fd == -1)
 607                 return (-1);
 608 
 609         /*
 610          * For now, zoneadmd doesn't do anything with the contract.
 611          * Deliver no events, don't inherit, and allow it to be orphaned.
 612          */
 613         err |= ct_tmpl_set_critical(fd, 0);
 614         err |= ct_tmpl_set_informative(fd, 0);


 627         zoneid_t        zoneid;
 628         boolean_t       mount_cmd;
 629 } fs_callback_t;
 630 
 631 static int
 632 mount_early_fs(void *data, const char *spec, const char *dir,
 633     const char *fstype, const char *opt)
 634 {
 635         zlog_t *zlogp = ((fs_callback_t *)data)->zlogp;
 636         zoneid_t zoneid = ((fs_callback_t *)data)->zoneid;
 637         boolean_t mount_cmd = ((fs_callback_t *)data)->mount_cmd;
 638         char rootpath[MAXPATHLEN];
 639         pid_t child;
 640         int child_status;
 641         int tmpl_fd;
 642         int rv;
 643         ctid_t ct;
 644 
 645         /* determine the zone rootpath */
 646         if (mount_cmd) {

 647                 char luroot[MAXPATHLEN];
 648 






 649                 (void) snprintf(luroot, sizeof (luroot), "%s/lu", zonepath);
 650                 resolve_lofs(zlogp, luroot, sizeof (luroot));
 651                 (void) strlcpy(rootpath, luroot, sizeof (rootpath));
 652         } else {
 653                 if (zone_get_rootpath(zone_name,
 654                     rootpath, sizeof (rootpath)) != Z_OK) {
 655                         zerror(zlogp, B_FALSE, "unable to determine zone root");
 656                         return (-1);
 657                 }
 658         }
 659 
 660         if ((rv = valid_mount_path(zlogp, rootpath, spec, dir, fstype)) < 0) {
 661                 zerror(zlogp, B_FALSE, "%s%s is not a valid mount point",
 662                     rootpath, dir);
 663                 return (-1);
 664         } else if (rv > 0) {
 665                 /* The mount point path doesn't exist, create it now. */
 666                 if (make_one_dir(zlogp, rootpath, dir,
 667                     DEFAULT_DIR_MODE, DEFAULT_DIR_USER,
 668                     DEFAULT_DIR_GROUP) != 0) {


 683                             "%s%s is not a valid mount point", rootpath, dir);
 684                         return (-1);
 685                 }
 686         }
 687 
 688         if ((tmpl_fd = init_template()) == -1) {
 689                 zerror(zlogp, B_TRUE, "failed to create contract");
 690                 return (-1);
 691         }
 692 
 693         if ((child = fork()) == -1) {
 694                 (void) ct_tmpl_clear(tmpl_fd);
 695                 (void) close(tmpl_fd);
 696                 zerror(zlogp, B_TRUE, "failed to fork");
 697                 return (-1);
 698 
 699         } else if (child == 0) {        /* child */
 700                 char opt_buf[MAX_MNTOPT_STR];
 701                 int optlen = 0;
 702                 int mflag = MS_DATA;
 703                 int i;
 704                 int ret;
 705 
 706                 (void) ct_tmpl_clear(tmpl_fd);
 707                 /*
 708                  * Even though there are no procs running in the zone, we
 709                  * do this for paranoia's sake.
 710                  */
 711                 (void) closefrom(0);
 712 
 713                 if (zone_enter(zoneid) == -1) {
 714                         _exit(errno);
 715                 }
 716                 if (opt != NULL) {
 717                         /*
 718                          * The mount() system call is incredibly annoying.
 719                          * If options are specified, we need to copy them
 720                          * into a temporary buffer since the mount() system
 721                          * call will overwrite the options string.  It will
 722                          * also fail if the new option string it wants to
 723                          * write is bigger than the one we passed in, so
 724                          * you must pass in a buffer of the maximum possible
 725                          * option string length.  sigh.
 726                          */
 727                         (void) strlcpy(opt_buf, opt, sizeof (opt_buf));
 728                         opt = opt_buf;
 729                         optlen = MAX_MNTOPT_STR;
 730                         mflag = MS_OPTIONSTR;
 731                 }
 732 
 733                 /*
 734                  * There is an obscure race condition which can cause mount
 735                  * to return EBUSY. This happens for example on the mount
 736                  * of the zone's /etc/svc/volatile file system if there is
 737                  * a GZ process running svcs -Z, which will touch the
 738                  * mountpoint, just as we're trying to do the mount. To cope
 739                  * with this, we retry up to 3 times to let this transient
 740                  * process get out of the way.
 741                  */
 742                 for (i = 0; i < 3; i++) {
 743                         ret = 0;
 744                         if (mount(spec, dir, mflag, fstype, NULL, 0, opt,
 745                             optlen) != 0)
 746                                 ret = errno;
 747                         if (ret != EBUSY)
 748                                 break;
 749                         (void) sleep(1);
 750                 }
 751                 _exit(ret);
 752         }
 753 
 754         /* parent */
 755         if (contract_latest(&ct) == -1)
 756                 ct = -1;
 757         (void) ct_tmpl_clear(tmpl_fd);
 758         (void) close(tmpl_fd);
 759         if (waitpid(child, &child_status, 0) != child) {
 760                 /* unexpected: we must have been signalled */
 761                 (void) contract_abandon_id(ct);
 762                 return (-1);
 763         }
 764         (void) contract_abandon_id(ct);
 765         if (WEXITSTATUS(child_status) != 0) {
 766                 errno = WEXITSTATUS(child_status);
 767                 zerror(zlogp, B_TRUE, "mount of %s failed", dir);
 768                 return (-1);
 769         }
 770 
 771         return (0);
 772 }
 773 
 774 /*
 775  * env variable name format
 776  *      _ZONECFG;{resource name};{identifying attr. name};{property name}
 777  */
 778 static void
 779 set_zonecfg_env(char *rsrc, char *attr, char *name, char *val)
 780 {
 781         char *p;
 782         /* Enough for maximal name, rsrc + attr, & slop for ZONECFG & _'s */
 783         char nm[2 * MAXNAMELEN + 32];
 784 
 785         if (attr == NULL)
 786                 (void) snprintf(nm, sizeof (nm), "_ZONECFG_%s_%s", rsrc,
 787                     name);
 788         else
 789                 (void) snprintf(nm, sizeof (nm), "_ZONECFG_%s_%s_%s", rsrc,
 790                     attr, name);
 791 
 792         p = nm;
 793         while ((p = strchr(p, '-')) != NULL)
 794                 *p++ = '_';
 795 
 796         (void) setenv(nm, val, 1);
 797 }
 798 
 799 /*
 800  * Export zonecfg network and device properties into environment for the boot
 801  * and state change hooks.
 802  * If debug is true, export the brand hook debug env. variable as well.
 803  *
 804  * We could export more of the config in the future, as necessary.
 805  */
 806 static int
 807 setup_subproc_env()
 808 {
 809         int res;
 810         zone_dochandle_t handle;
 811         struct zone_nwiftab ntab;
 812         struct zone_devtab dtab;
 813         char net_resources[MAXNAMELEN * 2];
 814         char dev_resources[MAXNAMELEN * 2];
 815 
 816         if ((handle = zonecfg_init_handle()) == NULL)
 817                 exit(Z_NOMEM);
 818 
 819         if ((res = zonecfg_get_handle(zone_name, handle)) != Z_OK)
 820                 goto done;
 821 
 822         if ((res = zonecfg_setnwifent(handle)) != Z_OK)
 823                 goto done;
 824 
 825         while (zonecfg_getnwifent(handle, &ntab) == Z_OK) {
 826                 struct zone_res_attrtab *rap;
 827                 char *phys;
 828 
 829                 phys = ntab.zone_nwif_physical;
 830 
 831                 (void) strlcat(net_resources, phys, sizeof (net_resources));
 832                 (void) strlcat(net_resources, " ", sizeof (net_resources));
 833 
 834                 set_zonecfg_env(RSRC_NET, phys, "physical", phys);
 835 
 836                 set_zonecfg_env(RSRC_NET, phys, "address",
 837                     ntab.zone_nwif_address);
 838                 set_zonecfg_env(RSRC_NET, phys, "allowed-address",
 839                     ntab.zone_nwif_allowed_address);
 840                 set_zonecfg_env(RSRC_NET, phys, "defrouter",
 841                     ntab.zone_nwif_defrouter);
 842                 set_zonecfg_env(RSRC_NET, phys, "global-nic",
 843                     ntab.zone_nwif_gnic);
 844                 set_zonecfg_env(RSRC_NET, phys, "mac-addr", ntab.zone_nwif_mac);
 845                 set_zonecfg_env(RSRC_NET, phys, "vlan-id",
 846                     ntab.zone_nwif_vlan_id);
 847 
 848                 for (rap = ntab.zone_nwif_attrp; rap != NULL;
 849                     rap = rap->zone_res_attr_next)
 850                         set_zonecfg_env(RSRC_NET, phys, rap->zone_res_attr_name,
 851                             rap->zone_res_attr_value);
 852         }
 853 
 854         (void) zonecfg_endnwifent(handle);
 855 
 856         if ((res = zonecfg_setdevent(handle)) != Z_OK)
 857                 goto done;
 858 
 859         while (zonecfg_getdevent(handle, &dtab) == Z_OK) {
 860                 struct zone_res_attrtab *rap;
 861                 char *match;
 862 
 863                 match = dtab.zone_dev_match;
 864 
 865                 (void) strlcat(dev_resources, match, sizeof (dev_resources));
 866                 (void) strlcat(dev_resources, " ", sizeof (dev_resources));
 867 
 868                 for (rap = dtab.zone_dev_attrp; rap != NULL;
 869                     rap = rap->zone_res_attr_next)
 870                         set_zonecfg_env(RSRC_DEV, match,
 871                             rap->zone_res_attr_name, rap->zone_res_attr_value);
 872         }
 873 
 874         (void) zonecfg_enddevent(handle);
 875 
 876         res = Z_OK;
 877 
 878 done:
 879         zonecfg_fini_handle(handle);
 880         return (res);
 881 }
 882 
 883 /*
 884  * If retstr is not NULL, the output of the subproc is returned in the str,
 885  * otherwise it is output using zerror().  Any memory allocated for retstr
 886  * should be freed by the caller.
 887  */
 888 int
 889 do_subproc(zlog_t *zlogp, char *cmdbuf, char **retstr)
 890 {
 891         char buf[1024];         /* arbitrary large amount */
 892         char *inbuf;
 893         FILE *file;
 894         int status;
 895         int rd_cnt;
 896 
 897         if (retstr != NULL) {
 898                 if ((*retstr = malloc(1024)) == NULL) {
 899                         zerror(zlogp, B_FALSE, "out of memory");
 900                         return (-1);
 901                 }
 902                 inbuf = *retstr;
 903                 rd_cnt = 0;
 904         } else {
 905                 inbuf = buf;
 906         }
 907 
 908         if (setup_subproc_env() != Z_OK) {
 909                 zerror(zlogp, B_FALSE, "failed to setup environment");
 910                 return (-1);
 911         }
 912 
 913         file = popen(cmdbuf, "r");
 914         if (file == NULL) {
 915                 zerror(zlogp, B_TRUE, "could not launch: %s", cmdbuf);
 916                 return (-1);
 917         }
 918 
 919         while (fgets(inbuf, 1024, file) != NULL) {
 920                 if (retstr == NULL) {
 921                         if (zlogp != &logsys)
 922                                 zerror(zlogp, B_FALSE, "%s", inbuf);
 923                 } else {
 924                         char *p;
 925 
 926                         rd_cnt += 1024 - 1;
 927                         if ((p = realloc(*retstr, rd_cnt + 1024)) == NULL) {
 928                                 zerror(zlogp, B_FALSE, "out of memory");
 929                                 (void) pclose(file);
 930                                 return (-1);
 931                         }
 932 
 933                         *retstr = p;
 934                         inbuf = *retstr + rd_cnt;
 935                 }
 936         }
 937         status = pclose(file);
 938 
 939         if (WIFSIGNALED(status)) {
 940                 zerror(zlogp, B_FALSE, "%s unexpectedly terminated due to "
 941                     "signal %d", cmdbuf, WTERMSIG(status));
 942                 return (-1);
 943         }
 944         assert(WIFEXITED(status));
 945         if (WEXITSTATUS(status) == ZEXIT_EXEC) {
 946                 zerror(zlogp, B_FALSE, "failed to exec %s", cmdbuf);
 947                 return (-1);
 948         }
 949         return (WEXITSTATUS(status));
 950 }
 951 
 952 /*
 953  * Get the app-svc-dependent flag for this zone's init process. This is a
 954  * zone-specific attr which controls the type of contract we create for the
 955  * zone's init. When true, the contract will include CT_PR_EV_EXIT in the fatal
 956  * set, so that when any service which is in the same contract exits, the init
 957  * application will be terminated.
 958  *
 959  * We use the global "snap_hndl", so no parameters get passed here.
 960  */
 961 static boolean_t
 962 is_app_svc_dep(void)
 963 {
 964         struct zone_attrtab a;
 965 
 966         bzero(&a, sizeof (a));
 967         (void) strlcpy(a.zone_attr_name, "app-svc-dependent",
 968             sizeof (a.zone_attr_name));
 969 
 970         if (zonecfg_lookup_attr(snap_hndl, &a) == Z_OK &&
 971             strcmp(a.zone_attr_value, "true") == 0) {
 972                 return (B_TRUE);
 973         }
 974 
 975         return (B_FALSE);
 976 }
 977 
 978 static int
 979 zone_bootup(zlog_t *zlogp, const char *bootargs, int zstate)
 980 {
 981         zoneid_t zoneid;
 982         struct stat st;
 983         char rpath[MAXPATHLEN], initpath[MAXPATHLEN], init_file[MAXPATHLEN];
 984         char nbootargs[BOOTARGS_MAX];
 985         char cmdbuf[MAXPATHLEN];
 986         fs_callback_t cb;
 987         brand_handle_t bh;
 988         zone_iptype_t iptype;

 989         dladm_status_t status;
 990         char errmsg[DLADM_STRSIZE];
 991         int err;
 992         boolean_t restart_init;
 993         boolean_t app_svc_dep;
 994 
 995         if (brand_prestatechg(zlogp, zstate, Z_BOOT) != 0)
 996                 return (-1);
 997 
 998         if ((zoneid = getzoneidbyname(zone_name)) == -1) {
 999                 zerror(zlogp, B_TRUE, "unable to get zoneid");
1000                 goto bad;
1001         }
1002 
1003         cb.zlogp = zlogp;
1004         cb.zoneid = zoneid;
1005         cb.mount_cmd = B_FALSE;
1006 
1007         /* Get a handle to the brand info for this zone */
1008         if ((bh = brand_open(brand_name)) == NULL) {
1009                 zerror(zlogp, B_FALSE, "unable to determine zone brand");
1010                 goto bad;
1011         }
1012 
1013         /*
1014          * Get the list of filesystems to mount from the brand
1015          * configuration.  These mounts are done via a thread that will
1016          * enter the zone, so they are done from within the context of the
1017          * zone.
1018          */
1019         if (brand_platform_iter_mounts(bh, mount_early_fs, &cb) != 0) {
1020                 zerror(zlogp, B_FALSE, "unable to mount filesystems");
1021                 brand_close(bh);
1022                 goto bad;
1023         }
1024 
1025         /*
1026          * Get the brand's boot callback if it exists.
1027          */





1028         (void) strcpy(cmdbuf, EXEC_PREFIX);
1029         if (brand_get_boot(bh, zone_name, zonepath, cmdbuf + EXEC_LEN,
1030             sizeof (cmdbuf) - EXEC_LEN) != 0) {
1031                 zerror(zlogp, B_FALSE,
1032                     "unable to determine branded zone's boot callback");
1033                 brand_close(bh);
1034                 goto bad;
1035         }
1036 
1037         /* Get the path for this zone's init(1M) (or equivalent) process.  */
1038         if (brand_get_initname(bh, init_file, MAXPATHLEN) != 0) {
1039                 zerror(zlogp, B_FALSE,
1040                     "unable to determine zone's init(1M) location");
1041                 brand_close(bh);
1042                 goto bad;
1043         }
1044 
1045         /* See if this zone's brand should restart init if it dies. */
1046         restart_init = brand_restartinit(bh);
1047 
1048         /*
1049          * See if we need to setup contract dependencies between the zone's
1050          * primary application and any of its services.
1051          */
1052         app_svc_dep = is_app_svc_dep();
1053 
1054         brand_close(bh);
1055 
1056         err = filter_bootargs(zlogp, bootargs, nbootargs, init_file);
1057         if (err != Z_OK)



1058                 goto bad;
1059 
1060         assert(init_file[0] != '\0');
1061 
1062         /*
1063          * Try to anticipate possible problems: If possible, make sure init is
1064          * executable.
1065          */
1066         if (zone_get_rootpath(zone_name, rpath, sizeof (rpath)) != Z_OK) {
1067                 zerror(zlogp, B_FALSE, "unable to determine zone root");
1068                 goto bad;
1069         }
1070 
1071         (void) snprintf(initpath, sizeof (initpath), "%s%s", rpath, init_file);
1072 
1073         if (lstat(initpath, &st) == -1) {
1074                 zerror(zlogp, B_TRUE, "could not stat %s", initpath);
1075                 goto bad;
1076         }
1077 
1078         /*
1079          * If a symlink, we'll have to wait and resolve when we boot,
1080          * otherwise check the executable bits now.
1081          */
1082         if ((st.st_mode & S_IFMT) != S_IFLNK && (st.st_mode & S_IXUSR) == 0) {
1083                 zerror(zlogp, B_FALSE, "%s is not executable", initpath);
1084                 goto bad;
1085         }
1086 
1087         /*
1088          * Exclusive stack zones interact with the dlmgmtd running in the
1089          * global zone.  dladm_zone_boot() tells dlmgmtd that this zone is
1090          * booting, and loads its datalinks from the zone's datalink
1091          * configuration file.
1092          */
1093         if (vplat_get_iptype(zlogp, &iptype) == 0 && iptype == ZS_EXCLUSIVE) {
1094                 status = dladm_zone_boot(dld_handle, zoneid);
1095                 if (status != DLADM_STATUS_OK) {
1096                         zerror(zlogp, B_FALSE, "unable to load zone datalinks: "
1097                             " %s", dladm_status2str(status, errmsg));
1098                         goto bad;
1099                 }

1100         }
1101 
1102         /*
1103          * If there is a brand 'boot' callback, execute it now to give the
1104          * brand one last chance to do any additional setup before the zone
1105          * is booted.
1106          */
1107         if ((strlen(cmdbuf) > EXEC_LEN) &&
1108             (do_subproc(zlogp, cmdbuf, NULL) != Z_OK)) {
1109                 zerror(zlogp, B_FALSE, "%s failed", cmdbuf);
1110                 goto bad;
1111         }
1112 
1113         if (zone_setattr(zoneid, ZONE_ATTR_INITNAME, init_file, 0) == -1) {
1114                 zerror(zlogp, B_TRUE, "could not set zone boot file");
1115                 goto bad;
1116         }
1117 
1118         if (zone_setattr(zoneid, ZONE_ATTR_BOOTARGS, nbootargs, 0) == -1) {
1119                 zerror(zlogp, B_TRUE, "could not set zone boot arguments");
1120                 goto bad;
1121         }
1122 
1123         if (!restart_init && zone_setattr(zoneid, ZONE_ATTR_INITNORESTART,
1124             NULL, 0) == -1) {
1125                 zerror(zlogp, B_TRUE, "could not set zone init-no-restart");
1126                 goto bad;
1127         }
1128 
1129         if (app_svc_dep && zone_setattr(zoneid, ZONE_ATTR_APP_SVC_CT,
1130             (void *)B_TRUE, sizeof (boolean_t)) == -1) {
1131                 zerror(zlogp, B_TRUE, "could not set zone app-die");
1132                 goto bad;
1133         }
1134 
1135         /*
1136          * Inform zonestatd of a new zone so that it can install a door for
1137          * the zone to contact it.
1138          */
1139         notify_zonestatd(zone_id);
1140 
1141         if (zone_boot(zoneid) == -1) {
1142                 zerror(zlogp, B_TRUE, "unable to boot zone");
1143                 goto bad;
1144         }
1145 
1146         if (brand_poststatechg(zlogp, zstate, Z_BOOT) != 0)
1147                 goto bad;
1148 
1149         /* Startup a thread to perform zfd logging/tty svc for the zone. */
1150         create_log_thread(zlogp, zone_id);
1151 
1152         /* Startup a thread to perform memory capping for the zone. */
1153         create_mcap_thread(zlogp, zone_id);
1154 
1155         return (0);
1156 
1157 bad:
1158         /*
1159          * If something goes wrong, we up the zones's state to the target
1160          * state, RUNNING, and then invoke the hook as if we're halting.
1161          */
1162         (void) brand_poststatechg(zlogp, ZONE_STATE_RUNNING, Z_HALT);
1163 

1164         return (-1);
1165 }
1166 
1167 static int
1168 zone_halt(zlog_t *zlogp, boolean_t unmount_cmd, boolean_t rebooting, int zstate)
1169 {
1170         int err;
1171 
1172         if (unmount_cmd == B_FALSE &&
1173             brand_prestatechg(zlogp, zstate, Z_HALT) != 0)
1174                 return (-1);
1175 
1176         /* Shutting down, stop the memcap thread */
1177         destroy_mcap_thread();
1178 
1179         if (vplat_teardown(zlogp, unmount_cmd, rebooting) != 0) {
1180                 if (!bringup_failure_recovery)
1181                         zerror(zlogp, B_FALSE, "unable to destroy zone");
1182                 destroy_log_thread();
1183                 return (-1);
1184         }
1185 
1186         /* Shut down is done, stop the log thread */
1187         destroy_log_thread();
1188 
1189         if (unmount_cmd == B_FALSE &&
1190             brand_poststatechg(zlogp, zstate, Z_HALT) != 0)
1191                 return (-1);
1192 
1193         if ((err = zonecfg_destroy_snapshot(zone_name)) != Z_OK)
1194                 zerror(zlogp, B_FALSE, "destroying snapshot: %s",
1195                     zonecfg_strerror(err));
1196 



1197         return (0);
1198 }
1199 
1200 static int
1201 zone_graceful_shutdown(zlog_t *zlogp)
1202 {
1203         zoneid_t zoneid;
1204         pid_t child;
1205         char cmdbuf[MAXPATHLEN];
1206         brand_handle_t bh = NULL;

1207         ctid_t ct;
1208         int tmpl_fd;
1209         int child_status;
1210 
1211         if (shutdown_in_progress) {
1212                 zerror(zlogp, B_FALSE, "shutdown already in progress");
1213                 return (-1);
1214         }
1215 
1216         if ((zoneid = getzoneidbyname(zone_name)) == -1) {
1217                 zerror(zlogp, B_TRUE, "unable to get zoneid");
1218                 return (-1);
1219         }
1220 
1221         /* Get a handle to the brand info for this zone */
1222         if ((bh = brand_open(brand_name)) == NULL) {
1223                 zerror(zlogp, B_FALSE, "unable to determine zone brand");
1224                 return (-1);
1225         }
1226 






1227         /*
1228          * If there is a brand 'shutdown' callback, execute it now to give the
1229          * brand a chance to cleanup any custom configuration.
1230          */
1231         (void) strcpy(cmdbuf, EXEC_PREFIX);
1232         if (brand_get_shutdown(bh, zone_name, zonepath, cmdbuf + EXEC_LEN,
1233             sizeof (cmdbuf) - EXEC_LEN) != 0 || strlen(cmdbuf) <= EXEC_LEN) {
1234                 (void) strcat(cmdbuf, SHUTDOWN_DEFAULT);
1235         }
1236         brand_close(bh);
1237 
1238         if ((tmpl_fd = init_template()) == -1) {
1239                 zerror(zlogp, B_TRUE, "failed to create contract");
1240                 return (-1);
1241         }
1242 
1243         if ((child = fork()) == -1) {
1244                 (void) ct_tmpl_clear(tmpl_fd);
1245                 (void) close(tmpl_fd);
1246                 zerror(zlogp, B_TRUE, "failed to fork");
1247                 return (-1);
1248         } else if (child == 0) {
1249                 (void) ct_tmpl_clear(tmpl_fd);
1250                 if (zone_enter(zoneid) == -1) {
1251                         _exit(errno);
1252                 }


1350         }
1351 
1352         event = adt_alloc_event(ah, ADT_zone_state);
1353         if (event == NULL) {
1354                 zerror(zlogp, B_TRUE, gettext("audit failure."));
1355                 (void) adt_end_session(ah);
1356                 return;
1357         }
1358         event->adt_zone_state.zonename = zone_name;
1359         event->adt_zone_state.new_state = new_state;
1360 
1361         if (adt_put_event(event, pass_fail, fail_reason))
1362                 zerror(zlogp, B_TRUE, gettext("audit failure."));
1363 
1364         adt_free_event(event);
1365 
1366         (void) adt_end_session(ah);
1367 }
1368 
1369 /*
1370  * Log the exit time and status of the zone's init process into
1371  * {zonepath}/lastexited. If the zone shutdown normally, the exit status will
1372  * be -1, otherwise it will be the exit status as described in wait.3c.
1373  * If the zone is configured to restart init, then nothing will be logged if
1374  * init exits unexpectedly (the kernel will never upcall in this case).
1375  */
1376 static void
1377 log_init_exit(int status)
1378 {
1379         char p[MAXPATHLEN];
1380         char buf[128];
1381         struct timeval t;
1382         int fd;
1383 
1384         if (snprintf(p, sizeof (p), "%s/lastexited", zonepath) > sizeof (p))
1385                 return;
1386         if (gettimeofday(&t, NULL) != 0)
1387                 return;
1388         if (snprintf(buf, sizeof (buf), "%ld.%ld %d\n", t.tv_sec, t.tv_usec,
1389             status) > sizeof (buf))
1390                 return;
1391         if ((fd = open(p, O_WRONLY | O_CREAT | O_TRUNC, 0644)) < 0)
1392                 return;
1393 
1394         (void) write(fd, buf, strlen(buf));
1395 
1396         (void) close(fd);
1397 }
1398 
1399 /*
1400  * The main routine for the door server that deals with zone state transitions.
1401  */
1402 /* ARGSUSED */
1403 static void
1404 server(void *cookie, char *args, size_t alen, door_desc_t *dp,
1405     uint_t n_desc)
1406 {
1407         ucred_t *uc = NULL;
1408         const priv_set_t *eset;
1409 
1410         zone_state_t zstate;
1411         zone_cmd_t cmd;
1412         int init_status;
1413         zone_cmd_arg_t *zargp;
1414 
1415         boolean_t kernelcall;
1416 
1417         int rval = -1;
1418         uint64_t uniqid;
1419         zoneid_t zoneid = -1;
1420         zlog_t zlog;
1421         zlog_t *zlogp;
1422         zone_cmd_rval_t *rvalp;
1423         size_t rlen = getpagesize(); /* conservative */
1424         fs_callback_t cb;
1425         brand_handle_t bh;
1426         boolean_t wait_shut = B_FALSE;
1427 
1428         /* LINTED E_BAD_PTR_CAST_ALIGN */
1429         zargp = (zone_cmd_arg_t *)args;
1430 
1431         /*
1432          * When we get the door unref message, we've fdetach'd the door, and


1445 
1446         rvalp = alloca(rlen);
1447         bzero(rvalp, rlen);
1448         zlog.logfile = NULL;
1449         zlog.buflen = zlog.loglen = rlen - sizeof (zone_cmd_rval_t) + 1;
1450         zlog.buf = rvalp->errbuf;
1451         zlog.log = zlog.buf;
1452         /* defer initialization of zlog.locale until after credential check */
1453         zlogp = &zlog;
1454 
1455         if (alen != sizeof (zone_cmd_arg_t)) {
1456                 /*
1457                  * This really shouldn't be happening.
1458                  */
1459                 zerror(&logsys, B_FALSE, "argument size (%d bytes) "
1460                     "unexpected (expected %d bytes)", alen,
1461                     sizeof (zone_cmd_arg_t));
1462                 goto out;
1463         }
1464         cmd = zargp->cmd;
1465         init_status = zargp->status;
1466 
1467         if (door_ucred(&uc) != 0) {
1468                 zerror(&logsys, B_TRUE, "door_ucred");
1469                 goto out;
1470         }
1471         eset = ucred_getprivset(uc, PRIV_EFFECTIVE);
1472         if (ucred_getzoneid(uc) != GLOBAL_ZONEID ||
1473             (eset != NULL ? !priv_ismember(eset, PRIV_SYS_CONFIG) :
1474             ucred_geteuid(uc) != 0)) {
1475                 zerror(&logsys, B_FALSE, "insufficient privileges");
1476                 goto out;
1477         }
1478 
1479         kernelcall = ucred_getpid(uc) == 0;
1480 
1481         /*
1482          * This is safe because we only use a zlog_t throughout the
1483          * duration of a door call; i.e., by the time the pointer
1484          * might become invalid, the door call would be over.
1485          */


1555                 zstate = ZONE_STATE_INSTALLED;
1556 
1557         switch (zstate) {
1558         case ZONE_STATE_CONFIGURED:
1559         case ZONE_STATE_INCOMPLETE:
1560                 /*
1561                  * Not our area of expertise; we just print a nice message
1562                  * and die off.
1563                  */
1564                 zerror(zlogp, B_FALSE,
1565                     "%s operation is invalid for zones in state '%s'",
1566                     z_cmd_name(cmd), zone_state_str(zstate));
1567                 break;
1568 
1569         case ZONE_STATE_INSTALLED:
1570                 switch (cmd) {
1571                 case Z_READY:
1572                         rval = zone_ready(zlogp, Z_MNT_BOOT, zstate);
1573                         if (rval == 0)
1574                                 eventstream_write(Z_EVT_ZONE_READIED);
1575                         zcons_statechanged();
1576                         break;
1577                 case Z_BOOT:
1578                 case Z_FORCEBOOT:
1579                         eventstream_write(Z_EVT_ZONE_BOOTING);
1580                         if ((rval = zone_ready(zlogp, Z_MNT_BOOT, zstate))
1581                             == 0) {
1582                                 rval = zone_bootup(zlogp, zargp->bootbuf,
1583                                     zstate);
1584                         }
1585                         audit_put_record(zlogp, uc, rval, "boot");
1586                         zcons_statechanged();
1587                         if (rval != 0) {
1588                                 bringup_failure_recovery = B_TRUE;
1589                                 (void) zone_halt(zlogp, B_FALSE, B_FALSE,
1590                                     zstate);
1591                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1592                         }
1593                         break;
1594                 case Z_SHUTDOWN:
1595                 case Z_HALT:
1596                         if (kernelcall) /* Invalid; can't happen */
1597                                 abort();
1598                         /*
1599                          * We could have two clients racing to halt this
1600                          * zone; the second client loses, but his request
1601                          * doesn't fail, since the zone is now in the desired
1602                          * state.
1603                          */
1604                         zerror(zlogp, B_FALSE, "zone is already halted");
1605                         rval = 0;
1606                         break;


1689                 break;
1690 
1691         case ZONE_STATE_READY:
1692                 switch (cmd) {
1693                 case Z_READY:
1694                         /*
1695                          * We could have two clients racing to ready this
1696                          * zone; the second client loses, but his request
1697                          * doesn't fail, since the zone is now in the desired
1698                          * state.
1699                          */
1700                         zerror(zlogp, B_FALSE, "zone is already ready");
1701                         rval = 0;
1702                         break;
1703                 case Z_BOOT:
1704                         (void) strlcpy(boot_args, zargp->bootbuf,
1705                             sizeof (boot_args));
1706                         eventstream_write(Z_EVT_ZONE_BOOTING);
1707                         rval = zone_bootup(zlogp, zargp->bootbuf, zstate);
1708                         audit_put_record(zlogp, uc, rval, "boot");
1709                         zcons_statechanged();
1710                         if (rval != 0) {
1711                                 bringup_failure_recovery = B_TRUE;
1712                                 (void) zone_halt(zlogp, B_FALSE, B_TRUE,
1713                                     zstate);
1714                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1715                         }
1716                         boot_args[0] = '\0';
1717                         break;
1718                 case Z_HALT:
1719                         if (kernelcall) /* Invalid; can't happen */
1720                                 abort();
1721                         if ((rval = zone_halt(zlogp, B_FALSE, B_FALSE, zstate))
1722                             != 0)
1723                                 break;
1724                         zcons_statechanged();
1725                         eventstream_write(Z_EVT_ZONE_HALTED);
1726                         break;
1727                 case Z_SHUTDOWN:
1728                 case Z_REBOOT:
1729                 case Z_NOTE_UNINSTALLING:
1730                 case Z_MOUNT:
1731                 case Z_UNMOUNT:
1732                         if (kernelcall) /* Invalid; can't happen */
1733                                 abort();
1734                         zerror(zlogp, B_FALSE, "%s operation is invalid "
1735                             "for zones in state '%s'", z_cmd_name(cmd),
1736                             zone_state_str(zstate));
1737                         rval = -1;
1738                         break;
1739                 }
1740                 break;
1741 
1742         case ZONE_STATE_MOUNTED:
1743                 switch (cmd) {
1744                 case Z_UNMOUNT:


1752                         break;
1753                 default:
1754                         if (kernelcall) /* Invalid; can't happen */
1755                                 abort();
1756                         zerror(zlogp, B_FALSE, "%s operation is invalid "
1757                             "for zones in state '%s'", z_cmd_name(cmd),
1758                             zone_state_str(zstate));
1759                         rval = -1;
1760                         break;
1761                 }
1762                 break;
1763 
1764         case ZONE_STATE_RUNNING:
1765         case ZONE_STATE_SHUTTING_DOWN:
1766         case ZONE_STATE_DOWN:
1767                 switch (cmd) {
1768                 case Z_READY:
1769                         if ((rval = zone_halt(zlogp, B_FALSE, B_TRUE, zstate))
1770                             != 0)
1771                                 break;
1772                         zcons_statechanged();
1773                         if ((rval = zone_ready(zlogp, Z_MNT_BOOT, zstate)) == 0)
1774                                 eventstream_write(Z_EVT_ZONE_READIED);
1775                         else
1776                                 eventstream_write(Z_EVT_ZONE_HALTED);
1777                         break;
1778                 case Z_BOOT:
1779                         /*
1780                          * We could have two clients racing to boot this
1781                          * zone; the second client loses, but his request
1782                          * doesn't fail, since the zone is now in the desired
1783                          * state.
1784                          */
1785                         zerror(zlogp, B_FALSE, "zone is already booted");
1786                         rval = 0;
1787                         break;
1788                 case Z_HALT:
1789                         if (kernelcall) {
1790                                 log_init_exit(init_status);
1791                         } else {
1792                                 log_init_exit(-1);
1793                         }
1794                         if ((rval = zone_halt(zlogp, B_FALSE, B_FALSE, zstate))
1795                             != 0)
1796                                 break;
1797                         eventstream_write(Z_EVT_ZONE_HALTED);
1798                         zcons_statechanged();
1799                         break;
1800                 case Z_REBOOT:
1801                         (void) strlcpy(boot_args, zargp->bootbuf,
1802                             sizeof (boot_args));
1803                         eventstream_write(Z_EVT_ZONE_REBOOTING);
1804                         if ((rval = zone_halt(zlogp, B_FALSE, B_TRUE, zstate))
1805                             != 0) {
1806                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1807                                 boot_args[0] = '\0';
1808                                 break;
1809                         }
1810                         zcons_statechanged();
1811                         if ((rval = zone_ready(zlogp, Z_MNT_BOOT, zstate)) !=
1812                             0) {
1813                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1814                                 boot_args[0] = '\0';
1815                                 break;
1816                         }
1817                         rval = zone_bootup(zlogp, zargp->bootbuf, zstate);
1818                         audit_put_record(zlogp, uc, rval, "reboot");
1819                         if (rval != 0) {
1820                                 (void) zone_halt(zlogp, B_FALSE, B_TRUE,
1821                                     zstate);
1822                                 eventstream_write(Z_EVT_ZONE_BOOTFAILED);
1823                         }
1824                         boot_args[0] = '\0';
1825                         break;
1826                 case Z_SHUTDOWN:
1827                         if ((rval = zone_graceful_shutdown(zlogp)) == 0) {
1828                                 wait_shut = B_TRUE;
1829                         }
1830                         break;
1831                 case Z_NOTE_UNINSTALLING:
1832                 case Z_MOUNT:


1976                          * which will try again (by which time zoneadmd
1977                          * will hopefully have exited).
1978                          */
1979                         goto out;
1980                 }
1981 
1982                 /*
1983                  * If we got this far, there's a fattach(3c)'ed door
1984                  * that belongs to a process that has exited, which can
1985                  * happen if the previous zoneadmd died unexpectedly.
1986                  *
1987                  * Let user know that something is amiss, but that we can
1988                  * recover; if the zone is in the installed state, then don't
1989                  * message, since having a running zoneadmd isn't really
1990                  * expected/needed.  We want to keep occurences of this message
1991                  * limited to times when zoneadmd is picking back up from a
1992                  * zoneadmd that died while the zone was in some non-trivial
1993                  * state.
1994                  */
1995                 if (zstate > ZONE_STATE_INSTALLED) {
1996                         static zoneid_t zid;
1997 
1998                         zerror(zlogp, B_FALSE,
1999                             "zone '%s': WARNING: zone is in state '%s', but "
2000                             "zoneadmd does not appear to be available; "
2001                             "restarted zoneadmd to recover.",
2002                             zone_name, zone_state_str(zstate));
2003 
2004                         /*
2005                          * Startup a thread to perform the zfd logging/tty svc
2006                          * and a thread to perform memory capping for the
2007                          * zone. zlogp won't be valid for much longer so use
2008                          * logsys.
2009                          */
2010                         if ((zid = getzoneidbyname(zone_name)) != -1) {
2011                                 create_log_thread(&logsys, zid);
2012                                 create_mcap_thread(&logsys, zid);
2013                         }
2014 
2015                         /* recover the global configuration snapshot */
2016                         if (snap_hndl == NULL) {
2017                                 if ((snap_hndl = zonecfg_init_handle())
2018                                     == NULL ||
2019                                     zonecfg_create_snapshot(zone_name)
2020                                     != Z_OK ||
2021                                     zonecfg_get_snapshot_handle(zone_name,
2022                                     snap_hndl) != Z_OK) {
2023                                         zerror(zlogp, B_FALSE, "recovering "
2024                                             "zone configuration handle");
2025                                         goto out;
2026                                 }
2027                         }
2028                 }
2029 
2030                 (void) fdetach(zone_door_path);
2031                 (void) close(doorfd);
2032                 goto top;
2033         }
2034         ret = 0;
2035 out:
2036         (void) close(doorfd);
2037         return (ret);
2038 }
2039 
2040 /*
2041  * Setup the brand's pre and post state change callbacks, as well as the
2042  * query callback, if any of these exist.
2043  */
2044 static int
2045 brand_callback_init(brand_handle_t bh, char *zone_name)
2046 {





2047         (void) strlcpy(pre_statechg_hook, EXEC_PREFIX,
2048             sizeof (pre_statechg_hook));
2049 
2050         if (brand_get_prestatechange(bh, zone_name, zonepath,
2051             pre_statechg_hook + EXEC_LEN,
2052             sizeof (pre_statechg_hook) - EXEC_LEN) != 0)
2053                 return (-1);
2054 
2055         if (strlen(pre_statechg_hook) <= EXEC_LEN)
2056                 pre_statechg_hook[0] = '\0';
2057 
2058         (void) strlcpy(post_statechg_hook, EXEC_PREFIX,
2059             sizeof (post_statechg_hook));
2060 
2061         if (brand_get_poststatechange(bh, zone_name, zonepath,
2062             post_statechg_hook + EXEC_LEN,
2063             sizeof (post_statechg_hook) - EXEC_LEN) != 0)
2064                 return (-1);
2065 
2066         if (strlen(post_statechg_hook) <= EXEC_LEN)
2067                 post_statechg_hook[0] = '\0';
2068 
2069         (void) strlcpy(query_hook, EXEC_PREFIX,
2070             sizeof (query_hook));
2071 
2072         if (brand_get_query(bh, zone_name, zonepath, query_hook + EXEC_LEN,
2073             sizeof (query_hook) - EXEC_LEN) != 0)
2074                 return (-1);
2075 
2076         if (strlen(query_hook) <= EXEC_LEN)
2077                 query_hook[0] = '\0';
2078 
2079         return (0);
2080 }
2081 
2082 int
2083 main(int argc, char *argv[])
2084 {
2085         int opt;
2086         zoneid_t zid;
2087         priv_set_t *privset;
2088         zone_state_t zstate;
2089         char parents_locale[MAXPATHLEN];
2090         brand_handle_t bh;
2091         int err;
2092 


2180         }
2181 
2182         if (zone_get_id(zone_name, &zid) != 0) {
2183                 zerror(zlogp, B_FALSE, "could not manage %s: %s", zone_name,
2184                     zonecfg_strerror(Z_NO_ZONE));
2185                 return (1);
2186         }
2187 
2188         if ((err = zone_get_state(zone_name, &zstate)) != Z_OK) {
2189                 zerror(zlogp, B_FALSE, "failed to get zone state: %s",
2190                     zonecfg_strerror(err));
2191                 return (1);
2192         }
2193         if (zstate < ZONE_STATE_INCOMPLETE) {
2194                 zerror(zlogp, B_FALSE,
2195                     "cannot manage a zone which is in state '%s'",
2196                     zone_state_str(zstate));
2197                 return (1);
2198         }
2199 
2200         if (zone_get_zonepath(zone_name, zonepath, sizeof (zonepath)) != Z_OK) {
2201                 zerror(zlogp, B_FALSE, "unable to determine zone path");
2202                 return (-1);
2203         }
2204 
2205         if (zonecfg_default_brand(default_brand,
2206             sizeof (default_brand)) != Z_OK) {
2207                 zerror(zlogp, B_FALSE, "unable to determine default brand");
2208                 return (1);
2209         }
2210 
2211         /* Get a handle to the brand info for this zone */
2212         if (zone_get_brand(zone_name, brand_name, sizeof (brand_name))
2213             != Z_OK) {
2214                 zerror(zlogp, B_FALSE, "unable to determine zone brand");
2215                 return (1);
2216         }
2217         zone_isnative = (strcmp(brand_name, NATIVE_BRAND_NAME) == 0);
2218         zone_islabeled = (strcmp(brand_name, LABELED_BRAND_NAME) == 0);
2219 
2220         /*
2221          * In the alternate root environment, the only supported
2222          * operations are mount and unmount.  In this case, just treat
2223          * the zone as native if it is cluster.  Cluster zones can be
2224          * native for the purpose of LU or upgrade, and the cluster