/* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2007 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ #pragma ident "%Z%%M% %I% %E% SMI" /* * zoneadm is a command interpreter for zone administration. It is all in * C (i.e., no lex/yacc), and all the argument passing is argc/argv based. * main() calls parse_and_run() which calls cmd_match(), then invokes the * appropriate command's handler function. The rest of the program is the * handler functions and their helper functions. * * Some of the helper functions are used largely to simplify I18N: reducing * the need for translation notes. This is particularly true of many of * the zerror() calls: doing e.g. zerror(gettext("%s failed"), "foo") rather * than zerror(gettext("foo failed")) with a translation note indicating * that "foo" need not be translated. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "zoneadm.h" #define MAXARGS 8 /* Reflects kernel zone entries */ typedef struct zone_entry { zoneid_t zid; char zname[ZONENAME_MAX]; char *zstate_str; zone_state_t zstate_num; char zbrand[MAXNAMELEN]; char zroot[MAXPATHLEN]; char zuuid[UUID_PRINTABLE_STRING_LENGTH]; zone_iptype_t ziptype; } zone_entry_t; static zone_entry_t *zents; static size_t nzents; static boolean_t is_native_zone = B_TRUE; #define LOOPBACK_IF "lo0" #define SOCKET_AF(af) (((af) == AF_UNSPEC) ? AF_INET : (af)) struct net_if { char *name; int af; }; /* 0755 is the default directory mode. */ #define DEFAULT_DIR_MODE \ (S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH) struct cmd { uint_t cmd_num; /* command number */ char *cmd_name; /* command name */ char *short_usage; /* short form help */ int (*handler)(int argc, char *argv[]); /* function to call */ }; #define SHELP_HELP "help" #define SHELP_BOOT "boot [-- boot_arguments]" #define SHELP_HALT "halt" #define SHELP_READY "ready" #define SHELP_REBOOT "reboot [-- boot_arguments]" #define SHELP_LIST "list [-cipv]" #define SHELP_VERIFY "verify" #define SHELP_INSTALL "install [-x nodataset] [brand-specific args]" #define SHELP_UNINSTALL "uninstall [-F]" #define SHELP_CLONE "clone [-m method] [-s ] zonename" #define SHELP_MOVE "move zonepath" #define SHELP_DETACH "detach [-n]" #define SHELP_ATTACH "attach [-F] [-n ]" #define SHELP_MARK "mark incomplete" #define EXEC_PREFIX "exec " #define EXEC_LEN (strlen(EXEC_PREFIX)) #define RMCOMMAND "/usr/bin/rm -rf" static int cleanup_zonepath(char *, boolean_t); extern int ifname_open(char *); static int help_func(int argc, char *argv[]); static int ready_func(int argc, char *argv[]); static int boot_func(int argc, char *argv[]); static int halt_func(int argc, char *argv[]); static int reboot_func(int argc, char *argv[]); static int list_func(int argc, char *argv[]); static int verify_func(int argc, char *argv[]); static int install_func(int argc, char *argv[]); static int uninstall_func(int argc, char *argv[]); static int mount_func(int argc, char *argv[]); static int unmount_func(int argc, char *argv[]); static int clone_func(int argc, char *argv[]); static int move_func(int argc, char *argv[]); static int detach_func(int argc, char *argv[]); static int attach_func(int argc, char *argv[]); static int mark_func(int argc, char *argv[]); static int apply_func(int argc, char *argv[]); static int sanity_check(char *zone, int cmd_num, boolean_t running, boolean_t unsafe_when_running, boolean_t force); static int cmd_match(char *cmd); static int verify_details(int, char *argv[]); static int verify_brand(zone_dochandle_t, int, char *argv[]); static int invoke_brand_handler(int, char *argv[]); static struct cmd cmdtab[] = { { CMD_HELP, "help", SHELP_HELP, help_func }, { CMD_BOOT, "boot", SHELP_BOOT, boot_func }, { CMD_HALT, "halt", SHELP_HALT, halt_func }, { CMD_READY, "ready", SHELP_READY, ready_func }, { CMD_REBOOT, "reboot", SHELP_REBOOT, reboot_func }, { CMD_LIST, "list", SHELP_LIST, list_func }, { CMD_VERIFY, "verify", SHELP_VERIFY, verify_func }, { CMD_INSTALL, "install", SHELP_INSTALL, install_func }, { CMD_UNINSTALL, "uninstall", SHELP_UNINSTALL, uninstall_func }, /* mount and unmount are private commands for admin/install */ { CMD_MOUNT, "mount", NULL, mount_func }, { CMD_UNMOUNT, "unmount", NULL, unmount_func }, { CMD_CLONE, "clone", SHELP_CLONE, clone_func }, { CMD_MOVE, "move", SHELP_MOVE, move_func }, { CMD_DETACH, "detach", SHELP_DETACH, detach_func }, { CMD_ATTACH, "attach", SHELP_ATTACH, attach_func }, { CMD_MARK, "mark", SHELP_MARK, mark_func }, { CMD_APPLY, "apply", NULL, apply_func } }; /* global variables */ /* set early in main(), never modified thereafter, used all over the place */ static char *execname; static char target_brand[MAXNAMELEN]; static char *locale; char *target_zone; static char *target_uuid; /* used in do_subproc() and signal handler */ static volatile boolean_t child_killed; static int do_subproc_cnt = 0; /* * Used to indicate whether this zoneadm instance has another zoneadm * instance in its ancestry. */ static boolean_t zoneadm_is_nested = B_FALSE; /* used to track nested zone-lock operations */ static int zone_lock_cnt = 0; /* used to communicate lock status to children */ #define LOCK_ENV_VAR "_ZONEADM_LOCK_HELD" static char zoneadm_lock_held[] = LOCK_ENV_VAR"=1"; static char zoneadm_lock_not_held[] = LOCK_ENV_VAR"=0"; char * cmd_to_str(int cmd_num) { assert(cmd_num >= CMD_MIN && cmd_num <= CMD_MAX); return (cmdtab[cmd_num].cmd_name); } /* This is a separate function because of gettext() wrapping. */ static char * long_help(int cmd_num) { assert(cmd_num >= CMD_MIN && cmd_num <= CMD_MAX); switch (cmd_num) { case CMD_HELP: return (gettext("Print usage message.")); case CMD_BOOT: return (gettext("Activates (boots) specified zone. See " "zoneadm(1m) for valid boot\n\targuments.")); case CMD_HALT: return (gettext("Halts specified zone, bypassing shutdown " "scripts and removing runtime\n\tresources of the zone.")); case CMD_READY: return (gettext("Prepares a zone for running applications but " "does not start any user\n\tprocesses in the zone.")); case CMD_REBOOT: return (gettext("Restarts the zone (equivalent to a halt / " "boot sequence).\n\tFails if the zone is not active. " "See zoneadm(1m) for valid boot\n\targuments.")); case CMD_LIST: return (gettext("Lists the current zones, or a " "specific zone if indicated. By default,\n\tall " "running zones are listed, though this can be " "expanded to all\n\tinstalled zones with the -i " "option or all configured zones with the\n\t-c " "option. When used with the general -z and/or -u " "\n\toptions, lists only the specified " "matching zone, but lists it\n\tregardless of its state, " "and the -i and -c options are disallowed. The\n\t-v " "option can be used to display verbose information: zone " "name, id,\n\tcurrent state, root directory and options. " "The -p option can be used\n\tto request machine-parsable " "output. The -v and -p options are mutually\n\texclusive." " If neither -v nor -p is used, just the zone name is " "listed.")); case CMD_VERIFY: return (gettext("Check to make sure the configuration " "can safely be instantiated\n\ton the machine: " "physical network interfaces exist, etc.")); case CMD_INSTALL: return (gettext("Install the configuration on to the system. " "The -x nodataset option\n\tcan be used to prevent the " "creation of a new ZFS file system for the\n\tzone " "(assuming the zonepath is within a ZFS file system).\n\t" "All other arguments are passed to the brand installation " "function;\n\tsee brand(4) for more information.")); case CMD_UNINSTALL: return (gettext("Uninstall the configuration from the system. " "The -F flag can be used\n\tto force the action.")); case CMD_CLONE: return (gettext("Clone the installation of another zone. " "The -m option can be used to\n\tspecify 'copy' which " "forces a copy of the source zone. The -s option\n\t" "can be used to specify the name of a ZFS snapshot " "that was taken from\n\ta previous clone command. The " "snapshot will be used as the source\n\tinstead of " "creating a new ZFS snapshot.")); case CMD_MOVE: return (gettext("Move the zone to a new zonepath.")); case CMD_DETACH: return (gettext("Detach the zone from the system. The zone " "state is changed to\n\t'configured' (but the files under " "the zonepath are untouched).\n\tThe zone can subsequently " "be attached, or can be moved to another\n\tsystem and " "attached there. The -n option can be used to specify\n\t" "'no-execute' mode. When -n is used, the information " "needed to attach\n\tthe zone is sent to standard output " "but the zone is not actually\n\tdetached.")); case CMD_ATTACH: return (gettext("Attach the zone to the system. The zone " "state must be 'configured'\n\tprior to attach; upon " "successful completion, the zone state will be\n\t" "'installed'. The system software on the current " "system must be\n\tcompatible with the software on the " "zone's original system.\n\tSpecify -F to force the attach " "and skip software compatibility tests.\n\tThe -n option " "can be used to specify 'no-execute' mode. When -n is\n\t" "used, the information needed to attach the zone is read " "from the\n\tspecified path and the configuration is only " "validated. The path can\n\tbe '-' to specify standard " "input.")); case CMD_MARK: return (gettext("Set the state of the zone. This can be used " "to force the zone\n\tstate to 'incomplete' " "administratively if some activity has rendered\n\tthe " "zone permanently unusable. The only valid state that " "may be\n\tspecified is 'incomplete'.")); default: return (""); } /* NOTREACHED */ return (NULL); } /* * Called with explicit B_TRUE when help is explicitly requested, B_FALSE for * unexpected errors. */ static int usage(boolean_t explicit) { int i; FILE *fd = explicit ? stdout : stderr; (void) fprintf(fd, "%s:\t%s help\n", gettext("usage"), execname); (void) fprintf(fd, "\t%s [-z ] [-u ] list\n", execname); (void) fprintf(fd, "\t%s {-z |-u } <%s>\n", execname, gettext("subcommand")); (void) fprintf(fd, "\n%s:\n\n", gettext("Subcommands")); for (i = CMD_MIN; i <= CMD_MAX; i++) { if (cmdtab[i].short_usage == NULL) continue; (void) fprintf(fd, "%s\n", cmdtab[i].short_usage); if (explicit) (void) fprintf(fd, "\t%s\n\n", long_help(i)); } if (!explicit) (void) fputs("\n", fd); return (Z_USAGE); } static void sub_usage(char *short_usage, int cmd_num) { (void) fprintf(stderr, "%s:\t%s\n", gettext("usage"), short_usage); (void) fprintf(stderr, "\t%s\n", long_help(cmd_num)); } /* * zperror() is like perror(3c) except that this also prints the executable * name at the start of the message, and takes a boolean indicating whether * to call libc'c strerror() or that from libzonecfg. */ void zperror(const char *str, boolean_t zonecfg_error) { (void) fprintf(stderr, "%s: %s: %s\n", execname, str, zonecfg_error ? zonecfg_strerror(errno) : strerror(errno)); } /* * zperror2() is very similar to zperror() above, except it also prints a * supplied zone name after the executable. * * All current consumers of this function want libzonecfg's strerror() rather * than libc's; if this ever changes, this function can be made more generic * like zperror() above. */ void zperror2(const char *zone, const char *str) { (void) fprintf(stderr, "%s: %s: %s: %s\n", execname, zone, str, zonecfg_strerror(errno)); } /* PRINTFLIKE1 */ void zerror(const char *fmt, ...) { va_list alist; va_start(alist, fmt); (void) fprintf(stderr, "%s: ", execname); if (target_zone != NULL) (void) fprintf(stderr, "zone '%s': ", target_zone); (void) vfprintf(stderr, fmt, alist); (void) fprintf(stderr, "\n"); va_end(alist); } static void * safe_calloc(size_t nelem, size_t elsize) { void *r = calloc(nelem, elsize); if (r == NULL) { zerror(gettext("failed to allocate %lu bytes: %s"), (ulong_t)nelem * elsize, strerror(errno)); exit(Z_ERR); } return (r); } static void zone_print(zone_entry_t *zent, boolean_t verbose, boolean_t parsable) { static boolean_t firsttime = B_TRUE; char *ip_type_str; if (zent->ziptype == ZS_EXCLUSIVE) ip_type_str = "excl"; else ip_type_str = "shared"; assert(!(verbose && parsable)); if (firsttime && verbose) { firsttime = B_FALSE; (void) printf("%*s %-16s %-10s %-30s %-8s %-6s\n", ZONEID_WIDTH, "ID", "NAME", "STATUS", "PATH", "BRAND", "IP"); } if (!verbose) { char *cp, *clim; if (!parsable) { (void) printf("%s\n", zent->zname); return; } if (zent->zid == ZONE_ID_UNDEFINED) (void) printf("-"); else (void) printf("%lu", zent->zid); (void) printf(":%s:%s:", zent->zname, zent->zstate_str); cp = zent->zroot; while ((clim = strchr(cp, ':')) != NULL) { (void) printf("%.*s\\:", clim - cp, cp); cp = clim + 1; } (void) printf("%s:%s:%s:%s\n", cp, zent->zuuid, zent->zbrand, ip_type_str); return; } if (zent->zstate_str != NULL) { if (zent->zid == ZONE_ID_UNDEFINED) (void) printf("%*s", ZONEID_WIDTH, "-"); else (void) printf("%*lu", ZONEID_WIDTH, zent->zid); (void) printf(" %-16s %-10s %-30s %-8s %-6s\n", zent->zname, zent->zstate_str, zent->zroot, zent->zbrand, ip_type_str); } } static int lookup_zone_info(const char *zone_name, zoneid_t zid, zone_entry_t *zent) { char root[MAXPATHLEN], *cp; int err; uuid_t uuid; (void) strlcpy(zent->zname, zone_name, sizeof (zent->zname)); (void) strlcpy(zent->zroot, "???", sizeof (zent->zroot)); (void) strlcpy(zent->zbrand, "???", sizeof (zent->zbrand)); zent->zstate_str = "???"; zent->zid = zid; if (zonecfg_get_uuid(zone_name, uuid) == Z_OK && !uuid_is_null(uuid)) uuid_unparse(uuid, zent->zuuid); else zent->zuuid[0] = '\0'; /* * For labeled zones which query the zone path of lower-level * zones, the path needs to be adjusted to drop the final * "/root" component. This adjusted path is then useful * for reading down any exported directories from the * lower-level zone. */ if (is_system_labeled() && zent->zid != ZONE_ID_UNDEFINED) { if (zone_getattr(zent->zid, ZONE_ATTR_ROOT, zent->zroot, sizeof (zent->zroot)) == -1) { zperror2(zent->zname, gettext("could not get zone path.")); return (Z_ERR); } cp = zent->zroot + strlen(zent->zroot) - 5; if (cp > zent->zroot && strcmp(cp, "/root") == 0) *cp = 0; } else { if ((err = zone_get_zonepath(zent->zname, root, sizeof (root))) != Z_OK) { errno = err; zperror2(zent->zname, gettext("could not get zone path.")); return (Z_ERR); } (void) strlcpy(zent->zroot, root, sizeof (zent->zroot)); } if ((err = zone_get_state(zent->zname, &zent->zstate_num)) != Z_OK) { errno = err; zperror2(zent->zname, gettext("could not get state")); return (Z_ERR); } zent->zstate_str = zone_state_str(zent->zstate_num); /* * A zone's brand is only available in the .xml file describing it, * which is only visible to the global zone. This causes * zone_get_brand() to fail when called from within a non-global * zone. Fortunately we only do this on labeled systems, where we * know all zones are native. */ if (getzoneid() != GLOBAL_ZONEID) { assert(is_system_labeled() != 0); (void) strlcpy(zent->zbrand, NATIVE_BRAND_NAME, sizeof (zent->zbrand)); } else if (zone_get_brand(zent->zname, zent->zbrand, sizeof (zent->zbrand)) != Z_OK) { zperror2(zent->zname, gettext("could not get brand name")); return (Z_ERR); } /* * Get ip type of the zone. * Note for global zone, ZS_SHARED is set always. */ if (zid == GLOBAL_ZONEID) { zent->ziptype = ZS_SHARED; } else { if (zent->zstate_num == ZONE_STATE_RUNNING) { ushort_t flags; if (zone_getattr(zid, ZONE_ATTR_FLAGS, &flags, sizeof (flags)) < 0) { zperror2(zent->zname, gettext("could not get zone flags")); return (Z_ERR); } if (flags & ZF_NET_EXCL) zent->ziptype = ZS_EXCLUSIVE; else zent->ziptype = ZS_SHARED; } else { zone_dochandle_t handle; if ((handle = zonecfg_init_handle()) == NULL) { zperror2(zent->zname, gettext("could not init handle")); return (Z_ERR); } if ((err = zonecfg_get_handle(zent->zname, handle)) != Z_OK) { zperror2(zent->zname, gettext("could not get handle")); zonecfg_fini_handle(handle); return (Z_ERR); } if ((err = zonecfg_get_iptype(handle, &zent->ziptype)) != Z_OK) { zperror2(zent->zname, gettext("could not get ip-type")); zonecfg_fini_handle(handle); return (Z_ERR); } zonecfg_fini_handle(handle); } } return (Z_OK); } /* * fetch_zents() calls zone_list(2) to find out how many zones are running * (which is stored in the global nzents), then calls zone_list(2) again * to fetch the list of running zones (stored in the global zents). This * function may be called multiple times, so if zents is already set, we * return immediately to save work. */ static int fetch_zents(void) { zoneid_t *zids = NULL; uint_t nzents_saved; int i, retv; FILE *fp; boolean_t inaltroot; zone_entry_t *zentp; if (nzents > 0) return (Z_OK); if (zone_list(NULL, &nzents) != 0) { zperror(gettext("failed to get zoneid list"), B_FALSE); return (Z_ERR); } again: if (nzents == 0) return (Z_OK); zids = safe_calloc(nzents, sizeof (zoneid_t)); nzents_saved = nzents; if (zone_list(zids, &nzents) != 0) { zperror(gettext("failed to get zone list"), B_FALSE); free(zids); return (Z_ERR); } if (nzents != nzents_saved) { /* list changed, try again */ free(zids); goto again; } zents = safe_calloc(nzents, sizeof (zone_entry_t)); inaltroot = zonecfg_in_alt_root(); if (inaltroot) fp = zonecfg_open_scratch("", B_FALSE); else fp = NULL; zentp = zents; retv = Z_OK; for (i = 0; i < nzents; i++) { char name[ZONENAME_MAX]; char altname[ZONENAME_MAX]; if (getzonenamebyid(zids[i], name, sizeof (name)) < 0) { zperror(gettext("failed to get zone name"), B_FALSE); retv = Z_ERR; continue; } if (zonecfg_is_scratch(name)) { /* Ignore scratch zones by default */ if (!inaltroot) continue; if (fp == NULL || zonecfg_reverse_scratch(fp, name, altname, sizeof (altname), NULL, 0) == -1) { zerror(gettext("could not resolve scratch " "zone %s"), name); retv = Z_ERR; continue; } (void) strcpy(name, altname); } else { /* Ignore non-scratch when in an alternate root */ if (inaltroot && strcmp(name, GLOBAL_ZONENAME) != 0) continue; } if (lookup_zone_info(name, zids[i], zentp) != Z_OK) { zerror(gettext("failed to get zone data")); retv = Z_ERR; continue; } zentp++; } nzents = zentp - zents; if (fp != NULL) zonecfg_close_scratch(fp); free(zids); return (retv); } static int zone_print_list(zone_state_t min_state, boolean_t verbose, boolean_t parsable) { int i; zone_entry_t zent; FILE *cookie; char *name; /* * First get the list of running zones from the kernel and print them. * If that is all we need, then return. */ if ((i = fetch_zents()) != Z_OK) { /* * No need for error messages; fetch_zents() has already taken * care of this. */ return (i); } for (i = 0; i < nzents; i++) zone_print(&zents[i], verbose, parsable); if (min_state >= ZONE_STATE_RUNNING) return (Z_OK); /* * Next, get the full list of zones from the configuration, skipping * any we have already printed. */ cookie = setzoneent(); while ((name = getzoneent(cookie)) != NULL) { for (i = 0; i < nzents; i++) { if (strcmp(zents[i].zname, name) == 0) break; } if (i < nzents) { free(name); continue; } if (lookup_zone_info(name, ZONE_ID_UNDEFINED, &zent) != Z_OK) { free(name); continue; } free(name); if (zent.zstate_num >= min_state) zone_print(&zent, verbose, parsable); } endzoneent(cookie); return (Z_OK); } static zone_entry_t * lookup_running_zone(char *str) { zoneid_t zoneid; char *cp; int i; if (fetch_zents() != Z_OK) return (NULL); for (i = 0; i < nzents; i++) { if (strcmp(str, zents[i].zname) == 0) return (&zents[i]); } errno = 0; zoneid = strtol(str, &cp, 0); if (zoneid < MIN_ZONEID || zoneid > MAX_ZONEID || errno != 0 || *cp != '\0') return (NULL); for (i = 0; i < nzents; i++) { if (zoneid == zents[i].zid) return (&zents[i]); } return (NULL); } /* * Check a bit in a mode_t: if on is B_TRUE, that bit should be on; if * B_FALSE, it should be off. Return B_TRUE if the mode is bad (incorrect). */ static boolean_t bad_mode_bit(mode_t mode, mode_t bit, boolean_t on, char *file) { char *str; assert(bit == S_IRUSR || bit == S_IWUSR || bit == S_IXUSR || bit == S_IRGRP || bit == S_IWGRP || bit == S_IXGRP || bit == S_IROTH || bit == S_IWOTH || bit == S_IXOTH); /* * TRANSLATION_NOTE * The strings below will be used as part of a larger message, * either: * (file name) must be (owner|group|world) (read|writ|execut)able * or * (file name) must not be (owner|group|world) (read|writ|execut)able */ switch (bit) { case S_IRUSR: str = gettext("owner readable"); break; case S_IWUSR: str = gettext("owner writable"); break; case S_IXUSR: str = gettext("owner executable"); break; case S_IRGRP: str = gettext("group readable"); break; case S_IWGRP: str = gettext("group writable"); break; case S_IXGRP: str = gettext("group executable"); break; case S_IROTH: str = gettext("world readable"); break; case S_IWOTH: str = gettext("world writable"); break; case S_IXOTH: str = gettext("world executable"); break; } if ((mode & bit) == (on ? 0 : bit)) { /* * TRANSLATION_NOTE * The first parameter below is a file name; the second * is one of the "(owner|group|world) (read|writ|execut)able" * strings from above. */ /* * The code below could be simplified but not in a way * that would easily translate to non-English locales. */ if (on) { (void) fprintf(stderr, gettext("%s must be %s.\n"), file, str); } else { (void) fprintf(stderr, gettext("%s must not be %s.\n"), file, str); } return (B_TRUE); } return (B_FALSE); } /* * We want to make sure that no zone has its zone path as a child node * (in the directory sense) of any other. We do that by comparing this * zone's path to the path of all other (non-global) zones. The comparison * in each case is simple: add '/' to the end of the path, then do a * strncmp() of the two paths, using the length of the shorter one. */ static int crosscheck_zonepaths(char *path) { char rpath[MAXPATHLEN]; /* resolved path */ char path_copy[MAXPATHLEN]; /* copy of original path */ char rpath_copy[MAXPATHLEN]; /* copy of original rpath */ struct zoneent *ze; int res, err; FILE *cookie; cookie = setzoneent(); while ((ze = getzoneent_private(cookie)) != NULL) { /* Skip zones which are not installed. */ if (ze->zone_state < ZONE_STATE_INSTALLED) { free(ze); continue; } /* Skip the global zone and the current target zone. */ if (strcmp(ze->zone_name, GLOBAL_ZONENAME) == 0 || strcmp(ze->zone_name, target_zone) == 0) { free(ze); continue; } if (strlen(ze->zone_path) == 0) { /* old index file without path, fall back */ if ((err = zone_get_zonepath(ze->zone_name, ze->zone_path, sizeof (ze->zone_path))) != Z_OK) { errno = err; zperror2(ze->zone_name, gettext("could not get zone path")); free(ze); continue; } } (void) snprintf(path_copy, sizeof (path_copy), "%s%s", zonecfg_get_root(), ze->zone_path); res = resolvepath(path_copy, rpath, sizeof (rpath)); if (res == -1) { if (errno != ENOENT) { zperror(path_copy, B_FALSE); free(ze); return (Z_ERR); } (void) printf(gettext("WARNING: zone %s is installed, " "but its %s %s does not exist.\n"), ze->zone_name, "zonepath", path_copy); free(ze); continue; } rpath[res] = '\0'; (void) snprintf(path_copy, sizeof (path_copy), "%s/", path); (void) snprintf(rpath_copy, sizeof (rpath_copy), "%s/", rpath); if (strncmp(path_copy, rpath_copy, min(strlen(path_copy), strlen(rpath_copy))) == 0) { /* * TRANSLATION_NOTE * zonepath is a literal that should not be translated. */ (void) fprintf(stderr, gettext("%s zonepath (%s) and " "%s zonepath (%s) overlap.\n"), target_zone, path, ze->zone_name, rpath); free(ze); return (Z_ERR); } free(ze); } endzoneent(cookie); return (Z_OK); } static int validate_zonepath(char *path, int cmd_num) { int res; /* result of last library/system call */ boolean_t err = B_FALSE; /* have we run into an error? */ struct stat stbuf; struct statvfs64 vfsbuf; char rpath[MAXPATHLEN]; /* resolved path */ char ppath[MAXPATHLEN]; /* parent path */ char rppath[MAXPATHLEN]; /* resolved parent path */ char rootpath[MAXPATHLEN]; /* root path */ zone_state_t state; if (path[0] != '/') { (void) fprintf(stderr, gettext("%s is not an absolute path.\n"), path); return (Z_ERR); } if ((res = resolvepath(path, rpath, sizeof (rpath))) == -1) { if ((errno != ENOENT) || (cmd_num != CMD_VERIFY && cmd_num != CMD_INSTALL && cmd_num != CMD_CLONE && cmd_num != CMD_MOVE)) { zperror(path, B_FALSE); return (Z_ERR); } if (cmd_num == CMD_VERIFY) { /* * TRANSLATION_NOTE * zoneadm is a literal that should not be translated. */ (void) fprintf(stderr, gettext("WARNING: %s does not " "exist, so it could not be verified.\nWhen " "'zoneadm %s' is run, '%s' will try to create\n%s, " "and '%s' will be tried again,\nbut the '%s' may " "fail if:\nthe parent directory of %s is group- or " "other-writable\nor\n%s overlaps with any other " "installed zones.\n"), path, cmd_to_str(CMD_INSTALL), cmd_to_str(CMD_INSTALL), path, cmd_to_str(CMD_VERIFY), cmd_to_str(CMD_VERIFY), path, path); return (Z_OK); } /* * The zonepath is supposed to be mode 700 but its * parent(s) 755. So use 755 on the mkdirp() then * chmod() the zonepath itself to 700. */ if (mkdirp(path, DEFAULT_DIR_MODE) < 0) { zperror(path, B_FALSE); return (Z_ERR); } /* * If the chmod() fails, report the error, but might * as well continue the verify procedure. */ if (chmod(path, S_IRWXU) != 0) zperror(path, B_FALSE); /* * Since the mkdir() succeeded, we should not have to * worry about a subsequent ENOENT, thus this should * only recurse once. */ return (validate_zonepath(path, cmd_num)); } rpath[res] = '\0'; if (strcmp(path, rpath) != 0) { errno = Z_RESOLVED_PATH; zperror(path, B_TRUE); return (Z_ERR); } if ((res = stat(rpath, &stbuf)) != 0) { zperror(rpath, B_FALSE); return (Z_ERR); } if (!S_ISDIR(stbuf.st_mode)) { (void) fprintf(stderr, gettext("%s is not a directory.\n"), rpath); return (Z_ERR); } if (strcmp(stbuf.st_fstype, MNTTYPE_TMPFS) == 0) { (void) printf(gettext("WARNING: %s is on a temporary " "file system.\n"), rpath); } if (crosscheck_zonepaths(rpath) != Z_OK) return (Z_ERR); /* * Try to collect and report as many minor errors as possible * before returning, so the user can learn everything that needs * to be fixed up front. */ if (stbuf.st_uid != 0) { (void) fprintf(stderr, gettext("%s is not owned by root.\n"), rpath); err = B_TRUE; } err |= bad_mode_bit(stbuf.st_mode, S_IRUSR, B_TRUE, rpath); err |= bad_mode_bit(stbuf.st_mode, S_IWUSR, B_TRUE, rpath); err |= bad_mode_bit(stbuf.st_mode, S_IXUSR, B_TRUE, rpath); err |= bad_mode_bit(stbuf.st_mode, S_IRGRP, B_FALSE, rpath); err |= bad_mode_bit(stbuf.st_mode, S_IWGRP, B_FALSE, rpath); err |= bad_mode_bit(stbuf.st_mode, S_IXGRP, B_FALSE, rpath); err |= bad_mode_bit(stbuf.st_mode, S_IROTH, B_FALSE, rpath); err |= bad_mode_bit(stbuf.st_mode, S_IWOTH, B_FALSE, rpath); err |= bad_mode_bit(stbuf.st_mode, S_IXOTH, B_FALSE, rpath); (void) snprintf(ppath, sizeof (ppath), "%s/..", path); if ((res = resolvepath(ppath, rppath, sizeof (rppath))) == -1) { zperror(ppath, B_FALSE); return (Z_ERR); } rppath[res] = '\0'; if ((res = stat(rppath, &stbuf)) != 0) { zperror(rppath, B_FALSE); return (Z_ERR); } /* theoretically impossible */ if (!S_ISDIR(stbuf.st_mode)) { (void) fprintf(stderr, gettext("%s is not a directory.\n"), rppath); return (Z_ERR); } if (stbuf.st_uid != 0) { (void) fprintf(stderr, gettext("%s is not owned by root.\n"), rppath); err = B_TRUE; } err |= bad_mode_bit(stbuf.st_mode, S_IRUSR, B_TRUE, rppath); err |= bad_mode_bit(stbuf.st_mode, S_IWUSR, B_TRUE, rppath); err |= bad_mode_bit(stbuf.st_mode, S_IXUSR, B_TRUE, rppath); err |= bad_mode_bit(stbuf.st_mode, S_IWGRP, B_FALSE, rppath); err |= bad_mode_bit(stbuf.st_mode, S_IWOTH, B_FALSE, rppath); if (strcmp(rpath, rppath) == 0) { (void) fprintf(stderr, gettext("%s is its own parent.\n"), rppath); err = B_TRUE; } if (statvfs64(rpath, &vfsbuf) != 0) { zperror(rpath, B_FALSE); return (Z_ERR); } if (strcmp(vfsbuf.f_basetype, MNTTYPE_NFS) == 0) { /* * TRANSLATION_NOTE * Zonepath and NFS are literals that should not be translated. */ (void) fprintf(stderr, gettext("Zonepath %s is on an NFS " "mounted file system.\n" "\tA local file system must be used.\n"), rpath); return (Z_ERR); } if (vfsbuf.f_flag & ST_NOSUID) { /* * TRANSLATION_NOTE * Zonepath and nosuid are literals that should not be * translated. */ (void) fprintf(stderr, gettext("Zonepath %s is on a nosuid " "file system.\n"), rpath); return (Z_ERR); } if ((res = zone_get_state(target_zone, &state)) != Z_OK) { errno = res; zperror2(target_zone, gettext("could not get state")); return (Z_ERR); } /* * The existence of the root path is only bad in the configured state, * as it is *supposed* to be there at the installed and later states. * However, the root path is expected to be there if the zone is * detached. * State/command mismatches are caught earlier in verify_details(). */ if (state == ZONE_STATE_CONFIGURED && cmd_num != CMD_ATTACH) { if (snprintf(rootpath, sizeof (rootpath), "%s/root", rpath) >= sizeof (rootpath)) { /* * TRANSLATION_NOTE * Zonepath is a literal that should not be translated. */ (void) fprintf(stderr, gettext("Zonepath %s is too long.\n"), rpath); return (Z_ERR); } if ((res = stat(rootpath, &stbuf)) == 0) { if (zonecfg_detached(rpath)) (void) fprintf(stderr, gettext("Cannot %s detached " "zone.\nUse attach or remove %s " "directory.\n"), cmd_to_str(cmd_num), rpath); else (void) fprintf(stderr, gettext("Rootpath %s exists; " "remove or move aside prior to %s.\n"), rootpath, cmd_to_str(cmd_num)); return (Z_ERR); } } return (err ? Z_ERR : Z_OK); } /* * The following two routines implement a simple locking mechanism to * ensure that only one instance of zoneadm at a time is able to manipulate * a given zone. The lock is built on top of an fcntl(2) lock of * []/var/run/zones/.zoneadm.lock. If a zoneadm instance * can grab that lock, it is allowed to manipulate the zone. * * Since zoneadm may call external applications which in turn invoke * zoneadm again, we introduce the notion of "lock inheritance". Any * instance of zoneadm that has another instance in its ancestry is assumed * to be acting on behalf of the original zoneadm, and is thus allowed to * manipulate its zone. * * This inheritance is implemented via the _ZONEADM_LOCK_HELD environment * variable. When zoneadm is granted a lock on its zone, this environment * variable is set to 1. When it releases the lock, the variable is set to * 0. Since a child process inherits its parent's environment, checking * the state of this variable indicates whether or not any ancestor owns * the lock. */ static void release_lock_file(int lockfd) { /* * If we are cleaning up from a failed attempt to lock the zone for * the first time, we might have a zone_lock_cnt of 0. In that * error case, we don't want to do anything but close the lock * file. */ assert(zone_lock_cnt >= 0); if (zone_lock_cnt > 0) { assert(getenv(LOCK_ENV_VAR) != NULL); assert(atoi(getenv(LOCK_ENV_VAR)) == 1); if (--zone_lock_cnt > 0) { assert(lockfd == -1); return; } if (putenv(zoneadm_lock_not_held) != 0) { zperror(target_zone, B_TRUE); exit(Z_ERR); } } assert(lockfd >= 0); (void) close(lockfd); } static int grab_lock_file(const char *zone_name, int *lockfd) { char pathbuf[PATH_MAX]; struct flock flock; /* * If we already have the lock, we can skip this expensive song * and dance. */ if (zone_lock_cnt > 0) { zone_lock_cnt++; *lockfd = -1; return (Z_OK); } assert(getenv(LOCK_ENV_VAR) != NULL); assert(atoi(getenv(LOCK_ENV_VAR)) == 0); if (snprintf(pathbuf, sizeof (pathbuf), "%s%s", zonecfg_get_root(), ZONES_TMPDIR) >= sizeof (pathbuf)) { zerror(gettext("alternate root path is too long")); return (Z_ERR); } if (mkdir(pathbuf, S_IRWXU) < 0 && errno != EEXIST) { zerror(gettext("could not mkdir %s: %s"), pathbuf, strerror(errno)); return (Z_ERR); } (void) chmod(pathbuf, S_IRWXU); /* * One of these lock files is created for each zone (when needed). * The lock files are not cleaned up (except on system reboot), * but since there is only one per zone, there is no resource * starvation issue. */ if (snprintf(pathbuf, sizeof (pathbuf), "%s%s/%s.zoneadm.lock", zonecfg_get_root(), ZONES_TMPDIR, zone_name) >= sizeof (pathbuf)) { zerror(gettext("alternate root path is too long")); return (Z_ERR); } if ((*lockfd = open(pathbuf, O_RDWR|O_CREAT, S_IRUSR|S_IWUSR)) < 0) { zerror(gettext("could not open %s: %s"), pathbuf, strerror(errno)); return (Z_ERR); } /* * Lock the file to synchronize with other zoneadmds */ flock.l_type = F_WRLCK; flock.l_whence = SEEK_SET; flock.l_start = (off_t)0; flock.l_len = (off_t)0; if ((fcntl(*lockfd, F_SETLKW, &flock) < 0) || (putenv(zoneadm_lock_held) != 0)) { zerror(gettext("unable to lock %s: %s"), pathbuf, strerror(errno)); release_lock_file(*lockfd); return (Z_ERR); } zone_lock_cnt = 1; return (Z_OK); } static boolean_t get_doorname(const char *zone_name, char *buffer) { return (snprintf(buffer, PATH_MAX, "%s" ZONE_DOOR_PATH, zonecfg_get_root(), zone_name) < PATH_MAX); } /* * system daemons are not audited. For the global zone, this occurs * "naturally" since init is started with the default audit * characteristics. Since zoneadmd is a system daemon and it starts * init for a zone, it is necessary to clear out the audit * characteristics inherited from whomever started zoneadmd. This is * indicated by the audit id, which is set from the ruid parameter of * adt_set_user(), below. */ static void prepare_audit_context() { adt_session_data_t *ah; char *failure = gettext("audit failure: %s"); if (adt_start_session(&ah, NULL, 0)) { zerror(failure, strerror(errno)); return; } if (adt_set_user(ah, ADT_NO_AUDIT, ADT_NO_AUDIT, ADT_NO_AUDIT, ADT_NO_AUDIT, NULL, ADT_NEW)) { zerror(failure, strerror(errno)); (void) adt_end_session(ah); return; } if (adt_set_proc(ah)) zerror(failure, strerror(errno)); (void) adt_end_session(ah); } static int start_zoneadmd(const char *zone_name) { char doorpath[PATH_MAX]; pid_t child_pid; int error = Z_ERR; int doorfd, lockfd; struct door_info info; if (!get_doorname(zone_name, doorpath)) return (Z_ERR); if (grab_lock_file(zone_name, &lockfd) != Z_OK) return (Z_ERR); /* * Now that we have the lock, re-confirm that the daemon is * *not* up and working fine. If it is still down, we have a green * light to start it. */ if ((doorfd = open(doorpath, O_RDONLY)) < 0) { if (errno != ENOENT) { zperror(doorpath, B_FALSE); goto out; } } else { if (door_info(doorfd, &info) == 0 && ((info.di_attributes & DOOR_REVOKED) == 0)) { error = Z_OK; (void) close(doorfd); goto out; } (void) close(doorfd); } if ((child_pid = fork()) == -1) { zperror(gettext("could not fork"), B_FALSE); goto out; } else if (child_pid == 0) { const char *argv[6], **ap; /* child process */ prepare_audit_context(); ap = argv; *ap++ = "zoneadmd"; *ap++ = "-z"; *ap++ = zone_name; if (zonecfg_in_alt_root()) { *ap++ = "-R"; *ap++ = zonecfg_get_root(); } *ap = NULL; (void) execv("/usr/lib/zones/zoneadmd", (char * const *)argv); /* * TRANSLATION_NOTE * zoneadmd is a literal that should not be translated. */ zperror(gettext("could not exec zoneadmd"), B_FALSE); _exit(Z_ERR); } else { /* parent process */ pid_t retval; int pstatus = 0; do { retval = waitpid(child_pid, &pstatus, 0); } while (retval != child_pid); if (WIFSIGNALED(pstatus) || (WIFEXITED(pstatus) && WEXITSTATUS(pstatus) != 0)) { zerror(gettext("could not start %s"), "zoneadmd"); goto out; } } error = Z_OK; out: release_lock_file(lockfd); return (error); } static int ping_zoneadmd(const char *zone_name) { char doorpath[PATH_MAX]; int doorfd; struct door_info info; if (!get_doorname(zone_name, doorpath)) return (Z_ERR); if ((doorfd = open(doorpath, O_RDONLY)) < 0) { return (Z_ERR); } if (door_info(doorfd, &info) == 0 && ((info.di_attributes & DOOR_REVOKED) == 0)) { (void) close(doorfd); return (Z_OK); } (void) close(doorfd); return (Z_ERR); } static int call_zoneadmd(const char *zone_name, zone_cmd_arg_t *arg) { char doorpath[PATH_MAX]; int doorfd, result; door_arg_t darg; zoneid_t zoneid; uint64_t uniqid = 0; zone_cmd_rval_t *rvalp; size_t rlen; char *cp, *errbuf; rlen = getpagesize(); if ((rvalp = malloc(rlen)) == NULL) { zerror(gettext("failed to allocate %lu bytes: %s"), rlen, strerror(errno)); return (-1); } if ((zoneid = getzoneidbyname(zone_name)) != ZONE_ID_UNDEFINED) { (void) zone_getattr(zoneid, ZONE_ATTR_UNIQID, &uniqid, sizeof (uniqid)); } arg->uniqid = uniqid; (void) strlcpy(arg->locale, locale, sizeof (arg->locale)); if (!get_doorname(zone_name, doorpath)) { zerror(gettext("alternate root path is too long")); free(rvalp); return (-1); } /* * Loop trying to start zoneadmd; if something goes seriously * wrong we break out and fail. */ for (;;) { if (start_zoneadmd(zone_name) != Z_OK) break; if ((doorfd = open(doorpath, O_RDONLY)) < 0) { zperror(gettext("failed to open zone door"), B_FALSE); break; } darg.data_ptr = (char *)arg; darg.data_size = sizeof (*arg); darg.desc_ptr = NULL; darg.desc_num = 0; darg.rbuf = (char *)rvalp; darg.rsize = rlen; if (door_call(doorfd, &darg) != 0) { (void) close(doorfd); /* * We'll get EBADF if the door has been revoked. */ if (errno != EBADF) { zperror(gettext("door_call failed"), B_FALSE); break; } continue; /* take another lap */ } (void) close(doorfd); if (darg.data_size == 0) { /* Door server is going away; kick it again. */ continue; } errbuf = rvalp->errbuf; while (*errbuf != '\0') { /* * Remove any newlines since zerror() * will append one automatically. */ cp = strchr(errbuf, '\n'); if (cp != NULL) *cp = '\0'; zerror("%s", errbuf); if (cp == NULL) break; errbuf = cp + 1; } result = rvalp->rval == 0 ? 0 : -1; free(rvalp); return (result); } free(rvalp); return (-1); } static int invoke_brand_handler(int cmd_num, char *argv[]) { zone_dochandle_t handle; int err; if ((handle = zonecfg_init_handle()) == NULL) { zperror(cmd_to_str(cmd_num), B_TRUE); return (Z_ERR); } if ((err = zonecfg_get_handle(target_zone, handle)) != Z_OK) { errno = err; zperror(cmd_to_str(cmd_num), B_TRUE); zonecfg_fini_handle(handle); return (Z_ERR); } if (verify_brand(handle, cmd_num, argv) != Z_OK) { zonecfg_fini_handle(handle); return (Z_ERR); } zonecfg_fini_handle(handle); return (Z_OK); } static int ready_func(int argc, char *argv[]) { zone_cmd_arg_t zarg; int arg; if (zonecfg_in_alt_root()) { zerror(gettext("cannot ready zone in alternate root")); return (Z_ERR); } optind = 0; if ((arg = getopt(argc, argv, "?")) != EOF) { switch (arg) { case '?': sub_usage(SHELP_READY, CMD_READY); return (optopt == '?' ? Z_OK : Z_USAGE); default: sub_usage(SHELP_READY, CMD_READY); return (Z_USAGE); } } if (argc > optind) { sub_usage(SHELP_READY, CMD_READY); return (Z_USAGE); } if (sanity_check(target_zone, CMD_READY, B_FALSE, B_FALSE, B_FALSE) != Z_OK) return (Z_ERR); if (verify_details(CMD_READY, argv) != Z_OK) return (Z_ERR); zarg.cmd = Z_READY; if (call_zoneadmd(target_zone, &zarg) != 0) { zerror(gettext("call to %s failed"), "zoneadmd"); return (Z_ERR); } return (Z_OK); } static int boot_func(int argc, char *argv[]) { zone_cmd_arg_t zarg; boolean_t force = B_FALSE; int arg; if (zonecfg_in_alt_root()) { zerror(gettext("cannot boot zone in alternate root")); return (Z_ERR); } zarg.bootbuf[0] = '\0'; /* * The following getopt processes arguments to zone boot; that * is to say, the [here] portion of the argument string: * * zoneadm -z myzone boot [here] -- -v -m verbose * * Where [here] can either be nothing, -? (in which case we bail * and print usage), -f (a private option to indicate that the * boot operation should be 'forced'), or -s. Support for -s is * vestigal and obsolete, but is retained because it was a * documented interface and there are known consumers including * admin/install; the proper way to specify boot arguments like -s * is: * * zoneadm -z myzone boot -- -s -v -m verbose. */ optind = 0; while ((arg = getopt(argc, argv, "?fs")) != EOF) { switch (arg) { case '?': sub_usage(SHELP_BOOT, CMD_BOOT); return (optopt == '?' ? Z_OK : Z_USAGE); case 's': (void) strlcpy(zarg.bootbuf, "-s", sizeof (zarg.bootbuf)); break; case 'f': force = B_TRUE; break; default: sub_usage(SHELP_BOOT, CMD_BOOT); return (Z_USAGE); } } for (; optind < argc; optind++) { if (strlcat(zarg.bootbuf, argv[optind], sizeof (zarg.bootbuf)) >= sizeof (zarg.bootbuf)) { zerror(gettext("Boot argument list too long")); return (Z_ERR); } if (optind < argc - 1) if (strlcat(zarg.bootbuf, " ", sizeof (zarg.bootbuf)) >= sizeof (zarg.bootbuf)) { zerror(gettext("Boot argument list too long")); return (Z_ERR); } } if (sanity_check(target_zone, CMD_BOOT, B_FALSE, B_FALSE, force) != Z_OK) return (Z_ERR); if (verify_details(CMD_BOOT, argv) != Z_OK) return (Z_ERR); zarg.cmd = force ? Z_FORCEBOOT : Z_BOOT; if (call_zoneadmd(target_zone, &zarg) != 0) { zerror(gettext("call to %s failed"), "zoneadmd"); return (Z_ERR); } return (Z_OK); } static void fake_up_local_zone(zoneid_t zid, zone_entry_t *zeptr) { ssize_t result; uuid_t uuid; FILE *fp; ushort_t flags; (void) memset(zeptr, 0, sizeof (*zeptr)); zeptr->zid = zid; /* * Since we're looking up our own (non-global) zone name, * we can be assured that it will succeed. */ result = getzonenamebyid(zid, zeptr->zname, sizeof (zeptr->zname)); assert(result >= 0); if (zonecfg_is_scratch(zeptr->zname) && (fp = zonecfg_open_scratch("", B_FALSE)) != NULL) { (void) zonecfg_reverse_scratch(fp, zeptr->zname, zeptr->zname, sizeof (zeptr->zname), NULL, 0); zonecfg_close_scratch(fp); } if (is_system_labeled()) { (void) zone_getattr(zid, ZONE_ATTR_ROOT, zeptr->zroot, sizeof (zeptr->zroot)); (void) strlcpy(zeptr->zbrand, NATIVE_BRAND_NAME, sizeof (zeptr->zbrand)); } else { (void) strlcpy(zeptr->zroot, "/", sizeof (zeptr->zroot)); (void) zone_getattr(zid, ZONE_ATTR_BRAND, zeptr->zbrand, sizeof (zeptr->zbrand)); } zeptr->zstate_str = "running"; if (zonecfg_get_uuid(zeptr->zname, uuid) == Z_OK && !uuid_is_null(uuid)) uuid_unparse(uuid, zeptr->zuuid); if (zone_getattr(zid, ZONE_ATTR_FLAGS, &flags, sizeof (flags)) < 0) { zperror2(zeptr->zname, gettext("could not get zone flags")); exit(Z_ERR); } if (flags & ZF_NET_EXCL) zeptr->ziptype = ZS_EXCLUSIVE; else zeptr->ziptype = ZS_SHARED; } static int list_func(int argc, char *argv[]) { zone_entry_t *zentp, zent; int arg, retv; boolean_t output = B_FALSE, verbose = B_FALSE, parsable = B_FALSE; zone_state_t min_state = ZONE_STATE_RUNNING; zoneid_t zone_id = getzoneid(); if (target_zone == NULL) { /* all zones: default view to running but allow override */ optind = 0; while ((arg = getopt(argc, argv, "?cipv")) != EOF) { switch (arg) { case '?': sub_usage(SHELP_LIST, CMD_LIST); return (optopt == '?' ? Z_OK : Z_USAGE); /* * The 'i' and 'c' options are not mutually * exclusive so if 'c' is given, then min_state * is set to 0 (ZONE_STATE_CONFIGURED) which is * the lowest possible state. If 'i' is given, * then min_state is set to be the lowest state * so far. */ case 'c': min_state = ZONE_STATE_CONFIGURED; break; case 'i': min_state = min(ZONE_STATE_INSTALLED, min_state); break; case 'p': parsable = B_TRUE; break; case 'v': verbose = B_TRUE; break; default: sub_usage(SHELP_LIST, CMD_LIST); return (Z_USAGE); } } if (parsable && verbose) { zerror(gettext("%s -p and -v are mutually exclusive."), cmd_to_str(CMD_LIST)); return (Z_ERR); } if (zone_id == GLOBAL_ZONEID || is_system_labeled()) { retv = zone_print_list(min_state, verbose, parsable); } else { fake_up_local_zone(zone_id, &zent); retv = Z_OK; zone_print(&zent, verbose, parsable); } return (retv); } /* * Specific target zone: disallow -i/-c suboptions. */ optind = 0; while ((arg = getopt(argc, argv, "?pv")) != EOF) { switch (arg) { case '?': sub_usage(SHELP_LIST, CMD_LIST); return (optopt == '?' ? Z_OK : Z_USAGE); case 'p': parsable = B_TRUE; break; case 'v': verbose = B_TRUE; break; default: sub_usage(SHELP_LIST, CMD_LIST); return (Z_USAGE); } } if (parsable && verbose) { zerror(gettext("%s -p and -v are mutually exclusive."), cmd_to_str(CMD_LIST)); return (Z_ERR); } if (argc > optind) { sub_usage(SHELP_LIST, CMD_LIST); return (Z_USAGE); } if (zone_id != GLOBAL_ZONEID) { fake_up_local_zone(zone_id, &zent); /* * main() will issue a Z_NO_ZONE error if it cannot get an * id for target_zone, which in a non-global zone should * happen for any zone name except `zonename`. Thus we * assert() that here but don't otherwise check. */ assert(strcmp(zent.zname, target_zone) == 0); zone_print(&zent, verbose, parsable); output = B_TRUE; } else if ((zentp = lookup_running_zone(target_zone)) != NULL) { zone_print(zentp, verbose, parsable); output = B_TRUE; } else if (lookup_zone_info(target_zone, ZONE_ID_UNDEFINED, &zent) == Z_OK) { zone_print(&zent, verbose, parsable); output = B_TRUE; } /* * Invoke brand-specific handler. Note that we do this * only if we're in the global zone, and target_zone is specified * and it is not the global zone. */ if (zone_id == GLOBAL_ZONEID && target_zone != NULL && strcmp(target_zone, GLOBAL_ZONENAME) != 0) if (invoke_brand_handler(CMD_LIST, argv) != Z_OK) return (Z_ERR); return (output ? Z_OK : Z_ERR); } static void sigterm(int sig) { /* * Ignore SIG{INT,TERM}, so we don't end up in an infinite loop, * then propagate the signal to our process group. */ assert(sig == SIGINT || sig == SIGTERM); (void) sigset(SIGINT, SIG_IGN); (void) sigset(SIGTERM, SIG_IGN); (void) kill(0, sig); child_killed = B_TRUE; } static int do_subproc(char *cmdbuf) { char inbuf[1024]; /* arbitrary large amount */ FILE *file; do_subproc_cnt++; child_killed = B_FALSE; /* * We use popen(3c) to launch child processes for [un]install; * this library call does not return a PID, so we have to kill * the whole process group. To avoid killing our parent, we * become a process group leader here. But doing so can wreak * havoc with reading from stdin when launched by a non-job-control * shell, so we close stdin and reopen it as /dev/null first. */ (void) close(STDIN_FILENO); (void) openat(STDIN_FILENO, "/dev/null", O_RDONLY); if (!zoneadm_is_nested) (void) setpgid(0, 0); (void) sigset(SIGINT, sigterm); (void) sigset(SIGTERM, sigterm); file = popen(cmdbuf, "r"); for (;;) { if (child_killed || fgets(inbuf, sizeof (inbuf), file) == NULL) break; (void) fputs(inbuf, stdout); } (void) sigset(SIGINT, SIG_DFL); (void) sigset(SIGTERM, SIG_DFL); return (pclose(file)); } static int do_subproc_interactive(char *cmdbuf) { void (*saveint)(int); void (*saveterm)(int); void (*savequit)(int); void (*savehup)(int); int pid, child, status; /* * do_subproc() links stdin to /dev/null, which would break any * interactive subprocess we try to launch here. Similarly, we * can't have been launched as a subprocess ourselves. */ assert(do_subproc_cnt == 0 && !zoneadm_is_nested); if ((child = vfork()) == 0) { (void) execl("/bin/sh", "sh", "-c", cmdbuf, (char *)NULL); } if (child == -1) return (-1); saveint = sigset(SIGINT, SIG_IGN); saveterm = sigset(SIGTERM, SIG_IGN); savequit = sigset(SIGQUIT, SIG_IGN); savehup = sigset(SIGHUP, SIG_IGN); while ((pid = waitpid(child, &status, 0)) != child && pid != -1) ; (void) sigset(SIGINT, saveint); (void) sigset(SIGTERM, saveterm); (void) sigset(SIGQUIT, savequit); (void) sigset(SIGHUP, savehup); return (pid == -1 ? -1 : status); } static int subproc_status(const char *cmd, int status, boolean_t verbose_failure) { if (WIFEXITED(status)) { int exit_code = WEXITSTATUS(status); if ((verbose_failure) && (exit_code != ZONE_SUBPROC_OK)) zerror(gettext("'%s' failed with exit code %d."), cmd, exit_code); return (exit_code); } else if (WIFSIGNALED(status)) { int signal = WTERMSIG(status); char sigstr[SIG2STR_MAX]; if (sig2str(signal, sigstr) == 0) { zerror(gettext("'%s' terminated by signal SIG%s."), cmd, sigstr); } else { zerror(gettext("'%s' terminated by an unknown signal."), cmd); } } else { zerror(gettext("'%s' failed for unknown reasons."), cmd); } /* * Assume a subprocess that died due to a signal or an unknown error * should be considered an exit code of ZONE_SUBPROC_FATAL, as the * user will likely need to do some manual cleanup. */ return (ZONE_SUBPROC_FATAL); } /* * Various sanity checks; make sure: * 1. We're in the global zone. * 2. The calling user has sufficient privilege. * 3. The target zone is neither the global zone nor anything starting with * "SUNW". * 4a. If we're looking for a 'not running' (i.e., configured or installed) * zone, the name service knows about it. * 4b. For some operations which expect a zone not to be running, that it is * not already running (or ready). */ static int sanity_check(char *zone, int cmd_num, boolean_t running, boolean_t unsafe_when_running, boolean_t force) { zone_entry_t *zent; priv_set_t *privset; zone_state_t state, min_state; char kernzone[ZONENAME_MAX]; FILE *fp; if (getzoneid() != GLOBAL_ZONEID) { switch (cmd_num) { case CMD_HALT: zerror(gettext("use %s to %s this zone."), "halt(1M)", cmd_to_str(cmd_num)); break; case CMD_REBOOT: zerror(gettext("use %s to %s this zone."), "reboot(1M)", cmd_to_str(cmd_num)); break; default: zerror(gettext("must be in the global zone to %s a " "zone."), cmd_to_str(cmd_num)); break; } return (Z_ERR); } if ((privset = priv_allocset()) == NULL) { zerror(gettext("%s failed"), "priv_allocset"); return (Z_ERR); } if (getppriv(PRIV_EFFECTIVE, privset) != 0) { zerror(gettext("%s failed"), "getppriv"); priv_freeset(privset); return (Z_ERR); } if (priv_isfullset(privset) == B_FALSE) { zerror(gettext("only a privileged user may %s a zone."), cmd_to_str(cmd_num)); priv_freeset(privset); return (Z_ERR); } priv_freeset(privset); if (zone == NULL) { zerror(gettext("no zone specified")); return (Z_ERR); } if (strcmp(zone, GLOBAL_ZONENAME) == 0) { zerror(gettext("%s operation is invalid for the global zone."), cmd_to_str(cmd_num)); return (Z_ERR); } if (strncmp(zone, "SUNW", 4) == 0) { zerror(gettext("%s operation is invalid for zones starting " "with SUNW."), cmd_to_str(cmd_num)); return (Z_ERR); } if (!is_native_zone && cmd_num == CMD_MOUNT) { zerror(gettext("%s operation is invalid for branded zones."), cmd_to_str(cmd_num)); return (Z_ERR); } if (!zonecfg_in_alt_root()) { zent = lookup_running_zone(zone); } else if ((fp = zonecfg_open_scratch("", B_FALSE)) == NULL) { zent = NULL; } else { if (zonecfg_find_scratch(fp, zone, zonecfg_get_root(), kernzone, sizeof (kernzone)) == 0) zent = lookup_running_zone(kernzone); else zent = NULL; zonecfg_close_scratch(fp); } /* * Look up from the kernel for 'running' zones. */ if (running && !force) { if (zent == NULL) { zerror(gettext("not running")); return (Z_ERR); } } else { int err; if (unsafe_when_running && zent != NULL) { /* check whether the zone is ready or running */ if ((err = zone_get_state(zent->zname, &zent->zstate_num)) != Z_OK) { errno = err; zperror2(zent->zname, gettext("could not get state")); /* can't tell, so hedge */ zent->zstate_str = "ready/running"; } else { zent->zstate_str = zone_state_str(zent->zstate_num); } zerror(gettext("%s operation is invalid for %s zones."), cmd_to_str(cmd_num), zent->zstate_str); return (Z_ERR); } if ((err = zone_get_state(zone, &state)) != Z_OK) { errno = err; zperror2(zone, gettext("could not get state")); return (Z_ERR); } switch (cmd_num) { case CMD_UNINSTALL: if (state == ZONE_STATE_CONFIGURED) { zerror(gettext("is already in state '%s'."), zone_state_str(ZONE_STATE_CONFIGURED)); return (Z_ERR); } break; case CMD_ATTACH: case CMD_CLONE: case CMD_INSTALL: if (state == ZONE_STATE_INSTALLED) { zerror(gettext("is already %s."), zone_state_str(ZONE_STATE_INSTALLED)); return (Z_ERR); } else if (state == ZONE_STATE_INCOMPLETE) { zerror(gettext("zone is %s; %s required."), zone_state_str(ZONE_STATE_INCOMPLETE), cmd_to_str(CMD_UNINSTALL)); return (Z_ERR); } break; case CMD_DETACH: case CMD_MOVE: case CMD_READY: case CMD_BOOT: case CMD_MOUNT: case CMD_MARK: if ((cmd_num == CMD_BOOT || cmd_num == CMD_MOUNT) && force) min_state = ZONE_STATE_INCOMPLETE; else min_state = ZONE_STATE_INSTALLED; if (force && cmd_num == CMD_BOOT && is_native_zone) { zerror(gettext("Only branded zones may be " "force-booted.")); return (Z_ERR); } if (state < min_state) { zerror(gettext("must be %s before %s."), zone_state_str(min_state), cmd_to_str(cmd_num)); return (Z_ERR); } break; case CMD_VERIFY: if (state == ZONE_STATE_INCOMPLETE) { zerror(gettext("zone is %s; %s required."), zone_state_str(ZONE_STATE_INCOMPLETE), cmd_to_str(CMD_UNINSTALL)); return (Z_ERR); } break; case CMD_UNMOUNT: if (state != ZONE_STATE_MOUNTED) { zerror(gettext("must be %s before %s."), zone_state_str(ZONE_STATE_MOUNTED), cmd_to_str(cmd_num)); return (Z_ERR); } break; } } return (Z_OK); } static int halt_func(int argc, char *argv[]) { zone_cmd_arg_t zarg; int arg; if (zonecfg_in_alt_root()) { zerror(gettext("cannot halt zone in alternate root")); return (Z_ERR); } optind = 0; if ((arg = getopt(argc, argv, "?")) != EOF) { switch (arg) { case '?': sub_usage(SHELP_HALT, CMD_HALT); return (optopt == '?' ? Z_OK : Z_USAGE); default: sub_usage(SHELP_HALT, CMD_HALT); return (Z_USAGE); } } if (argc > optind) { sub_usage(SHELP_HALT, CMD_HALT); return (Z_USAGE); } /* * zoneadmd should be the one to decide whether or not to proceed, * so even though it seems that the fourth parameter below should * perhaps be B_TRUE, it really shouldn't be. */ if (sanity_check(target_zone, CMD_HALT, B_FALSE, B_FALSE, B_FALSE) != Z_OK) return (Z_ERR); /* * Invoke brand-specific handler. */ if (invoke_brand_handler(CMD_HALT, argv) != Z_OK) return (Z_ERR); zarg.cmd = Z_HALT; return ((call_zoneadmd(target_zone, &zarg) == 0) ? Z_OK : Z_ERR); } static int reboot_func(int argc, char *argv[]) { zone_cmd_arg_t zarg; int arg; if (zonecfg_in_alt_root()) { zerror(gettext("cannot reboot zone in alternate root")); return (Z_ERR); } optind = 0; if ((arg = getopt(argc, argv, "?")) != EOF) { switch (arg) { case '?': sub_usage(SHELP_REBOOT, CMD_REBOOT); return (optopt == '?' ? Z_OK : Z_USAGE); default: sub_usage(SHELP_REBOOT, CMD_REBOOT); return (Z_USAGE); } } zarg.bootbuf[0] = '\0'; for (; optind < argc; optind++) { if (strlcat(zarg.bootbuf, argv[optind], sizeof (zarg.bootbuf)) >= sizeof (zarg.bootbuf)) { zerror(gettext("Boot argument list too long")); return (Z_ERR); } if (optind < argc - 1) if (strlcat(zarg.bootbuf, " ", sizeof (zarg.bootbuf)) >= sizeof (zarg.bootbuf)) { zerror(gettext("Boot argument list too long")); return (Z_ERR); } } /* * zoneadmd should be the one to decide whether or not to proceed, * so even though it seems that the fourth parameter below should * perhaps be B_TRUE, it really shouldn't be. */ if (sanity_check(target_zone, CMD_REBOOT, B_TRUE, B_FALSE, B_FALSE) != Z_OK) return (Z_ERR); if (verify_details(CMD_REBOOT, argv) != Z_OK) return (Z_ERR); zarg.cmd = Z_REBOOT; return ((call_zoneadmd(target_zone, &zarg) == 0) ? Z_OK : Z_ERR); } static int verify_brand(zone_dochandle_t handle, int cmd_num, char *argv[]) { char cmdbuf[MAXPATHLEN]; int err; char zonepath[MAXPATHLEN]; brand_handle_t bh = NULL; int status, i; /* * Fetch the verify command from the brand configuration. * "exec" the command so that the returned status is that of * the command and not the shell. */ if ((err = zonecfg_get_zonepath(handle, zonepath, sizeof (zonepath))) != Z_OK) { errno = err; zperror(cmd_to_str(cmd_num), B_TRUE); return (Z_ERR); } if ((bh = brand_open(target_brand)) == NULL) { zerror(gettext("missing or invalid brand")); return (Z_ERR); } /* * If the brand has its own verification routine, execute it now. * The verification routine validates the intended zoneadm * operation for the specific brand. The zoneadm subcommand and * all its arguments are passed to the routine. */ (void) strcpy(cmdbuf, EXEC_PREFIX); err = brand_get_verify_adm(bh, target_zone, zonepath, cmdbuf + EXEC_LEN, sizeof (cmdbuf) - EXEC_LEN, 0, NULL); brand_close(bh); if (err != 0) return (Z_BRAND_ERROR); if (strlen(cmdbuf) <= EXEC_LEN) return (Z_OK); if (strlcat(cmdbuf, cmd_to_str(cmd_num), sizeof (cmdbuf)) >= sizeof (cmdbuf)) return (Z_ERR); /* Build the argv string */ i = 0; while (argv[i] != NULL) { if ((strlcat(cmdbuf, " ", sizeof (cmdbuf)) >= sizeof (cmdbuf)) || (strlcat(cmdbuf, argv[i++], sizeof (cmdbuf)) >= sizeof (cmdbuf))) return (Z_ERR); } if (zoneadm_is_nested) status = do_subproc(cmdbuf); else status = do_subproc_interactive(cmdbuf); err = subproc_status(gettext("brand-specific verification"), status, B_FALSE); return ((err == ZONE_SUBPROC_OK) ? Z_OK : Z_BRAND_ERROR); } static int verify_rctls(zone_dochandle_t handle) { struct zone_rctltab rctltab; size_t rbs = rctlblk_size(); rctlblk_t *rctlblk; int error = Z_INVAL; if ((rctlblk = malloc(rbs)) == NULL) { zerror(gettext("failed to allocate %lu bytes: %s"), rbs, strerror(errno)); return (Z_NOMEM); } if (zonecfg_setrctlent(handle) != Z_OK) { zerror(gettext("zonecfg_setrctlent failed")); free(rctlblk); return (error); } rctltab.zone_rctl_valptr = NULL; while (zonecfg_getrctlent(handle, &rctltab) == Z_OK) { struct zone_rctlvaltab *rctlval; const char *name = rctltab.zone_rctl_name; if (!zonecfg_is_rctl(name)) { zerror(gettext("WARNING: Ignoring unrecognized rctl " "'%s'."), name); zonecfg_free_rctl_value_list(rctltab.zone_rctl_valptr); rctltab.zone_rctl_valptr = NULL; continue; } for (rctlval = rctltab.zone_rctl_valptr; rctlval != NULL; rctlval = rctlval->zone_rctlval_next) { if (zonecfg_construct_rctlblk(rctlval, rctlblk) != Z_OK) { zerror(gettext("invalid rctl value: " "(priv=%s,limit=%s,action%s)"), rctlval->zone_rctlval_priv, rctlval->zone_rctlval_limit, rctlval->zone_rctlval_action); goto out; } if (!zonecfg_valid_rctl(name, rctlblk)) { zerror(gettext("(priv=%s,limit=%s,action=%s) " "is not a valid value for rctl '%s'"), rctlval->zone_rctlval_priv, rctlval->zone_rctlval_limit, rctlval->zone_rctlval_action, name); goto out; } } zonecfg_free_rctl_value_list(rctltab.zone_rctl_valptr); } rctltab.zone_rctl_valptr = NULL; error = Z_OK; out: zonecfg_free_rctl_value_list(rctltab.zone_rctl_valptr); (void) zonecfg_endrctlent(handle); free(rctlblk); return (error); } static int verify_pool(zone_dochandle_t handle) { char poolname[MAXPATHLEN]; pool_conf_t *poolconf; pool_t *pool; int status; int error; /* * This ends up being very similar to the check done in zoneadmd. */ error = zonecfg_get_pool(handle, poolname, sizeof (poolname)); if (error == Z_NO_ENTRY || (error == Z_OK && strlen(poolname) == 0)) { /* * No pool specified. */ return (0); } if (error != Z_OK) { zperror(gettext("Unable to retrieve pool name from " "configuration"), B_TRUE); return (error); } /* * Don't do anything if pools aren't enabled. */ if (pool_get_status(&status) != PO_SUCCESS || status != POOL_ENABLED) { zerror(gettext("WARNING: pools facility not active; " "zone will not be bound to pool '%s'."), poolname); return (Z_OK); } /* * Try to provide a sane error message if the requested pool doesn't * exist. It isn't clear that pools-related failures should * necessarily translate to a failure to verify the zone configuration, * hence they are not considered errors. */ if ((poolconf = pool_conf_alloc()) == NULL) { zerror(gettext("WARNING: pool_conf_alloc failed; " "using default pool")); return (Z_OK); } if (pool_conf_open(poolconf, pool_dynamic_location(), PO_RDONLY) != PO_SUCCESS) { zerror(gettext("WARNING: pool_conf_open failed; " "using default pool")); pool_conf_free(poolconf); return (Z_OK); } pool = pool_get_pool(poolconf, poolname); (void) pool_conf_close(poolconf); pool_conf_free(poolconf); if (pool == NULL) { zerror(gettext("WARNING: pool '%s' not found. " "using default pool"), poolname); } return (Z_OK); } static int verify_ipd(zone_dochandle_t handle) { int return_code = Z_OK; struct zone_fstab fstab; struct stat st; char specdir[MAXPATHLEN]; if (zonecfg_setipdent(handle) != Z_OK) { /* * TRANSLATION_NOTE * inherit-pkg-dirs is a literal that should not be translated. */ (void) fprintf(stderr, gettext("could not verify " "inherit-pkg-dirs: unable to enumerate mounts\n")); return (Z_ERR); } while (zonecfg_getipdent(handle, &fstab) == Z_OK) { /* * Verify fs_dir exists. */ (void) snprintf(specdir, sizeof (specdir), "%s%s", zonecfg_get_root(), fstab.zone_fs_dir); if (stat(specdir, &st) != 0) { /* * TRANSLATION_NOTE * inherit-pkg-dir is a literal that should not be * translated. */ (void) fprintf(stderr, gettext("could not verify " "inherit-pkg-dir %s: %s\n"), fstab.zone_fs_dir, strerror(errno)); return_code = Z_ERR; } if (strcmp(st.st_fstype, MNTTYPE_NFS) == 0) { /* * TRANSLATION_NOTE * inherit-pkg-dir and NFS are literals that should * not be translated. */ (void) fprintf(stderr, gettext("cannot verify " "inherit-pkg-dir %s: NFS mounted file system.\n" "\tA local file system must be used.\n"), fstab.zone_fs_dir); return_code = Z_ERR; } } (void) zonecfg_endipdent(handle); return (return_code); } /* * Verify that the special device/file system exists and is valid. */ static int verify_fs_special(struct zone_fstab *fstab) { struct stat st; /* * This validation is really intended for standard zone administration. * If we are in a mini-root or some other upgrade situation where * we are using the scratch zone, just by-pass this. */ if (zonecfg_in_alt_root()) return (Z_OK); if (strcmp(fstab->zone_fs_type, MNTTYPE_ZFS) == 0) return (verify_fs_zfs(fstab)); if (stat(fstab->zone_fs_special, &st) != 0) { (void) fprintf(stderr, gettext("could not verify fs " "%s: could not access %s: %s\n"), fstab->zone_fs_dir, fstab->zone_fs_special, strerror(errno)); return (Z_ERR); } if (strcmp(st.st_fstype, MNTTYPE_NFS) == 0) { /* * TRANSLATION_NOTE * fs and NFS are literals that should * not be translated. */ (void) fprintf(stderr, gettext("cannot verify " "fs %s: NFS mounted file system.\n" "\tA local file system must be used.\n"), fstab->zone_fs_special); return (Z_ERR); } return (Z_OK); } static int verify_filesystems(zone_dochandle_t handle) { int return_code = Z_OK; struct zone_fstab fstab; char cmdbuf[MAXPATHLEN]; struct stat st; /* * No need to verify inherit-pkg-dir fs types, as their type is * implicitly lofs, which is known. Therefore, the types are only * verified for regular file systems below. * * Since the actual mount point is not known until the dependent mounts * are performed, we don't attempt any path validation here: that will * happen later when zoneadmd actually does the mounts. */ if (zonecfg_setfsent(handle) != Z_OK) { (void) fprintf(stderr, gettext("could not verify file systems: " "unable to enumerate mounts\n")); return (Z_ERR); } while (zonecfg_getfsent(handle, &fstab) == Z_OK) { if (!zonecfg_valid_fs_type(fstab.zone_fs_type)) { (void) fprintf(stderr, gettext("cannot verify fs %s: " "type %s is not allowed.\n"), fstab.zone_fs_dir, fstab.zone_fs_type); return_code = Z_ERR; goto next_fs; } /* * Verify /usr/lib/fs//mount exists. */ if (snprintf(cmdbuf, sizeof (cmdbuf), "/usr/lib/fs/%s/mount", fstab.zone_fs_type) > sizeof (cmdbuf)) { (void) fprintf(stderr, gettext("cannot verify fs %s: " "type %s is too long.\n"), fstab.zone_fs_dir, fstab.zone_fs_type); return_code = Z_ERR; goto next_fs; } if (stat(cmdbuf, &st) != 0) { (void) fprintf(stderr, gettext("could not verify fs " "%s: could not access %s: %s\n"), fstab.zone_fs_dir, cmdbuf, strerror(errno)); return_code = Z_ERR; goto next_fs; } if (!S_ISREG(st.st_mode)) { (void) fprintf(stderr, gettext("could not verify fs " "%s: %s is not a regular file\n"), fstab.zone_fs_dir, cmdbuf); return_code = Z_ERR; goto next_fs; } /* * Verify /usr/lib/fs//fsck exists iff zone_fs_raw is * set. */ if (snprintf(cmdbuf, sizeof (cmdbuf), "/usr/lib/fs/%s/fsck", fstab.zone_fs_type) > sizeof (cmdbuf)) { (void) fprintf(stderr, gettext("cannot verify fs %s: " "type %s is too long.\n"), fstab.zone_fs_dir, fstab.zone_fs_type); return_code = Z_ERR; goto next_fs; } if (fstab.zone_fs_raw[0] == '\0' && stat(cmdbuf, &st) == 0) { (void) fprintf(stderr, gettext("could not verify fs " "%s: must specify 'raw' device for %s " "file systems\n"), fstab.zone_fs_dir, fstab.zone_fs_type); return_code = Z_ERR; goto next_fs; } if (fstab.zone_fs_raw[0] != '\0' && (stat(cmdbuf, &st) != 0 || !S_ISREG(st.st_mode))) { (void) fprintf(stderr, gettext("cannot verify fs %s: " "'raw' device specified but " "no fsck executable exists for %s\n"), fstab.zone_fs_dir, fstab.zone_fs_type); return_code = Z_ERR; goto next_fs; } /* Verify fs_special. */ if ((return_code = verify_fs_special(&fstab)) != Z_OK) goto next_fs; /* Verify fs_raw. */ if (fstab.zone_fs_raw[0] != '\0' && stat(fstab.zone_fs_raw, &st) != 0) { /* * TRANSLATION_NOTE * fs is a literal that should not be translated. */ (void) fprintf(stderr, gettext("could not verify fs " "%s: could not access %s: %s\n"), fstab.zone_fs_dir, fstab.zone_fs_raw, strerror(errno)); return_code = Z_ERR; goto next_fs; } next_fs: zonecfg_free_fs_option_list(fstab.zone_fs_options); } (void) zonecfg_endfsent(handle); return (return_code); } static int verify_limitpriv(zone_dochandle_t handle) { char *privname = NULL; int err; priv_set_t *privs; if ((privs = priv_allocset()) == NULL) { zperror(gettext("failed to allocate privilege set"), B_FALSE); return (Z_NOMEM); } err = zonecfg_get_privset(handle, privs, &privname); switch (err) { case Z_OK: break; case Z_PRIV_PROHIBITED: (void) fprintf(stderr, gettext("privilege \"%s\" is not " "permitted within the zone's privilege set\n"), privname); break; case Z_PRIV_REQUIRED: (void) fprintf(stderr, gettext("required privilege \"%s\" is " "missing from the zone's privilege set\n"), privname); break; case Z_PRIV_UNKNOWN: (void) fprintf(stderr, gettext("unknown privilege \"%s\" " "specified in the zone's privilege set\n"), privname); break; default: zperror( gettext("failed to determine the zone's privilege set"), B_TRUE); break; } free(privname); priv_freeset(privs); return (err); } static void free_local_netifs(int if_cnt, struct net_if **if_list) { int i; for (i = 0; i < if_cnt; i++) { free(if_list[i]->name); free(if_list[i]); } free(if_list); } /* * Get a list of the network interfaces, along with their address families, * that are plumbed in the global zone. See if_tcp(7p) for a description * of the ioctls used here. */ static int get_local_netifs(int *if_cnt, struct net_if ***if_list) { int s; int i; int res = Z_OK; int space_needed; int cnt = 0; struct lifnum if_num; struct lifconf if_conf; struct lifreq *if_reqp; char *if_buf; struct net_if **local_ifs = NULL; *if_cnt = 0; *if_list = NULL; if ((s = socket(SOCKET_AF(AF_INET), SOCK_DGRAM, 0)) < 0) return (Z_ERR); /* * Come back here in the unlikely event that the number of interfaces * increases between the time we get the count and the time we do the * SIOCGLIFCONF ioctl. */ retry: /* Get the number of interfaces. */ if_num.lifn_family = AF_UNSPEC; if_num.lifn_flags = LIFC_NOXMIT; if (ioctl(s, SIOCGLIFNUM, &if_num) < 0) { (void) close(s); return (Z_ERR); } /* Get the interface configuration list. */ space_needed = if_num.lifn_count * sizeof (struct lifreq); if ((if_buf = malloc(space_needed)) == NULL) { (void) close(s); return (Z_ERR); } if_conf.lifc_family = AF_UNSPEC; if_conf.lifc_flags = LIFC_NOXMIT; if_conf.lifc_len = space_needed; if_conf.lifc_buf = if_buf; if (ioctl(s, SIOCGLIFCONF, &if_conf) < 0) { free(if_buf); /* * SIOCGLIFCONF returns EINVAL if the buffer we passed in is * too small. In this case go back and get the new if cnt. */ if (errno == EINVAL) goto retry; (void) close(s); return (Z_ERR); } (void) close(s); /* Get the name and address family for each interface. */ if_reqp = if_conf.lifc_req; for (i = 0; i < (if_conf.lifc_len / sizeof (struct lifreq)); i++) { struct net_if **p; struct lifreq req; if (strcmp(LOOPBACK_IF, if_reqp->lifr_name) == 0) { if_reqp++; continue; } if ((s = socket(SOCKET_AF(if_reqp->lifr_addr.ss_family), SOCK_DGRAM, 0)) == -1) { res = Z_ERR; break; } (void) strncpy(req.lifr_name, if_reqp->lifr_name, sizeof (req.lifr_name)); if (ioctl(s, SIOCGLIFADDR, &req) < 0) { (void) close(s); if_reqp++; continue; } if ((p = (struct net_if **)realloc(local_ifs, sizeof (struct net_if *) * (cnt + 1))) == NULL) { res = Z_ERR; break; } local_ifs = p; if ((local_ifs[cnt] = malloc(sizeof (struct net_if))) == NULL) { res = Z_ERR; break; } if ((local_ifs[cnt]->name = strdup(if_reqp->lifr_name)) == NULL) { free(local_ifs[cnt]); res = Z_ERR; break; } local_ifs[cnt]->af = req.lifr_addr.ss_family; cnt++; (void) close(s); if_reqp++; } free(if_buf); if (res != Z_OK) { free_local_netifs(cnt, local_ifs); } else { *if_cnt = cnt; *if_list = local_ifs; } return (res); } static char * af2str(int af) { switch (af) { case AF_INET: return ("IPv4"); case AF_INET6: return ("IPv6"); default: return ("Unknown"); } } /* * Cross check the network interface name and address family with the * interfaces that are set up in the global zone so that we can print the * appropriate error message. */ static void print_net_err(char *phys, char *addr, int af, char *msg) { int i; int local_if_cnt = 0; struct net_if **local_ifs = NULL; boolean_t found_if = B_FALSE; boolean_t found_af = B_FALSE; if (get_local_netifs(&local_if_cnt, &local_ifs) != Z_OK) { (void) fprintf(stderr, gettext("could not verify %s %s=%s %s=%s\n\t%s\n"), "net", "address", addr, "physical", phys, msg); return; } for (i = 0; i < local_if_cnt; i++) { if (strcmp(phys, local_ifs[i]->name) == 0) { found_if = B_TRUE; if (af == local_ifs[i]->af) { found_af = B_TRUE; break; } } } free_local_netifs(local_if_cnt, local_ifs); if (!found_if) { (void) fprintf(stderr, gettext("could not verify %s %s=%s\n\t" "network interface %s is not plumbed in the global zone\n"), "net", "physical", phys, phys); return; } /* * Print this error if we were unable to find the address family * for this interface. If the af variable is not initialized to * to something meaningful by the caller (not AF_UNSPEC) then we * also skip this message since it wouldn't be informative. */ if (!found_af && af != AF_UNSPEC) { (void) fprintf(stderr, gettext("could not verify %s %s=%s %s=%s\n\tthe %s address " "family is not configured on this network interface in " "the\n\tglobal zone\n"), "net", "address", addr, "physical", phys, af2str(af)); return; } (void) fprintf(stderr, gettext("could not verify %s %s=%s %s=%s\n\t%s\n"), "net", "address", addr, "physical", phys, msg); } static int verify_handle(int cmd_num, zone_dochandle_t handle, char *argv[]) { struct zone_nwiftab nwiftab; int return_code = Z_OK; int err; boolean_t in_alt_root; zone_iptype_t iptype; int fd; in_alt_root = zonecfg_in_alt_root(); if (in_alt_root) goto no_net; if ((err = zonecfg_get_iptype(handle, &iptype)) != Z_OK) { errno = err; zperror(cmd_to_str(cmd_num), B_TRUE); zonecfg_fini_handle(handle); return (Z_ERR); } if ((err = zonecfg_setnwifent(handle)) != Z_OK) { errno = err; zperror(cmd_to_str(cmd_num), B_TRUE); zonecfg_fini_handle(handle); return (Z_ERR); } while (zonecfg_getnwifent(handle, &nwiftab) == Z_OK) { struct lifreq lifr; sa_family_t af = AF_UNSPEC; char dl_owner_zname[ZONENAME_MAX]; zoneid_t dl_owner_zid; zoneid_t target_zid; int res; /* skip any loopback interfaces */ if (strcmp(nwiftab.zone_nwif_physical, "lo0") == 0) continue; switch (iptype) { case ZS_SHARED: if ((res = zonecfg_valid_net_address( nwiftab.zone_nwif_address, &lifr)) != Z_OK) { print_net_err(nwiftab.zone_nwif_physical, nwiftab.zone_nwif_address, af, zonecfg_strerror(res)); return_code = Z_ERR; continue; } af = lifr.lifr_addr.ss_family; if (!zonecfg_ifname_exists(af, nwiftab.zone_nwif_physical)) { /* * The interface failed to come up. We continue * on anyway for the sake of consistency: a * zone is not shut down if the interface fails * any time after boot, nor does the global zone * fail to boot if an interface fails. */ (void) fprintf(stderr, gettext("WARNING: skipping network " "interface '%s' which may not be " "present/plumbed in the global " "zone.\n"), nwiftab.zone_nwif_physical); } break; case ZS_EXCLUSIVE: /* Warning if it exists for either IPv4 or IPv6 */ if (zonecfg_ifname_exists(AF_INET, nwiftab.zone_nwif_physical) || zonecfg_ifname_exists(AF_INET6, nwiftab.zone_nwif_physical)) { (void) fprintf(stderr, gettext("WARNING: skipping network " "interface '%s' which is used in the " "global zone.\n"), nwiftab.zone_nwif_physical); break; } /* * Verify that the physical interface can * be opened */ fd = ifname_open(nwiftab.zone_nwif_physical); if (fd == -1) { (void) fprintf(stderr, gettext("WARNING: skipping network " "interface '%s' which cannot be opened.\n"), nwiftab.zone_nwif_physical); break; } else { (void) close(fd); } /* * Verify whether the physical interface is already * used by a zone. */ dl_owner_zid = ALL_ZONES; if (zone_check_datalink(&dl_owner_zid, nwiftab.zone_nwif_physical) != 0) break; /* * If the zone being verified is * running and owns the interface */ target_zid = getzoneidbyname(target_zone); if (target_zid == dl_owner_zid) break; /* Zone id match failed, use name to check */ if (getzonenamebyid(dl_owner_zid, dl_owner_zname, ZONENAME_MAX) < 0) { /* No name, show ID instead */ (void) snprintf(dl_owner_zname, ZONENAME_MAX, "<%d>", dl_owner_zid); } else if (strcmp(dl_owner_zname, target_zone) == 0) break; /* * Note here we only report a warning that * the interface is already in use by another * running zone, and the verify process just * goes on, if the interface is still in use * when this zone really boots up, zoneadmd * will find it. If the name of the zone which * owns this interface cannot be determined, * then it is not possible to determine if there * is a conflict so just report it as a warning. */ (void) fprintf(stderr, gettext("WARNING: skipping network interface " "'%s' which is used by the non-global zone " "'%s'.\n"), nwiftab.zone_nwif_physical, dl_owner_zname); break; } } (void) zonecfg_endnwifent(handle); no_net: /* verify that lofs has not been excluded from the kernel */ if (!(cmd_num == CMD_DETACH || cmd_num == CMD_ATTACH || cmd_num == CMD_MOVE || cmd_num == CMD_CLONE) && modctl(MODLOAD, 1, "fs/lofs", NULL) != 0) { if (errno == ENXIO) (void) fprintf(stderr, gettext("could not verify " "lofs(7FS): possibly excluded in /etc/system\n")); else (void) fprintf(stderr, gettext("could not verify " "lofs(7FS): %s\n"), strerror(errno)); return_code = Z_ERR; } if (verify_filesystems(handle) != Z_OK) return_code = Z_ERR; if (verify_ipd(handle) != Z_OK) return_code = Z_ERR; if (!in_alt_root && verify_rctls(handle) != Z_OK) return_code = Z_ERR; if (!in_alt_root && verify_pool(handle) != Z_OK) return_code = Z_ERR; if (!in_alt_root && verify_brand(handle, cmd_num, argv) != Z_OK) return_code = Z_ERR; if (!in_alt_root && verify_datasets(handle) != Z_OK) return_code = Z_ERR; /* * As the "mount" command is used for patching/upgrading of zones * or other maintenance processes, the zone's privilege set is not * checked in this case. Instead, the default, safe set of * privileges will be used when this zone is created in the * kernel. */ if (!in_alt_root && cmd_num != CMD_MOUNT && verify_limitpriv(handle) != Z_OK) return_code = Z_ERR; return (return_code); } static int verify_details(int cmd_num, char *argv[]) { zone_dochandle_t handle; char zonepath[MAXPATHLEN], checkpath[MAXPATHLEN]; int return_code = Z_OK; int err; if ((handle = zonecfg_init_handle()) == NULL) { zperror(cmd_to_str(cmd_num), B_TRUE); return (Z_ERR); } if ((err = zonecfg_get_handle(target_zone, handle)) != Z_OK) { errno = err; zperror(cmd_to_str(cmd_num), B_TRUE); zonecfg_fini_handle(handle); return (Z_ERR); } if ((err = zonecfg_get_zonepath(handle, zonepath, sizeof (zonepath))) != Z_OK) { errno = err; zperror(cmd_to_str(cmd_num), B_TRUE); zonecfg_fini_handle(handle); return (Z_ERR); } /* * zonecfg_get_zonepath() gets its data from the XML repository. * Verify this against the index file, which is checked first by * zone_get_zonepath(). If they don't match, bail out. */ if ((err = zone_get_zonepath(target_zone, checkpath, sizeof (checkpath))) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not get zone path")); zonecfg_fini_handle(handle); return (Z_ERR); } if (strcmp(zonepath, checkpath) != 0) { /* * TRANSLATION_NOTE * XML and zonepath are literals that should not be translated. */ (void) fprintf(stderr, gettext("The XML repository has " "zonepath '%s',\nbut the index file has zonepath '%s'.\n" "These must match, so fix the incorrect entry.\n"), zonepath, checkpath); zonecfg_fini_handle(handle); return (Z_ERR); } if (validate_zonepath(zonepath, cmd_num) != Z_OK) { (void) fprintf(stderr, gettext("could not verify zonepath %s " "because of the above errors.\n"), zonepath); return_code = Z_ERR; } if (verify_handle(cmd_num, handle, argv) != Z_OK) return_code = Z_ERR; zonecfg_fini_handle(handle); if (return_code == Z_ERR) (void) fprintf(stderr, gettext("%s: zone %s failed to verify\n"), execname, target_zone); return (return_code); } static int verify_func(int argc, char *argv[]) { int arg; optind = 0; if ((arg = getopt(argc, argv, "?")) != EOF) { switch (arg) { case '?': sub_usage(SHELP_VERIFY, CMD_VERIFY); return (optopt == '?' ? Z_OK : Z_USAGE); default: sub_usage(SHELP_VERIFY, CMD_VERIFY); return (Z_USAGE); } } if (argc > optind) { sub_usage(SHELP_VERIFY, CMD_VERIFY); return (Z_USAGE); } if (sanity_check(target_zone, CMD_VERIFY, B_FALSE, B_FALSE, B_FALSE) != Z_OK) return (Z_ERR); return (verify_details(CMD_VERIFY, argv)); } static int addopt(char *buf, int opt, char *optarg, size_t bufsize) { char optstring[4]; if (opt > 0) (void) sprintf(optstring, " -%c", opt); else (void) strcpy(optstring, " "); if ((strlcat(buf, optstring, bufsize) > bufsize)) return (Z_ERR); if ((optarg != NULL) && (strlcat(buf, optarg, bufsize) > bufsize)) return (Z_ERR); return (Z_OK); } static int install_func(int argc, char *argv[]) { char cmdbuf[MAXPATHLEN]; int lockfd; int arg, err, subproc_err; char zonepath[MAXPATHLEN]; brand_handle_t bh = NULL; int status; boolean_t nodataset = B_FALSE; char opts[128]; if (target_zone == NULL) { sub_usage(SHELP_INSTALL, CMD_INSTALL); return (Z_USAGE); } if (zonecfg_in_alt_root()) { zerror(gettext("cannot install zone in alternate root")); return (Z_ERR); } if ((err = zone_get_zonepath(target_zone, zonepath, sizeof (zonepath))) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not get zone path")); return (Z_ERR); } /* Fetch the install command from the brand configuration. */ if ((bh = brand_open(target_brand)) == NULL) { zerror(gettext("missing or invalid brand")); return (Z_ERR); } (void) strcpy(cmdbuf, EXEC_PREFIX); if (brand_get_install(bh, target_zone, zonepath, cmdbuf + EXEC_LEN, sizeof (cmdbuf) - EXEC_LEN, 0, NULL) != 0) { zerror("invalid brand configuration: missing install resource"); brand_close(bh); return (Z_ERR); } (void) strcpy(opts, "?x:"); if (!is_native_zone) { /* * Fetch the list of recognized command-line options from * the brand configuration file. */ if (brand_get_installopts(bh, opts + strlen(opts), sizeof (opts) - strlen(opts)) != 0) { zerror("invalid brand configuration: missing " "install options resource"); brand_close(bh); return (Z_ERR); } } brand_close(bh); optind = 0; while ((arg = getopt(argc, argv, opts)) != EOF) { switch (arg) { case '?': sub_usage(SHELP_INSTALL, CMD_INSTALL); return (optopt == '?' ? Z_OK : Z_USAGE); case 'x': if (strcmp(optarg, "nodataset") != 0) { sub_usage(SHELP_INSTALL, CMD_INSTALL); return (Z_USAGE); } nodataset = B_TRUE; break; default: if (is_native_zone) { sub_usage(SHELP_INSTALL, CMD_INSTALL); return (Z_USAGE); } /* * This option isn't for zoneadm, so append it to * the command line passed to the brand-specific * install routine. */ if (addopt(cmdbuf, optopt, optarg, sizeof (cmdbuf)) != Z_OK) { zerror("Install command line too long"); return (Z_ERR); } break; } } if (!is_native_zone) { for (; optind < argc; optind++) { if (addopt(cmdbuf, 0, argv[optind], sizeof (cmdbuf)) != Z_OK) { zerror("Install command line too long"); return (Z_ERR); } } } if (sanity_check(target_zone, CMD_INSTALL, B_FALSE, B_TRUE, B_FALSE) != Z_OK) return (Z_ERR); if (verify_details(CMD_INSTALL, argv) != Z_OK) return (Z_ERR); if (grab_lock_file(target_zone, &lockfd) != Z_OK) { zerror(gettext("another %s may have an operation in progress."), "zoneadm"); return (Z_ERR); } err = zone_set_state(target_zone, ZONE_STATE_INCOMPLETE); if (err != Z_OK) { errno = err; zperror2(target_zone, gettext("could not set state")); goto done; } if (!nodataset) create_zfs_zonepath(zonepath); /* * According to the Application Packaging Developer's Guide, a * "checkinstall" script when included in a package is executed as * the user "install", if such a user exists, or by the user * "nobody". In order to support this dubious behavior, the path * to the zone being constructed is opened up during the life of * the command laying down the zone's root file system. Once this * has completed, regardless of whether it was successful, the * path to the zone is again restricted. */ if (chmod(zonepath, DEFAULT_DIR_MODE) != 0) { zperror(zonepath, B_FALSE); err = Z_ERR; goto done; } if (is_native_zone) status = do_subproc(cmdbuf); else status = do_subproc_interactive(cmdbuf); if (chmod(zonepath, S_IRWXU) != 0) { zperror(zonepath, B_FALSE); err = Z_ERR; goto done; } if ((subproc_err = subproc_status(gettext("brand-specific installation"), status, B_FALSE)) != ZONE_SUBPROC_OK) { err = Z_ERR; goto done; } if ((err = zone_set_state(target_zone, ZONE_STATE_INSTALLED)) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not set state")); goto done; } done: /* * If the install script exited with ZONE_SUBPROC_USAGE or * ZONE_SUBPROC_NOTCOMPLETE, try to clean up the zone and leave the * zone in the CONFIGURED state so that another install can be * attempted without requiring an uninstall first. */ if ((subproc_err == ZONE_SUBPROC_USAGE) || (subproc_err == ZONE_SUBPROC_NOTCOMPLETE)) { if ((err = cleanup_zonepath(zonepath, B_FALSE)) != Z_OK) { errno = err; zperror2(target_zone, gettext("cleaning up zonepath failed")); } else if ((err = zone_set_state(target_zone, ZONE_STATE_CONFIGURED)) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not set state")); } } release_lock_file(lockfd); return ((err == Z_OK) ? Z_OK : Z_ERR); } /* * Check that the inherited pkg dirs are the same for the clone and its source. * The easiest way to do that is check that the list of ipds is the same * by matching each one against the other. This algorithm should be fine since * the list of ipds should not be that long. */ static int valid_ipd_clone(zone_dochandle_t s_handle, char *source_zone, zone_dochandle_t t_handle, char *target_zone) { int err; int res = Z_OK; int s_cnt = 0; int t_cnt = 0; struct zone_fstab s_fstab; struct zone_fstab t_fstab; /* * First check the source of the clone against the target. */ if ((err = zonecfg_setipdent(s_handle)) != Z_OK) { errno = err; zperror2(source_zone, gettext("could not enumerate " "inherit-pkg-dirs")); return (Z_ERR); } while (zonecfg_getipdent(s_handle, &s_fstab) == Z_OK) { boolean_t match = B_FALSE; s_cnt++; if ((err = zonecfg_setipdent(t_handle)) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not enumerate " "inherit-pkg-dirs")); (void) zonecfg_endipdent(s_handle); return (Z_ERR); } while (zonecfg_getipdent(t_handle, &t_fstab) == Z_OK) { if (strcmp(s_fstab.zone_fs_dir, t_fstab.zone_fs_dir) == 0) { match = B_TRUE; break; } } (void) zonecfg_endipdent(t_handle); if (!match) { (void) fprintf(stderr, gettext("inherit-pkg-dir " "'%s' is not configured in zone %s.\n"), s_fstab.zone_fs_dir, target_zone); res = Z_ERR; } } (void) zonecfg_endipdent(s_handle); /* skip the next check if we already have errors */ if (res == Z_ERR) return (res); /* * Now check the number of ipds in the target so we can verify * that the source is not a subset of the target. */ if ((err = zonecfg_setipdent(t_handle)) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not enumerate " "inherit-pkg-dirs")); return (Z_ERR); } while (zonecfg_getipdent(t_handle, &t_fstab) == Z_OK) t_cnt++; (void) zonecfg_endipdent(t_handle); if (t_cnt != s_cnt) { (void) fprintf(stderr, gettext("Zone %s is configured " "with inherit-pkg-dirs that are not configured in zone " "%s.\n"), target_zone, source_zone); res = Z_ERR; } return (res); } static void warn_dev_match(zone_dochandle_t s_handle, char *source_zone, zone_dochandle_t t_handle, char *target_zone) { int err; struct zone_devtab s_devtab; struct zone_devtab t_devtab; if ((err = zonecfg_setdevent(t_handle)) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not enumerate devices")); return; } while (zonecfg_getdevent(t_handle, &t_devtab) == Z_OK) { if ((err = zonecfg_setdevent(s_handle)) != Z_OK) { errno = err; zperror2(source_zone, gettext("could not enumerate devices")); (void) zonecfg_enddevent(t_handle); return; } while (zonecfg_getdevent(s_handle, &s_devtab) == Z_OK) { /* * Use fnmatch to catch the case where wildcards * were used in one zone and the other has an * explicit entry (e.g. /dev/dsk/c0t0d0s6 vs. * /dev/\*dsk/c0t0d0s6). */ if (fnmatch(t_devtab.zone_dev_match, s_devtab.zone_dev_match, FNM_PATHNAME) == 0 || fnmatch(s_devtab.zone_dev_match, t_devtab.zone_dev_match, FNM_PATHNAME) == 0) { (void) fprintf(stderr, gettext("WARNING: device '%s' " "is configured in both zones.\n"), t_devtab.zone_dev_match); break; } } (void) zonecfg_enddevent(s_handle); } (void) zonecfg_enddevent(t_handle); } /* * Check if the specified mount option (opt) is contained within the * options string. */ static boolean_t opt_match(char *opt, char *options) { char *p; char *lastp; if ((p = strtok_r(options, ",", &lastp)) != NULL) { if (strcmp(p, opt) == 0) return (B_TRUE); while ((p = strtok_r(NULL, ",", &lastp)) != NULL) { if (strcmp(p, opt) == 0) return (B_TRUE); } } return (B_FALSE); } #define RW_LOFS "WARNING: read-write lofs file system on '%s' is configured " \ "in both zones.\n" static void print_fs_warnings(struct zone_fstab *s_fstab, struct zone_fstab *t_fstab) { /* * It is ok to have shared lofs mounted fs but we want to warn if * either is rw since this will effect the other zone. */ if (strcmp(t_fstab->zone_fs_type, "lofs") == 0) { zone_fsopt_t *optp; /* The default is rw so no options means rw */ if (t_fstab->zone_fs_options == NULL || s_fstab->zone_fs_options == NULL) { (void) fprintf(stderr, gettext(RW_LOFS), t_fstab->zone_fs_special); return; } for (optp = s_fstab->zone_fs_options; optp != NULL; optp = optp->zone_fsopt_next) { if (opt_match("rw", optp->zone_fsopt_opt)) { (void) fprintf(stderr, gettext(RW_LOFS), s_fstab->zone_fs_special); return; } } for (optp = t_fstab->zone_fs_options; optp != NULL; optp = optp->zone_fsopt_next) { if (opt_match("rw", optp->zone_fsopt_opt)) { (void) fprintf(stderr, gettext(RW_LOFS), t_fstab->zone_fs_special); return; } } return; } /* * TRANSLATION_NOTE * The first variable is the file system type and the second is * the file system special device. For example, * WARNING: ufs file system on '/dev/dsk/c0t0d0s0' ... */ (void) fprintf(stderr, gettext("WARNING: %s file system on '%s' " "is configured in both zones.\n"), t_fstab->zone_fs_type, t_fstab->zone_fs_special); } static void warn_fs_match(zone_dochandle_t s_handle, char *source_zone, zone_dochandle_t t_handle, char *target_zone) { int err; struct zone_fstab s_fstab; struct zone_fstab t_fstab; if ((err = zonecfg_setfsent(t_handle)) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not enumerate file systems")); return; } while (zonecfg_getfsent(t_handle, &t_fstab) == Z_OK) { if ((err = zonecfg_setfsent(s_handle)) != Z_OK) { errno = err; zperror2(source_zone, gettext("could not enumerate file systems")); (void) zonecfg_endfsent(t_handle); return; } while (zonecfg_getfsent(s_handle, &s_fstab) == Z_OK) { if (strcmp(t_fstab.zone_fs_special, s_fstab.zone_fs_special) == 0) { print_fs_warnings(&s_fstab, &t_fstab); break; } } (void) zonecfg_endfsent(s_handle); } (void) zonecfg_endfsent(t_handle); } /* * We don't catch the case where you used the same IP address but * it is not an exact string match. For example, 192.9.0.128 vs. 192.09.0.128. * However, we're not going to worry about that but we will check for * a possible netmask on one of the addresses (e.g. 10.0.0.1 and 10.0.0.1/24) * and handle that case as a match. */ static void warn_ip_match(zone_dochandle_t s_handle, char *source_zone, zone_dochandle_t t_handle, char *target_zone) { int err; struct zone_nwiftab s_nwiftab; struct zone_nwiftab t_nwiftab; if ((err = zonecfg_setnwifent(t_handle)) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not enumerate network interfaces")); return; } while (zonecfg_getnwifent(t_handle, &t_nwiftab) == Z_OK) { char *p; /* remove an (optional) netmask from the address */ if ((p = strchr(t_nwiftab.zone_nwif_address, '/')) != NULL) *p = '\0'; if ((err = zonecfg_setnwifent(s_handle)) != Z_OK) { errno = err; zperror2(source_zone, gettext("could not enumerate network interfaces")); (void) zonecfg_endnwifent(t_handle); return; } while (zonecfg_getnwifent(s_handle, &s_nwiftab) == Z_OK) { /* remove an (optional) netmask from the address */ if ((p = strchr(s_nwiftab.zone_nwif_address, '/')) != NULL) *p = '\0'; /* For exclusive-IP zones, address is not specified. */ if (strlen(s_nwiftab.zone_nwif_address) == 0) continue; if (strcmp(t_nwiftab.zone_nwif_address, s_nwiftab.zone_nwif_address) == 0) { (void) fprintf(stderr, gettext("WARNING: network address '%s' " "is configured in both zones.\n"), t_nwiftab.zone_nwif_address); break; } } (void) zonecfg_endnwifent(s_handle); } (void) zonecfg_endnwifent(t_handle); } static void warn_dataset_match(zone_dochandle_t s_handle, char *source, zone_dochandle_t t_handle, char *target) { int err; struct zone_dstab s_dstab; struct zone_dstab t_dstab; if ((err = zonecfg_setdsent(t_handle)) != Z_OK) { errno = err; zperror2(target, gettext("could not enumerate datasets")); return; } while (zonecfg_getdsent(t_handle, &t_dstab) == Z_OK) { if ((err = zonecfg_setdsent(s_handle)) != Z_OK) { errno = err; zperror2(source, gettext("could not enumerate datasets")); (void) zonecfg_enddsent(t_handle); return; } while (zonecfg_getdsent(s_handle, &s_dstab) == Z_OK) { if (strcmp(t_dstab.zone_dataset_name, s_dstab.zone_dataset_name) == 0) { target_zone = source; zerror(gettext("WARNING: dataset '%s' " "is configured in both zones.\n"), t_dstab.zone_dataset_name); break; } } (void) zonecfg_enddsent(s_handle); } (void) zonecfg_enddsent(t_handle); } /* * Check that the clone and its source have the same brand type. */ static int valid_brand_clone(char *source_zone, char *target_zone) { brand_handle_t bh; char source_brand[MAXNAMELEN]; if ((zone_get_brand(source_zone, source_brand, sizeof (source_brand))) != Z_OK) { (void) fprintf(stderr, "%s: zone '%s': %s\n", execname, source_zone, gettext("missing or invalid brand")); return (Z_ERR); } if (strcmp(source_brand, target_brand) != NULL) { (void) fprintf(stderr, gettext("%s: Zones '%s' and '%s' have different brand " "types.\n"), execname, source_zone, target_zone); return (Z_ERR); } if ((bh = brand_open(target_brand)) == NULL) { zerror(gettext("missing or invalid brand")); return (Z_ERR); } brand_close(bh); return (Z_OK); } static int validate_clone(char *source_zone, char *target_zone) { int err = Z_OK; zone_dochandle_t s_handle; zone_dochandle_t t_handle; if ((t_handle = zonecfg_init_handle()) == NULL) { zperror(cmd_to_str(CMD_CLONE), B_TRUE); return (Z_ERR); } if ((err = zonecfg_get_handle(target_zone, t_handle)) != Z_OK) { errno = err; zperror(cmd_to_str(CMD_CLONE), B_TRUE); zonecfg_fini_handle(t_handle); return (Z_ERR); } if ((s_handle = zonecfg_init_handle()) == NULL) { zperror(cmd_to_str(CMD_CLONE), B_TRUE); zonecfg_fini_handle(t_handle); return (Z_ERR); } if ((err = zonecfg_get_handle(source_zone, s_handle)) != Z_OK) { errno = err; zperror(cmd_to_str(CMD_CLONE), B_TRUE); goto done; } /* verify new zone has same brand type */ err = valid_brand_clone(source_zone, target_zone); if (err != Z_OK) goto done; /* verify new zone has same inherit-pkg-dirs */ err = valid_ipd_clone(s_handle, source_zone, t_handle, target_zone); /* warn about imported fs's which are the same */ warn_fs_match(s_handle, source_zone, t_handle, target_zone); /* warn about imported IP addresses which are the same */ warn_ip_match(s_handle, source_zone, t_handle, target_zone); /* warn about imported devices which are the same */ warn_dev_match(s_handle, source_zone, t_handle, target_zone); /* warn about imported datasets which are the same */ warn_dataset_match(s_handle, source_zone, t_handle, target_zone); done: zonecfg_fini_handle(t_handle); zonecfg_fini_handle(s_handle); return ((err == Z_OK) ? Z_OK : Z_ERR); } static int copy_zone(char *src, char *dst) { boolean_t out_null = B_FALSE; int status; char *outfile; char cmdbuf[MAXPATHLEN * 2 + 128]; if ((outfile = tempnam("/var/log", "zone")) == NULL) { outfile = "/dev/null"; out_null = B_TRUE; } /* * Use find to get the list of files to copy. We need to skip * files of type "socket" since cpio can't handle those but that * should be ok since the app will recreate the socket when it runs. * We also need to filter out anything under the .zfs subdir. Since * find is running depth-first, we need the extra egrep to filter .zfs. */ (void) snprintf(cmdbuf, sizeof (cmdbuf), "cd %s && /usr/bin/find . -type s -prune -o -depth -print | " "/usr/bin/egrep -v '^\\./\\.zfs$|^\\./\\.zfs/' | " "/usr/bin/cpio -pdmuP@ %s > %s 2>&1", src, dst, outfile); status = do_subproc(cmdbuf); if (subproc_status("copy", status, B_TRUE) != ZONE_SUBPROC_OK) { if (!out_null) (void) fprintf(stderr, gettext("\nThe copy failed.\n" "More information can be found in %s\n"), outfile); return (Z_ERR); } if (!out_null) (void) unlink(outfile); return (Z_OK); } static int zone_postclone(char *zonepath) { char cmdbuf[MAXPATHLEN]; int status; brand_handle_t bh; int err = Z_OK; /* * Fetch the post-clone command, if any, from the brand * configuration. */ if ((bh = brand_open(target_brand)) == NULL) { zerror(gettext("missing or invalid brand")); return (Z_ERR); } (void) strcpy(cmdbuf, EXEC_PREFIX); err = brand_get_postclone(bh, target_zone, zonepath, cmdbuf + EXEC_LEN, sizeof (cmdbuf) - EXEC_LEN, 0, NULL); brand_close(bh); if (err == 0 && strlen(cmdbuf) > EXEC_LEN) { status = do_subproc(cmdbuf); if ((err = subproc_status("postclone", status, B_FALSE)) != ZONE_SUBPROC_OK) { zerror(gettext("post-clone configuration failed.")); err = Z_ERR; } } return (err); } /* ARGSUSED */ static int zfm_print(const char *p, void *r) { zerror(" %s\n", p); return (0); } int clone_copy(char *source_zonepath, char *zonepath) { int err; /* Don't clone the zone if anything is still mounted there */ if (zonecfg_find_mounts(source_zonepath, NULL, NULL)) { zerror(gettext("These file systems are mounted on " "subdirectories of %s.\n"), source_zonepath); (void) zonecfg_find_mounts(source_zonepath, zfm_print, NULL); return (Z_ERR); } /* * Attempt to create a ZFS fs for the zonepath. As usual, we don't * care if this works or not since we always have the default behavior * of a simple directory for the zonepath. */ create_zfs_zonepath(zonepath); (void) printf(gettext("Copying %s..."), source_zonepath); (void) fflush(stdout); err = copy_zone(source_zonepath, zonepath); (void) printf("\n"); return (err); } static int clone_func(int argc, char *argv[]) { char *source_zone = NULL; int lockfd; int err, arg; char zonepath[MAXPATHLEN]; char source_zonepath[MAXPATHLEN]; zone_state_t state; zone_entry_t *zent; char *method = NULL; char *snapshot = NULL; if (zonecfg_in_alt_root()) { zerror(gettext("cannot clone zone in alternate root")); return (Z_ERR); } optind = 0; if ((arg = getopt(argc, argv, "?m:s:")) != EOF) { switch (arg) { case '?': sub_usage(SHELP_CLONE, CMD_CLONE); return (optopt == '?' ? Z_OK : Z_USAGE); case 'm': method = optarg; break; case 's': snapshot = optarg; break; default: sub_usage(SHELP_CLONE, CMD_CLONE); return (Z_USAGE); } } if (argc != (optind + 1) || (method != NULL && strcmp(method, "copy") != 0)) { sub_usage(SHELP_CLONE, CMD_CLONE); return (Z_USAGE); } source_zone = argv[optind]; if (sanity_check(target_zone, CMD_CLONE, B_FALSE, B_TRUE, B_FALSE) != Z_OK) return (Z_ERR); if (verify_details(CMD_CLONE, argv) != Z_OK) return (Z_ERR); /* * We also need to do some extra validation on the source zone. */ if (strcmp(source_zone, GLOBAL_ZONENAME) == 0) { zerror(gettext("%s operation is invalid for the global zone."), cmd_to_str(CMD_CLONE)); return (Z_ERR); } if (strncmp(source_zone, "SUNW", 4) == 0) { zerror(gettext("%s operation is invalid for zones starting " "with SUNW."), cmd_to_str(CMD_CLONE)); return (Z_ERR); } zent = lookup_running_zone(source_zone); if (zent != NULL) { /* check whether the zone is ready or running */ if ((err = zone_get_state(zent->zname, &zent->zstate_num)) != Z_OK) { errno = err; zperror2(zent->zname, gettext("could not get state")); /* can't tell, so hedge */ zent->zstate_str = "ready/running"; } else { zent->zstate_str = zone_state_str(zent->zstate_num); } zerror(gettext("%s operation is invalid for %s zones."), cmd_to_str(CMD_CLONE), zent->zstate_str); return (Z_ERR); } if ((err = zone_get_state(source_zone, &state)) != Z_OK) { errno = err; zperror2(source_zone, gettext("could not get state")); return (Z_ERR); } if (state != ZONE_STATE_INSTALLED) { (void) fprintf(stderr, gettext("%s: zone %s is %s; %s is required.\n"), execname, source_zone, zone_state_str(state), zone_state_str(ZONE_STATE_INSTALLED)); return (Z_ERR); } /* * The source zone checks out ok, continue with the clone. */ if (validate_clone(source_zone, target_zone) != Z_OK) return (Z_ERR); if (grab_lock_file(target_zone, &lockfd) != Z_OK) { zerror(gettext("another %s may have an operation in progress."), "zoneadm"); return (Z_ERR); } if ((err = zone_get_zonepath(source_zone, source_zonepath, sizeof (source_zonepath))) != Z_OK) { errno = err; zperror2(source_zone, gettext("could not get zone path")); goto done; } if ((err = zone_get_zonepath(target_zone, zonepath, sizeof (zonepath))) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not get zone path")); goto done; } if ((err = zone_set_state(target_zone, ZONE_STATE_INCOMPLETE)) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not set state")); goto done; } if (snapshot != NULL) { err = clone_snapshot_zfs(snapshot, zonepath); } else { /* * We always copy the clone unless the source is ZFS and a * ZFS clone worked. We fallback to copying if the ZFS clone * fails for some reason. */ err = Z_ERR; if (method == NULL && is_zonepath_zfs(source_zonepath)) err = clone_zfs(source_zone, source_zonepath, zonepath); if (err != Z_OK) err = clone_copy(source_zonepath, zonepath); } /* * Trusted Extensions requires that cloned zones use the same sysid * configuration, so it is not appropriate to perform any * post-clone reconfiguration. */ if ((err == Z_OK) && !is_system_labeled()) err = zone_postclone(zonepath); done: /* * If everything went well, we mark the zone as installed. */ if (err == Z_OK) { err = zone_set_state(target_zone, ZONE_STATE_INSTALLED); if (err != Z_OK) { errno = err; zperror2(target_zone, gettext("could not set state")); } } release_lock_file(lockfd); return ((err == Z_OK) ? Z_OK : Z_ERR); } /* * Used when removing a zonepath after uninstalling or cleaning up after * the move subcommand. This handles a zonepath that has non-standard * contents so that we will only cleanup the stuff we know about and leave * any user data alone. * * If the "all" parameter is true then we should remove the whole zonepath * even if it has non-standard files/directories in it. This can be used when * we need to cleanup after moving the zonepath across file systems. * * We "exec" the RMCOMMAND so that the returned status is that of RMCOMMAND * and not the shell. */ static int cleanup_zonepath(char *zonepath, boolean_t all) { int status; int i; boolean_t non_std = B_FALSE; struct dirent *dp; DIR *dirp; /* * The SUNWattached.xml file is expected since it might * exist if the zone was force-attached after a * migration. */ char *std_entries[] = {"dev", "lu", "root", "SUNWattached.xml", NULL}; /* (MAXPATHLEN * 3) is for the 3 std_entries dirs */ char cmdbuf[sizeof (RMCOMMAND) + (MAXPATHLEN * 3) + 64]; /* * We shouldn't need these checks but lets be paranoid since we * could blow away the whole system here if we got the wrong zonepath. */ if (*zonepath == NULL || strcmp(zonepath, "/") == 0) { (void) fprintf(stderr, "invalid zonepath '%s'\n", zonepath); return (Z_INVAL); } /* * If the dirpath is already gone (maybe it was manually removed) then * we just return Z_OK so that the cleanup is successful. */ if ((dirp = opendir(zonepath)) == NULL) return (Z_OK); /* * Look through the zonepath directory to see if there are any * non-standard files/dirs. Also skip .zfs since that might be * there but we'll handle ZFS file systems as a special case. */ while ((dp = readdir(dirp)) != NULL) { if (strcmp(dp->d_name, ".") == 0 || strcmp(dp->d_name, "..") == 0 || strcmp(dp->d_name, ".zfs") == 0) continue; for (i = 0; std_entries[i] != NULL; i++) if (strcmp(dp->d_name, std_entries[i]) == 0) break; if (std_entries[i] == NULL) non_std = B_TRUE; } (void) closedir(dirp); if (!all && non_std) { /* * There are extra, non-standard directories/files in the * zonepath so we don't want to remove the zonepath. We * just want to remove the standard directories and leave * the user data alone. */ (void) snprintf(cmdbuf, sizeof (cmdbuf), "exec " RMCOMMAND); for (i = 0; std_entries[i] != NULL; i++) { char tmpbuf[MAXPATHLEN]; if (snprintf(tmpbuf, sizeof (tmpbuf), " %s/%s", zonepath, std_entries[i]) >= sizeof (tmpbuf) || strlcat(cmdbuf, tmpbuf, sizeof (cmdbuf)) >= sizeof (cmdbuf)) { (void) fprintf(stderr, gettext("path is too long\n")); return (Z_INVAL); } } status = do_subproc(cmdbuf); (void) fprintf(stderr, gettext("WARNING: Unable to completely " "remove %s\nbecause it contains additional user data. " "Only the standard directory\nentries have been " "removed.\n"), zonepath); return ((subproc_status(RMCOMMAND, status, B_TRUE) == ZONE_SUBPROC_OK) ? Z_OK : Z_ERR); } /* * There is nothing unexpected in the zonepath, try to get rid of the * whole zonepath directory. * * If the zonepath is its own zfs file system, try to destroy the * file system. If that fails for some reason (e.g. it has clones) * then we'll just remove the contents of the zonepath. */ if (is_zonepath_zfs(zonepath)) { if (destroy_zfs(zonepath) == Z_OK) return (Z_OK); (void) snprintf(cmdbuf, sizeof (cmdbuf), "exec " RMCOMMAND " %s/*", zonepath); status = do_subproc(cmdbuf); return ((subproc_status(RMCOMMAND, status, B_TRUE) == ZONE_SUBPROC_OK) ? Z_OK : Z_ERR); } (void) snprintf(cmdbuf, sizeof (cmdbuf), "exec " RMCOMMAND " %s", zonepath); status = do_subproc(cmdbuf); return ((subproc_status(RMCOMMAND, status, B_TRUE) == ZONE_SUBPROC_OK) ? Z_OK : Z_ERR); } static int move_func(int argc, char *argv[]) { char *new_zonepath = NULL; int lockfd; int err, arg; char zonepath[MAXPATHLEN]; zone_dochandle_t handle; boolean_t fast; boolean_t is_zfs = B_FALSE; struct dirent *dp; DIR *dirp; boolean_t empty = B_TRUE; boolean_t revert; struct stat zonepath_buf; struct stat new_zonepath_buf; if (zonecfg_in_alt_root()) { zerror(gettext("cannot move zone in alternate root")); return (Z_ERR); } optind = 0; if ((arg = getopt(argc, argv, "?")) != EOF) { switch (arg) { case '?': sub_usage(SHELP_MOVE, CMD_MOVE); return (optopt == '?' ? Z_OK : Z_USAGE); default: sub_usage(SHELP_MOVE, CMD_MOVE); return (Z_USAGE); } } if (argc != (optind + 1)) { sub_usage(SHELP_MOVE, CMD_MOVE); return (Z_USAGE); } new_zonepath = argv[optind]; if (sanity_check(target_zone, CMD_MOVE, B_FALSE, B_TRUE, B_FALSE) != Z_OK) return (Z_ERR); if (verify_details(CMD_MOVE, argv) != Z_OK) return (Z_ERR); /* * Check out the new zonepath. This has the side effect of creating * a directory for the new zonepath. We depend on this later when we * stat to see if we are doing a cross file system move or not. */ if (validate_zonepath(new_zonepath, CMD_MOVE) != Z_OK) return (Z_ERR); if ((err = zone_get_zonepath(target_zone, zonepath, sizeof (zonepath))) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not get zone path")); return (Z_ERR); } if (stat(zonepath, &zonepath_buf) == -1) { zperror(gettext("could not stat zone path"), B_FALSE); return (Z_ERR); } if (stat(new_zonepath, &new_zonepath_buf) == -1) { zperror(gettext("could not stat new zone path"), B_FALSE); return (Z_ERR); } /* * Check if the destination directory is empty. */ if ((dirp = opendir(new_zonepath)) == NULL) { zperror(gettext("could not open new zone path"), B_FALSE); return (Z_ERR); } while ((dp = readdir(dirp)) != (struct dirent *)0) { if (strcmp(dp->d_name, ".") == 0 || strcmp(dp->d_name, "..") == 0) continue; empty = B_FALSE; break; } (void) closedir(dirp); /* Error if there is anything in the destination directory. */ if (!empty) { (void) fprintf(stderr, gettext("could not move zone to %s: " "directory not empty\n"), new_zonepath); return (Z_ERR); } /* Don't move the zone if anything is still mounted there */ if (zonecfg_find_mounts(zonepath, NULL, NULL)) { zerror(gettext("These file systems are mounted on " "subdirectories of %s.\n"), zonepath); (void) zonecfg_find_mounts(zonepath, zfm_print, NULL); return (Z_ERR); } /* * Check if we are moving in the same file system and can do a fast * move or if we are crossing file systems and have to copy the data. */ fast = (zonepath_buf.st_dev == new_zonepath_buf.st_dev); if ((handle = zonecfg_init_handle()) == NULL) { zperror(cmd_to_str(CMD_MOVE), B_TRUE); return (Z_ERR); } if ((err = zonecfg_get_handle(target_zone, handle)) != Z_OK) { errno = err; zperror(cmd_to_str(CMD_MOVE), B_TRUE); zonecfg_fini_handle(handle); return (Z_ERR); } if (grab_lock_file(target_zone, &lockfd) != Z_OK) { zerror(gettext("another %s may have an operation in progress."), "zoneadm"); zonecfg_fini_handle(handle); return (Z_ERR); } /* * We're making some file system changes now so we have to clean up * the file system before we are done. This will either clean up the * new zonepath if the zonecfg update failed or it will clean up the * old zonepath if everything is ok. */ revert = B_TRUE; if (is_zonepath_zfs(zonepath) && move_zfs(zonepath, new_zonepath) != Z_ERR) { is_zfs = B_TRUE; } else if (fast) { /* same file system, use rename for a quick move */ /* * Remove the new_zonepath directory that got created above * during the validation. It gets in the way of the rename. */ if (rmdir(new_zonepath) != 0) { zperror(gettext("could not rmdir new zone path"), B_FALSE); zonecfg_fini_handle(handle); release_lock_file(lockfd); return (Z_ERR); } if (rename(zonepath, new_zonepath) != 0) { /* * If this fails we don't need to do all of the * cleanup that happens for the rest of the code * so just return from this error. */ zperror(gettext("could not move zone"), B_FALSE); zonecfg_fini_handle(handle); release_lock_file(lockfd); return (Z_ERR); } } else { /* * Attempt to create a ZFS fs for the new zonepath. As usual, * we don't care if this works or not since we always have the * default behavior of a simple directory for the zonepath. */ create_zfs_zonepath(new_zonepath); (void) printf(gettext( "Moving across file systems; copying zonepath %s..."), zonepath); (void) fflush(stdout); err = copy_zone(zonepath, new_zonepath); (void) printf("\n"); if (err != Z_OK) goto done; } if ((err = zonecfg_set_zonepath(handle, new_zonepath)) != Z_OK) { errno = err; zperror(gettext("could not set new zonepath"), B_TRUE); goto done; } if ((err = zonecfg_save(handle)) != Z_OK) { errno = err; zperror(gettext("zonecfg save failed"), B_TRUE); goto done; } revert = B_FALSE; done: zonecfg_fini_handle(handle); release_lock_file(lockfd); /* * Clean up the file system based on how things went. We either * clean up the new zonepath if the operation failed for some reason * or we clean up the old zonepath if everything is ok. */ if (revert) { /* The zonecfg update failed, cleanup the new zonepath. */ if (is_zfs) { if (move_zfs(new_zonepath, zonepath) == Z_ERR) { (void) fprintf(stderr, gettext("could not " "restore zonepath, the zfs mountpoint is " "set as:\n%s\n"), new_zonepath); /* * err is already != Z_OK since we're reverting */ } } else if (fast) { if (rename(new_zonepath, zonepath) != 0) { zperror(gettext("could not restore zonepath"), B_FALSE); /* * err is already != Z_OK since we're reverting */ } } else { (void) printf(gettext("Cleaning up zonepath %s..."), new_zonepath); (void) fflush(stdout); err = cleanup_zonepath(new_zonepath, B_TRUE); (void) printf("\n"); if (err != Z_OK) { errno = err; zperror(gettext("could not remove new " "zonepath"), B_TRUE); } else { /* * Because we're reverting we know the mainline * code failed but we just reused the err * variable so we reset it back to Z_ERR. */ err = Z_ERR; } } } else { /* The move was successful, cleanup the old zonepath. */ if (!is_zfs && !fast) { (void) printf( gettext("Cleaning up zonepath %s..."), zonepath); (void) fflush(stdout); err = cleanup_zonepath(zonepath, B_TRUE); (void) printf("\n"); if (err != Z_OK) { errno = err; zperror(gettext("could not remove zonepath"), B_TRUE); } } } return ((err == Z_OK) ? Z_OK : Z_ERR); } static int detach_func(int argc, char *argv[]) { int lockfd; int err, arg; char zonepath[MAXPATHLEN]; zone_dochandle_t handle; boolean_t execute = B_TRUE; if (zonecfg_in_alt_root()) { zerror(gettext("cannot detach zone in alternate root")); return (Z_ERR); } optind = 0; if ((arg = getopt(argc, argv, "?n")) != EOF) { switch (arg) { case '?': sub_usage(SHELP_DETACH, CMD_DETACH); return (optopt == '?' ? Z_OK : Z_USAGE); case 'n': execute = B_FALSE; break; default: sub_usage(SHELP_DETACH, CMD_DETACH); return (Z_USAGE); } } if (execute) { if (sanity_check(target_zone, CMD_DETACH, B_FALSE, B_TRUE, B_FALSE) != Z_OK) return (Z_ERR); if (verify_details(CMD_DETACH, argv) != Z_OK) return (Z_ERR); } else { /* * We want a dry-run to work for a non-privileged user so we * only do minimal validation. */ if (getzoneid() != GLOBAL_ZONEID) { zerror(gettext("must be in the global zone to %s a " "zone."), cmd_to_str(CMD_DETACH)); return (Z_ERR); } if (target_zone == NULL) { zerror(gettext("no zone specified")); return (Z_ERR); } if (strcmp(target_zone, GLOBAL_ZONENAME) == 0) { zerror(gettext("%s operation is invalid for the " "global zone."), cmd_to_str(CMD_DETACH)); return (Z_ERR); } } if ((err = zone_get_zonepath(target_zone, zonepath, sizeof (zonepath))) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not get zone path")); return (Z_ERR); } /* Don't detach the zone if anything is still mounted there */ if (execute && zonecfg_find_mounts(zonepath, NULL, NULL)) { zerror(gettext("These file systems are mounted on " "subdirectories of %s.\n"), zonepath); (void) zonecfg_find_mounts(zonepath, zfm_print, NULL); return (Z_ERR); } if ((handle = zonecfg_init_handle()) == NULL) { zperror(cmd_to_str(CMD_DETACH), B_TRUE); return (Z_ERR); } if ((err = zonecfg_get_handle(target_zone, handle)) != Z_OK) { errno = err; zperror(cmd_to_str(CMD_DETACH), B_TRUE); zonecfg_fini_handle(handle); return (Z_ERR); } if (execute && grab_lock_file(target_zone, &lockfd) != Z_OK) { zerror(gettext("another %s may have an operation in progress."), "zoneadm"); zonecfg_fini_handle(handle); return (Z_ERR); } if ((err = zonecfg_get_detach_info(handle, B_TRUE)) != Z_OK) { errno = err; zperror(gettext("getting the detach information failed"), B_TRUE); goto done; } if ((err = zonecfg_detach_save(handle, (execute ? 0 : ZONE_DRY_RUN))) != Z_OK) { errno = err; zperror(gettext("saving the detach manifest failed"), B_TRUE); goto done; } /* * Set the zone state back to configured unless we are running with the * no-execute option. */ if (execute && (err = zone_set_state(target_zone, ZONE_STATE_CONFIGURED)) != Z_OK) { errno = err; zperror(gettext("could not reset state"), B_TRUE); } done: zonecfg_fini_handle(handle); if (execute) release_lock_file(lockfd); return ((err == Z_OK) ? Z_OK : Z_ERR); } /* * During attach we go through and fix up the /dev entries for the zone * we are attaching. In order to regenerate /dev with the correct devices, * the old /dev will be removed, the zone readied (which generates a new * /dev) then halted, then we use the info from the manifest to update * the modes, owners, etc. on the new /dev. */ static int dev_fix(zone_dochandle_t handle) { int res; int err; int status; struct zone_devpermtab devtab; zone_cmd_arg_t zarg; char devpath[MAXPATHLEN]; /* 6: "exec " and " " */ char cmdbuf[sizeof (RMCOMMAND) + MAXPATHLEN + 6]; if ((res = zonecfg_get_zonepath(handle, devpath, sizeof (devpath))) != Z_OK) return (res); if (strlcat(devpath, "/dev", sizeof (devpath)) >= sizeof (devpath)) return (Z_TOO_BIG); /* * "exec" the command so that the returned status is that of * RMCOMMAND and not the shell. */ (void) snprintf(cmdbuf, sizeof (cmdbuf), EXEC_PREFIX RMCOMMAND " %s", devpath); status = do_subproc(cmdbuf); if ((err = subproc_status(RMCOMMAND, status, B_TRUE)) != ZONE_SUBPROC_OK) { (void) fprintf(stderr, gettext("could not remove existing /dev\n")); return (Z_ERR); } /* In order to ready the zone, it must be in the installed state */ if ((err = zone_set_state(target_zone, ZONE_STATE_INSTALLED)) != Z_OK) { errno = err; zperror(gettext("could not reset state"), B_TRUE); return (Z_ERR); } /* We have to ready the zone to regen the dev tree */ zarg.cmd = Z_READY; if (call_zoneadmd(target_zone, &zarg) != 0) { zerror(gettext("call to %s failed"), "zoneadmd"); /* attempt to restore zone to configured state */ (void) zone_set_state(target_zone, ZONE_STATE_CONFIGURED); return (Z_ERR); } zarg.cmd = Z_HALT; if (call_zoneadmd(target_zone, &zarg) != 0) { zerror(gettext("call to %s failed"), "zoneadmd"); /* attempt to restore zone to configured state */ (void) zone_set_state(target_zone, ZONE_STATE_CONFIGURED); return (Z_ERR); } /* attempt to restore zone to configured state */ (void) zone_set_state(target_zone, ZONE_STATE_CONFIGURED); if (zonecfg_setdevperment(handle) != Z_OK) { (void) fprintf(stderr, gettext("unable to enumerate device entries\n")); return (Z_ERR); } while (zonecfg_getdevperment(handle, &devtab) == Z_OK) { int err; if ((err = zonecfg_devperms_apply(handle, devtab.zone_devperm_name, devtab.zone_devperm_uid, devtab.zone_devperm_gid, devtab.zone_devperm_mode, devtab.zone_devperm_acl)) != Z_OK && err != Z_INVAL) (void) fprintf(stderr, gettext("error updating device " "%s: %s\n"), devtab.zone_devperm_name, zonecfg_strerror(err)); free(devtab.zone_devperm_acl); } (void) zonecfg_enddevperment(handle); return (Z_OK); } /* * Validate attaching a zone but don't actually do the work. The zone * does not have to exist, so there is some complexity getting a new zone * configuration set up so that we can perform the validation. This is * handled within zonecfg_attach_manifest() which returns two handles; one * for the the full configuration to validate (rem_handle) and the other * (local_handle) containing only the zone configuration derived from the * manifest. */ static int dryrun_attach(char *manifest_path, char *argv[]) { int fd; int err; int res; zone_dochandle_t local_handle; zone_dochandle_t rem_handle = NULL; if (strcmp(manifest_path, "-") == 0) { fd = 0; } else if ((fd = open(manifest_path, O_RDONLY)) < 0) { zperror(gettext("could not open manifest path"), B_FALSE); return (Z_ERR); } if ((local_handle = zonecfg_init_handle()) == NULL) { zperror(cmd_to_str(CMD_ATTACH), B_TRUE); res = Z_ERR; goto done; } if ((rem_handle = zonecfg_init_handle()) == NULL) { zperror(cmd_to_str(CMD_ATTACH), B_TRUE); res = Z_ERR; goto done; } if ((err = zonecfg_attach_manifest(fd, local_handle, rem_handle)) != Z_OK) { res = Z_ERR; if (err == Z_INVALID_DOCUMENT) { struct stat st; char buf[6]; if (strcmp(manifest_path, "-") == 0) { zerror(gettext("Input is not a valid XML " "file")); goto done; } if (fstat(fd, &st) == -1 || !S_ISREG(st.st_mode)) { zerror(gettext("%s is not an XML file"), manifest_path); goto done; } bzero(buf, sizeof (buf)); (void) lseek(fd, 0L, SEEK_SET); if (read(fd, buf, sizeof (buf) - 1) < 0 || strncmp(buf, " yes, FALSE => no. * On return, TRUE => 1, FALSE => 0, could not ask => -1. */ static int ask_yesno(boolean_t default_answer, const char *question) { char line[64]; /* should be large enough to answer yes or no */ if (!isatty(STDIN_FILENO)) return (-1); for (;;) { (void) printf("%s (%s)? ", question, default_answer ? "[y]/n" : "y/[n]"); if (fgets(line, sizeof (line), stdin) == NULL || line[0] == '\n') return (default_answer ? 1 : 0); if (tolower(line[0]) == 'y') return (1); if (tolower(line[0]) == 'n') return (0); } } static int uninstall_func(int argc, char *argv[]) { char line[ZONENAME_MAX + 128]; /* Enough for "Are you sure ..." */ char rootpath[MAXPATHLEN], zonepath[MAXPATHLEN]; boolean_t force = B_FALSE; int lockfd, answer; int err, arg; if (zonecfg_in_alt_root()) { zerror(gettext("cannot uninstall zone in alternate root")); return (Z_ERR); } optind = 0; while ((arg = getopt(argc, argv, "?F")) != EOF) { switch (arg) { case '?': sub_usage(SHELP_UNINSTALL, CMD_UNINSTALL); return (optopt == '?' ? Z_OK : Z_USAGE); case 'F': force = B_TRUE; break; default: sub_usage(SHELP_UNINSTALL, CMD_UNINSTALL); return (Z_USAGE); } } if (argc > optind) { sub_usage(SHELP_UNINSTALL, CMD_UNINSTALL); return (Z_USAGE); } if (sanity_check(target_zone, CMD_UNINSTALL, B_FALSE, B_TRUE, B_FALSE) != Z_OK) return (Z_ERR); /* * Invoke brand-specific handler. */ if (invoke_brand_handler(CMD_UNINSTALL, argv) != Z_OK) return (Z_ERR); if (!force) { (void) snprintf(line, sizeof (line), gettext("Are you sure you want to %s zone %s"), cmd_to_str(CMD_UNINSTALL), target_zone); if ((answer = ask_yesno(B_FALSE, line)) == 0) { return (Z_OK); } else if (answer == -1) { zerror(gettext("Input not from terminal and -F " "not specified: %s not done."), cmd_to_str(CMD_UNINSTALL)); return (Z_ERR); } } if ((err = zone_get_zonepath(target_zone, zonepath, sizeof (zonepath))) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not get zone path")); return (Z_ERR); } if ((err = zone_get_rootpath(target_zone, rootpath, sizeof (rootpath))) != Z_OK) { errno = err; zperror2(target_zone, gettext("could not get root path")); return (Z_ERR); } /* * If there seems to be a zoneadmd running for this zone, call it * to tell it that an uninstall is happening; if all goes well it * will then shut itself down. */ if (ping_zoneadmd(target_zone) == Z_OK) { zone_cmd_arg_t zarg; zarg.cmd = Z_NOTE_UNINSTALLING; /* we don't care too much if this fails... just plow on */ (void) call_zoneadmd(target_zone, &zarg); } if (grab_lock_file(target_zone, &lockfd) != Z_OK) { zerror(gettext("another %s may have an operation in progress."), "zoneadm"); return (Z_ERR); } /* Don't uninstall the zone if anything is mounted there */ err = zonecfg_find_mounts(rootpath, NULL, NULL); if (err) { zerror(gettext("These file systems are mounted on " "subdirectories of %s.\n"), rootpath); (void) zonecfg_find_mounts(rootpath, zfm_print, NULL); return (Z_ERR); } err = zone_set_state(target_zone, ZONE_STATE_INCOMPLETE); if (err != Z_OK) { errno = err; zperror2(target_zone, gettext("could not set state")); goto bad; } if ((err = cleanup_zonepath(zonepath, B_FALSE)) != Z_OK) { errno = err; zperror2(target_zone, gettext("cleaning up zonepath failed")); goto bad; } err = zone_set_state(target_zone, ZONE_STATE_CONFIGURED); if (err != Z_OK) { errno = err; zperror2(target_zone, gettext("could not reset state")); } bad: release_lock_file(lockfd); return (err); } /* ARGSUSED */ static int mount_func(int argc, char *argv[]) { zone_cmd_arg_t zarg; boolean_t force = B_FALSE; int arg; /* * The only supported subargument to the "mount" subcommand is * "-f", which forces us to mount a zone in the INCOMPLETE state. */ optind = 0; if ((arg = getopt(argc, argv, "f")) != EOF) { switch (arg) { case 'f': force = B_TRUE; break; default: return (Z_USAGE); } } if (argc > optind) return (Z_USAGE); if (sanity_check(target_zone, CMD_MOUNT, B_FALSE, B_FALSE, force) != Z_OK) return (Z_ERR); if (verify_details(CMD_MOUNT, argv) != Z_OK) return (Z_ERR); zarg.cmd = force ? Z_FORCEMOUNT : Z_MOUNT; if (call_zoneadmd(target_zone, &zarg) != 0) { zerror(gettext("call to %s failed"), "zoneadmd"); return (Z_ERR); } return (Z_OK); } /* ARGSUSED */ static int unmount_func(int argc, char *argv[]) { zone_cmd_arg_t zarg; if (argc > 0) return (Z_USAGE); if (sanity_check(target_zone, CMD_UNMOUNT, B_FALSE, B_FALSE, B_FALSE) != Z_OK) return (Z_ERR); zarg.cmd = Z_UNMOUNT; if (call_zoneadmd(target_zone, &zarg) != 0) { zerror(gettext("call to %s failed"), "zoneadmd"); return (Z_ERR); } return (Z_OK); } static int mark_func(int argc, char *argv[]) { int err, lockfd; if (argc != 1 || strcmp(argv[0], "incomplete") != 0) return (Z_USAGE); if (sanity_check(target_zone, CMD_MARK, B_FALSE, B_FALSE, B_FALSE) != Z_OK) return (Z_ERR); /* * Invoke brand-specific handler. */ if (invoke_brand_handler(CMD_MARK, argv) != Z_OK) return (Z_ERR); if (grab_lock_file(target_zone, &lockfd) != Z_OK) { zerror(gettext("another %s may have an operation in progress."), "zoneadm"); return (Z_ERR); } err = zone_set_state(target_zone, ZONE_STATE_INCOMPLETE); if (err != Z_OK) { errno = err; zperror2(target_zone, gettext("could not set state")); } release_lock_file(lockfd); return (err); } /* * Check what scheduling class we're running under and print a warning if * we're not using FSS. */ static int check_sched_fss(zone_dochandle_t handle) { char class_name[PC_CLNMSZ]; if (zonecfg_get_dflt_sched_class(handle, class_name, sizeof (class_name)) != Z_OK) { zerror(gettext("WARNING: unable to determine the zone's " "scheduling class")); } else if (strcmp("FSS", class_name) != 0) { zerror(gettext("WARNING: The zone.cpu-shares rctl is set but\n" "FSS is not the default scheduling class for this zone. " "FSS will be\nused for processes in the zone but to get " "the full benefit of FSS,\nit should be the default " "scheduling class. See dispadmin(1M) for\nmore details.")); return (Z_SYSTEM); } return (Z_OK); } static int check_cpu_shares_sched(zone_dochandle_t handle) { int err; int res = Z_OK; struct zone_rctltab rctl; if ((err = zonecfg_setrctlent(handle)) != Z_OK) { errno = err; zperror(cmd_to_str(CMD_APPLY), B_TRUE); return (err); } while (zonecfg_getrctlent(handle, &rctl) == Z_OK) { if (strcmp(rctl.zone_rctl_name, "zone.cpu-shares") == 0) { if (check_sched_fss(handle) != Z_OK) res = Z_SYSTEM; break; } } (void) zonecfg_endrctlent(handle); return (res); } /* * Check if there is a mix of processes running in different pools within the * zone. This is currently only going to be called for the global zone from * apply_func but that could be generalized in the future. */ static boolean_t mixed_pools(zoneid_t zoneid) { DIR *dirp; dirent_t *dent; boolean_t mixed = B_FALSE; boolean_t poolid_set = B_FALSE; poolid_t last_poolid = 0; if ((dirp = opendir("/proc")) == NULL) { zerror(gettext("could not open /proc")); return (B_FALSE); } while ((dent = readdir(dirp)) != NULL) { int procfd; psinfo_t ps; char procpath[MAXPATHLEN]; if (dent->d_name[0] == '.') continue; (void) snprintf(procpath, sizeof (procpath), "/proc/%s/psinfo", dent->d_name); if ((procfd = open(procpath, O_RDONLY)) == -1) continue; if (read(procfd, &ps, sizeof (ps)) == sizeof (psinfo_t)) { /* skip processes in other zones and system processes */ if (zoneid != ps.pr_zoneid || ps.pr_flag & SSYS) { (void) close(procfd); continue; } if (poolid_set) { if (ps.pr_poolid != last_poolid) mixed = B_TRUE; } else { last_poolid = ps.pr_poolid; poolid_set = B_TRUE; } } (void) close(procfd); if (mixed) break; } (void) closedir(dirp); return (mixed); } /* * Check if a persistent or temporary pool is configured for the zone. * This is currently only going to be called for the global zone from * apply_func but that could be generalized in the future. */ static boolean_t pool_configured(zone_dochandle_t handle) { int err1, err2; struct zone_psettab pset_tab; char poolname[MAXPATHLEN]; err1 = zonecfg_lookup_pset(handle, &pset_tab); err2 = zonecfg_get_pool(handle, poolname, sizeof (poolname)); if (err1 == Z_NO_ENTRY && (err2 == Z_NO_ENTRY || (err2 == Z_OK && strlen(poolname) == 0))) return (B_FALSE); return (B_TRUE); } /* * This is an undocumented interface which is currently only used to apply * the global zone resource management settings when the system boots. * This function does not yet properly handle updating a running system so * any projects running in the zone would be trashed if this function * were to run after the zone had booted. It also does not reset any * rctl settings that were removed from zonecfg. There is still work to be * done before we can properly support dynamically updating the resource * management settings for a running zone (global or non-global). Thus, this * functionality is undocumented for now. */ /* ARGSUSED */ static int apply_func(int argc, char *argv[]) { int err; int res = Z_OK; priv_set_t *privset; zoneid_t zoneid; zone_dochandle_t handle; struct zone_mcaptab mcap; char pool_err[128]; zoneid = getzoneid(); if (zonecfg_in_alt_root() || zoneid != GLOBAL_ZONEID || target_zone == NULL || strcmp(target_zone, GLOBAL_ZONENAME) != 0) return (usage(B_FALSE)); if ((privset = priv_allocset()) == NULL) { zerror(gettext("%s failed"), "priv_allocset"); return (Z_ERR); } if (getppriv(PRIV_EFFECTIVE, privset) != 0) { zerror(gettext("%s failed"), "getppriv"); priv_freeset(privset); return (Z_ERR); } if (priv_isfullset(privset) == B_FALSE) { (void) usage(B_FALSE); priv_freeset(privset); return (Z_ERR); } priv_freeset(privset); if ((handle = zonecfg_init_handle()) == NULL) { zperror(cmd_to_str(CMD_APPLY), B_TRUE); return (Z_ERR); } if ((err = zonecfg_get_handle(target_zone, handle)) != Z_OK) { errno = err; zperror(cmd_to_str(CMD_APPLY), B_TRUE); zonecfg_fini_handle(handle); return (Z_ERR); } /* specific error msgs are printed within apply_rctls */ if ((err = zonecfg_apply_rctls(target_zone, handle)) != Z_OK) { errno = err; zperror(cmd_to_str(CMD_APPLY), B_TRUE); res = Z_ERR; } if ((err = check_cpu_shares_sched(handle)) != Z_OK) res = Z_ERR; if (pool_configured(handle)) { if (mixed_pools(zoneid)) { zerror(gettext("Zone is using multiple resource " "pools. The pool\nconfiguration cannot be " "applied without rebooting.")); res = Z_ERR; } else { /* * The next two blocks of code attempt to set up * temporary pools as well as persistent pools. In * both cases we call the functions unconditionally. * Within each funtion the code will check if the zone * is actually configured for a temporary pool or * persistent pool and just return if there is nothing * to do. */ if ((err = zonecfg_bind_tmp_pool(handle, zoneid, pool_err, sizeof (pool_err))) != Z_OK) { if (err == Z_POOL || err == Z_POOL_CREATE || err == Z_POOL_BIND) zerror("%s: %s", zonecfg_strerror(err), pool_err); else zerror(gettext("could not bind zone to " "temporary pool: %s"), zonecfg_strerror(err)); res = Z_ERR; } if ((err = zonecfg_bind_pool(handle, zoneid, pool_err, sizeof (pool_err))) != Z_OK) { if (err == Z_POOL || err == Z_POOL_BIND) zerror("%s: %s", zonecfg_strerror(err), pool_err); else zerror("%s", zonecfg_strerror(err)); } } } /* * If a memory cap is configured, set the cap in the kernel using * zone_setattr() and make sure the rcapd SMF service is enabled. */ if (zonecfg_getmcapent(handle, &mcap) == Z_OK) { uint64_t num; char smf_err[128]; num = (uint64_t)strtoll(mcap.zone_physmem_cap, NULL, 10); if (zone_setattr(zoneid, ZONE_ATTR_PHYS_MCAP, &num, 0) == -1) { zerror(gettext("could not set zone memory cap")); res = Z_ERR; } if (zonecfg_enable_rcapd(smf_err, sizeof (smf_err)) != Z_OK) { zerror(gettext("enabling system/rcap service failed: " "%s"), smf_err); res = Z_ERR; } } zonecfg_fini_handle(handle); return (res); } static int help_func(int argc, char *argv[]) { int arg, cmd_num; if (argc == 0) { (void) usage(B_TRUE); return (Z_OK); } optind = 0; if ((arg = getopt(argc, argv, "?")) != EOF) { switch (arg) { case '?': sub_usage(SHELP_HELP, CMD_HELP); return (optopt == '?' ? Z_OK : Z_USAGE); default: sub_usage(SHELP_HELP, CMD_HELP); return (Z_USAGE); } } while (optind < argc) { /* Private commands have NULL short_usage; omit them */ if ((cmd_num = cmd_match(argv[optind])) < 0 || cmdtab[cmd_num].short_usage == NULL) { sub_usage(SHELP_HELP, CMD_HELP); return (Z_USAGE); } sub_usage(cmdtab[cmd_num].short_usage, cmd_num); optind++; } return (Z_OK); } /* * Returns: CMD_MIN thru CMD_MAX on success, -1 on error */ static int cmd_match(char *cmd) { int i; for (i = CMD_MIN; i <= CMD_MAX; i++) { /* return only if there is an exact match */ if (strcmp(cmd, cmdtab[i].cmd_name) == 0) return (cmdtab[i].cmd_num); } return (-1); } static int parse_and_run(int argc, char *argv[]) { int i = cmd_match(argv[0]); if (i < 0) return (usage(B_FALSE)); return (cmdtab[i].handler(argc - 1, &(argv[1]))); } static char * get_execbasename(char *execfullname) { char *last_slash, *execbasename; /* guard against '/' at end of command invocation */ for (;;) { last_slash = strrchr(execfullname, '/'); if (last_slash == NULL) { execbasename = execfullname; break; } else { execbasename = last_slash + 1; if (*execbasename == '\0') { *last_slash = '\0'; continue; } break; } } return (execbasename); } int main(int argc, char **argv) { int arg; zoneid_t zid; struct stat st; char *zone_lock_env; int err; if ((locale = setlocale(LC_ALL, "")) == NULL) locale = "C"; (void) textdomain(TEXT_DOMAIN); setbuf(stdout, NULL); (void) sigset(SIGHUP, SIG_IGN); execname = get_execbasename(argv[0]); target_zone = NULL; if (chdir("/") != 0) { zerror(gettext("could not change directory to /.")); exit(Z_ERR); } if (init_zfs() != Z_OK) exit(Z_ERR); while ((arg = getopt(argc, argv, "?u:z:R:")) != EOF) { switch (arg) { case '?': return (usage(B_TRUE)); case 'u': target_uuid = optarg; break; case 'z': target_zone = optarg; break; case 'R': /* private option for admin/install use */ if (*optarg != '/') { zerror(gettext("root path must be absolute.")); exit(Z_ERR); } if (stat(optarg, &st) == -1 || !S_ISDIR(st.st_mode)) { zerror( gettext("root path must be a directory.")); exit(Z_ERR); } zonecfg_set_root(optarg); break; default: return (usage(B_FALSE)); } } if (optind >= argc) return (usage(B_FALSE)); if (target_uuid != NULL && *target_uuid != '\0') { uuid_t uuid; static char newtarget[ZONENAME_MAX]; if (uuid_parse(target_uuid, uuid) == -1) { zerror(gettext("illegal UUID value specified")); exit(Z_ERR); } if (zonecfg_get_name_by_uuid(uuid, newtarget, sizeof (newtarget)) == Z_OK) target_zone = newtarget; } if (target_zone != NULL && zone_get_id(target_zone, &zid) != 0) { errno = Z_NO_ZONE; zperror(target_zone, B_TRUE); exit(Z_ERR); } /* * See if we have inherited the right to manipulate this zone from * a zoneadm instance in our ancestry. If so, set zone_lock_cnt to * indicate it. If not, make that explicit in our environment. */ zone_lock_env = getenv(LOCK_ENV_VAR); if (zone_lock_env == NULL) { if (putenv(zoneadm_lock_not_held) != 0) { zperror(target_zone, B_TRUE); exit(Z_ERR); } } else { zoneadm_is_nested = B_TRUE; if (atoi(zone_lock_env) == 1) zone_lock_cnt = 1; } /* * If we are going to be operating on a single zone, retrieve its * brand type and determine whether it is native or not. */ if ((target_zone != NULL) && (strcmp(target_zone, GLOBAL_ZONENAME) != NULL)) { if (zone_get_brand(target_zone, target_brand, sizeof (target_brand)) != Z_OK) { zerror(gettext("missing or invalid brand")); exit(Z_ERR); } is_native_zone = (strcmp(target_brand, NATIVE_BRAND_NAME) == 0); } err = parse_and_run(argc - optind, &argv[optind]); return (err); }