xref: /illumos-gate/usr/src/uts/common/os/zone.c (revision d2a70789)

/*
 * 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 (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright 2015, Joyent Inc. All rights reserved.
 */

/*
 * Zones
 *
 *   A zone is a named collection of processes, namespace constraints,
 *   and other system resources which comprise a secure and manageable
 *   application containment facility.
 *
 *   Zones (represented by the reference counted zone_t) are tracked in
 *   the kernel in the zonehash.  Elsewhere in the kernel, Zone IDs
 *   (zoneid_t) are used to track zone association.  Zone IDs are
 *   dynamically generated when the zone is created; if a persistent
 *   identifier is needed (core files, accounting logs, audit trail,
 *   etc.), the zone name should be used.
 *
 *
 *   Global Zone:
 *
 *   The global zone (zoneid 0) is automatically associated with all
 *   system resources that have not been bound to a user-created zone.
 *   This means that even systems where zones are not in active use
 *   have a global zone, and all processes, mounts, etc. are
 *   associated with that zone.  The global zone is generally
 *   unconstrained in terms of privileges and access, though the usual
 *   credential and privilege based restrictions apply.
 *
 *
 *   Zone States:
 *
 *   The states in which a zone may be in and the transitions are as
 *   follows:
 *
 *   ZONE_IS_UNINITIALIZED: primordial state for a zone. The partially
 *   initialized zone is added to the list of active zones on the system but
 *   isn't accessible.
 *
 *   ZONE_IS_INITIALIZED: Initialization complete except the ZSD callbacks are
 *   not yet completed. Not possible to enter the zone, but attributes can
 *   be retrieved.
 *
 *   ZONE_IS_READY: zsched (the kernel dummy process for a zone) is
 *   ready.  The zone is made visible after the ZSD constructor callbacks are
 *   executed.  A zone remains in this state until it transitions into
 *   the ZONE_IS_BOOTING state as a result of a call to zone_boot().
 *
 *   ZONE_IS_BOOTING: in this shortlived-state, zsched attempts to start
 *   init.  Should that fail, the zone proceeds to the ZONE_IS_SHUTTING_DOWN
 *   state.
 *
 *   ZONE_IS_RUNNING: The zone is open for business: zsched has
 *   successfully started init.   A zone remains in this state until
 *   zone_shutdown() is called.
 *
 *   ZONE_IS_SHUTTING_DOWN: zone_shutdown() has been called, the system is
 *   killing all processes running in the zone. The zone remains
 *   in this state until there are no more user processes running in the zone.
 *   zone_create(), zone_enter(), and zone_destroy() on this zone will fail.
 *   Since zone_shutdown() is restartable, it may be called successfully
 *   multiple times for the same zone_t.  Setting of the zone's state to
 *   ZONE_IS_SHUTTING_DOWN is synchronized with mounts, so VOP_MOUNT() may check
 *   the zone's status without worrying about it being a moving target.
 *
 *   ZONE_IS_EMPTY: zone_shutdown() has been called, and there
 *   are no more user processes in the zone.  The zone remains in this
 *   state until there are no more kernel threads associated with the
 *   zone.  zone_create(), zone_enter(), and zone_destroy() on this zone will
 *   fail.
 *
 *   ZONE_IS_DOWN: All kernel threads doing work on behalf of the zone
 *   have exited.  zone_shutdown() returns.  Henceforth it is not possible to
 *   join the zone or create kernel threads therein.
 *
 *   ZONE_IS_DYING: zone_destroy() has been called on the zone; zone
 *   remains in this state until zsched exits.  Calls to zone_find_by_*()
 *   return NULL from now on.
 *
 *   ZONE_IS_DEAD: zsched has exited (zone_ntasks == 0).  There are no
 *   processes or threads doing work on behalf of the zone.  The zone is
 *   removed from the list of active zones.  zone_destroy() returns, and
 *   the zone can be recreated.
 *
 *   ZONE_IS_FREE (internal state): zone_ref goes to 0, ZSD destructor
 *   callbacks are executed, and all memory associated with the zone is
 *   freed.
 *
 *   Threads can wait for the zone to enter a requested state by using
 *   zone_status_wait() or zone_status_timedwait() with the desired
 *   state passed in as an argument.  Zone state transitions are
 *   uni-directional; it is not possible to move back to an earlier state.
 *
 *
 *   Zone-Specific Data:
 *
 *   Subsystems needing to maintain zone-specific data can store that
 *   data using the ZSD mechanism.  This provides a zone-specific data
 *   store, similar to thread-specific data (see pthread_getspecific(3C)
 *   or the TSD code in uts/common/disp/thread.c.  Also, ZSD can be used
 *   to register callbacks to be invoked when a zone is created, shut
 *   down, or destroyed.  This can be used to initialize zone-specific
 *   data for new zones and to clean up when zones go away.
 *
 *
 *   Data Structures:
 *
 *   The per-zone structure (zone_t) is reference counted, and freed
 *   when all references are released.  zone_hold and zone_rele can be
 *   used to adjust the reference count.  In addition, reference counts
 *   associated with the cred_t structure are tracked separately using
 *   zone_cred_hold and zone_cred_rele.
 *
 *   Pointers to active zone_t's are stored in two hash tables; one
 *   for searching by id, the other for searching by name.  Lookups
 *   can be performed on either basis, using zone_find_by_id and
 *   zone_find_by_name.  Both return zone_t pointers with the zone
 *   held, so zone_rele should be called when the pointer is no longer
 *   needed.  Zones can also be searched by path; zone_find_by_path
 *   returns the zone with which a path name is associated (global
 *   zone if the path is not within some other zone's file system
 *   hierarchy).  This currently requires iterating through each zone,
 *   so it is slower than an id or name search via a hash table.
 *
 *
 *   Locking:
 *
 *   zonehash_lock: This is a top-level global lock used to protect the
 *       zone hash tables and lists.  Zones cannot be created or destroyed
 *       while this lock is held.
 *   zone_status_lock: This is a global lock protecting zone state.
 *       Zones cannot change state while this lock is held.  It also
 *       protects the list of kernel threads associated with a zone.
 *   zone_lock: This is a per-zone lock used to protect several fields of
 *       the zone_t (see <sys/zone.h> for details).  In addition, holding
 *       this lock means that the zone cannot go away.
 *   zone_nlwps_lock: This is a per-zone lock used to protect the fields
 *	 related to the zone.max-lwps rctl.
 *   zone_mem_lock: This is a per-zone lock used to protect the fields
 *	 related to the zone.max-locked-memory and zone.max-swap rctls.
 *   zone_rctl_lock: This is a per-zone lock used to protect other rctls,
 *       currently just max_lofi
 *   zsd_key_lock: This is a global lock protecting the key state for ZSD.
 *   zone_deathrow_lock: This is a global lock protecting the "deathrow"
 *       list (a list of zones in the ZONE_IS_DEAD state).
 *
 *   Ordering requirements:
 *       pool_lock --> cpu_lock --> zonehash_lock --> zone_status_lock -->
 *       	zone_lock --> zsd_key_lock --> pidlock --> p_lock
 *
 *   When taking zone_mem_lock or zone_nlwps_lock, the lock ordering is:
 *	zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_mem_lock
 *	zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_nlwps_lock
 *
 *   Blocking memory allocations are permitted while holding any of the
 *   zone locks.
 *
 *
 *   System Call Interface:
 *
 *   The zone subsystem can be managed and queried from user level with
 *   the following system calls (all subcodes of the primary "zone"
 *   system call):
 *   - zone_create: creates a zone with selected attributes (name,
 *     root path, privileges, resource controls, ZFS datasets)
 *   - zone_enter: allows the current process to enter a zone
 *   - zone_getattr: reports attributes of a zone
 *   - zone_setattr: set attributes of a zone
 *   - zone_boot: set 'init' running for the zone
 *   - zone_list: lists all zones active in the system
 *   - zone_lookup: looks up zone id based on name
 *   - zone_shutdown: initiates shutdown process (see states above)
 *   - zone_destroy: completes shutdown process (see states above)
 *
 */

#include <sys/priv_impl.h>
#include <sys/cred.h>
#include <c2/audit.h>
#include <sys/debug.h>
#include <sys/file.h>
#include <sys/kmem.h>
#include <sys/kstat.h>
#include <sys/mutex.h>
#include <sys/note.h>
#include <sys/pathname.h>
#include <sys/proc.h>
#include <sys/project.h>
#include <sys/sysevent.h>
#include <sys/task.h>
#include <sys/systm.h>
#include <sys/types.h>
#include <sys/utsname.h>
#include <sys/vnode.h>
#include <sys/vfs.h>
#include <sys/systeminfo.h>
#include <sys/policy.h>
#include <sys/cred_impl.h>
#include <sys/contract_impl.h>
#include <sys/contract/process_impl.h>
#include <sys/class.h>
#include <sys/pool.h>
#include <sys/pool_pset.h>
#include <sys/pset.h>
#include <sys/strlog.h>
#include <sys/sysmacros.h>
#include <sys/callb.h>
#include <sys/vmparam.h>
#include <sys/corectl.h>
#include <sys/ipc_impl.h>
#include <sys/klpd.h>

#include <sys/door.h>
#include <sys/cpuvar.h>
#include <sys/sdt.h>

#include <sys/uadmin.h>
#include <sys/session.h>
#include <sys/cmn_err.h>
#include <sys/modhash.h>
#include <sys/sunddi.h>
#include <sys/nvpair.h>
#include <sys/rctl.h>
#include <sys/fss.h>
#include <sys/brand.h>
#include <sys/zone.h>
#include <net/if.h>
#include <sys/cpucaps.h>
#include <vm/seg.h>
#include <sys/mac.h>

/*
 * This constant specifies the number of seconds that threads waiting for
 * subsystems to release a zone's general-purpose references will wait before
 * they log the zone's reference counts.  The constant's value shouldn't
 * be so small that reference counts are unnecessarily reported for zones
 * whose references are slowly released.  On the other hand, it shouldn't be so
 * large that users reboot their systems out of frustration over hung zones
 * before the system logs the zones' reference counts.
 */
#define	ZONE_DESTROY_TIMEOUT_SECS	60

/* List of data link IDs which are accessible from the zone */
typedef struct zone_dl {
	datalink_id_t	zdl_id;
	nvlist_t	*zdl_net;
	list_node_t	zdl_linkage;
} zone_dl_t;

/*
 * cv used to signal that all references to the zone have been released.  This
 * needs to be global since there may be multiple waiters, and the first to
 * wake up will free the zone_t, hence we cannot use zone->zone_cv.
 */
static kcondvar_t zone_destroy_cv;
/*
 * Lock used to serialize access to zone_cv.  This could have been per-zone,
 * but then we'd need another lock for zone_destroy_cv, and why bother?
 */
static kmutex_t zone_status_lock;

/*
 * ZSD-related global variables.
 */
static kmutex_t zsd_key_lock;	/* protects the following two */
/*
 * The next caller of zone_key_create() will be assigned a key of ++zsd_keyval.
 */
static zone_key_t zsd_keyval = 0;
/*
 * Global list of registered keys.  We use this when a new zone is created.
 */
static list_t zsd_registered_keys;

int zone_hash_size = 256;
static mod_hash_t *zonehashbyname, *zonehashbyid, *zonehashbylabel;
static kmutex_t zonehash_lock;
static uint_t zonecount;
static id_space_t *zoneid_space;

/*
 * The global zone (aka zone0) is the all-seeing, all-knowing zone in which the
 * kernel proper runs, and which manages all other zones.
 *
 * Although not declared as static, the variable "zone0" should not be used
 * except for by code that needs to reference the global zone early on in boot,
 * before it is fully initialized.  All other consumers should use
 * 'global_zone'.
 */
zone_t zone0;
zone_t *global_zone = NULL;	/* Set when the global zone is initialized */

/*
 * List of active zones, protected by zonehash_lock.
 */
static list_t zone_active;

/*
 * List of destroyed zones that still have outstanding cred references.
 * Used for debugging.  Uses a separate lock to avoid lock ordering
 * problems in zone_free.
 */
static list_t zone_deathrow;
static kmutex_t zone_deathrow_lock;

/* number of zones is limited by virtual interface limit in IP */
uint_t maxzones = 8192;

/* Event channel to sent zone state change notifications */
evchan_t *zone_event_chan;

/*
 * This table holds the mapping from kernel zone states to
 * states visible in the state notification API.
 * The idea is that we only expose "obvious" states and
 * do not expose states which are just implementation details.
 */
const char  *zone_status_table[] = {
	ZONE_EVENT_UNINITIALIZED,	/* uninitialized */
	ZONE_EVENT_INITIALIZED,		/* initialized */
	ZONE_EVENT_READY,		/* ready */
	ZONE_EVENT_READY,		/* booting */
	ZONE_EVENT_RUNNING,		/* running */
	ZONE_EVENT_SHUTTING_DOWN,	/* shutting_down */
	ZONE_EVENT_SHUTTING_DOWN,	/* empty */
	ZONE_EVENT_SHUTTING_DOWN,	/* down */
	ZONE_EVENT_SHUTTING_DOWN,	/* dying */
	ZONE_EVENT_UNINITIALIZED,	/* dead */
};

/*
 * This array contains the names of the subsystems listed in zone_ref_subsys_t
 * (see sys/zone.h).
 */
static char *zone_ref_subsys_names[] = {
	"NFS",		/* ZONE_REF_NFS */
	"NFSv4",	/* ZONE_REF_NFSV4 */
	"SMBFS",	/* ZONE_REF_SMBFS */
	"MNTFS",	/* ZONE_REF_MNTFS */
	"LOFI",		/* ZONE_REF_LOFI */
	"VFS",		/* ZONE_REF_VFS */
	"IPC"		/* ZONE_REF_IPC */
};

/*
 * This isn't static so lint doesn't complain.
 */
rctl_hndl_t rc_zone_cpu_shares;
rctl_hndl_t rc_zone_locked_mem;
rctl_hndl_t rc_zone_max_swap;
rctl_hndl_t rc_zone_max_lofi;
rctl_hndl_t rc_zone_cpu_cap;
rctl_hndl_t rc_zone_nlwps;
rctl_hndl_t rc_zone_nprocs;
rctl_hndl_t rc_zone_shmmax;
rctl_hndl_t rc_zone_shmmni;
rctl_hndl_t rc_zone_semmni;
rctl_hndl_t rc_zone_msgmni;

const char * const zone_default_initname = "/sbin/init";
static char * const zone_prefix = "/zone/";
static int zone_shutdown(zoneid_t zoneid);
static int zone_add_datalink(zoneid_t, datalink_id_t);
static int zone_remove_datalink(zoneid_t, datalink_id_t);
static int zone_list_datalink(zoneid_t, int *, datalink_id_t *);
static int zone_set_network(zoneid_t, zone_net_data_t *);
static int zone_get_network(zoneid_t, zone_net_data_t *);

typedef boolean_t zsd_applyfn_t(kmutex_t *, boolean_t, zone_t *, zone_key_t);

static void zsd_apply_all_zones(zsd_applyfn_t *, zone_key_t);
static void zsd_apply_all_keys(zsd_applyfn_t *, zone_t *);
static boolean_t zsd_apply_create(kmutex_t *, boolean_t, zone_t *, zone_key_t);
static boolean_t zsd_apply_shutdown(kmutex_t *, boolean_t, zone_t *,
    zone_key_t);
static boolean_t zsd_apply_destroy(kmutex_t *, boolean_t, zone_t *, zone_key_t);
static boolean_t zsd_wait_for_creator(zone_t *, struct zsd_entry *,
    kmutex_t *);
static boolean_t zsd_wait_for_inprogress(zone_t *, struct zsd_entry *,
    kmutex_t *);

/*
 * Bump this number when you alter the zone syscall interfaces; this is
 * because we need to have support for previous API versions in libc
 * to support patching; libc calls into the kernel to determine this number.
 *
 * Version 1 of the API is the version originally shipped with Solaris 10
 * Version 2 alters the zone_create system call in order to support more
 *     arguments by moving the args into a structure; and to do better
 *     error reporting when zone_create() fails.
 * Version 3 alters the zone_create system call in order to support the
 *     import of ZFS datasets to zones.
 * Version 4 alters the zone_create system call in order to support
 *     Trusted Extensions.
 * Version 5 alters the zone_boot system call, and converts its old
 *     bootargs parameter to be set by the zone_setattr API instead.
 * Version 6 adds the flag argument to zone_create.
 */
static const int ZONE_SYSCALL_API_VERSION = 6;

/*
 * Certain filesystems (such as NFS and autofs) need to know which zone
 * the mount is being placed in.  Because of this, we need to be able to
 * ensure that a zone isn't in the process of being created/destroyed such
 * that nfs_mount() thinks it is in the global/NGZ zone, while by the time
 * it gets added the list of mounted zones, it ends up on the wrong zone's
 * mount list. Since a zone can't reside on an NFS file system, we don't
 * have to worry about the zonepath itself.
 *
 * The following functions: block_mounts()/resume_mounts() and
 * mount_in_progress()/mount_completed() are used by zones and the VFS
 * layer (respectively) to synchronize zone state transitions and new
 * mounts within a zone. This syncronization is on a per-zone basis, so
 * activity for one zone will not interfere with activity for another zone.
 *
 * The semantics are like a reader-reader lock such that there may
 * either be multiple mounts (or zone state transitions, if that weren't
 * serialized by zonehash_lock) in progress at the same time, but not
 * both.
 *
 * We use cv's so the user can ctrl-C out of the operation if it's
 * taking too long.
 *
 * The semantics are such that there is unfair bias towards the
 * "current" operation.  This means that zone halt may starve if
 * there is a rapid succession of new mounts coming in to the zone.
 */
/*
 * Prevent new mounts from progressing to the point of calling
 * VFS_MOUNT().  If there are already mounts in this "region", wait for
 * them to complete.
 */
static int
block_mounts(zone_t *zp)
{
	int retval = 0;

	/*
	 * Since it may block for a long time, block_mounts() shouldn't be
	 * called with zonehash_lock held.
	 */
	ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
	mutex_enter(&zp->zone_mount_lock);
	while (zp->zone_mounts_in_progress > 0) {
		if (cv_wait_sig(&zp->zone_mount_cv, &zp->zone_mount_lock) == 0)
			goto signaled;
	}
	/*
	 * A negative value of mounts_in_progress indicates that mounts
	 * have been blocked by (-mounts_in_progress) different callers
	 * (remotely possible if two threads enter zone_shutdown at the same
	 * time).
	 */
	zp->zone_mounts_in_progress--;
	retval = 1;
signaled:
	mutex_exit(&zp->zone_mount_lock);
	return (retval);
}

/*
 * The VFS layer may progress with new mounts as far as we're concerned.
 * Allow them to progress if we were the last obstacle.
 */
static void
resume_mounts(zone_t *zp)
{
	mutex_enter(&zp->zone_mount_lock);
	if (++zp->zone_mounts_in_progress == 0)
		cv_broadcast(&zp->zone_mount_cv);
	mutex_exit(&zp->zone_mount_lock);
}

/*
 * The VFS layer is busy with a mount; this zone should wait until all
 * of its mounts are completed to progress.
 */
void
mount_in_progress(zone_t *zp)
{
	mutex_enter(&zp->zone_mount_lock);
	while (zp->zone_mounts_in_progress < 0)
		cv_wait(&zp->zone_mount_cv, &zp->zone_mount_lock);
	zp->zone_mounts_in_progress++;
	mutex_exit(&zp->zone_mount_lock);
}

/*
 * VFS is done with one mount; wake up any waiting block_mounts()
 * callers if this is the last mount.
 */
void
mount_completed(zone_t *zp)
{
	mutex_enter(&zp->zone_mount_lock);
	if (--zp->zone_mounts_in_progress == 0)
		cv_broadcast(&zp->zone_mount_cv);
	mutex_exit(&zp->zone_mount_lock);
}

/*
 * ZSD routines.
 *
 * Zone Specific Data (ZSD) is modeled after Thread Specific Data as
 * defined by the pthread_key_create() and related interfaces.
 *
 * Kernel subsystems may register one or more data items and/or
 * callbacks to be executed when a zone is created, shutdown, or
 * destroyed.
 *
 * Unlike the thread counterpart, destructor callbacks will be executed
 * even if the data pointer is NULL and/or there are no constructor
 * callbacks, so it is the responsibility of such callbacks to check for
 * NULL data values if necessary.
 *
 * The locking strategy and overall picture is as follows:
 *
 * When someone calls zone_key_create(), a template ZSD entry is added to the
 * global list "zsd_registered_keys", protected by zsd_key_lock.  While
 * holding that lock all the existing zones are marked as
 * ZSD_CREATE_NEEDED and a copy of the ZSD entry added to the per-zone
 * zone_zsd list (protected by zone_lock). The global list is updated first
 * (under zone_key_lock) to make sure that newly created zones use the
 * most recent list of keys. Then under zonehash_lock we walk the zones
 * and mark them.  Similar locking is used in zone_key_delete().
 *
 * The actual create, shutdown, and destroy callbacks are done without
 * holding any lock. And zsd_flags are used to ensure that the operations
 * completed so that when zone_key_create (and zone_create) is done, as well as
 * zone_key_delete (and zone_destroy) is done, all the necessary callbacks
 * are completed.
 *
 * When new zones are created constructor callbacks for all registered ZSD
 * entries will be called. That also uses the above two phases of marking
 * what needs to be done, and then running the callbacks without holding
 * any locks.
 *
 * The framework does not provide any locking around zone_getspecific() and
 * zone_setspecific() apart from that needed for internal consistency, so
 * callers interested in atomic "test-and-set" semantics will need to provide
 * their own locking.
 */

/*
 * Helper function to find the zsd_entry associated with the key in the
 * given list.
 */
static struct zsd_entry *
zsd_find(list_t *l, zone_key_t key)
{
	struct zsd_entry *zsd;

	for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
		if (zsd->zsd_key == key) {
			return (zsd);
		}
	}
	return (NULL);
}

/*
 * Helper function to find the zsd_entry associated with the key in the
 * given list. Move it to the front of the list.
 */
static struct zsd_entry *
zsd_find_mru(list_t *l, zone_key_t key)
{
	struct zsd_entry *zsd;

	for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
		if (zsd->zsd_key == key) {
			/*
			 * Move to head of list to keep list in MRU order.
			 */
			if (zsd != list_head(l)) {
				list_remove(l, zsd);
				list_insert_head(l, zsd);
			}
			return (zsd);
		}
	}
	return (NULL);
}

void
zone_key_create(zone_key_t *keyp, void *(*create)(zoneid_t),
    void (*shutdown)(zoneid_t, void *), void (*destroy)(zoneid_t, void *))
{
	struct zsd_entry *zsdp;
	struct zsd_entry *t;
	struct zone *zone;
	zone_key_t  key;

	zsdp = kmem_zalloc(sizeof (*zsdp), KM_SLEEP);
	zsdp->zsd_data = NULL;
	zsdp->zsd_create = create;
	zsdp->zsd_shutdown = shutdown;
	zsdp->zsd_destroy = destroy;

	/*
	 * Insert in global list of callbacks. Makes future zone creations
	 * see it.
	 */
	mutex_enter(&zsd_key_lock);
	key = zsdp->zsd_key = ++zsd_keyval;
	ASSERT(zsd_keyval != 0);
	list_insert_tail(&zsd_registered_keys, zsdp);
	mutex_exit(&zsd_key_lock);

	/*
	 * Insert for all existing zones and mark them as needing
	 * a create callback.
	 */
	mutex_enter(&zonehash_lock);	/* stop the world */
	for (zone = list_head(&zone_active); zone != NULL;
	    zone = list_next(&zone_active, zone)) {
		zone_status_t status;

		mutex_enter(&zone->zone_lock);

		/* Skip zones that are on the way down or not yet up */
		status = zone_status_get(zone);
		if (status >= ZONE_IS_DOWN ||
		    status == ZONE_IS_UNINITIALIZED) {
			mutex_exit(&zone->zone_lock);
			continue;
		}

		t = zsd_find_mru(&zone->zone_zsd, key);
		if (t != NULL) {
			/*
			 * A zsd_configure already inserted it after
			 * we dropped zsd_key_lock above.
			 */
			mutex_exit(&zone->zone_lock);
			continue;
		}
		t = kmem_zalloc(sizeof (*t), KM_SLEEP);
		t->zsd_key = key;
		t->zsd_create = create;
		t->zsd_shutdown = shutdown;
		t->zsd_destroy = destroy;
		if (create != NULL) {
			t->zsd_flags = ZSD_CREATE_NEEDED;
			DTRACE_PROBE2(zsd__create__needed,
			    zone_t *, zone, zone_key_t, key);
		}
		list_insert_tail(&zone->zone_zsd, t);
		mutex_exit(&zone->zone_lock);
	}
	mutex_exit(&zonehash_lock);

	if (create != NULL) {
		/* Now call the create callback for this key */
		zsd_apply_all_zones(zsd_apply_create, key);
	}
	/*
	 * It is safe for consumers to use the key now, make it
	 * globally visible. Specifically zone_getspecific() will
	 * always successfully return the zone specific data associated
	 * with the key.
	 */
	*keyp = key;

}

/*
 * Function called when a module is being unloaded, or otherwise wishes
 * to unregister its ZSD key and callbacks.
 *
 * Remove from the global list and determine the functions that need to
 * be called under a global lock. Then call the functions without
 * holding any locks. Finally free up the zone_zsd entries. (The apply
 * functions need to access the zone_zsd entries to find zsd_data etc.)
 */
int
zone_key_delete(zone_key_t key)
{
	struct zsd_entry *zsdp = NULL;
	zone_t *zone;

	mutex_enter(&zsd_key_lock);
	zsdp = zsd_find_mru(&zsd_registered_keys, key);
	if (zsdp == NULL) {
		mutex_exit(&zsd_key_lock);
		return (-1);
	}
	list_remove(&zsd_registered_keys, zsdp);
	mutex_exit(&zsd_key_lock);

	mutex_enter(&zonehash_lock);
	for (zone = list_head(&zone_active); zone != NULL;
	    zone = list_next(&zone_active, zone)) {
		struct zsd_entry *del;

		mutex_enter(&zone->zone_lock);
		del = zsd_find_mru(&zone->zone_zsd, key);
		if (del == NULL) {
			/*
			 * Somebody else got here first e.g the zone going
			 * away.
			 */
			mutex_exit(&zone->zone_lock);
			continue;
		}
		ASSERT(del->zsd_shutdown == zsdp->zsd_shutdown);
		ASSERT(del->zsd_destroy == zsdp->zsd_destroy);
		if (del->zsd_shutdown != NULL &&
		    (del->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
			del->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
			DTRACE_PROBE2(zsd__shutdown__needed,
			    zone_t *, zone, zone_key_t, key);
		}
		if (del->zsd_destroy != NULL &&
		    (del->zsd_flags & ZSD_DESTROY_ALL) == 0) {
			del->zsd_flags |= ZSD_DESTROY_NEEDED;
			DTRACE_PROBE2(zsd__destroy__needed,
			    zone_t *, zone, zone_key_t, key);
		}
		mutex_exit(&zone->zone_lock);
	}
	mutex_exit(&zonehash_lock);
	kmem_free(zsdp, sizeof (*zsdp));

	/* Now call the shutdown and destroy callback for this key */
	zsd_apply_all_zones(zsd_apply_shutdown, key);
	zsd_apply_all_zones(zsd_apply_destroy, key);

	/* Now we can free up the zsdp structures in each zone */
	mutex_enter(&zonehash_lock);
	for (zone = list_head(&zone_active); zone != NULL;
	    zone = list_next(&zone_active, zone)) {
		struct zsd_entry *del;

		mutex_enter(&zone->zone_lock);
		del = zsd_find(&zone->zone_zsd, key);
		if (del != NULL) {
			list_remove(&zone->zone_zsd, del);
			ASSERT(!(del->zsd_flags & ZSD_ALL_INPROGRESS));
			kmem_free(del, sizeof (*del));
		}
		mutex_exit(&zone->zone_lock);
	}
	mutex_exit(&zonehash_lock);

	return (0);
}

/*
 * ZSD counterpart of pthread_setspecific().
 *
 * Since all zsd callbacks, including those with no create function,
 * have an entry in zone_zsd, if the key is registered it is part of
 * the zone_zsd list.
 * Return an error if the key wasn't registerd.
 */
int
zone_setspecific(zone_key_t key, zone_t *zone, const void *data)
{
	struct zsd_entry *t;

	mutex_enter(&zone->zone_lock);
	t = zsd_find_mru(&zone->zone_zsd, key);
	if (t != NULL) {
		/*
		 * Replace old value with new
		 */
		t->zsd_data = (void *)data;
		mutex_exit(&zone->zone_lock);
		return (0);
	}
	mutex_exit(&zone->zone_lock);
	return (-1);
}

/*
 * ZSD counterpart of pthread_getspecific().
 */
void *
zone_getspecific(zone_key_t key, zone_t *zone)
{
	struct zsd_entry *t;
	void *data;

	mutex_enter(&zone->zone_lock);
	t = zsd_find_mru(&zone->zone_zsd, key);
	data = (t == NULL ? NULL : t->zsd_data);
	mutex_exit(&zone->zone_lock);
	return (data);
}

/*
 * Function used to initialize a zone's list of ZSD callbacks and data
 * when the zone is being created.  The callbacks are initialized from
 * the template list (zsd_registered_keys). The constructor callback is
 * executed later (once the zone exists and with locks dropped).
 */
static void
zone_zsd_configure(zone_t *zone)
{
	struct zsd_entry *zsdp;
	struct zsd_entry *t;

	ASSERT(MUTEX_HELD(&zonehash_lock));
	ASSERT(list_head(&zone->zone_zsd) == NULL);
	mutex_enter(&zone->zone_lock);
	mutex_enter(&zsd_key_lock);
	for (zsdp = list_head(&zsd_registered_keys); zsdp != NULL;
	    zsdp = list_next(&zsd_registered_keys, zsdp)) {
		/*
		 * Since this zone is ZONE_IS_UNCONFIGURED, zone_key_create
		 * should not have added anything to it.
		 */
		ASSERT(zsd_find(&zone->zone_zsd, zsdp->zsd_key) == NULL);

		t = kmem_zalloc(sizeof (*t), KM_SLEEP);
		t->zsd_key = zsdp->zsd_key;
		t->zsd_create = zsdp->zsd_create;
		t->zsd_shutdown = zsdp->zsd_shutdown;
		t->zsd_destroy = zsdp->zsd_destroy;
		if (zsdp->zsd_create != NULL) {
			t->zsd_flags = ZSD_CREATE_NEEDED;
			DTRACE_PROBE2(zsd__create__needed,
			    zone_t *, zone, zone_key_t, zsdp->zsd_key);
		}
		list_insert_tail(&zone->zone_zsd, t);
	}
	mutex_exit(&zsd_key_lock);
	mutex_exit(&zone->zone_lock);
}

enum zsd_callback_type { ZSD_CREATE, ZSD_SHUTDOWN, ZSD_DESTROY };

/*
 * Helper function to execute shutdown or destructor callbacks.
 */
static void
zone_zsd_callbacks(zone_t *zone, enum zsd_callback_type ct)
{
	struct zsd_entry *t;

	ASSERT(ct == ZSD_SHUTDOWN || ct == ZSD_DESTROY);
	ASSERT(ct != ZSD_SHUTDOWN || zone_status_get(zone) >= ZONE_IS_EMPTY);
	ASSERT(ct != ZSD_DESTROY || zone_status_get(zone) >= ZONE_IS_DOWN);

	/*
	 * Run the callback solely based on what is registered for the zone
	 * in zone_zsd. The global list can change independently of this
	 * as keys are registered and unregistered and we don't register new
	 * callbacks for a zone that is in the process of going away.
	 */
	mutex_enter(&zone->zone_lock);
	for (t = list_head(&zone->zone_zsd); t != NULL;
	    t = list_next(&zone->zone_zsd, t)) {
		zone_key_t key = t->zsd_key;

		/* Skip if no callbacks registered */

		if (ct == ZSD_SHUTDOWN) {
			if (t->zsd_shutdown != NULL &&
			    (t->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
				t->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
				DTRACE_PROBE2(zsd__shutdown__needed,
				    zone_t *, zone, zone_key_t, key);
			}
		} else {
			if (t->zsd_destroy != NULL &&
			    (t->zsd_flags & ZSD_DESTROY_ALL) == 0) {
				t->zsd_flags |= ZSD_DESTROY_NEEDED;
				DTRACE_PROBE2(zsd__destroy__needed,
				    zone_t *, zone, zone_key_t, key);
			}
		}
	}
	mutex_exit(&zone->zone_lock);

	/* Now call the shutdown and destroy callback for this key */
	zsd_apply_all_keys(zsd_apply_shutdown, zone);
	zsd_apply_all_keys(zsd_apply_destroy, zone);

}

/*
 * Called when the zone is going away; free ZSD-related memory, and
 * destroy the zone_zsd list.
 */
static void
zone_free_zsd(zone_t *zone)
{
	struct zsd_entry *t, *next;

	/*
	 * Free all the zsd_entry's we had on this zone.
	 */
	mutex_enter(&zone->zone_lock);
	for (t = list_head(&zone->zone_zsd); t != NULL; t = next) {
		next = list_next(&zone->zone_zsd, t);
		list_remove(&zone->zone_zsd, t);
		ASSERT(!(t->zsd_flags & ZSD_ALL_INPROGRESS));
		kmem_free(t, sizeof (*t));
	}
	list_destroy(&zone->zone_zsd);
	mutex_exit(&zone->zone_lock);

}

/*
 * Apply a function to all zones for particular key value.
 *
 * The applyfn has to drop zonehash_lock if it does some work, and
 * then reacquire it before it returns.
 * When the lock is dropped we don't follow list_next even
 * if it is possible to do so without any hazards. This is
 * because we want the design to allow for the list of zones
 * to change in any arbitrary way during the time the
 * lock was dropped.
 *
 * It is safe to restart the loop at list_head since the applyfn
 * changes the zsd_flags as it does work, so a subsequent
 * pass through will have no effect in applyfn, hence the loop will terminate
 * in at worst O(N^2).
 */
static void
zsd_apply_all_zones(zsd_applyfn_t *applyfn, zone_key_t key)
{
	zone_t *zone;

	mutex_enter(&zonehash_lock);
	zone = list_head(&zone_active);
	while (zone != NULL) {
		if ((applyfn)(&zonehash_lock, B_FALSE, zone, key)) {
			/* Lock dropped - restart at head */
			zone = list_head(&zone_active);
		} else {
			zone = list_next(&zone_active, zone);
		}
	}
	mutex_exit(&zonehash_lock);
}

/*
 * Apply a function to all keys for a particular zone.
 *
 * The applyfn has to drop zonehash_lock if it does some work, and
 * then reacquire it before it returns.
 * When the lock is dropped we don't follow list_next even
 * if it is possible to do so without any hazards. This is
 * because we want the design to allow for the list of zsd callbacks
 * to change in any arbitrary way during the time the
 * lock was dropped.
 *
 * It is safe to restart the loop at list_head since the applyfn
 * changes the zsd_flags as it does work, so a subsequent
 * pass through will have no effect in applyfn, hence the loop will terminate
 * in at worst O(N^2).
 */
static void
zsd_apply_all_keys(zsd_applyfn_t *applyfn, zone_t *zone)
{
	struct zsd_entry *t;

	mutex_enter(&zone->zone_lock);
	t = list_head(&zone->zone_zsd);
	while (t != NULL) {
		if ((applyfn)(NULL, B_TRUE, zone, t->zsd_key)) {
			/* Lock dropped - restart at head */
			t = list_head(&zone->zone_zsd);
		} else {
			t = list_next(&zone->zone_zsd, t);
		}
	}
	mutex_exit(&zone->zone_lock);
}

/*
 * Call the create function for the zone and key if CREATE_NEEDED
 * is set.
 * If some other thread gets here first and sets CREATE_INPROGRESS, then
 * we wait for that thread to complete so that we can ensure that
 * all the callbacks are done when we've looped over all zones/keys.
 *
 * When we call the create function, we drop the global held by the
 * caller, and return true to tell the caller it needs to re-evalute the
 * state.
 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
 * remains held on exit.
 */
static boolean_t
zsd_apply_create(kmutex_t *lockp, boolean_t zone_lock_held,
    zone_t *zone, zone_key_t key)
{
	void *result;
	struct zsd_entry *t;
	boolean_t dropped;

	if (lockp != NULL) {
		ASSERT(MUTEX_HELD(lockp));
	}
	if (zone_lock_held) {
		ASSERT(MUTEX_HELD(&zone->zone_lock));
	} else {
		mutex_enter(&zone->zone_lock);
	}

	t = zsd_find(&zone->zone_zsd, key);
	if (t == NULL) {
		/*
		 * Somebody else got here first e.g the zone going
		 * away.
		 */
		if (!zone_lock_held)
			mutex_exit(&zone->zone_lock);
		return (B_FALSE);
	}
	dropped = B_FALSE;
	if (zsd_wait_for_inprogress(zone, t, lockp))
		dropped = B_TRUE;

	if (t->zsd_flags & ZSD_CREATE_NEEDED) {
		t->zsd_flags &= ~ZSD_CREATE_NEEDED;
		t->zsd_flags |= ZSD_CREATE_INPROGRESS;
		DTRACE_PROBE2(zsd__create__inprogress,
		    zone_t *, zone, zone_key_t, key);
		mutex_exit(&zone->zone_lock);
		if (lockp != NULL)
			mutex_exit(lockp);

		dropped = B_TRUE;
		ASSERT(t->zsd_create != NULL);
		DTRACE_PROBE2(zsd__create__start,
		    zone_t *, zone, zone_key_t, key);

		result = (*t->zsd_create)(zone->zone_id);

		DTRACE_PROBE2(zsd__create__end,
		    zone_t *, zone, voidn *, result);

		ASSERT(result != NULL);
		if (lockp != NULL)
			mutex_enter(lockp);
		mutex_enter(&zone->zone_lock);
		t->zsd_data = result;
		t->zsd_flags &= ~ZSD_CREATE_INPROGRESS;
		t->zsd_flags |= ZSD_CREATE_COMPLETED;
		cv_broadcast(&t->zsd_cv);
		DTRACE_PROBE2(zsd__create__completed,
		    zone_t *, zone, zone_key_t, key);
	}
	if (!zone_lock_held)
		mutex_exit(&zone->zone_lock);
	return (dropped);
}

/*
 * Call the shutdown function for the zone and key if SHUTDOWN_NEEDED
 * is set.
 * If some other thread gets here first and sets *_INPROGRESS, then
 * we wait for that thread to complete so that we can ensure that
 * all the callbacks are done when we've looped over all zones/keys.
 *
 * When we call the shutdown function, we drop the global held by the
 * caller, and return true to tell the caller it needs to re-evalute the
 * state.
 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
 * remains held on exit.
 */
static boolean_t
zsd_apply_shutdown(kmutex_t *lockp, boolean_t zone_lock_held,
    zone_t *zone, zone_key_t key)
{
	struct zsd_entry *t;
	void *data;
	boolean_t dropped;

	if (lockp != NULL) {
		ASSERT(MUTEX_HELD(lockp));
	}
	if (zone_lock_held) {
		ASSERT(MUTEX_HELD(&zone->zone_lock));
	} else {
		mutex_enter(&zone->zone_lock);
	}

	t = zsd_find(&zone->zone_zsd, key);
	if (t == NULL) {
		/*
		 * Somebody else got here first e.g the zone going
		 * away.
		 */
		if (!zone_lock_held)
			mutex_exit(&zone->zone_lock);
		return (B_FALSE);
	}
	dropped = B_FALSE;
	if (zsd_wait_for_creator(zone, t, lockp))
		dropped = B_TRUE;

	if (zsd_wait_for_inprogress(zone, t, lockp))
		dropped = B_TRUE;

	if (t->zsd_flags & ZSD_SHUTDOWN_NEEDED) {
		t->zsd_flags &= ~ZSD_SHUTDOWN_NEEDED;
		t->zsd_flags |= ZSD_SHUTDOWN_INPROGRESS;
		DTRACE_PROBE2(zsd__shutdown__inprogress,
		    zone_t *, zone, zone_key_t, key);
		mutex_exit(&zone->zone_lock);
		if (lockp != NULL)
			mutex_exit(lockp);
		dropped = B_TRUE;

		ASSERT(t->zsd_shutdown != NULL);
		data = t->zsd_data;

		DTRACE_PROBE2(zsd__shutdown__start,
		    zone_t *, zone, zone_key_t, key);

		(t->zsd_shutdown)(zone->zone_id, data);
		DTRACE_PROBE2(zsd__shutdown__end,
		    zone_t *, zone, zone_key_t, key);

		if (lockp != NULL)
			mutex_enter(lockp);
		mutex_enter(&zone->zone_lock);
		t->zsd_flags &= ~ZSD_SHUTDOWN_INPROGRESS;
		t->zsd_flags |= ZSD_SHUTDOWN_COMPLETED;
		cv_broadcast(&t->zsd_cv);
		DTRACE_PROBE2(zsd__shutdown__completed,
		    zone_t *, zone, zone_key_t, key);
	}
	if (!zone_lock_held)
		mutex_exit(&zone->zone_lock);
	return (dropped);
}

/*
 * Call the destroy function for the zone and key if DESTROY_NEEDED
 * is set.
 * If some other thread gets here first and sets *_INPROGRESS, then
 * we wait for that thread to complete so that we can ensure that
 * all the callbacks are done when we've looped over all zones/keys.
 *
 * When we call the destroy function, we drop the global held by the
 * caller, and return true to tell the caller it needs to re-evalute the
 * state.
 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
 * remains held on exit.
 */
static boolean_t
zsd_apply_destroy(kmutex_t *lockp, boolean_t zone_lock_held,
    zone_t *zone, zone_key_t key)
{
	struct zsd_entry *t;
	void *data;
	boolean_t dropped;

	if (lockp != NULL) {
		ASSERT(MUTEX_HELD(lockp));
	}
	if (zone_lock_held) {
		ASSERT(MUTEX_HELD(&zone->zone_lock));
	} else {
		mutex_enter(&zone->zone_lock);
	}

	t = zsd_find(&zone->zone_zsd, key);
	if (t == NULL) {
		/*
		 * Somebody else got here first e.g the zone going
		 * away.
		 */
		if (!zone_lock_held)
			mutex_exit(&zone->zone_lock);
		return (B_FALSE);
	}
	dropped = B_FALSE;
	if (zsd_wait_for_creator(zone, t, lockp))
		dropped = B_TRUE;

	if (zsd_wait_for_inprogress(zone, t, lockp))
		dropped = B_TRUE;

	if (t->zsd_flags & ZSD_DESTROY_NEEDED) {
		t->zsd_flags &= ~ZSD_DESTROY_NEEDED;
		t->zsd_flags |= ZSD_DESTROY_INPROGRESS;
		DTRACE_PROBE2(zsd__destroy__inprogress,
		    zone_t *, zone, zone_key_t, key);
		mutex_exit(&zone->zone_lock);
		if (lockp != NULL)
			mutex_exit(lockp);
		dropped = B_TRUE;

		ASSERT(t->zsd_destroy != NULL);
		data = t->zsd_data;
		DTRACE_PROBE2(zsd__destroy__start,
		    zone_t *, zone, zone_key_t, key);

		(t->zsd_destroy)(zone->zone_id, data);
		DTRACE_PROBE2(zsd__destroy__end,
		    zone_t *, zone, zone_key_t, key);

		if (lockp != NULL)
			mutex_enter(lockp);
		mutex_enter(&zone->zone_lock);
		t->zsd_data = NULL;
		t->zsd_flags &= ~ZSD_DESTROY_INPROGRESS;
		t->zsd_flags |= ZSD_DESTROY_COMPLETED;
		cv_broadcast(&t->zsd_cv);
		DTRACE_PROBE2(zsd__destroy__completed,
		    zone_t *, zone, zone_key_t, key);
	}
	if (!zone_lock_held)
		mutex_exit(&zone->zone_lock);
	return (dropped);
}

/*
 * Wait for any CREATE_NEEDED flag to be cleared.
 * Returns true if lockp was temporarily dropped while waiting.
 */
static boolean_t
zsd_wait_for_creator(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
{
	boolean_t dropped = B_FALSE;

	while (t->zsd_flags & ZSD_CREATE_NEEDED) {
		DTRACE_PROBE2(zsd__wait__for__creator,
		    zone_t *, zone, struct zsd_entry *, t);
		if (lockp != NULL) {
			dropped = B_TRUE;
			mutex_exit(lockp);
		}
		cv_wait(&t->zsd_cv, &zone->zone_lock);
		if (lockp != NULL) {
			/* First drop zone_lock to preserve order */
			mutex_exit(&zone->zone_lock);
			mutex_enter(lockp);
			mutex_enter(&zone->zone_lock);
		}
	}
	return (dropped);
}

/*
 * Wait for any INPROGRESS flag to be cleared.
 * Returns true if lockp was temporarily dropped while waiting.
 */
static boolean_t
zsd_wait_for_inprogress(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
{
	boolean_t dropped = B_FALSE;

	while (t->zsd_flags & ZSD_ALL_INPROGRESS) {
		DTRACE_PROBE2(zsd__wait__for__inprogress,
		    zone_t *, zone, struct zsd_entry *, t);
		if (lockp != NULL) {
			dropped = B_TRUE;
			mutex_exit(lockp);
		}
		cv_wait(&t->zsd_cv, &zone->zone_lock);
		if (lockp != NULL) {
			/* First drop zone_lock to preserve order */
			mutex_exit(&zone->zone_lock);
			mutex_enter(lockp);
			mutex_enter(&zone->zone_lock);
		}
	}
	return (dropped);
}

/*
 * Frees memory associated with the zone dataset list.
 */
static void
zone_free_datasets(zone_t *zone)
{
	zone_dataset_t *t, *next;

	for (t = list_head(&zone->zone_datasets); t != NULL; t = next) {
		next = list_next(&zone->zone_datasets, t);
		list_remove(&zone->zone_datasets, t);
		kmem_free(t->zd_dataset, strlen(t->zd_dataset) + 1);
		kmem_free(t, sizeof (*t));
	}
	list_destroy(&zone->zone_datasets);
}

/*
 * zone.cpu-shares resource control support.
 */
/*ARGSUSED*/
static rctl_qty_t
zone_cpu_shares_usage(rctl_t *rctl, struct proc *p)
{
	ASSERT(MUTEX_HELD(&p->p_lock));
	return (p->p_zone->zone_shares);
}

/*ARGSUSED*/
static int
zone_cpu_shares_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
    rctl_qty_t nv)
{
	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(e->rcep_t == RCENTITY_ZONE);
	if (e->rcep_p.zone == NULL)
		return (0);

	e->rcep_p.zone->zone_shares = nv;
	return (0);
}

static rctl_ops_t zone_cpu_shares_ops = {
	rcop_no_action,
	zone_cpu_shares_usage,
	zone_cpu_shares_set,
	rcop_no_test
};

/*
 * zone.cpu-cap resource control support.
 */
/*ARGSUSED*/
static rctl_qty_t
zone_cpu_cap_get(rctl_t *rctl, struct proc *p)
{
	ASSERT(MUTEX_HELD(&p->p_lock));
	return (cpucaps_zone_get(p->p_zone));
}

/*ARGSUSED*/
static int
zone_cpu_cap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
    rctl_qty_t nv)
{
	zone_t *zone = e->rcep_p.zone;

	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(e->rcep_t == RCENTITY_ZONE);

	if (zone == NULL)
		return (0);

	/*
	 * set cap to the new value.
	 */
	return (cpucaps_zone_set(zone, nv));
}

static rctl_ops_t zone_cpu_cap_ops = {
	rcop_no_action,
	zone_cpu_cap_get,
	zone_cpu_cap_set,
	rcop_no_test
};

/*ARGSUSED*/
static rctl_qty_t
zone_lwps_usage(rctl_t *r, proc_t *p)
{
	rctl_qty_t nlwps;
	zone_t *zone = p->p_zone;

	ASSERT(MUTEX_HELD(&p->p_lock));

	mutex_enter(&zone->zone_nlwps_lock);
	nlwps = zone->zone_nlwps;
	mutex_exit(&zone->zone_nlwps_lock);

	return (nlwps);
}

/*ARGSUSED*/
static int
zone_lwps_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
    rctl_qty_t incr, uint_t flags)
{
	rctl_qty_t nlwps;

	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(e->rcep_t == RCENTITY_ZONE);
	if (e->rcep_p.zone == NULL)
		return (0);
	ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
	nlwps = e->rcep_p.zone->zone_nlwps;

	if (nlwps + incr > rcntl->rcv_value)
		return (1);

	return (0);
}

/*ARGSUSED*/
static int
zone_lwps_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv)
{
	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(e->rcep_t == RCENTITY_ZONE);
	if (e->rcep_p.zone == NULL)
		return (0);
	e->rcep_p.zone->zone_nlwps_ctl = nv;
	return (0);
}

static rctl_ops_t zone_lwps_ops = {
	rcop_no_action,
	zone_lwps_usage,
	zone_lwps_set,
	zone_lwps_test,
};

/*ARGSUSED*/
static rctl_qty_t
zone_procs_usage(rctl_t *r, proc_t *p)
{
	rctl_qty_t nprocs;
	zone_t *zone = p->p_zone;

	ASSERT(MUTEX_HELD(&p->p_lock));

	mutex_enter(&zone->zone_nlwps_lock);
	nprocs = zone->zone_nprocs;
	mutex_exit(&zone->zone_nlwps_lock);

	return (nprocs);
}

/*ARGSUSED*/
static int
zone_procs_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
    rctl_qty_t incr, uint_t flags)
{
	rctl_qty_t nprocs;

	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(e->rcep_t == RCENTITY_ZONE);
	if (e->rcep_p.zone == NULL)
		return (0);
	ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
	nprocs = e->rcep_p.zone->zone_nprocs;

	if (nprocs + incr > rcntl->rcv_value)
		return (1);

	return (0);
}

/*ARGSUSED*/
static int
zone_procs_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv)
{
	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(e->rcep_t == RCENTITY_ZONE);
	if (e->rcep_p.zone == NULL)
		return (0);
	e->rcep_p.zone->zone_nprocs_ctl = nv;
	return (0);
}

static rctl_ops_t zone_procs_ops = {
	rcop_no_action,
	zone_procs_usage,
	zone_procs_set,
	zone_procs_test,
};

/*ARGSUSED*/
static rctl_qty_t
zone_shmmax_usage(rctl_t *rctl, struct proc *p)
{
	ASSERT(MUTEX_HELD(&p->p_lock));
	return (p->p_zone->zone_shmmax);
}

/*ARGSUSED*/
static int
zone_shmmax_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
    rctl_qty_t incr, uint_t flags)
{
	rctl_qty_t v;
	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(e->rcep_t == RCENTITY_ZONE);
	v = e->rcep_p.zone->zone_shmmax + incr;
	if (v > rval->rcv_value)
		return (1);
	return (0);
}

static rctl_ops_t zone_shmmax_ops = {
	rcop_no_action,
	zone_shmmax_usage,
	rcop_no_set,
	zone_shmmax_test
};

/*ARGSUSED*/
static rctl_qty_t
zone_shmmni_usage(rctl_t *rctl, struct proc *p)
{
	ASSERT(MUTEX_HELD(&p->p_lock));
	return (p->p_zone->zone_ipc.ipcq_shmmni);
}

/*ARGSUSED*/
static int
zone_shmmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
    rctl_qty_t incr, uint_t flags)
{
	rctl_qty_t v;
	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(e->rcep_t == RCENTITY_ZONE);
	v = e->rcep_p.zone->zone_ipc.ipcq_shmmni + incr;
	if (v > rval->rcv_value)
		return (1);
	return (0);
}

static rctl_ops_t zone_shmmni_ops = {
	rcop_no_action,
	zone_shmmni_usage,
	rcop_no_set,
	zone_shmmni_test
};

/*ARGSUSED*/
static rctl_qty_t
zone_semmni_usage(rctl_t *rctl, struct proc *p)
{
	ASSERT(MUTEX_HELD(&p->p_lock));
	return (p->p_zone->zone_ipc.ipcq_semmni);
}

/*ARGSUSED*/
static int
zone_semmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
    rctl_qty_t incr, uint_t flags)
{
	rctl_qty_t v;
	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(e->rcep_t == RCENTITY_ZONE);
	v = e->rcep_p.zone->zone_ipc.ipcq_semmni + incr;
	if (v > rval->rcv_value)
		return (1);
	return (0);
}

static rctl_ops_t zone_semmni_ops = {
	rcop_no_action,
	zone_semmni_usage,
	rcop_no_set,
	zone_semmni_test
};

/*ARGSUSED*/
static rctl_qty_t
zone_msgmni_usage(rctl_t *rctl, struct proc *p)
{
	ASSERT(MUTEX_HELD(&p->p_lock));
	return (p->p_zone->zone_ipc.ipcq_msgmni);
}

/*ARGSUSED*/
static int
zone_msgmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
    rctl_qty_t incr, uint_t flags)
{
	rctl_qty_t v;
	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(e->rcep_t == RCENTITY_ZONE);
	v = e->rcep_p.zone->zone_ipc.ipcq_msgmni + incr;
	if (v > rval->rcv_value)
		return (1);
	return (0);
}

static rctl_ops_t zone_msgmni_ops = {
	rcop_no_action,
	zone_msgmni_usage,
	rcop_no_set,
	zone_msgmni_test
};

/*ARGSUSED*/
static rctl_qty_t
zone_locked_mem_usage(rctl_t *rctl, struct proc *p)
{
	rctl_qty_t q;
	ASSERT(MUTEX_HELD(&p->p_lock));
	mutex_enter(&p->p_zone->zone_mem_lock);
	q = p->p_zone->zone_locked_mem;
	mutex_exit(&p->p_zone->zone_mem_lock);
	return (q);
}

/*ARGSUSED*/
static int
zone_locked_mem_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
    rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
{
	rctl_qty_t q;
	zone_t *z;

	z = e->rcep_p.zone;
	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(MUTEX_HELD(&z->zone_mem_lock));
	q = z->zone_locked_mem;
	if (q + incr > rcntl->rcv_value)
		return (1);
	return (0);
}

/*ARGSUSED*/
static int
zone_locked_mem_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
    rctl_qty_t nv)
{
	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(e->rcep_t == RCENTITY_ZONE);
	if (e->rcep_p.zone == NULL)
		return (0);
	e->rcep_p.zone->zone_locked_mem_ctl = nv;
	return (0);
}

static rctl_ops_t zone_locked_mem_ops = {
	rcop_no_action,
	zone_locked_mem_usage,
	zone_locked_mem_set,
	zone_locked_mem_test
};

/*ARGSUSED*/
static rctl_qty_t
zone_max_swap_usage(rctl_t *rctl, struct proc *p)
{
	rctl_qty_t q;
	zone_t *z = p->p_zone;

	ASSERT(MUTEX_HELD(&p->p_lock));
	mutex_enter(&z->zone_mem_lock);
	q = z->zone_max_swap;
	mutex_exit(&z->zone_mem_lock);
	return (q);
}

/*ARGSUSED*/
static int
zone_max_swap_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
    rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
{
	rctl_qty_t q;
	zone_t *z;

	z = e->rcep_p.zone;
	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(MUTEX_HELD(&z->zone_mem_lock));
	q = z->zone_max_swap;
	if (q + incr > rcntl->rcv_value)
		return (1);
	return (0);
}

/*ARGSUSED*/
static int
zone_max_swap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
    rctl_qty_t nv)
{
	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(e->rcep_t == RCENTITY_ZONE);
	if (e->rcep_p.zone == NULL)
		return (0);
	e->rcep_p.zone->zone_max_swap_ctl = nv;
	return (0);
}

static rctl_ops_t zone_max_swap_ops = {
	rcop_no_action,
	zone_max_swap_usage,
	zone_max_swap_set,
	zone_max_swap_test
};

/*ARGSUSED*/
static rctl_qty_t
zone_max_lofi_usage(rctl_t *rctl, struct proc *p)
{
	rctl_qty_t q;
	zone_t *z = p->p_zone;

	ASSERT(MUTEX_HELD(&p->p_lock));
	mutex_enter(&z->zone_rctl_lock);
	q = z->zone_max_lofi;
	mutex_exit(&z->zone_rctl_lock);
	return (q);
}

/*ARGSUSED*/
static int
zone_max_lofi_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
    rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
{
	rctl_qty_t q;
	zone_t *z;

	z = e->rcep_p.zone;
	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(MUTEX_HELD(&z->zone_rctl_lock));
	q = z->zone_max_lofi;
	if (q + incr > rcntl->rcv_value)
		return (1);
	return (0);
}

/*ARGSUSED*/
static int
zone_max_lofi_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
    rctl_qty_t nv)
{
	ASSERT(MUTEX_HELD(&p->p_lock));
	ASSERT(e->rcep_t == RCENTITY_ZONE);
	if (e->rcep_p.zone == NULL)
		return (0);
	e->rcep_p.zone->zone_max_lofi_ctl = nv;
	return (0);
}

static rctl_ops_t zone_max_lofi_ops = {
	rcop_no_action,
	zone_max_lofi_usage,
	zone_max_lofi_set,
	zone_max_lofi_test
};

/*
 * Helper function to brand the zone with a unique ID.
 */
static void
zone_uniqid(zone_t *zone)
{
	static uint64_t uniqid = 0;

	ASSERT(MUTEX_HELD(&zonehash_lock));
	zone->zone_uniqid = uniqid++;
}

/*
 * Returns a held pointer to the "kcred" for the specified zone.
 */
struct cred *
zone_get_kcred(zoneid_t zoneid)
{
	zone_t *zone;
	cred_t *cr;

	if ((zone = zone_find_by_id(zoneid)) == NULL)
		return (NULL);
	cr = zone->zone_kcred;
	crhold(cr);
	zone_rele(zone);
	return (cr);
}

static int
zone_lockedmem_kstat_update(kstat_t *ksp, int rw)
{
	zone_t *zone = ksp->ks_private;
	zone_kstat_t *zk = ksp->ks_data;

	if (rw == KSTAT_WRITE)
		return (EACCES);

	zk->zk_usage.value.ui64 = zone->zone_locked_mem;
	zk->zk_value.value.ui64 = zone->zone_locked_mem_ctl;
	return (0);
}

static int
zone_nprocs_kstat_update(kstat_t *ksp, int rw)
{
	zone_t *zone = ksp->ks_private;
	zone_kstat_t *zk = ksp->ks_data;

	if (rw == KSTAT_WRITE)
		return (EACCES);

	zk->zk_usage.value.ui64 = zone->zone_nprocs;
	zk->zk_value.value.ui64 = zone->zone_nprocs_ctl;
	return (0);
}

static int
zone_swapresv_kstat_update(kstat_t *ksp, int rw)
{
	zone_t *zone = ksp->ks_private;
	zone_kstat_t *zk = ksp->ks_data;

	if (rw == KSTAT_WRITE)
		return (EACCES);

	zk->zk_usage.value.ui64 = zone->zone_max_swap;
	zk->zk_value.value.ui64 = zone->zone_max_swap_ctl;
	return (0);
}

static kstat_t *
zone_kstat_create_common(zone_t *zone, char *name,
    int (*updatefunc) (kstat_t *, int))
{
	kstat_t *ksp;
	zone_kstat_t *zk;

	ksp = rctl_kstat_create_zone(zone, name, KSTAT_TYPE_NAMED,
	    sizeof (zone_kstat_t) / sizeof (kstat_named_t),
	    KSTAT_FLAG_VIRTUAL);

	if (ksp == NULL)
		return (NULL);

	zk = ksp->ks_data = kmem_alloc(sizeof (zone_kstat_t), KM_SLEEP);
	ksp->ks_data_size += strlen(zone->zone_name) + 1;
	kstat_named_init(&zk->zk_zonename, "zonename", KSTAT_DATA_STRING);
	kstat_named_setstr(&zk->zk_zonename, zone->zone_name);
	kstat_named_init(&zk->zk_usage, "usage", KSTAT_DATA_UINT64);
	kstat_named_init(&zk->zk_value, "value", KSTAT_DATA_UINT64);
	ksp->ks_update = updatefunc;
	ksp->ks_private = zone;
	kstat_install(ksp);
	return (ksp);
}


static int
zone_mcap_kstat_update(kstat_t *ksp, int rw)
{
	zone_t *zone = ksp->ks_private;
	zone_mcap_kstat_t *zmp = ksp->ks_data;

	if (rw == KSTAT_WRITE)
		return (EACCES);

	zmp->zm_pgpgin.value.ui64 = zone->zone_pgpgin;
	zmp->zm_anonpgin.value.ui64 = zone->zone_anonpgin;
	zmp->zm_execpgin.value.ui64 = zone->zone_execpgin;
	zmp->zm_fspgin.value.ui64 = zone->zone_fspgin;
	zmp->zm_anon_alloc_fail.value.ui64 = zone->zone_anon_alloc_fail;

	return (0);
}

static kstat_t *
zone_mcap_kstat_create(zone_t *zone)
{
	kstat_t *ksp;
	zone_mcap_kstat_t *zmp;

	if ((ksp = kstat_create_zone("memory_cap", zone->zone_id,
	    zone->zone_name, "zone_memory_cap", KSTAT_TYPE_NAMED,
	    sizeof (zone_mcap_kstat_t) / sizeof (kstat_named_t),
	    KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
		return (NULL);

	if (zone->zone_id != GLOBAL_ZONEID)
		kstat_zone_add(ksp, GLOBAL_ZONEID);

	zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_mcap_kstat_t), KM_SLEEP);
	ksp->ks_data_size += strlen(zone->zone_name) + 1;
	ksp->ks_lock = &zone->zone_mcap_lock;
	zone->zone_mcap_stats = zmp;

	/* The kstat "name" field is not large enough for a full zonename */
	kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
	kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
	kstat_named_init(&zmp->zm_pgpgin, "pgpgin", KSTAT_DATA_UINT64);
	kstat_named_init(&zmp->zm_anonpgin, "anonpgin", KSTAT_DATA_UINT64);
	kstat_named_init(&zmp->zm_execpgin, "execpgin", KSTAT_DATA_UINT64);
	kstat_named_init(&zmp->zm_fspgin, "fspgin", KSTAT_DATA_UINT64);
	kstat_named_init(&zmp->zm_anon_alloc_fail, "anon_alloc_fail",
	    KSTAT_DATA_UINT64);

	ksp->ks_update = zone_mcap_kstat_update;
	ksp->ks_private = zone;

	kstat_install(ksp);
	return (ksp);
}

static int
zone_misc_kstat_update(kstat_t *ksp, int rw)
{
	zone_t *zone = ksp->ks_private;
	zone_misc_kstat_t *zmp = ksp->ks_data;
	hrtime_t tmp;

	if (rw == KSTAT_WRITE)
		return (EACCES);

	tmp = zone->zone_utime;
	scalehrtime(&tmp);
	zmp->zm_utime.value.ui64 = tmp;
	tmp = zone->zone_stime;
	scalehrtime(&tmp);
	zmp->zm_stime.value.ui64 = tmp;
	tmp = zone->zone_wtime;
	scalehrtime(&tmp);
	zmp->zm_wtime.value.ui64 = tmp;

	zmp->zm_avenrun1.value.ui32 = zone->zone_avenrun[0];
	zmp->zm_avenrun5.value.ui32 = zone->zone_avenrun[1];
	zmp->zm_avenrun15.value.ui32 = zone->zone_avenrun[2];

	zmp->zm_ffcap.value.ui32 = zone->zone_ffcap;
	zmp->zm_ffnoproc.value.ui32 = zone->zone_ffnoproc;
	zmp->zm_ffnomem.value.ui32 = zone->zone_ffnomem;
	zmp->zm_ffmisc.value.ui32 = zone->zone_ffmisc;

	zmp->zm_nested_intp.value.ui32 = zone->zone_nested_intp;

	zmp->zm_init_pid.value.ui32 = zone->zone_proc_initpid;
	zmp->zm_boot_time.value.ui64 = (uint64_t)zone->zone_boot_time;

	return (0);
}

static kstat_t *
zone_misc_kstat_create(zone_t *zone)
{
	kstat_t *ksp;
	zone_misc_kstat_t *zmp;

	if ((ksp = kstat_create_zone("zones", zone->zone_id,
	    zone->zone_name, "zone_misc", KSTAT_TYPE_NAMED,
	    sizeof (zone_misc_kstat_t) / sizeof (kstat_named_t),
	    KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
		return (NULL);

	if (zone->zone_id != GLOBAL_ZONEID)
		kstat_zone_add(ksp, GLOBAL_ZONEID);

	zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_misc_kstat_t), KM_SLEEP);
	ksp->ks_data_size += strlen(zone->zone_name) + 1;
	ksp->ks_lock = &zone->zone_misc_lock;
	zone->zone_misc_stats = zmp;

	/* The kstat "name" field is not large enough for a full zonename */
	kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
	kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
	kstat_named_init(&zmp->zm_utime, "nsec_user", KSTAT_DATA_UINT64);
	kstat_named_init(&zmp->zm_stime, "nsec_sys", KSTAT_DATA_UINT64);
	kstat_named_init(&zmp->zm_wtime, "nsec_waitrq", KSTAT_DATA_UINT64);
	kstat_named_init(&zmp->zm_avenrun1, "avenrun_1min", KSTAT_DATA_UINT32);
	kstat_named_init(&zmp->zm_avenrun5, "avenrun_5min", KSTAT_DATA_UINT32);
	kstat_named_init(&zmp->zm_avenrun15, "avenrun_15min",
	    KSTAT_DATA_UINT32);
	kstat_named_init(&zmp->zm_ffcap, "forkfail_cap", KSTAT_DATA_UINT32);
	kstat_named_init(&zmp->zm_ffnoproc, "forkfail_noproc",
	    KSTAT_DATA_UINT32);
	kstat_named_init(&zmp->zm_ffnomem, "forkfail_nomem", KSTAT_DATA_UINT32);
	kstat_named_init(&zmp->zm_ffmisc, "forkfail_misc", KSTAT_DATA_UINT32);
	kstat_named_init(&zmp->zm_nested_intp, "nested_interp",
	    KSTAT_DATA_UINT32);
	kstat_named_init(&zmp->zm_init_pid, "init_pid", KSTAT_DATA_UINT32);
	kstat_named_init(&zmp->zm_boot_time, "boot_time", KSTAT_DATA_UINT64);

	ksp->ks_update = zone_misc_kstat_update;
	ksp->ks_private = zone;

	kstat_install(ksp);
	return (ksp);
}

static void
zone_kstat_create(zone_t *zone)
{
	zone->zone_lockedmem_kstat = zone_kstat_create_common(zone,
	    "lockedmem", zone_lockedmem_kstat_update);
	zone->zone_swapresv_kstat = zone_kstat_create_common(zone,
	    "swapresv", zone_swapresv_kstat_update);
	zone->zone_nprocs_kstat = zone_kstat_create_common(zone,
	    "nprocs", zone_nprocs_kstat_update);

	if ((zone->zone_mcap_ksp = zone_mcap_kstat_create(zone)) == NULL) {
		zone->zone_mcap_stats = kmem_zalloc(
		    sizeof (zone_mcap_kstat_t), KM_SLEEP);
	}

	if ((zone->zone_misc_ksp = zone_misc_kstat_create(zone)) == NULL) {
		zone->zone_misc_stats = kmem_zalloc(
		    sizeof (zone_misc_kstat_t), KM_SLEEP);
	}
}

static void
zone_kstat_delete_common(kstat_t **pkstat, size_t datasz)
{
	void *data;

	if (*pkstat != NULL) {
		data = (*pkstat)->ks_data;
		kstat_delete(*pkstat);
		kmem_free(data, datasz);
		*pkstat = NULL;
	}
}

static void
zone_kstat_delete(zone_t *zone)
{
	zone_kstat_delete_common(&zone->zone_lockedmem_kstat,
	    sizeof (zone_kstat_t));
	zone_kstat_delete_common(&zone->zone_swapresv_kstat,
	    sizeof (zone_kstat_t));
	zone_kstat_delete_common(&zone->zone_nprocs_kstat,
	    sizeof (zone_kstat_t));
	zone_kstat_delete_common(&zone->zone_mcap_ksp,
	    sizeof (zone_mcap_kstat_t));
	zone_kstat_delete_common(&zone->zone_misc_ksp,
	    sizeof (zone_misc_kstat_t));
}

/*
 * Called very early on in boot to initialize the ZSD list so that
 * zone_key_create() can be called before zone_init().  It also initializes
 * portions of zone0 which may be used before zone_init() is called.  The
 * variable "global_zone" will be set when zone0 is fully initialized by
 * zone_init().
 */
void
zone_zsd_init(void)
{
	mutex_init(&zonehash_lock, NULL, MUTEX_DEFAULT, NULL);
	mutex_init(&zsd_key_lock, NULL, MUTEX_DEFAULT, NULL);
	list_create(&zsd_registered_keys, sizeof (struct zsd_entry),
	    offsetof(struct zsd_entry, zsd_linkage));
	list_create(&zone_active, sizeof (zone_t),
	    offsetof(zone_t, zone_linkage));
	list_create(&zone_deathrow, sizeof (zone_t),
	    offsetof(zone_t, zone_linkage));

	mutex_init(&zone0.zone_lock, NULL, MUTEX_DEFAULT, NULL);
	mutex_init(&zone0.zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
	mutex_init(&zone0.zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
	zone0.zone_shares = 1;
	zone0.zone_nlwps = 0;
	zone0.zone_nlwps_ctl = INT_MAX;
	zone0.zone_nprocs = 0;
	zone0.zone_nprocs_ctl = INT_MAX;
	zone0.zone_locked_mem = 0;
	zone0.zone_locked_mem_ctl = UINT64_MAX;
	ASSERT(zone0.zone_max_swap == 0);
	zone0.zone_max_swap_ctl = UINT64_MAX;
	zone0.zone_max_lofi = 0;
	zone0.zone_max_lofi_ctl = UINT64_MAX;
	zone0.zone_shmmax = 0;
	zone0.zone_ipc.ipcq_shmmni = 0;
	zone0.zone_ipc.ipcq_semmni = 0;
	zone0.zone_ipc.ipcq_msgmni = 0;
	zone0.zone_name = GLOBAL_ZONENAME;
	zone0.zone_nodename = utsname.nodename;
	zone0.zone_domain = srpc_domain;
	zone0.zone_hostid = HW_INVALID_HOSTID;
	zone0.zone_fs_allowed = NULL;
	psecflags_default(&zone0.zone_secflags);
	zone0.zone_ref = 1;
	zone0.zone_id = GLOBAL_ZONEID;
	zone0.zone_status = ZONE_IS_RUNNING;
	zone0.zone_rootpath = "/";
	zone0.zone_rootpathlen = 2;
	zone0.zone_psetid = ZONE_PS_INVAL;
	zone0.zone_ncpus = 0;
	zone0.zone_ncpus_online = 0;
	zone0.zone_proc_initpid = 1;
	zone0.zone_initname = initname;
	zone0.zone_lockedmem_kstat = NULL;
	zone0.zone_swapresv_kstat = NULL;
	zone0.zone_nprocs_kstat = NULL;

	zone0.zone_stime = 0;
	zone0.zone_utime = 0;
	zone0.zone_wtime = 0;

	list_create(&zone0.zone_ref_list, sizeof (zone_ref_t),
	    offsetof(zone_ref_t, zref_linkage));
	list_create(&zone0.zone_zsd, sizeof (struct zsd_entry),
	    offsetof(struct zsd_entry, zsd_linkage));
	list_insert_head(&zone_active, &zone0);

	/*
	 * The root filesystem is not mounted yet, so zone_rootvp cannot be set
	 * to anything meaningful.  It is assigned to be 'rootdir' in
	 * vfs_mountroot().
	 */
	zone0.zone_rootvp = NULL;
	zone0.zone_vfslist = NULL;
	zone0.zone_bootargs = initargs;
	zone0.zone_privset = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
	/*
	 * The global zone has all privileges
	 */
	priv_fillset(zone0.zone_privset);
	/*
	 * Add p0 to the global zone
	 */
	zone0.zone_zsched = &p0;
	p0.p_zone = &zone0;
}

/*
 * Compute a hash value based on the contents of the label and the DOI.  The
 * hash algorithm is somewhat arbitrary, but is based on the observation that
 * humans will likely pick labels that differ by amounts that work out to be
 * multiples of the number of hash chains, and thus stirring in some primes
 * should help.
 */
static uint_t
hash_bylabel(void *hdata, mod_hash_key_t key)
{
	const ts_label_t *lab = (ts_label_t *)key;
	const uint32_t *up, *ue;
	uint_t hash;
	int i;

	_NOTE(ARGUNUSED(hdata));

	hash = lab->tsl_doi + (lab->tsl_doi << 1);
	/* we depend on alignment of label, but not representation */
	up = (const uint32_t *)&lab->tsl_label;
	ue = up + sizeof (lab->tsl_label) / sizeof (*up);
	i = 1;
	while (up < ue) {
		/* using 2^n + 1, 1 <= n <= 16 as source of many primes */
		hash += *up + (*up << ((i % 16) + 1));
		up++;
		i++;
	}
	return (hash);
}

/*
 * All that mod_hash cares about here is zero (equal) versus non-zero (not
 * equal).  This may need to be changed if less than / greater than is ever
 * needed.
 */
static int
hash_labelkey_cmp(mod_hash_key_t key1, mod_hash_key_t key2)
{
	ts_label_t *lab1 = (ts_label_t *)key1;
	ts_label_t *lab2 = (ts_label_t *)key2;

	return (label_equal(lab1, lab2) ? 0 : 1);
}

/*
 * Called by main() to initialize the zones framework.
 */
void
zone_init(void)
{
	rctl_dict_entry_t *rde;
	rctl_val_t *dval;
	rctl_set_t *set;
	rctl_alloc_gp_t *gp;
	rctl_entity_p_t e;
	int res;

	ASSERT(curproc == &p0);

	/*
	 * Create ID space for zone IDs.  ID 0 is reserved for the
	 * global zone.
	 */
	zoneid_space = id_space_create("zoneid_space", 1, MAX_ZONEID);

	/*
	 * Initialize generic zone resource controls, if any.
	 */
	rc_zone_cpu_shares = rctl_register("zone.cpu-shares",
	    RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_NEVER |
	    RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT | RCTL_GLOBAL_SYSLOG_NEVER,
	    FSS_MAXSHARES, FSS_MAXSHARES, &zone_cpu_shares_ops);

	rc_zone_cpu_cap = rctl_register("zone.cpu-cap",
	    RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_ALWAYS |
	    RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |RCTL_GLOBAL_SYSLOG_NEVER |
	    RCTL_GLOBAL_INFINITE,
	    MAXCAP, MAXCAP, &zone_cpu_cap_ops);

	rc_zone_nlwps = rctl_register("zone.max-lwps", RCENTITY_ZONE,
	    RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
	    INT_MAX, INT_MAX, &zone_lwps_ops);

	rc_zone_nprocs = rctl_register("zone.max-processes", RCENTITY_ZONE,
	    RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
	    INT_MAX, INT_MAX, &zone_procs_ops);

	/*
	 * System V IPC resource controls
	 */
	rc_zone_msgmni = rctl_register("zone.max-msg-ids",
	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
	    RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_msgmni_ops);

	rc_zone_semmni = rctl_register("zone.max-sem-ids",
	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
	    RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_semmni_ops);

	rc_zone_shmmni = rctl_register("zone.max-shm-ids",
	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
	    RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_shmmni_ops);

	rc_zone_shmmax = rctl_register("zone.max-shm-memory",
	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
	    RCTL_GLOBAL_BYTES, UINT64_MAX, UINT64_MAX, &zone_shmmax_ops);

	/*
	 * Create a rctl_val with PRIVILEGED, NOACTION, value = 1.  Then attach
	 * this at the head of the rctl_dict_entry for ``zone.cpu-shares''.
	 */
	dval = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
	bzero(dval, sizeof (rctl_val_t));
	dval->rcv_value = 1;
	dval->rcv_privilege = RCPRIV_PRIVILEGED;
	dval->rcv_flagaction = RCTL_LOCAL_NOACTION;
	dval->rcv_action_recip_pid = -1;

	rde = rctl_dict_lookup("zone.cpu-shares");
	(void) rctl_val_list_insert(&rde->rcd_default_value, dval);

	rc_zone_locked_mem = rctl_register("zone.max-locked-memory",
	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
	    &zone_locked_mem_ops);

	rc_zone_max_swap = rctl_register("zone.max-swap",
	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
	    &zone_max_swap_ops);

	rc_zone_max_lofi = rctl_register("zone.max-lofi",
	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |
	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
	    &zone_max_lofi_ops);

	/*
	 * Initialize the ``global zone''.
	 */
	set = rctl_set_create();
	gp = rctl_set_init_prealloc(RCENTITY_ZONE);
	mutex_enter(&p0.p_lock);
	e.rcep_p.zone = &zone0;
	e.rcep_t = RCENTITY_ZONE;
	zone0.zone_rctls = rctl_set_init(RCENTITY_ZONE, &p0, &e, set,
	    gp);

	zone0.zone_nlwps = p0.p_lwpcnt;
	zone0.zone_nprocs = 1;
	zone0.zone_ntasks = 1;
	mutex_exit(&p0.p_lock);
	zone0.zone_restart_init = B_TRUE;
	zone0.zone_brand = &native_brand;
	rctl_prealloc_destroy(gp);
	/*
	 * pool_default hasn't been initialized yet, so we let pool_init()
	 * take care of making sure the global zone is in the default pool.
	 */

	/*
	 * Initialize global zone kstats
	 */
	zone_kstat_create(&zone0);

	/*
	 * Initialize zone label.
	 * mlp are initialized when tnzonecfg is loaded.
	 */
	zone0.zone_slabel = l_admin_low;
	rw_init(&zone0.zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
	label_hold(l_admin_low);

	/*
	 * Initialise the lock for the database structure used by mntfs.
	 */
	rw_init(&zone0.zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL);

	mutex_enter(&zonehash_lock);
	zone_uniqid(&zone0);
	ASSERT(zone0.zone_uniqid == GLOBAL_ZONEUNIQID);

	zonehashbyid = mod_hash_create_idhash("zone_by_id", zone_hash_size,
	    mod_hash_null_valdtor);
	zonehashbyname = mod_hash_create_strhash("zone_by_name",
	    zone_hash_size, mod_hash_null_valdtor);
	/*
	 * maintain zonehashbylabel only for labeled systems
	 */
	if (is_system_labeled())
		zonehashbylabel = mod_hash_create_extended("zone_by_label",
		    zone_hash_size, mod_hash_null_keydtor,
		    mod_hash_null_valdtor, hash_bylabel, NULL,
		    hash_labelkey_cmp, KM_SLEEP);
	zonecount = 1;

	(void) mod_hash_insert(zonehashbyid, (mod_hash_key_t)GLOBAL_ZONEID,
	    (mod_hash_val_t)&zone0);
	(void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)zone0.zone_name,
	    (mod_hash_val_t)&zone0);
	if (is_system_labeled()) {
		zone0.zone_flags |= ZF_HASHED_LABEL;
		(void) mod_hash_insert(zonehashbylabel,
		    (mod_hash_key_t)zone0.zone_slabel, (mod_hash_val_t)&zone0);
	}
	mutex_exit(&zonehash_lock);

	/*
	 * We avoid setting zone_kcred until now, since kcred is initialized
	 * sometime after zone_zsd_init() and before zone_init().
	 */
	zone0.zone_kcred = kcred;
	/*
	 * The global zone is fully initialized (except for zone_rootvp which
	 * will be set when the root filesystem is mounted).
	 */
	global_zone = &zone0;

	/*
	 * Setup an event channel to send zone status change notifications on
	 */
	res = sysevent_evc_bind(ZONE_EVENT_CHANNEL, &zone_event_chan,
	    EVCH_CREAT);

	if (res)
		panic("Sysevent_evc_bind failed during zone setup.\n");

}

static void
zone_free(zone_t *zone)
{
	ASSERT(zone != global_zone);
	ASSERT(zone->zone_ntasks == 0);
	ASSERT(zone->zone_nlwps == 0);
	ASSERT(zone->zone_nprocs == 0);
	ASSERT(zone->zone_cred_ref == 0);
	ASSERT(zone->zone_kcred == NULL);
	ASSERT(zone_status_get(zone) == ZONE_IS_DEAD ||
	    zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
	ASSERT(list_is_empty(&zone->zone_ref_list));

	/*
	 * Remove any zone caps.
	 */
	cpucaps_zone_remove(zone);

	ASSERT(zone->zone_cpucap == NULL);

	/* remove from deathrow list */
	if (zone_status_get(zone) == ZONE_IS_DEAD) {
		ASSERT(zone->zone_ref == 0);
		mutex_enter(&zone_deathrow_lock);
		list_remove(&zone_deathrow, zone);
		mutex_exit(&zone_deathrow_lock);
	}

	list_destroy(&zone->zone_ref_list);
	zone_free_zsd(zone);
	zone_free_datasets(zone);
	list_destroy(&zone->zone_dl_list);

	if (zone->zone_rootvp != NULL)
		VN_RELE(zone->zone_rootvp);
	if (zone->zone_rootpath)
		kmem_free(zone->zone_rootpath, zone->zone_rootpathlen);
	if (zone->zone_name != NULL)
		kmem_free(zone->zone_name, ZONENAME_MAX);
	if (zone->zone_slabel != NULL)
		label_rele(zone->zone_slabel);
	if (zone->zone_nodename != NULL)
		kmem_free(zone->zone_nodename, _SYS_NMLN);
	if (zone->zone_domain != NULL)
		kmem_free(zone->zone_domain, _SYS_NMLN);
	if (zone->zone_privset != NULL)
		kmem_free(zone->zone_privset, sizeof (priv_set_t));
	if (zone->zone_rctls != NULL)
		rctl_set_free(zone->zone_rctls);
	if (zone->zone_bootargs != NULL)
		strfree(zone->zone_bootargs);
	if (zone->zone_initname != NULL)
		strfree(zone->zone_initname);
	if (zone->zone_fs_allowed != NULL)
		strfree(zone->zone_fs_allowed);
	if (zone->zone_pfexecd != NULL)
		klpd_freelist(&zone->zone_pfexecd);
	id_free(zoneid_space, zone->zone_id);
	mutex_destroy(&zone->zone_lock);
	cv_destroy(&zone->zone_cv);
	rw_destroy(&zone->zone_mlps.mlpl_rwlock);
	rw_destroy(&zone->zone_mntfs_db_lock);
	kmem_free(zone, sizeof (zone_t));
}

/*
 * See block comment at the top of this file for information about zone
 * status values.
 */
/*
 * Convenience function for setting zone status.
 */
static void
zone_status_set(zone_t *zone, zone_status_t status)
{

	nvlist_t *nvl = NULL;
	ASSERT(MUTEX_HELD(&zone_status_lock));
	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE &&
	    status >= zone_status_get(zone));

	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) ||
	    nvlist_add_string(nvl, ZONE_CB_NAME, zone->zone_name) ||
	    nvlist_add_string(nvl, ZONE_CB_NEWSTATE,
	    zone_status_table[status]) ||
	    nvlist_add_string(nvl, ZONE_CB_OLDSTATE,
	    zone_status_table[zone->zone_status]) ||
	    nvlist_add_int32(nvl, ZONE_CB_ZONEID, zone->zone_id) ||
	    nvlist_add_uint64(nvl, ZONE_CB_TIMESTAMP, (uint64_t)gethrtime()) ||
	    sysevent_evc_publish(zone_event_chan, ZONE_EVENT_STATUS_CLASS,
	    ZONE_EVENT_STATUS_SUBCLASS, "sun.com", "kernel", nvl, EVCH_SLEEP)) {
#ifdef DEBUG
		(void) printf(
		    "Failed to allocate and send zone state change event.\n");
#endif
	}
	nvlist_free(nvl);

	zone->zone_status = status;

	cv_broadcast(&zone->zone_cv);
}

/*
 * Public function to retrieve the zone status.  The zone status may
 * change after it is retrieved.
 */
zone_status_t
zone_status_get(zone_t *zone)
{
	return (zone->zone_status);
}

static int
zone_set_bootargs(zone_t *zone, const char *zone_bootargs)
{
	char *buf = kmem_zalloc(BOOTARGS_MAX, KM_SLEEP);
	int err = 0;

	ASSERT(zone != global_zone);
	if ((err = copyinstr(zone_bootargs, buf, BOOTARGS_MAX, NULL)) != 0)
		goto done;	/* EFAULT or ENAMETOOLONG */

	if (zone->zone_bootargs != NULL)
		strfree(zone->zone_bootargs);

	zone->zone_bootargs = strdup(buf);

done:
	kmem_free(buf, BOOTARGS_MAX);
	return (err);
}

static int
zone_set_brand(zone_t *zone, const char *brand)
{
	struct brand_attr *attrp;
	brand_t *bp;

	attrp = kmem_alloc(sizeof (struct brand_attr), KM_SLEEP);
	if (copyin(brand, attrp, sizeof (struct brand_attr)) != 0) {
		kmem_free(attrp, sizeof (struct brand_attr));
		return (EFAULT);
	}

	bp = brand_register_zone(attrp);
	kmem_free(attrp, sizeof (struct brand_attr));
	if (bp == NULL)
		return (EINVAL);

	/*
	 * This is the only place where a zone can change it's brand.
	 * We already need to hold zone_status_lock to check the zone
	 * status, so we'll just use that lock to serialize zone
	 * branding requests as well.
	 */
	mutex_enter(&zone_status_lock);

	/* Re-Branding is not allowed and the zone can't be booted yet */
	if ((ZONE_IS_BRANDED(zone)) ||
	    (zone_status_get(zone) >= ZONE_IS_BOOTING)) {
		mutex_exit(&zone_status_lock);
		brand_unregister_zone(bp);
		return (EINVAL);
	}

	/* set up the brand specific data */
	zone->zone_brand = bp;
	ZBROP(zone)->b_init_brand_data(zone);

	mutex_exit(&zone_status_lock);
	return (0);
}

static int
zone_set_secflags(zone_t *zone, const psecflags_t *zone_secflags)
{
	int err = 0;
	psecflags_t psf;

	ASSERT(zone != global_zone);

	if ((err = copyin(zone_secflags, &psf, sizeof (psf))) != 0)
		return (err);

	if (zone_status_get(zone) > ZONE_IS_READY)
		return (EINVAL);

	if (!psecflags_validate(&psf))
		return (EINVAL);

	(void) memcpy(&zone->zone_secflags, &psf, sizeof (psf));

	/* Set security flags on the zone's zsched */
	(void) memcpy(&zone->zone_zsched->p_secflags, &zone->zone_secflags,
	    sizeof (zone->zone_zsched->p_secflags));

	return (0);
}

static int
zone_set_fs_allowed(zone_t *zone, const char *zone_fs_allowed)
{
	char *buf = kmem_zalloc(ZONE_FS_ALLOWED_MAX, KM_SLEEP);
	int err = 0;

	ASSERT(zone != global_zone);
	if ((err = copyinstr(zone_fs_allowed, buf,
	    ZONE_FS_ALLOWED_MAX, NULL)) != 0)
		goto done;

	if (zone->zone_fs_allowed != NULL)
		strfree(zone->zone_fs_allowed);

	zone->zone_fs_allowed = strdup(buf);

done:
	kmem_free(buf, ZONE_FS_ALLOWED_MAX);
	return (err);
}

static int
zone_set_initname(zone_t *zone, const char *zone_initname)
{
	char initname[INITNAME_SZ];
	size_t len;
	int err = 0;

	ASSERT(zone != global_zone);
	if ((err = copyinstr(zone_initname, initname, INITNAME_SZ, &len)) != 0)
		return (err);	/* EFAULT or ENAMETOOLONG */

	if (zone->zone_initname != NULL)
		strfree(zone->zone_initname);

	zone->zone_initname = kmem_alloc(strlen(initname) + 1, KM_SLEEP);
	(void) strcpy(zone->zone_initname, initname);
	return (0);
}

static int
zone_set_phys_mcap(zone_t *zone, const uint64_t *zone_mcap)
{
	uint64_t mcap;
	int err = 0;

	if ((err = copyin(zone_mcap, &mcap, sizeof (uint64_t))) == 0)
		zone->zone_phys_mcap = mcap;

	return (err);
}

static int
zone_set_sched_class(zone_t *zone, const char *new_class)
{
	char sched_class[PC_CLNMSZ];
	id_t classid;
	int err;

	ASSERT(zone != global_zone);
	if ((err = copyinstr(new_class, sched_class, PC_CLNMSZ, NULL)) != 0)
		return (err);	/* EFAULT or ENAMETOOLONG */

	if (getcid(sched_class, &classid) != 0 || CLASS_KERNEL(classid))
		return (set_errno(EINVAL));
	zone->zone_defaultcid = classid;
	ASSERT(zone->zone_defaultcid > 0 &&
	    zone->zone_defaultcid < loaded_classes);

	return (0);
}

/*
 * Block indefinitely waiting for (zone_status >= status)
 */
void
zone_status_wait(zone_t *zone, zone_status_t status)
{
	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);

	mutex_enter(&zone_status_lock);
	while (zone->zone_status < status) {
		cv_wait(&zone->zone_cv, &zone_status_lock);
	}
	mutex_exit(&zone_status_lock);
}

/*
 * Private CPR-safe version of zone_status_wait().
 */
static void
zone_status_wait_cpr(zone_t *zone, zone_status_t status, char *str)
{
	callb_cpr_t cprinfo;

	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);

	CALLB_CPR_INIT(&cprinfo, &zone_status_lock, callb_generic_cpr,
	    str);
	mutex_enter(&zone_status_lock);
	while (zone->zone_status < status) {
		CALLB_CPR_SAFE_BEGIN(&cprinfo);
		cv_wait(&zone->zone_cv, &zone_status_lock);
		CALLB_CPR_SAFE_END(&cprinfo, &zone_status_lock);
	}
	/*
	 * zone_status_lock is implicitly released by the following.
	 */
	CALLB_CPR_EXIT(&cprinfo);
}

/*
 * Block until zone enters requested state or signal is received.  Return (0)
 * if signaled, non-zero otherwise.
 */
int
zone_status_wait_sig(zone_t *zone, zone_status_t status)
{
	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);

	mutex_enter(&zone_status_lock);
	while (zone->zone_status < status) {
		if (!cv_wait_sig(&zone->zone_cv, &zone_status_lock)) {
			mutex_exit(&zone_status_lock);
			return (0);
		}
	}
	mutex_exit(&zone_status_lock);
	return (1);
}

/*
 * Block until the zone enters the requested state or the timeout expires,
 * whichever happens first.  Return (-1) if operation timed out, time remaining
 * otherwise.
 */
clock_t
zone_status_timedwait(zone_t *zone, clock_t tim, zone_status_t status)
{
	clock_t timeleft = 0;

	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);

	mutex_enter(&zone_status_lock);
	while (zone->zone_status < status && timeleft != -1) {
		timeleft = cv_timedwait(&zone->zone_cv, &zone_status_lock, tim);
	}
	mutex_exit(&zone_status_lock);
	return (timeleft);
}

/*
 * Block until the zone enters the requested state, the current process is
 * signaled,  or the timeout expires, whichever happens first.  Return (-1) if
 * operation timed out, 0 if signaled, time remaining otherwise.
 */
clock_t
zone_status_timedwait_sig(zone_t *zone, clock_t tim, zone_status_t status)
{
	clock_t timeleft = tim - ddi_get_lbolt();

	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);

	mutex_enter(&zone_status_lock);
	while (zone->zone_status < status) {
		timeleft = cv_timedwait_sig(&zone->zone_cv, &zone_status_lock,
		    tim);
		if (timeleft <= 0)
			break;
	}
	mutex_exit(&zone_status_lock);
	return (timeleft);
}

/*
 * Zones have two reference counts: one for references from credential
 * structures (zone_cred_ref), and one (zone_ref) for everything else.
 * This is so we can allow a zone to be rebooted while there are still
 * outstanding cred references, since certain drivers cache dblks (which
 * implicitly results in cached creds).  We wait for zone_ref to drop to
 * 0 (actually 1), but not zone_cred_ref.  The zone structure itself is
 * later freed when the zone_cred_ref drops to 0, though nothing other
 * than the zone id and privilege set should be accessed once the zone
 * is "dead".
 *
 * A debugging flag, zone_wait_for_cred, can be set to a non-zero value
 * to force halt/reboot to block waiting for the zone_cred_ref to drop
 * to 0.  This can be useful to flush out other sources of cached creds
 * that may be less innocuous than the driver case.
 *
 * Zones also provide a tracked reference counting mechanism in which zone
 * references are represented by "crumbs" (zone_ref structures).  Crumbs help
 * debuggers determine the sources of leaked zone references.  See
 * zone_hold_ref() and zone_rele_ref() below for more information.
 */

int zone_wait_for_cred = 0;

static void
zone_hold_locked(zone_t *z)
{
	ASSERT(MUTEX_HELD(&z->zone_lock));
	z->zone_ref++;
	ASSERT(z->zone_ref != 0);
}

/*
 * Increment the specified zone's reference count.  The zone's zone_t structure
 * will not be freed as long as the zone's reference count is nonzero.
 * Decrement the zone's reference count via zone_rele().
 *
 * NOTE: This function should only be used to hold zones for short periods of
 * time.  Use zone_hold_ref() if the zone must be held for a long time.
 */
void
zone_hold(zone_t *z)
{
	mutex_enter(&z->zone_lock);
	zone_hold_locked(z);
	mutex_exit(&z->zone_lock);
}

/*
 * If the non-cred ref count drops to 1 and either the cred ref count
 * is 0 or we aren't waiting for cred references, the zone is ready to
 * be destroyed.
 */
#define	ZONE_IS_UNREF(zone)	((zone)->zone_ref == 1 && \
	    (!zone_wait_for_cred || (zone)->zone_cred_ref == 0))

/*
 * Common zone reference release function invoked by zone_rele() and
 * zone_rele_ref().  If subsys is ZONE_REF_NUM_SUBSYS, then the specified
 * zone's subsystem-specific reference counters are not affected by the
 * release.  If ref is not NULL, then the zone_ref_t to which it refers is
 * removed from the specified zone's reference list.  ref must be non-NULL iff
 * subsys is not ZONE_REF_NUM_SUBSYS.
 */
static void
zone_rele_common(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
{
	boolean_t wakeup;

	mutex_enter(&z->zone_lock);
	ASSERT(z->zone_ref != 0);
	z->zone_ref--;
	if (subsys != ZONE_REF_NUM_SUBSYS) {
		ASSERT(z->zone_subsys_ref[subsys] != 0);
		z->zone_subsys_ref[subsys]--;
		list_remove(&z->zone_ref_list, ref);
	}
	if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
		/* no more refs, free the structure */
		mutex_exit(&z->zone_lock);
		zone_free(z);
		return;
	}
	/* signal zone_destroy so the zone can finish halting */
	wakeup = (ZONE_IS_UNREF(z) && zone_status_get(z) >= ZONE_IS_DEAD);
	mutex_exit(&z->zone_lock);

	if (wakeup) {
		/*
		 * Grabbing zonehash_lock here effectively synchronizes with
		 * zone_destroy() to avoid missed signals.
		 */
		mutex_enter(&zonehash_lock);
		cv_broadcast(&zone_destroy_cv);
		mutex_exit(&zonehash_lock);
	}
}

/*
 * Decrement the specified zone's reference count.  The specified zone will
 * cease to exist after this function returns if the reference count drops to
 * zero.  This function should be paired with zone_hold().
 */
void
zone_rele(zone_t *z)
{
	zone_rele_common(z, NULL, ZONE_REF_NUM_SUBSYS);
}

/*
 * Initialize a zone reference structure.  This function must be invoked for
 * a reference structure before the structure is passed to zone_hold_ref().
 */
void
zone_init_ref(zone_ref_t *ref)
{
	ref->zref_zone = NULL;
	list_link_init(&ref->zref_linkage);
}

/*
 * Acquire a reference to zone z.  The caller must specify the
 * zone_ref_subsys_t constant associated with its subsystem.  The specified
 * zone_ref_t structure will represent a reference to the specified zone.  Use
 * zone_rele_ref() to release the reference.
 *
 * The referenced zone_t structure will not be freed as long as the zone_t's
 * zone_status field is not ZONE_IS_DEAD and the zone has outstanding
 * references.
 *
 * NOTE: The zone_ref_t structure must be initialized before it is used.
 * See zone_init_ref() above.
 */
void
zone_hold_ref(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
{
	ASSERT(subsys >= 0 && subsys < ZONE_REF_NUM_SUBSYS);

	/*
	 * Prevent consumers from reusing a reference structure before
	 * releasing it.
	 */
	VERIFY(ref->zref_zone == NULL);

	ref->zref_zone = z;
	mutex_enter(&z->zone_lock);
	zone_hold_locked(z);
	z->zone_subsys_ref[subsys]++;
	ASSERT(z->zone_subsys_ref[subsys] != 0);
	list_insert_head(&z->zone_ref_list, ref);
	mutex_exit(&z->zone_lock);
}

/*
 * Release the zone reference represented by the specified zone_ref_t.
 * The reference is invalid after it's released; however, the zone_ref_t
 * structure can be reused without having to invoke zone_init_ref().
 * subsys should be the same value that was passed to zone_hold_ref()
 * when the reference was acquired.
 */
void
zone_rele_ref(zone_ref_t *ref, zone_ref_subsys_t subsys)
{
	zone_rele_common(ref->zref_zone, ref, subsys);

	/*
	 * Set the zone_ref_t's zref_zone field to NULL to generate panics
	 * when consumers dereference the reference.  This helps us catch
	 * consumers who use released references.  Furthermore, this lets
	 * consumers reuse the zone_ref_t structure without having to
	 * invoke zone_init_ref().
	 */
	ref->zref_zone = NULL;
}

void
zone_cred_hold(zone_t *z)
{
	mutex_enter(&z->zone_lock);
	z->zone_cred_ref++;
	ASSERT(z->zone_cred_ref != 0);
	mutex_exit(&z->zone_lock);
}

void
zone_cred_rele(zone_t *z)
{
	boolean_t wakeup;

	mutex_enter(&z->zone_lock);
	ASSERT(z->zone_cred_ref != 0);
	z->zone_cred_ref--;
	if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
		/* no more refs, free the structure */
		mutex_exit(&z->zone_lock);
		zone_free(z);
		return;
	}
	/*
	 * If zone_destroy is waiting for the cred references to drain
	 * out, and they have, signal it.
	 */
	wakeup = (zone_wait_for_cred && ZONE_IS_UNREF(z) &&
	    zone_status_get(z) >= ZONE_IS_DEAD);
	mutex_exit(&z->zone_lock);

	if (wakeup) {
		/*
		 * Grabbing zonehash_lock here effectively synchronizes with
		 * zone_destroy() to avoid missed signals.
		 */
		mutex_enter(&zonehash_lock);
		cv_broadcast(&zone_destroy_cv);
		mutex_exit(&zonehash_lock);
	}
}

void
zone_task_hold(zone_t *z)
{
	mutex_enter(&z->zone_lock);
	z->zone_ntasks++;
	ASSERT(z->zone_ntasks != 0);
	mutex_exit(&z->zone_lock);
}

void
zone_task_rele(zone_t *zone)
{
	uint_t refcnt;

	mutex_enter(&zone->zone_lock);
	ASSERT(zone->zone_ntasks != 0);
	refcnt = --zone->zone_ntasks;
	if (refcnt > 1)	{	/* Common case */
		mutex_exit(&zone->zone_lock);
		return;
	}
	zone_hold_locked(zone);	/* so we can use the zone_t later */
	mutex_exit(&zone->zone_lock);
	if (refcnt == 1) {
		/*
		 * See if the zone is shutting down.
		 */
		mutex_enter(&zone_status_lock);
		if (zone_status_get(zone) != ZONE_IS_SHUTTING_DOWN) {
			goto out;
		}

		/*
		 * Make sure the ntasks didn't change since we
		 * dropped zone_lock.
		 */
		mutex_enter(&zone->zone_lock);
		if (refcnt != zone->zone_ntasks) {
			mutex_exit(&zone->zone_lock);
			goto out;
		}
		mutex_exit(&zone->zone_lock);

		/*
		 * No more user processes in the zone.  The zone is empty.
		 */
		zone_status_set(zone, ZONE_IS_EMPTY);
		goto out;
	}

	ASSERT(refcnt == 0);
	/*
	 * zsched has exited; the zone is dead.
	 */
	zone->zone_zsched = NULL;		/* paranoia */
	mutex_enter(&zone_status_lock);
	zone_status_set(zone, ZONE_IS_DEAD);
out:
	mutex_exit(&zone_status_lock);
	zone_rele(zone);
}

zoneid_t
getzoneid(void)
{
	return (curproc->p_zone->zone_id);
}

/*
 * Internal versions of zone_find_by_*().  These don't zone_hold() or
 * check the validity of a zone's state.
 */
static zone_t *
zone_find_all_by_id(zoneid_t zoneid)
{
	mod_hash_val_t hv;
	zone_t *zone = NULL;

	ASSERT(MUTEX_HELD(&zonehash_lock));

	if (mod_hash_find(zonehashbyid,
	    (mod_hash_key_t)(uintptr_t)zoneid, &hv) == 0)
		zone = (zone_t *)hv;
	return (zone);
}

static zone_t *
zone_find_all_by_label(const ts_label_t *label)
{
	mod_hash_val_t hv;
	zone_t *zone = NULL;

	ASSERT(MUTEX_HELD(&zonehash_lock));

	/*
	 * zonehashbylabel is not maintained for unlabeled systems
	 */
	if (!is_system_labeled())
		return (NULL);
	if (mod_hash_find(zonehashbylabel, (mod_hash_key_t)label, &hv) == 0)
		zone = (zone_t *)hv;
	return (zone);
}

static zone_t *
zone_find_all_by_name(char *name)
{
	mod_hash_val_t hv;
	zone_t *zone = NULL;

	ASSERT(MUTEX_HELD(&zonehash_lock));

	if (mod_hash_find(zonehashbyname, (mod_hash_key_t)name, &hv) == 0)
		zone = (zone_t *)hv;
	return (zone);
}

/*
 * Public interface for looking up a zone by zoneid.  Only returns the zone if
 * it is fully initialized, and has not yet begun the zone_destroy() sequence.
 * Caller must call zone_rele() once it is done with the zone.
 *
 * The zone may begin the zone_destroy() sequence immediately after this
 * function returns, but may be safely used until zone_rele() is called.
 */
zone_t *
zone_find_by_id(zoneid_t zoneid)
{
	zone_t *zone;
	zone_status_t status;

	mutex_enter(&zonehash_lock);
	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
		mutex_exit(&zonehash_lock);
		return (NULL);
	}
	status = zone_status_get(zone);
	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
		/*
		 * For all practical purposes the zone doesn't exist.
		 */
		mutex_exit(&zonehash_lock);
		return (NULL);
	}
	zone_hold(zone);
	mutex_exit(&zonehash_lock);
	return (zone);
}

/*
 * Similar to zone_find_by_id, but using zone label as the key.
 */
zone_t *
zone_find_by_label(const ts_label_t *label)
{
	zone_t *zone;
	zone_status_t status;

	mutex_enter(&zonehash_lock);
	if ((zone = zone_find_all_by_label(label)) == NULL) {
		mutex_exit(&zonehash_lock);
		return (NULL);
	}

	status = zone_status_get(zone);
	if (status > ZONE_IS_DOWN) {
		/*
		 * For all practical purposes the zone doesn't exist.
		 */
		mutex_exit(&zonehash_lock);
		return (NULL);
	}
	zone_hold(zone);
	mutex_exit(&zonehash_lock);
	return (zone);
}

/*
 * Similar to zone_find_by_id, but using zone name as the key.
 */
zone_t *
zone_find_by_name(char *name)
{
	zone_t *zone;
	zone_status_t status;

	mutex_enter(&zonehash_lock);
	if ((zone = zone_find_all_by_name(name)) == NULL) {
		mutex_exit(&zonehash_lock);
		return (NULL);
	}
	status = zone_status_get(zone);
	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
		/*
		 * For all practical purposes the zone doesn't exist.
		 */
		mutex_exit(&zonehash_lock);
		return (NULL);
	}
	zone_hold(zone);
	mutex_exit(&zonehash_lock);
	return (zone);
}

/*
 * Similar to zone_find_by_id(), using the path as a key.  For instance,
 * if there is a zone "foo" rooted at /foo/root, and the path argument
 * is "/foo/root/proc", it will return the held zone_t corresponding to
 * zone "foo".
 *
 * zone_find_by_path() always returns a non-NULL value, since at the
 * very least every path will be contained in the global zone.
 *
 * As with the other zone_find_by_*() functions, the caller is
 * responsible for zone_rele()ing the return value of this function.
 */
zone_t *
zone_find_by_path(const char *path)
{
	zone_t *zone;
	zone_t *zret = NULL;
	zone_status_t status;

	if (path == NULL) {
		/*
		 * Call from rootconf().
		 */
		zone_hold(global_zone);
		return (global_zone);
	}
	ASSERT(*path == '/');
	mutex_enter(&zonehash_lock);
	for (zone = list_head(&zone_active); zone != NULL;
	    zone = list_next(&zone_active, zone)) {
		if (ZONE_PATH_VISIBLE(path, zone))
			zret = zone;
	}
	ASSERT(zret != NULL);
	status = zone_status_get(zret);
	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
		/*
		 * Zone practically doesn't exist.
		 */
		zret = global_zone;
	}
	zone_hold(zret);
	mutex_exit(&zonehash_lock);
	return (zret);
}

/*
 * Public interface for updating per-zone load averages.  Called once per
 * second.
 *
 * Based on loadavg_update(), genloadavg() and calcloadavg() from clock.c.
 */
void
zone_loadavg_update()
{
	zone_t *zp;
	zone_status_t status;
	struct loadavg_s *lavg;
	hrtime_t zone_total;
	int i;
	hrtime_t hr_avg;
	int nrun;
	static int64_t f[3] = { 135, 27, 9 };
	int64_t q, r;

	mutex_enter(&zonehash_lock);
	for (zp = list_head(&zone_active); zp != NULL;
	    zp = list_next(&zone_active, zp)) {
		mutex_enter(&zp->zone_lock);

		/* Skip zones that are on the way down or not yet up */
		status = zone_status_get(zp);
		if (status < ZONE_IS_READY || status >= ZONE_IS_DOWN) {
			/* For all practical purposes the zone doesn't exist. */
			mutex_exit(&zp->zone_lock);
			continue;
		}

		/*
		 * Update the 10 second moving average data in zone_loadavg.
		 */
		lavg = &zp->zone_loadavg;

		zone_total = zp->zone_utime + zp->zone_stime + zp->zone_wtime;
		scalehrtime(&zone_total);

		/* The zone_total should always be increasing. */
		lavg->lg_loads[lavg->lg_cur] = (zone_total > lavg->lg_total) ?
		    zone_total - lavg->lg_total : 0;
		lavg->lg_cur = (lavg->lg_cur + 1) % S_LOADAVG_SZ;
		/* lg_total holds the prev. 1 sec. total */
		lavg->lg_total = zone_total;

		/*
		 * To simplify the calculation, we don't calculate the load avg.
		 * until the zone has been up for at least 10 seconds and our
		 * moving average is thus full.
		 */
		if ((lavg->lg_len + 1) < S_LOADAVG_SZ) {
			lavg->lg_len++;
			mutex_exit(&zp->zone_lock);
			continue;
		}

		/* Now calculate the 1min, 5min, 15 min load avg. */
		hr_avg = 0;
		for (i = 0; i < S_LOADAVG_SZ; i++)
			hr_avg += lavg->lg_loads[i];
		hr_avg = hr_avg / S_LOADAVG_SZ;
		nrun = hr_avg / (NANOSEC / LGRP_LOADAVG_IN_THREAD_MAX);

		/* Compute load avg. See comment in calcloadavg() */
		for (i = 0; i < 3; i++) {
			q = (zp->zone_hp_avenrun[i] >> 16) << 7;
			r = (zp->zone_hp_avenrun[i] & 0xffff) << 7;
			zp->zone_hp_avenrun[i] +=
			    ((nrun - q) * f[i] - ((r * f[i]) >> 16)) >> 4;

			/* avenrun[] can only hold 31 bits of load avg. */
			if (zp->zone_hp_avenrun[i] <
			    ((uint64_t)1<<(31+16-FSHIFT)))
				zp->zone_avenrun[i] = (int32_t)
				    (zp->zone_hp_avenrun[i] >> (16 - FSHIFT));
			else
				zp->zone_avenrun[i] = 0x7fffffff;
		}

		mutex_exit(&zp->zone_lock);
	}
	mutex_exit(&zonehash_lock);
}

/*
 * Get the number of cpus visible to this zone.  The system-wide global
 * 'ncpus' is returned if pools are disabled, the caller is in the
 * global zone, or a NULL zone argument is passed in.
 */
int
zone_ncpus_get(zone_t *zone)
{
	int myncpus = zone == NULL ? 0 : zone->zone_ncpus;

	return (myncpus != 0 ? myncpus : ncpus);
}

/*
 * Get the number of online cpus visible to this zone.  The system-wide
 * global 'ncpus_online' is returned if pools are disabled, the caller
 * is in the global zone, or a NULL zone argument is passed in.
 */
int
zone_ncpus_online_get(zone_t *zone)
{
	int myncpus_online = zone == NULL ? 0 : zone->zone_ncpus_online;

	return (myncpus_online != 0 ? myncpus_online : ncpus_online);
}

/*
 * Return the pool to which the zone is currently bound.
 */
pool_t *
zone_pool_get(zone_t *zone)
{
	ASSERT(pool_lock_held());

	return (zone->zone_pool);
}

/*
 * Set the zone's pool pointer and update the zone's visibility to match
 * the resources in the new pool.
 */
void
zone_pool_set(zone_t *zone, pool_t *pool)
{
	ASSERT(pool_lock_held());
	ASSERT(MUTEX_HELD(&cpu_lock));

	zone->zone_pool = pool;
	zone_pset_set(zone, pool->pool_pset->pset_id);
}

/*
 * Return the cached value of the id of the processor set to which the
 * zone is currently bound.  The value will be ZONE_PS_INVAL if the pools
 * facility is disabled.
 */
psetid_t
zone_pset_get(zone_t *zone)
{
	ASSERT(MUTEX_HELD(&cpu_lock));

	return (zone->zone_psetid);
}

/*
 * Set the cached value of the id of the processor set to which the zone
 * is currently bound.  Also update the zone's visibility to match the
 * resources in the new processor set.
 */
void
zone_pset_set(zone_t *zone, psetid_t newpsetid)
{
	psetid_t oldpsetid;

	ASSERT(MUTEX_HELD(&cpu_lock));
	oldpsetid = zone_pset_get(zone);

	if (oldpsetid == newpsetid)
		return;
	/*
	 * Global zone sees all.
	 */
	if (zone != global_zone) {
		zone->zone_psetid = newpsetid;
		if (newpsetid != ZONE_PS_INVAL)
			pool_pset_visibility_add(newpsetid, zone);
		if (oldpsetid != ZONE_PS_INVAL)
			pool_pset_visibility_remove(oldpsetid, zone);
	}
	/*
	 * Disabling pools, so we should start using the global values
	 * for ncpus and ncpus_online.
	 */
	if (newpsetid == ZONE_PS_INVAL) {
		zone->zone_ncpus = 0;
		zone->zone_ncpus_online = 0;
	}
}

/*
 * Walk the list of active zones and issue the provided callback for
 * each of them.
 *
 * Caller must not be holding any locks that may be acquired under
 * zonehash_lock.  See comment at the beginning of the file for a list of
 * common locks and their interactions with zones.
 */
int
zone_walk(int (*cb)(zone_t *, void *), void *data)
{
	zone_t *zone;
	int ret = 0;
	zone_status_t status;

	mutex_enter(&zonehash_lock);
	for (zone = list_head(&zone_active); zone != NULL;
	    zone = list_next(&zone_active, zone)) {
		/*
		 * Skip zones that shouldn't be externally visible.
		 */
		status = zone_status_get(zone);
		if (status < ZONE_IS_READY || status > ZONE_IS_DOWN)
			continue;
		/*
		 * Bail immediately if any callback invocation returns a
		 * non-zero value.
		 */
		ret = (*cb)(zone, data);
		if (ret != 0)
			break;
	}
	mutex_exit(&zonehash_lock);
	return (ret);
}

static int
zone_set_root(zone_t *zone, const char *upath)
{
	vnode_t *vp;
	int trycount;
	int error = 0;
	char *path;
	struct pathname upn, pn;
	size_t pathlen;

	if ((error = pn_get((char *)upath, UIO_USERSPACE, &upn)) != 0)
		return (error);

	pn_alloc(&pn);

	/* prevent infinite loop */
	trycount = 10;
	for (;;) {
		if (--trycount <= 0) {
			error = ESTALE;
			goto out;
		}

		if ((error = lookuppn(&upn, &pn, FOLLOW, NULLVPP, &vp)) == 0) {
			/*
			 * VOP_ACCESS() may cover 'vp' with a new
			 * filesystem, if 'vp' is an autoFS vnode.
			 * Get the new 'vp' if so.
			 */
			if ((error =
			    VOP_ACCESS(vp, VEXEC, 0,