xref: /illumos-gate/usr/src/uts/common/os/zone.c (revision 93cf283a)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright 2015, Joyent Inc. All rights reserved.
25  */
26 
27 /*
28  * Zones
29  *
30  *   A zone is a named collection of processes, namespace constraints,
31  *   and other system resources which comprise a secure and manageable
32  *   application containment facility.
33  *
34  *   Zones (represented by the reference counted zone_t) are tracked in
35  *   the kernel in the zonehash.  Elsewhere in the kernel, Zone IDs
36  *   (zoneid_t) are used to track zone association.  Zone IDs are
37  *   dynamically generated when the zone is created; if a persistent
38  *   identifier is needed (core files, accounting logs, audit trail,
39  *   etc.), the zone name should be used.
40  *
41  *
42  *   Global Zone:
43  *
44  *   The global zone (zoneid 0) is automatically associated with all
45  *   system resources that have not been bound to a user-created zone.
46  *   This means that even systems where zones are not in active use
47  *   have a global zone, and all processes, mounts, etc. are
48  *   associated with that zone.  The global zone is generally
49  *   unconstrained in terms of privileges and access, though the usual
50  *   credential and privilege based restrictions apply.
51  *
52  *
53  *   Zone States:
54  *
55  *   The states in which a zone may be in and the transitions are as
56  *   follows:
57  *
58  *   ZONE_IS_UNINITIALIZED: primordial state for a zone. The partially
59  *   initialized zone is added to the list of active zones on the system but
60  *   isn't accessible.
61  *
62  *   ZONE_IS_INITIALIZED: Initialization complete except the ZSD callbacks are
63  *   not yet completed. Not possible to enter the zone, but attributes can
64  *   be retrieved.
65  *
66  *   ZONE_IS_READY: zsched (the kernel dummy process for a zone) is
67  *   ready.  The zone is made visible after the ZSD constructor callbacks are
68  *   executed.  A zone remains in this state until it transitions into
69  *   the ZONE_IS_BOOTING state as a result of a call to zone_boot().
70  *
71  *   ZONE_IS_BOOTING: in this shortlived-state, zsched attempts to start
72  *   init.  Should that fail, the zone proceeds to the ZONE_IS_SHUTTING_DOWN
73  *   state.
74  *
75  *   ZONE_IS_RUNNING: The zone is open for business: zsched has
76  *   successfully started init.   A zone remains in this state until
77  *   zone_shutdown() is called.
78  *
79  *   ZONE_IS_SHUTTING_DOWN: zone_shutdown() has been called, the system is
80  *   killing all processes running in the zone. The zone remains
81  *   in this state until there are no more user processes running in the zone.
82  *   zone_create(), zone_enter(), and zone_destroy() on this zone will fail.
83  *   Since zone_shutdown() is restartable, it may be called successfully
84  *   multiple times for the same zone_t.  Setting of the zone's state to
85  *   ZONE_IS_SHUTTING_DOWN is synchronized with mounts, so VOP_MOUNT() may check
86  *   the zone's status without worrying about it being a moving target.
87  *
88  *   ZONE_IS_EMPTY: zone_shutdown() has been called, and there
89  *   are no more user processes in the zone.  The zone remains in this
90  *   state until there are no more kernel threads associated with the
91  *   zone.  zone_create(), zone_enter(), and zone_destroy() on this zone will
92  *   fail.
93  *
94  *   ZONE_IS_DOWN: All kernel threads doing work on behalf of the zone
95  *   have exited.  zone_shutdown() returns.  Henceforth it is not possible to
96  *   join the zone or create kernel threads therein.
97  *
98  *   ZONE_IS_DYING: zone_destroy() has been called on the zone; zone
99  *   remains in this state until zsched exits.  Calls to zone_find_by_*()
100  *   return NULL from now on.
101  *
102  *   ZONE_IS_DEAD: zsched has exited (zone_ntasks == 0).  There are no
103  *   processes or threads doing work on behalf of the zone.  The zone is
104  *   removed from the list of active zones.  zone_destroy() returns, and
105  *   the zone can be recreated.
106  *
107  *   ZONE_IS_FREE (internal state): zone_ref goes to 0, ZSD destructor
108  *   callbacks are executed, and all memory associated with the zone is
109  *   freed.
110  *
111  *   Threads can wait for the zone to enter a requested state by using
112  *   zone_status_wait() or zone_status_timedwait() with the desired
113  *   state passed in as an argument.  Zone state transitions are
114  *   uni-directional; it is not possible to move back to an earlier state.
115  *
116  *
117  *   Zone-Specific Data:
118  *
119  *   Subsystems needing to maintain zone-specific data can store that
120  *   data using the ZSD mechanism.  This provides a zone-specific data
121  *   store, similar to thread-specific data (see pthread_getspecific(3C)
122  *   or the TSD code in uts/common/disp/thread.c.  Also, ZSD can be used
123  *   to register callbacks to be invoked when a zone is created, shut
124  *   down, or destroyed.  This can be used to initialize zone-specific
125  *   data for new zones and to clean up when zones go away.
126  *
127  *
128  *   Data Structures:
129  *
130  *   The per-zone structure (zone_t) is reference counted, and freed
131  *   when all references are released.  zone_hold and zone_rele can be
132  *   used to adjust the reference count.  In addition, reference counts
133  *   associated with the cred_t structure are tracked separately using
134  *   zone_cred_hold and zone_cred_rele.
135  *
136  *   Pointers to active zone_t's are stored in two hash tables; one
137  *   for searching by id, the other for searching by name.  Lookups
138  *   can be performed on either basis, using zone_find_by_id and
139  *   zone_find_by_name.  Both return zone_t pointers with the zone
140  *   held, so zone_rele should be called when the pointer is no longer
141  *   needed.  Zones can also be searched by path; zone_find_by_path
142  *   returns the zone with which a path name is associated (global
143  *   zone if the path is not within some other zone's file system
144  *   hierarchy).  This currently requires iterating through each zone,
145  *   so it is slower than an id or name search via a hash table.
146  *
147  *
148  *   Locking:
149  *
150  *   zonehash_lock: This is a top-level global lock used to protect the
151  *       zone hash tables and lists.  Zones cannot be created or destroyed
152  *       while this lock is held.
153  *   zone_status_lock: This is a global lock protecting zone state.
154  *       Zones cannot change state while this lock is held.  It also
155  *       protects the list of kernel threads associated with a zone.
156  *   zone_lock: This is a per-zone lock used to protect several fields of
157  *       the zone_t (see <sys/zone.h> for details).  In addition, holding
158  *       this lock means that the zone cannot go away.
159  *   zone_nlwps_lock: This is a per-zone lock used to protect the fields
160  *	 related to the zone.max-lwps rctl.
161  *   zone_mem_lock: This is a per-zone lock used to protect the fields
162  *	 related to the zone.max-locked-memory and zone.max-swap rctls.
163  *   zone_rctl_lock: This is a per-zone lock used to protect other rctls,
164  *       currently just max_lofi
165  *   zsd_key_lock: This is a global lock protecting the key state for ZSD.
166  *   zone_deathrow_lock: This is a global lock protecting the "deathrow"
167  *       list (a list of zones in the ZONE_IS_DEAD state).
168  *
169  *   Ordering requirements:
170  *       pool_lock --> cpu_lock --> zonehash_lock --> zone_status_lock -->
171  *       	zone_lock --> zsd_key_lock --> pidlock --> p_lock
172  *
173  *   When taking zone_mem_lock or zone_nlwps_lock, the lock ordering is:
174  *	zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_mem_lock
175  *	zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_nlwps_lock
176  *
177  *   Blocking memory allocations are permitted while holding any of the
178  *   zone locks.
179  *
180  *
181  *   System Call Interface:
182  *
183  *   The zone subsystem can be managed and queried from user level with
184  *   the following system calls (all subcodes of the primary "zone"
185  *   system call):
186  *   - zone_create: creates a zone with selected attributes (name,
187  *     root path, privileges, resource controls, ZFS datasets)
188  *   - zone_enter: allows the current process to enter a zone
189  *   - zone_getattr: reports attributes of a zone
190  *   - zone_setattr: set attributes of a zone
191  *   - zone_boot: set 'init' running for the zone
192  *   - zone_list: lists all zones active in the system
193  *   - zone_lookup: looks up zone id based on name
194  *   - zone_shutdown: initiates shutdown process (see states above)
195  *   - zone_destroy: completes shutdown process (see states above)
196  *
197  */
198 
199 #include <sys/priv_impl.h>
200 #include <sys/cred.h>
201 #include <c2/audit.h>
202 #include <sys/debug.h>
203 #include <sys/file.h>
204 #include <sys/kmem.h>
205 #include <sys/kstat.h>
206 #include <sys/mutex.h>
207 #include <sys/note.h>
208 #include <sys/pathname.h>
209 #include <sys/proc.h>
210 #include <sys/project.h>
211 #include <sys/sysevent.h>
212 #include <sys/task.h>
213 #include <sys/systm.h>
214 #include <sys/types.h>
215 #include <sys/utsname.h>
216 #include <sys/vnode.h>
217 #include <sys/vfs.h>
218 #include <sys/systeminfo.h>
219 #include <sys/policy.h>
220 #include <sys/cred_impl.h>
221 #include <sys/contract_impl.h>
222 #include <sys/contract/process_impl.h>
223 #include <sys/class.h>
224 #include <sys/pool.h>
225 #include <sys/pool_pset.h>
226 #include <sys/pset.h>
227 #include <sys/strlog.h>
228 #include <sys/sysmacros.h>
229 #include <sys/callb.h>
230 #include <sys/vmparam.h>
231 #include <sys/corectl.h>
232 #include <sys/ipc_impl.h>
233 #include <sys/klpd.h>
234 
235 #include <sys/door.h>
236 #include <sys/cpuvar.h>
237 #include <sys/sdt.h>
238 
239 #include <sys/uadmin.h>
240 #include <sys/session.h>
241 #include <sys/cmn_err.h>
242 #include <sys/modhash.h>
243 #include <sys/sunddi.h>
244 #include <sys/nvpair.h>
245 #include <sys/rctl.h>
246 #include <sys/fss.h>
247 #include <sys/brand.h>
248 #include <sys/zone.h>
249 #include <net/if.h>
250 #include <sys/cpucaps.h>
251 #include <vm/seg.h>
252 #include <sys/mac.h>
253 
254 /*
255  * This constant specifies the number of seconds that threads waiting for
256  * subsystems to release a zone's general-purpose references will wait before
257  * they log the zone's reference counts.  The constant's value shouldn't
258  * be so small that reference counts are unnecessarily reported for zones
259  * whose references are slowly released.  On the other hand, it shouldn't be so
260  * large that users reboot their systems out of frustration over hung zones
261  * before the system logs the zones' reference counts.
262  */
263 #define	ZONE_DESTROY_TIMEOUT_SECS	60
264 
265 /* List of data link IDs which are accessible from the zone */
266 typedef struct zone_dl {
267 	datalink_id_t	zdl_id;
268 	nvlist_t	*zdl_net;
269 	list_node_t	zdl_linkage;
270 } zone_dl_t;
271 
272 /*
273  * cv used to signal that all references to the zone have been released.  This
274  * needs to be global since there may be multiple waiters, and the first to
275  * wake up will free the zone_t, hence we cannot use zone->zone_cv.
276  */
277 static kcondvar_t zone_destroy_cv;
278 /*
279  * Lock used to serialize access to zone_cv.  This could have been per-zone,
280  * but then we'd need another lock for zone_destroy_cv, and why bother?
281  */
282 static kmutex_t zone_status_lock;
283 
284 /*
285  * ZSD-related global variables.
286  */
287 static kmutex_t zsd_key_lock;	/* protects the following two */
288 /*
289  * The next caller of zone_key_create() will be assigned a key of ++zsd_keyval.
290  */
291 static zone_key_t zsd_keyval = 0;
292 /*
293  * Global list of registered keys.  We use this when a new zone is created.
294  */
295 static list_t zsd_registered_keys;
296 
297 int zone_hash_size = 256;
298 static mod_hash_t *zonehashbyname, *zonehashbyid, *zonehashbylabel;
299 static kmutex_t zonehash_lock;
300 static uint_t zonecount;
301 static id_space_t *zoneid_space;
302 
303 /*
304  * The global zone (aka zone0) is the all-seeing, all-knowing zone in which the
305  * kernel proper runs, and which manages all other zones.
306  *
307  * Although not declared as static, the variable "zone0" should not be used
308  * except for by code that needs to reference the global zone early on in boot,
309  * before it is fully initialized.  All other consumers should use
310  * 'global_zone'.
311  */
312 zone_t zone0;
313 zone_t *global_zone = NULL;	/* Set when the global zone is initialized */
314 
315 /*
316  * List of active zones, protected by zonehash_lock.
317  */
318 static list_t zone_active;
319 
320 /*
321  * List of destroyed zones that still have outstanding cred references.
322  * Used for debugging.  Uses a separate lock to avoid lock ordering
323  * problems in zone_free.
324  */
325 static list_t zone_deathrow;
326 static kmutex_t zone_deathrow_lock;
327 
328 /* number of zones is limited by virtual interface limit in IP */
329 uint_t maxzones = 8192;
330 
331 /* Event channel to sent zone state change notifications */
332 evchan_t *zone_event_chan;
333 
334 /*
335  * This table holds the mapping from kernel zone states to
336  * states visible in the state notification API.
337  * The idea is that we only expose "obvious" states and
338  * do not expose states which are just implementation details.
339  */
340 const char  *zone_status_table[] = {
341 	ZONE_EVENT_UNINITIALIZED,	/* uninitialized */
342 	ZONE_EVENT_INITIALIZED,		/* initialized */
343 	ZONE_EVENT_READY,		/* ready */
344 	ZONE_EVENT_READY,		/* booting */
345 	ZONE_EVENT_RUNNING,		/* running */
346 	ZONE_EVENT_SHUTTING_DOWN,	/* shutting_down */
347 	ZONE_EVENT_SHUTTING_DOWN,	/* empty */
348 	ZONE_EVENT_SHUTTING_DOWN,	/* down */
349 	ZONE_EVENT_SHUTTING_DOWN,	/* dying */
350 	ZONE_EVENT_UNINITIALIZED,	/* dead */
351 };
352 
353 /*
354  * This array contains the names of the subsystems listed in zone_ref_subsys_t
355  * (see sys/zone.h).
356  */
357 static char *zone_ref_subsys_names[] = {
358 	"NFS",		/* ZONE_REF_NFS */
359 	"NFSv4",	/* ZONE_REF_NFSV4 */
360 	"SMBFS",	/* ZONE_REF_SMBFS */
361 	"MNTFS",	/* ZONE_REF_MNTFS */
362 	"LOFI",		/* ZONE_REF_LOFI */
363 	"VFS",		/* ZONE_REF_VFS */
364 	"IPC"		/* ZONE_REF_IPC */
365 };
366 
367 /*
368  * This isn't static so lint doesn't complain.
369  */
370 rctl_hndl_t rc_zone_cpu_shares;
371 rctl_hndl_t rc_zone_locked_mem;
372 rctl_hndl_t rc_zone_max_swap;
373 rctl_hndl_t rc_zone_max_lofi;
374 rctl_hndl_t rc_zone_cpu_cap;
375 rctl_hndl_t rc_zone_nlwps;
376 rctl_hndl_t rc_zone_nprocs;
377 rctl_hndl_t rc_zone_shmmax;
378 rctl_hndl_t rc_zone_shmmni;
379 rctl_hndl_t rc_zone_semmni;
380 rctl_hndl_t rc_zone_msgmni;
381 
382 const char * const zone_default_initname = "/sbin/init";
383 static char * const zone_prefix = "/zone/";
384 static int zone_shutdown(zoneid_t zoneid);
385 static int zone_add_datalink(zoneid_t, datalink_id_t);
386 static int zone_remove_datalink(zoneid_t, datalink_id_t);
387 static int zone_list_datalink(zoneid_t, int *, datalink_id_t *);
388 static int zone_set_network(zoneid_t, zone_net_data_t *);
389 static int zone_get_network(zoneid_t, zone_net_data_t *);
390 
391 typedef boolean_t zsd_applyfn_t(kmutex_t *, boolean_t, zone_t *, zone_key_t);
392 
393 static void zsd_apply_all_zones(zsd_applyfn_t *, zone_key_t);
394 static void zsd_apply_all_keys(zsd_applyfn_t *, zone_t *);
395 static boolean_t zsd_apply_create(kmutex_t *, boolean_t, zone_t *, zone_key_t);
396 static boolean_t zsd_apply_shutdown(kmutex_t *, boolean_t, zone_t *,
397     zone_key_t);
398 static boolean_t zsd_apply_destroy(kmutex_t *, boolean_t, zone_t *, zone_key_t);
399 static boolean_t zsd_wait_for_creator(zone_t *, struct zsd_entry *,
400     kmutex_t *);
401 static boolean_t zsd_wait_for_inprogress(zone_t *, struct zsd_entry *,
402     kmutex_t *);
403 
404 /*
405  * Bump this number when you alter the zone syscall interfaces; this is
406  * because we need to have support for previous API versions in libc
407  * to support patching; libc calls into the kernel to determine this number.
408  *
409  * Version 1 of the API is the version originally shipped with Solaris 10
410  * Version 2 alters the zone_create system call in order to support more
411  *     arguments by moving the args into a structure; and to do better
412  *     error reporting when zone_create() fails.
413  * Version 3 alters the zone_create system call in order to support the
414  *     import of ZFS datasets to zones.
415  * Version 4 alters the zone_create system call in order to support
416  *     Trusted Extensions.
417  * Version 5 alters the zone_boot system call, and converts its old
418  *     bootargs parameter to be set by the zone_setattr API instead.
419  * Version 6 adds the flag argument to zone_create.
420  */
421 static const int ZONE_SYSCALL_API_VERSION = 6;
422 
423 /*
424  * Certain filesystems (such as NFS and autofs) need to know which zone
425  * the mount is being placed in.  Because of this, we need to be able to
426  * ensure that a zone isn't in the process of being created/destroyed such
427  * that nfs_mount() thinks it is in the global/NGZ zone, while by the time
428  * it gets added the list of mounted zones, it ends up on the wrong zone's
429  * mount list. Since a zone can't reside on an NFS file system, we don't
430  * have to worry about the zonepath itself.
431  *
432  * The following functions: block_mounts()/resume_mounts() and
433  * mount_in_progress()/mount_completed() are used by zones and the VFS
434  * layer (respectively) to synchronize zone state transitions and new
435  * mounts within a zone. This syncronization is on a per-zone basis, so
436  * activity for one zone will not interfere with activity for another zone.
437  *
438  * The semantics are like a reader-reader lock such that there may
439  * either be multiple mounts (or zone state transitions, if that weren't
440  * serialized by zonehash_lock) in progress at the same time, but not
441  * both.
442  *
443  * We use cv's so the user can ctrl-C out of the operation if it's
444  * taking too long.
445  *
446  * The semantics are such that there is unfair bias towards the
447  * "current" operation.  This means that zone halt may starve if
448  * there is a rapid succession of new mounts coming in to the zone.
449  */
450 /*
451  * Prevent new mounts from progressing to the point of calling
452  * VFS_MOUNT().  If there are already mounts in this "region", wait for
453  * them to complete.
454  */
455 static int
456 block_mounts(zone_t *zp)
457 {
458 	int retval = 0;
459 
460 	/*
461 	 * Since it may block for a long time, block_mounts() shouldn't be
462 	 * called with zonehash_lock held.
463 	 */
464 	ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
465 	mutex_enter(&zp->zone_mount_lock);
466 	while (zp->zone_mounts_in_progress > 0) {
467 		if (cv_wait_sig(&zp->zone_mount_cv, &zp->zone_mount_lock) == 0)
468 			goto signaled;
469 	}
470 	/*
471 	 * A negative value of mounts_in_progress indicates that mounts
472 	 * have been blocked by (-mounts_in_progress) different callers
473 	 * (remotely possible if two threads enter zone_shutdown at the same
474 	 * time).
475 	 */
476 	zp->zone_mounts_in_progress--;
477 	retval = 1;
478 signaled:
479 	mutex_exit(&zp->zone_mount_lock);
480 	return (retval);
481 }
482 
483 /*
484  * The VFS layer may progress with new mounts as far as we're concerned.
485  * Allow them to progress if we were the last obstacle.
486  */
487 static void
488 resume_mounts(zone_t *zp)
489 {
490 	mutex_enter(&zp->zone_mount_lock);
491 	if (++zp->zone_mounts_in_progress == 0)
492 		cv_broadcast(&zp->zone_mount_cv);
493 	mutex_exit(&zp->zone_mount_lock);
494 }
495 
496 /*
497  * The VFS layer is busy with a mount; this zone should wait until all
498  * of its mounts are completed to progress.
499  */
500 void
501 mount_in_progress(zone_t *zp)
502 {
503 	mutex_enter(&zp->zone_mount_lock);
504 	while (zp->zone_mounts_in_progress < 0)
505 		cv_wait(&zp->zone_mount_cv, &zp->zone_mount_lock);
506 	zp->zone_mounts_in_progress++;
507 	mutex_exit(&zp->zone_mount_lock);
508 }
509 
510 /*
511  * VFS is done with one mount; wake up any waiting block_mounts()
512  * callers if this is the last mount.
513  */
514 void
515 mount_completed(zone_t *zp)
516 {
517 	mutex_enter(&zp->zone_mount_lock);
518 	if (--zp->zone_mounts_in_progress == 0)
519 		cv_broadcast(&zp->zone_mount_cv);
520 	mutex_exit(&zp->zone_mount_lock);
521 }
522 
523 /*
524  * ZSD routines.
525  *
526  * Zone Specific Data (ZSD) is modeled after Thread Specific Data as
527  * defined by the pthread_key_create() and related interfaces.
528  *
529  * Kernel subsystems may register one or more data items and/or
530  * callbacks to be executed when a zone is created, shutdown, or
531  * destroyed.
532  *
533  * Unlike the thread counterpart, destructor callbacks will be executed
534  * even if the data pointer is NULL and/or there are no constructor
535  * callbacks, so it is the responsibility of such callbacks to check for
536  * NULL data values if necessary.
537  *
538  * The locking strategy and overall picture is as follows:
539  *
540  * When someone calls zone_key_create(), a template ZSD entry is added to the
541  * global list "zsd_registered_keys", protected by zsd_key_lock.  While
542  * holding that lock all the existing zones are marked as
543  * ZSD_CREATE_NEEDED and a copy of the ZSD entry added to the per-zone
544  * zone_zsd list (protected by zone_lock). The global list is updated first
545  * (under zone_key_lock) to make sure that newly created zones use the
546  * most recent list of keys. Then under zonehash_lock we walk the zones
547  * and mark them.  Similar locking is used in zone_key_delete().
548  *
549  * The actual create, shutdown, and destroy callbacks are done without
550  * holding any lock. And zsd_flags are used to ensure that the operations
551  * completed so that when zone_key_create (and zone_create) is done, as well as
552  * zone_key_delete (and zone_destroy) is done, all the necessary callbacks
553  * are completed.
554  *
555  * When new zones are created constructor callbacks for all registered ZSD
556  * entries will be called. That also uses the above two phases of marking
557  * what needs to be done, and then running the callbacks without holding
558  * any locks.
559  *
560  * The framework does not provide any locking around zone_getspecific() and
561  * zone_setspecific() apart from that needed for internal consistency, so
562  * callers interested in atomic "test-and-set" semantics will need to provide
563  * their own locking.
564  */
565 
566 /*
567  * Helper function to find the zsd_entry associated with the key in the
568  * given list.
569  */
570 static struct zsd_entry *
571 zsd_find(list_t *l, zone_key_t key)
572 {
573 	struct zsd_entry *zsd;
574 
575 	for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
576 		if (zsd->zsd_key == key) {
577 			return (zsd);
578 		}
579 	}
580 	return (NULL);
581 }
582 
583 /*
584  * Helper function to find the zsd_entry associated with the key in the
585  * given list. Move it to the front of the list.
586  */
587 static struct zsd_entry *
588 zsd_find_mru(list_t *l, zone_key_t key)
589 {
590 	struct zsd_entry *zsd;
591 
592 	for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
593 		if (zsd->zsd_key == key) {
594 			/*
595 			 * Move to head of list to keep list in MRU order.
596 			 */
597 			if (zsd != list_head(l)) {
598 				list_remove(l, zsd);
599 				list_insert_head(l, zsd);
600 			}
601 			return (zsd);
602 		}
603 	}
604 	return (NULL);
605 }
606 
607 void
608 zone_key_create(zone_key_t *keyp, void *(*create)(zoneid_t),
609     void (*shutdown)(zoneid_t, void *), void (*destroy)(zoneid_t, void *))
610 {
611 	struct zsd_entry *zsdp;
612 	struct zsd_entry *t;
613 	struct zone *zone;
614 	zone_key_t  key;
615 
616 	zsdp = kmem_zalloc(sizeof (*zsdp), KM_SLEEP);
617 	zsdp->zsd_data = NULL;
618 	zsdp->zsd_create = create;
619 	zsdp->zsd_shutdown = shutdown;
620 	zsdp->zsd_destroy = destroy;
621 
622 	/*
623 	 * Insert in global list of callbacks. Makes future zone creations
624 	 * see it.
625 	 */
626 	mutex_enter(&zsd_key_lock);
627 	key = zsdp->zsd_key = ++zsd_keyval;
628 	ASSERT(zsd_keyval != 0);
629 	list_insert_tail(&zsd_registered_keys, zsdp);
630 	mutex_exit(&zsd_key_lock);
631 
632 	/*
633 	 * Insert for all existing zones and mark them as needing
634 	 * a create callback.
635 	 */
636 	mutex_enter(&zonehash_lock);	/* stop the world */
637 	for (zone = list_head(&zone_active); zone != NULL;
638 	    zone = list_next(&zone_active, zone)) {
639 		zone_status_t status;
640 
641 		mutex_enter(&zone->zone_lock);
642 
643 		/* Skip zones that are on the way down or not yet up */
644 		status = zone_status_get(zone);
645 		if (status >= ZONE_IS_DOWN ||
646 		    status == ZONE_IS_UNINITIALIZED) {
647 			mutex_exit(&zone->zone_lock);
648 			continue;
649 		}
650 
651 		t = zsd_find_mru(&zone->zone_zsd, key);
652 		if (t != NULL) {
653 			/*
654 			 * A zsd_configure already inserted it after
655 			 * we dropped zsd_key_lock above.
656 			 */
657 			mutex_exit(&zone->zone_lock);
658 			continue;
659 		}
660 		t = kmem_zalloc(sizeof (*t), KM_SLEEP);
661 		t->zsd_key = key;
662 		t->zsd_create = create;
663 		t->zsd_shutdown = shutdown;
664 		t->zsd_destroy = destroy;
665 		if (create != NULL) {
666 			t->zsd_flags = ZSD_CREATE_NEEDED;
667 			DTRACE_PROBE2(zsd__create__needed,
668 			    zone_t *, zone, zone_key_t, key);
669 		}
670 		list_insert_tail(&zone->zone_zsd, t);
671 		mutex_exit(&zone->zone_lock);
672 	}
673 	mutex_exit(&zonehash_lock);
674 
675 	if (create != NULL) {
676 		/* Now call the create callback for this key */
677 		zsd_apply_all_zones(zsd_apply_create, key);
678 	}
679 	/*
680 	 * It is safe for consumers to use the key now, make it
681 	 * globally visible. Specifically zone_getspecific() will
682 	 * always successfully return the zone specific data associated
683 	 * with the key.
684 	 */
685 	*keyp = key;
686 
687 }
688 
689 /*
690  * Function called when a module is being unloaded, or otherwise wishes
691  * to unregister its ZSD key and callbacks.
692  *
693  * Remove from the global list and determine the functions that need to
694  * be called under a global lock. Then call the functions without
695  * holding any locks. Finally free up the zone_zsd entries. (The apply
696  * functions need to access the zone_zsd entries to find zsd_data etc.)
697  */
698 int
699 zone_key_delete(zone_key_t key)
700 {
701 	struct zsd_entry *zsdp = NULL;
702 	zone_t *zone;
703 
704 	mutex_enter(&zsd_key_lock);
705 	zsdp = zsd_find_mru(&zsd_registered_keys, key);
706 	if (zsdp == NULL) {
707 		mutex_exit(&zsd_key_lock);
708 		return (-1);
709 	}
710 	list_remove(&zsd_registered_keys, zsdp);
711 	mutex_exit(&zsd_key_lock);
712 
713 	mutex_enter(&zonehash_lock);
714 	for (zone = list_head(&zone_active); zone != NULL;
715 	    zone = list_next(&zone_active, zone)) {
716 		struct zsd_entry *del;
717 
718 		mutex_enter(&zone->zone_lock);
719 		del = zsd_find_mru(&zone->zone_zsd, key);
720 		if (del == NULL) {
721 			/*
722 			 * Somebody else got here first e.g the zone going
723 			 * away.
724 			 */
725 			mutex_exit(&zone->zone_lock);
726 			continue;
727 		}
728 		ASSERT(del->zsd_shutdown == zsdp->zsd_shutdown);
729 		ASSERT(del->zsd_destroy == zsdp->zsd_destroy);
730 		if (del->zsd_shutdown != NULL &&
731 		    (del->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
732 			del->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
733 			DTRACE_PROBE2(zsd__shutdown__needed,
734 			    zone_t *, zone, zone_key_t, key);
735 		}
736 		if (del->zsd_destroy != NULL &&
737 		    (del->zsd_flags & ZSD_DESTROY_ALL) == 0) {
738 			del->zsd_flags |= ZSD_DESTROY_NEEDED;
739 			DTRACE_PROBE2(zsd__destroy__needed,
740 			    zone_t *, zone, zone_key_t, key);
741 		}
742 		mutex_exit(&zone->zone_lock);
743 	}
744 	mutex_exit(&zonehash_lock);
745 	kmem_free(zsdp, sizeof (*zsdp));
746 
747 	/* Now call the shutdown and destroy callback for this key */
748 	zsd_apply_all_zones(zsd_apply_shutdown, key);
749 	zsd_apply_all_zones(zsd_apply_destroy, key);
750 
751 	/* Now we can free up the zsdp structures in each zone */
752 	mutex_enter(&zonehash_lock);
753 	for (zone = list_head(&zone_active); zone != NULL;
754 	    zone = list_next(&zone_active, zone)) {
755 		struct zsd_entry *del;
756 
757 		mutex_enter(&zone->zone_lock);
758 		del = zsd_find(&zone->zone_zsd, key);
759 		if (del != NULL) {
760 			list_remove(&zone->zone_zsd, del);
761 			ASSERT(!(del->zsd_flags & ZSD_ALL_INPROGRESS));
762 			kmem_free(del, sizeof (*del));
763 		}
764 		mutex_exit(&zone->zone_lock);
765 	}
766 	mutex_exit(&zonehash_lock);
767 
768 	return (0);
769 }
770 
771 /*
772  * ZSD counterpart of pthread_setspecific().
773  *
774  * Since all zsd callbacks, including those with no create function,
775  * have an entry in zone_zsd, if the key is registered it is part of
776  * the zone_zsd list.
777  * Return an error if the key wasn't registerd.
778  */
779 int
780 zone_setspecific(zone_key_t key, zone_t *zone, const void *data)
781 {
782 	struct zsd_entry *t;
783 
784 	mutex_enter(&zone->zone_lock);
785 	t = zsd_find_mru(&zone->zone_zsd, key);
786 	if (t != NULL) {
787 		/*
788 		 * Replace old value with new
789 		 */
790 		t->zsd_data = (void *)data;
791 		mutex_exit(&zone->zone_lock);
792 		return (0);
793 	}
794 	mutex_exit(&zone->zone_lock);
795 	return (-1);
796 }
797 
798 /*
799  * ZSD counterpart of pthread_getspecific().
800  */
801 void *
802 zone_getspecific(zone_key_t key, zone_t *zone)
803 {
804 	struct zsd_entry *t;
805 	void *data;
806 
807 	mutex_enter(&zone->zone_lock);
808 	t = zsd_find_mru(&zone->zone_zsd, key);
809 	data = (t == NULL ? NULL : t->zsd_data);
810 	mutex_exit(&zone->zone_lock);
811 	return (data);
812 }
813 
814 /*
815  * Function used to initialize a zone's list of ZSD callbacks and data
816  * when the zone is being created.  The callbacks are initialized from
817  * the template list (zsd_registered_keys). The constructor callback is
818  * executed later (once the zone exists and with locks dropped).
819  */
820 static void
821 zone_zsd_configure(zone_t *zone)
822 {
823 	struct zsd_entry *zsdp;
824 	struct zsd_entry *t;
825 
826 	ASSERT(MUTEX_HELD(&zonehash_lock));
827 	ASSERT(list_head(&zone->zone_zsd) == NULL);
828 	mutex_enter(&zone->zone_lock);
829 	mutex_enter(&zsd_key_lock);
830 	for (zsdp = list_head(&zsd_registered_keys); zsdp != NULL;
831 	    zsdp = list_next(&zsd_registered_keys, zsdp)) {
832 		/*
833 		 * Since this zone is ZONE_IS_UNCONFIGURED, zone_key_create
834 		 * should not have added anything to it.
835 		 */
836 		ASSERT(zsd_find(&zone->zone_zsd, zsdp->zsd_key) == NULL);
837 
838 		t = kmem_zalloc(sizeof (*t), KM_SLEEP);
839 		t->zsd_key = zsdp->zsd_key;
840 		t->zsd_create = zsdp->zsd_create;
841 		t->zsd_shutdown = zsdp->zsd_shutdown;
842 		t->zsd_destroy = zsdp->zsd_destroy;
843 		if (zsdp->zsd_create != NULL) {
844 			t->zsd_flags = ZSD_CREATE_NEEDED;
845 			DTRACE_PROBE2(zsd__create__needed,
846 			    zone_t *, zone, zone_key_t, zsdp->zsd_key);
847 		}
848 		list_insert_tail(&zone->zone_zsd, t);
849 	}
850 	mutex_exit(&zsd_key_lock);
851 	mutex_exit(&zone->zone_lock);
852 }
853 
854 enum zsd_callback_type { ZSD_CREATE, ZSD_SHUTDOWN, ZSD_DESTROY };
855 
856 /*
857  * Helper function to execute shutdown or destructor callbacks.
858  */
859 static void
860 zone_zsd_callbacks(zone_t *zone, enum zsd_callback_type ct)
861 {
862 	struct zsd_entry *t;
863 
864 	ASSERT(ct == ZSD_SHUTDOWN || ct == ZSD_DESTROY);
865 	ASSERT(ct != ZSD_SHUTDOWN || zone_status_get(zone) >= ZONE_IS_EMPTY);
866 	ASSERT(ct != ZSD_DESTROY || zone_status_get(zone) >= ZONE_IS_DOWN);
867 
868 	/*
869 	 * Run the callback solely based on what is registered for the zone
870 	 * in zone_zsd. The global list can change independently of this
871 	 * as keys are registered and unregistered and we don't register new
872 	 * callbacks for a zone that is in the process of going away.
873 	 */
874 	mutex_enter(&zone->zone_lock);
875 	for (t = list_head(&zone->zone_zsd); t != NULL;
876 	    t = list_next(&zone->zone_zsd, t)) {
877 		zone_key_t key = t->zsd_key;
878 
879 		/* Skip if no callbacks registered */
880 
881 		if (ct == ZSD_SHUTDOWN) {
882 			if (t->zsd_shutdown != NULL &&
883 			    (t->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
884 				t->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
885 				DTRACE_PROBE2(zsd__shutdown__needed,
886 				    zone_t *, zone, zone_key_t, key);
887 			}
888 		} else {
889 			if (t->zsd_destroy != NULL &&
890 			    (t->zsd_flags & ZSD_DESTROY_ALL) == 0) {
891 				t->zsd_flags |= ZSD_DESTROY_NEEDED;
892 				DTRACE_PROBE2(zsd__destroy__needed,
893 				    zone_t *, zone, zone_key_t, key);
894 			}
895 		}
896 	}
897 	mutex_exit(&zone->zone_lock);
898 
899 	/* Now call the shutdown and destroy callback for this key */
900 	zsd_apply_all_keys(zsd_apply_shutdown, zone);
901 	zsd_apply_all_keys(zsd_apply_destroy, zone);
902 
903 }
904 
905 /*
906  * Called when the zone is going away; free ZSD-related memory, and
907  * destroy the zone_zsd list.
908  */
909 static void
910 zone_free_zsd(zone_t *zone)
911 {
912 	struct zsd_entry *t, *next;
913 
914 	/*
915 	 * Free all the zsd_entry's we had on this zone.
916 	 */
917 	mutex_enter(&zone->zone_lock);
918 	for (t = list_head(&zone->zone_zsd); t != NULL; t = next) {
919 		next = list_next(&zone->zone_zsd, t);
920 		list_remove(&zone->zone_zsd, t);
921 		ASSERT(!(t->zsd_flags & ZSD_ALL_INPROGRESS));
922 		kmem_free(t, sizeof (*t));
923 	}
924 	list_destroy(&zone->zone_zsd);
925 	mutex_exit(&zone->zone_lock);
926 
927 }
928 
929 /*
930  * Apply a function to all zones for particular key value.
931  *
932  * The applyfn has to drop zonehash_lock if it does some work, and
933  * then reacquire it before it returns.
934  * When the lock is dropped we don't follow list_next even
935  * if it is possible to do so without any hazards. This is
936  * because we want the design to allow for the list of zones
937  * to change in any arbitrary way during the time the
938  * lock was dropped.
939  *
940  * It is safe to restart the loop at list_head since the applyfn
941  * changes the zsd_flags as it does work, so a subsequent
942  * pass through will have no effect in applyfn, hence the loop will terminate
943  * in at worst O(N^2).
944  */
945 static void
946 zsd_apply_all_zones(zsd_applyfn_t *applyfn, zone_key_t key)
947 {
948 	zone_t *zone;
949 
950 	mutex_enter(&zonehash_lock);
951 	zone = list_head(&zone_active);
952 	while (zone != NULL) {
953 		if ((applyfn)(&zonehash_lock, B_FALSE, zone, key)) {
954 			/* Lock dropped - restart at head */
955 			zone = list_head(&zone_active);
956 		} else {
957 			zone = list_next(&zone_active, zone);
958 		}
959 	}
960 	mutex_exit(&zonehash_lock);
961 }
962 
963 /*
964  * Apply a function to all keys for a particular zone.
965  *
966  * The applyfn has to drop zonehash_lock if it does some work, and
967  * then reacquire it before it returns.
968  * When the lock is dropped we don't follow list_next even
969  * if it is possible to do so without any hazards. This is
970  * because we want the design to allow for the list of zsd callbacks
971  * to change in any arbitrary way during the time the
972  * lock was dropped.
973  *
974  * It is safe to restart the loop at list_head since the applyfn
975  * changes the zsd_flags as it does work, so a subsequent
976  * pass through will have no effect in applyfn, hence the loop will terminate
977  * in at worst O(N^2).
978  */
979 static void
980 zsd_apply_all_keys(zsd_applyfn_t *applyfn, zone_t *zone)
981 {
982 	struct zsd_entry *t;
983 
984 	mutex_enter(&zone->zone_lock);
985 	t = list_head(&zone->zone_zsd);
986 	while (t != NULL) {
987 		if ((applyfn)(NULL, B_TRUE, zone, t->zsd_key)) {
988 			/* Lock dropped - restart at head */
989 			t = list_head(&zone->zone_zsd);
990 		} else {
991 			t = list_next(&zone->zone_zsd, t);
992 		}
993 	}
994 	mutex_exit(&zone->zone_lock);
995 }
996 
997 /*
998  * Call the create function for the zone and key if CREATE_NEEDED
999  * is set.
1000  * If some other thread gets here first and sets CREATE_INPROGRESS, then
1001  * we wait for that thread to complete so that we can ensure that
1002  * all the callbacks are done when we've looped over all zones/keys.
1003  *
1004  * When we call the create function, we drop the global held by the
1005  * caller, and return true to tell the caller it needs to re-evalute the
1006  * state.
1007  * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1008  * remains held on exit.
1009  */
1010 static boolean_t
1011 zsd_apply_create(kmutex_t *lockp, boolean_t zone_lock_held,
1012     zone_t *zone, zone_key_t key)
1013 {
1014 	void *result;
1015 	struct zsd_entry *t;
1016 	boolean_t dropped;
1017 
1018 	if (lockp != NULL) {
1019 		ASSERT(MUTEX_HELD(lockp));
1020 	}
1021 	if (zone_lock_held) {
1022 		ASSERT(MUTEX_HELD(&zone->zone_lock));
1023 	} else {
1024 		mutex_enter(&zone->zone_lock);
1025 	}
1026 
1027 	t = zsd_find(&zone->zone_zsd, key);
1028 	if (t == NULL) {
1029 		/*
1030 		 * Somebody else got here first e.g the zone going
1031 		 * away.
1032 		 */
1033 		if (!zone_lock_held)
1034 			mutex_exit(&zone->zone_lock);
1035 		return (B_FALSE);
1036 	}
1037 	dropped = B_FALSE;
1038 	if (zsd_wait_for_inprogress(zone, t, lockp))
1039 		dropped = B_TRUE;
1040 
1041 	if (t->zsd_flags & ZSD_CREATE_NEEDED) {
1042 		t->zsd_flags &= ~ZSD_CREATE_NEEDED;
1043 		t->zsd_flags |= ZSD_CREATE_INPROGRESS;
1044 		DTRACE_PROBE2(zsd__create__inprogress,
1045 		    zone_t *, zone, zone_key_t, key);
1046 		mutex_exit(&zone->zone_lock);
1047 		if (lockp != NULL)
1048 			mutex_exit(lockp);
1049 
1050 		dropped = B_TRUE;
1051 		ASSERT(t->zsd_create != NULL);
1052 		DTRACE_PROBE2(zsd__create__start,
1053 		    zone_t *, zone, zone_key_t, key);
1054 
1055 		result = (*t->zsd_create)(zone->zone_id);
1056 
1057 		DTRACE_PROBE2(zsd__create__end,
1058 		    zone_t *, zone, voidn *, result);
1059 
1060 		ASSERT(result != NULL);
1061 		if (lockp != NULL)
1062 			mutex_enter(lockp);
1063 		mutex_enter(&zone->zone_lock);
1064 		t->zsd_data = result;
1065 		t->zsd_flags &= ~ZSD_CREATE_INPROGRESS;
1066 		t->zsd_flags |= ZSD_CREATE_COMPLETED;
1067 		cv_broadcast(&t->zsd_cv);
1068 		DTRACE_PROBE2(zsd__create__completed,
1069 		    zone_t *, zone, zone_key_t, key);
1070 	}
1071 	if (!zone_lock_held)
1072 		mutex_exit(&zone->zone_lock);
1073 	return (dropped);
1074 }
1075 
1076 /*
1077  * Call the shutdown function for the zone and key if SHUTDOWN_NEEDED
1078  * is set.
1079  * If some other thread gets here first and sets *_INPROGRESS, then
1080  * we wait for that thread to complete so that we can ensure that
1081  * all the callbacks are done when we've looped over all zones/keys.
1082  *
1083  * When we call the shutdown function, we drop the global held by the
1084  * caller, and return true to tell the caller it needs to re-evalute the
1085  * state.
1086  * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1087  * remains held on exit.
1088  */
1089 static boolean_t
1090 zsd_apply_shutdown(kmutex_t *lockp, boolean_t zone_lock_held,
1091     zone_t *zone, zone_key_t key)
1092 {
1093 	struct zsd_entry *t;
1094 	void *data;
1095 	boolean_t dropped;
1096 
1097 	if (lockp != NULL) {
1098 		ASSERT(MUTEX_HELD(lockp));
1099 	}
1100 	if (zone_lock_held) {
1101 		ASSERT(MUTEX_HELD(&zone->zone_lock));
1102 	} else {
1103 		mutex_enter(&zone->zone_lock);
1104 	}
1105 
1106 	t = zsd_find(&zone->zone_zsd, key);
1107 	if (t == NULL) {
1108 		/*
1109 		 * Somebody else got here first e.g the zone going
1110 		 * away.
1111 		 */
1112 		if (!zone_lock_held)
1113 			mutex_exit(&zone->zone_lock);
1114 		return (B_FALSE);
1115 	}
1116 	dropped = B_FALSE;
1117 	if (zsd_wait_for_creator(zone, t, lockp))
1118 		dropped = B_TRUE;
1119 
1120 	if (zsd_wait_for_inprogress(zone, t, lockp))
1121 		dropped = B_TRUE;
1122 
1123 	if (t->zsd_flags & ZSD_SHUTDOWN_NEEDED) {
1124 		t->zsd_flags &= ~ZSD_SHUTDOWN_NEEDED;
1125 		t->zsd_flags |= ZSD_SHUTDOWN_INPROGRESS;
1126 		DTRACE_PROBE2(zsd__shutdown__inprogress,
1127 		    zone_t *, zone, zone_key_t, key);
1128 		mutex_exit(&zone->zone_lock);
1129 		if (lockp != NULL)
1130 			mutex_exit(lockp);
1131 		dropped = B_TRUE;
1132 
1133 		ASSERT(t->zsd_shutdown != NULL);
1134 		data = t->zsd_data;
1135 
1136 		DTRACE_PROBE2(zsd__shutdown__start,
1137 		    zone_t *, zone, zone_key_t, key);
1138 
1139 		(t->zsd_shutdown)(zone->zone_id, data);
1140 		DTRACE_PROBE2(zsd__shutdown__end,
1141 		    zone_t *, zone, zone_key_t, key);
1142 
1143 		if (lockp != NULL)
1144 			mutex_enter(lockp);
1145 		mutex_enter(&zone->zone_lock);
1146 		t->zsd_flags &= ~ZSD_SHUTDOWN_INPROGRESS;
1147 		t->zsd_flags |= ZSD_SHUTDOWN_COMPLETED;
1148 		cv_broadcast(&t->zsd_cv);
1149 		DTRACE_PROBE2(zsd__shutdown__completed,
1150 		    zone_t *, zone, zone_key_t, key);
1151 	}
1152 	if (!zone_lock_held)
1153 		mutex_exit(&zone->zone_lock);
1154 	return (dropped);
1155 }
1156 
1157 /*
1158  * Call the destroy function for the zone and key if DESTROY_NEEDED
1159  * is set.
1160  * If some other thread gets here first and sets *_INPROGRESS, then
1161  * we wait for that thread to complete so that we can ensure that
1162  * all the callbacks are done when we've looped over all zones/keys.
1163  *
1164  * When we call the destroy function, we drop the global held by the
1165  * caller, and return true to tell the caller it needs to re-evalute the
1166  * state.
1167  * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1168  * remains held on exit.
1169  */
1170 static boolean_t
1171 zsd_apply_destroy(kmutex_t *lockp, boolean_t zone_lock_held,
1172     zone_t *zone, zone_key_t key)
1173 {
1174 	struct zsd_entry *t;
1175 	void *data;
1176 	boolean_t dropped;
1177 
1178 	if (lockp != NULL) {
1179 		ASSERT(MUTEX_HELD(lockp));
1180 	}
1181 	if (zone_lock_held) {
1182 		ASSERT(MUTEX_HELD(&zone->zone_lock));
1183 	} else {
1184 		mutex_enter(&zone->zone_lock);
1185 	}
1186 
1187 	t = zsd_find(&zone->zone_zsd, key);
1188 	if (t == NULL) {
1189 		/*
1190 		 * Somebody else got here first e.g the zone going
1191 		 * away.
1192 		 */
1193 		if (!zone_lock_held)
1194 			mutex_exit(&zone->zone_lock);
1195 		return (B_FALSE);
1196 	}
1197 	dropped = B_FALSE;
1198 	if (zsd_wait_for_creator(zone, t, lockp))
1199 		dropped = B_TRUE;
1200 
1201 	if (zsd_wait_for_inprogress(zone, t, lockp))
1202 		dropped = B_TRUE;
1203 
1204 	if (t->zsd_flags & ZSD_DESTROY_NEEDED) {
1205 		t->zsd_flags &= ~ZSD_DESTROY_NEEDED;
1206 		t->zsd_flags |= ZSD_DESTROY_INPROGRESS;
1207 		DTRACE_PROBE2(zsd__destroy__inprogress,
1208 		    zone_t *, zone, zone_key_t, key);
1209 		mutex_exit(&zone->zone_lock);
1210 		if (lockp != NULL)
1211 			mutex_exit(lockp);
1212 		dropped = B_TRUE;
1213 
1214 		ASSERT(t->zsd_destroy != NULL);
1215 		data = t->zsd_data;
1216 		DTRACE_PROBE2(zsd__destroy__start,
1217 		    zone_t *, zone, zone_key_t, key);
1218 
1219 		(t->zsd_destroy)(zone->zone_id, data);
1220 		DTRACE_PROBE2(zsd__destroy__end,
1221 		    zone_t *, zone, zone_key_t, key);
1222 
1223 		if (lockp != NULL)
1224 			mutex_enter(lockp);
1225 		mutex_enter(&zone->zone_lock);
1226 		t->zsd_data = NULL;
1227 		t->zsd_flags &= ~ZSD_DESTROY_INPROGRESS;
1228 		t->zsd_flags |= ZSD_DESTROY_COMPLETED;
1229 		cv_broadcast(&t->zsd_cv);
1230 		DTRACE_PROBE2(zsd__destroy__completed,
1231 		    zone_t *, zone, zone_key_t, key);
1232 	}
1233 	if (!zone_lock_held)
1234 		mutex_exit(&zone->zone_lock);
1235 	return (dropped);
1236 }
1237 
1238 /*
1239  * Wait for any CREATE_NEEDED flag to be cleared.
1240  * Returns true if lockp was temporarily dropped while waiting.
1241  */
1242 static boolean_t
1243 zsd_wait_for_creator(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
1244 {
1245 	boolean_t dropped = B_FALSE;
1246 
1247 	while (t->zsd_flags & ZSD_CREATE_NEEDED) {
1248 		DTRACE_PROBE2(zsd__wait__for__creator,
1249 		    zone_t *, zone, struct zsd_entry *, t);
1250 		if (lockp != NULL) {
1251 			dropped = B_TRUE;
1252 			mutex_exit(lockp);
1253 		}
1254 		cv_wait(&t->zsd_cv, &zone->zone_lock);
1255 		if (lockp != NULL) {
1256 			/* First drop zone_lock to preserve order */
1257 			mutex_exit(&zone->zone_lock);
1258 			mutex_enter(lockp);
1259 			mutex_enter(&zone->zone_lock);
1260 		}
1261 	}
1262 	return (dropped);
1263 }
1264 
1265 /*
1266  * Wait for any INPROGRESS flag to be cleared.
1267  * Returns true if lockp was temporarily dropped while waiting.
1268  */
1269 static boolean_t
1270 zsd_wait_for_inprogress(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
1271 {
1272 	boolean_t dropped = B_FALSE;
1273 
1274 	while (t->zsd_flags & ZSD_ALL_INPROGRESS) {
1275 		DTRACE_PROBE2(zsd__wait__for__inprogress,
1276 		    zone_t *, zone, struct zsd_entry *, t);
1277 		if (lockp != NULL) {
1278 			dropped = B_TRUE;
1279 			mutex_exit(lockp);
1280 		}
1281 		cv_wait(&t->zsd_cv, &zone->zone_lock);
1282 		if (lockp != NULL) {
1283 			/* First drop zone_lock to preserve order */
1284 			mutex_exit(&zone->zone_lock);
1285 			mutex_enter(lockp);
1286 			mutex_enter(&zone->zone_lock);
1287 		}
1288 	}
1289 	return (dropped);
1290 }
1291 
1292 /*
1293  * Frees memory associated with the zone dataset list.
1294  */
1295 static void
1296 zone_free_datasets(zone_t *zone)
1297 {
1298 	zone_dataset_t *t, *next;
1299 
1300 	for (t = list_head(&zone->zone_datasets); t != NULL; t = next) {
1301 		next = list_next(&zone->zone_datasets, t);
1302 		list_remove(&zone->zone_datasets, t);
1303 		kmem_free(t->zd_dataset, strlen(t->zd_dataset) + 1);
1304 		kmem_free(t, sizeof (*t));
1305 	}
1306 	list_destroy(&zone->zone_datasets);
1307 }
1308 
1309 /*
1310  * zone.cpu-shares resource control support.
1311  */
1312 /*ARGSUSED*/
1313 static rctl_qty_t
1314 zone_cpu_shares_usage(rctl_t *rctl, struct proc *p)
1315 {
1316 	ASSERT(MUTEX_HELD(&p->p_lock));
1317 	return (p->p_zone->zone_shares);
1318 }
1319 
1320 /*ARGSUSED*/
1321 static int
1322 zone_cpu_shares_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1323     rctl_qty_t nv)
1324 {
1325 	ASSERT(MUTEX_HELD(&p->p_lock));
1326 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1327 	if (e->rcep_p.zone == NULL)
1328 		return (0);
1329 
1330 	e->rcep_p.zone->zone_shares = nv;
1331 	return (0);
1332 }
1333 
1334 static rctl_ops_t zone_cpu_shares_ops = {
1335 	rcop_no_action,
1336 	zone_cpu_shares_usage,
1337 	zone_cpu_shares_set,
1338 	rcop_no_test
1339 };
1340 
1341 /*
1342  * zone.cpu-cap resource control support.
1343  */
1344 /*ARGSUSED*/
1345 static rctl_qty_t
1346 zone_cpu_cap_get(rctl_t *rctl, struct proc *p)
1347 {
1348 	ASSERT(MUTEX_HELD(&p->p_lock));
1349 	return (cpucaps_zone_get(p->p_zone));
1350 }
1351 
1352 /*ARGSUSED*/
1353 static int
1354 zone_cpu_cap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1355     rctl_qty_t nv)
1356 {
1357 	zone_t *zone = e->rcep_p.zone;
1358 
1359 	ASSERT(MUTEX_HELD(&p->p_lock));
1360 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1361 
1362 	if (zone == NULL)
1363 		return (0);
1364 
1365 	/*
1366 	 * set cap to the new value.
1367 	 */
1368 	return (cpucaps_zone_set(zone, nv));
1369 }
1370 
1371 static rctl_ops_t zone_cpu_cap_ops = {
1372 	rcop_no_action,
1373 	zone_cpu_cap_get,
1374 	zone_cpu_cap_set,
1375 	rcop_no_test
1376 };
1377 
1378 /*ARGSUSED*/
1379 static rctl_qty_t
1380 zone_lwps_usage(rctl_t *r, proc_t *p)
1381 {
1382 	rctl_qty_t nlwps;
1383 	zone_t *zone = p->p_zone;
1384 
1385 	ASSERT(MUTEX_HELD(&p->p_lock));
1386 
1387 	mutex_enter(&zone->zone_nlwps_lock);
1388 	nlwps = zone->zone_nlwps;
1389 	mutex_exit(&zone->zone_nlwps_lock);
1390 
1391 	return (nlwps);
1392 }
1393 
1394 /*ARGSUSED*/
1395 static int
1396 zone_lwps_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
1397     rctl_qty_t incr, uint_t flags)
1398 {
1399 	rctl_qty_t nlwps;
1400 
1401 	ASSERT(MUTEX_HELD(&p->p_lock));
1402 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1403 	if (e->rcep_p.zone == NULL)
1404 		return (0);
1405 	ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
1406 	nlwps = e->rcep_p.zone->zone_nlwps;
1407 
1408 	if (nlwps + incr > rcntl->rcv_value)
1409 		return (1);
1410 
1411 	return (0);
1412 }
1413 
1414 /*ARGSUSED*/
1415 static int
1416 zone_lwps_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv)
1417 {
1418 	ASSERT(MUTEX_HELD(&p->p_lock));
1419 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1420 	if (e->rcep_p.zone == NULL)
1421 		return (0);
1422 	e->rcep_p.zone->zone_nlwps_ctl = nv;
1423 	return (0);
1424 }
1425 
1426 static rctl_ops_t zone_lwps_ops = {
1427 	rcop_no_action,
1428 	zone_lwps_usage,
1429 	zone_lwps_set,
1430 	zone_lwps_test,
1431 };
1432 
1433 /*ARGSUSED*/
1434 static rctl_qty_t
1435 zone_procs_usage(rctl_t *r, proc_t *p)
1436 {
1437 	rctl_qty_t nprocs;
1438 	zone_t *zone = p->p_zone;
1439 
1440 	ASSERT(MUTEX_HELD(&p->p_lock));
1441 
1442 	mutex_enter(&zone->zone_nlwps_lock);
1443 	nprocs = zone->zone_nprocs;
1444 	mutex_exit(&zone->zone_nlwps_lock);
1445 
1446 	return (nprocs);
1447 }
1448 
1449 /*ARGSUSED*/
1450 static int
1451 zone_procs_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
1452     rctl_qty_t incr, uint_t flags)
1453 {
1454 	rctl_qty_t nprocs;
1455 
1456 	ASSERT(MUTEX_HELD(&p->p_lock));
1457 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1458 	if (e->rcep_p.zone == NULL)
1459 		return (0);
1460 	ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
1461 	nprocs = e->rcep_p.zone->zone_nprocs;
1462 
1463 	if (nprocs + incr > rcntl->rcv_value)
1464 		return (1);
1465 
1466 	return (0);
1467 }
1468 
1469 /*ARGSUSED*/
1470 static int
1471 zone_procs_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv)
1472 {
1473 	ASSERT(MUTEX_HELD(&p->p_lock));
1474 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1475 	if (e->rcep_p.zone == NULL)
1476 		return (0);
1477 	e->rcep_p.zone->zone_nprocs_ctl = nv;
1478 	return (0);
1479 }
1480 
1481 static rctl_ops_t zone_procs_ops = {
1482 	rcop_no_action,
1483 	zone_procs_usage,
1484 	zone_procs_set,
1485 	zone_procs_test,
1486 };
1487 
1488 /*ARGSUSED*/
1489 static int
1490 zone_shmmax_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1491     rctl_qty_t incr, uint_t flags)
1492 {
1493 	rctl_qty_t v;
1494 	ASSERT(MUTEX_HELD(&p->p_lock));
1495 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1496 	v = e->rcep_p.zone->zone_shmmax + incr;
1497 	if (v > rval->rcv_value)
1498 		return (1);
1499 	return (0);
1500 }
1501 
1502 static rctl_ops_t zone_shmmax_ops = {
1503 	rcop_no_action,
1504 	rcop_no_usage,
1505 	rcop_no_set,
1506 	zone_shmmax_test
1507 };
1508 
1509 /*ARGSUSED*/
1510 static int
1511 zone_shmmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1512     rctl_qty_t incr, uint_t flags)
1513 {
1514 	rctl_qty_t v;
1515 	ASSERT(MUTEX_HELD(&p->p_lock));
1516 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1517 	v = e->rcep_p.zone->zone_ipc.ipcq_shmmni + incr;
1518 	if (v > rval->rcv_value)
1519 		return (1);
1520 	return (0);
1521 }
1522 
1523 static rctl_ops_t zone_shmmni_ops = {
1524 	rcop_no_action,
1525 	rcop_no_usage,
1526 	rcop_no_set,
1527 	zone_shmmni_test
1528 };
1529 
1530 /*ARGSUSED*/
1531 static int
1532 zone_semmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1533     rctl_qty_t incr, uint_t flags)
1534 {
1535 	rctl_qty_t v;
1536 	ASSERT(MUTEX_HELD(&p->p_lock));
1537 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1538 	v = e->rcep_p.zone->zone_ipc.ipcq_semmni + incr;
1539 	if (v > rval->rcv_value)
1540 		return (1);
1541 	return (0);
1542 }
1543 
1544 static rctl_ops_t zone_semmni_ops = {
1545 	rcop_no_action,
1546 	rcop_no_usage,
1547 	rcop_no_set,
1548 	zone_semmni_test
1549 };
1550 
1551 /*ARGSUSED*/
1552 static int
1553 zone_msgmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1554     rctl_qty_t incr, uint_t flags)
1555 {
1556 	rctl_qty_t v;
1557 	ASSERT(MUTEX_HELD(&p->p_lock));
1558 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1559 	v = e->rcep_p.zone->zone_ipc.ipcq_msgmni + incr;
1560 	if (v > rval->rcv_value)
1561 		return (1);
1562 	return (0);
1563 }
1564 
1565 static rctl_ops_t zone_msgmni_ops = {
1566 	rcop_no_action,
1567 	rcop_no_usage,
1568 	rcop_no_set,
1569 	zone_msgmni_test
1570 };
1571 
1572 /*ARGSUSED*/
1573 static rctl_qty_t
1574 zone_locked_mem_usage(rctl_t *rctl, struct proc *p)
1575 {
1576 	rctl_qty_t q;
1577 	ASSERT(MUTEX_HELD(&p->p_lock));
1578 	mutex_enter(&p->p_zone->zone_mem_lock);
1579 	q = p->p_zone->zone_locked_mem;
1580 	mutex_exit(&p->p_zone->zone_mem_lock);
1581 	return (q);
1582 }
1583 
1584 /*ARGSUSED*/
1585 static int
1586 zone_locked_mem_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1587     rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1588 {
1589 	rctl_qty_t q;
1590 	zone_t *z;
1591 
1592 	z = e->rcep_p.zone;
1593 	ASSERT(MUTEX_HELD(&p->p_lock));
1594 	ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1595 	q = z->zone_locked_mem;
1596 	if (q + incr > rcntl->rcv_value)
1597 		return (1);
1598 	return (0);
1599 }
1600 
1601 /*ARGSUSED*/
1602 static int
1603 zone_locked_mem_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1604     rctl_qty_t nv)
1605 {
1606 	ASSERT(MUTEX_HELD(&p->p_lock));
1607 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1608 	if (e->rcep_p.zone == NULL)
1609 		return (0);
1610 	e->rcep_p.zone->zone_locked_mem_ctl = nv;
1611 	return (0);
1612 }
1613 
1614 static rctl_ops_t zone_locked_mem_ops = {
1615 	rcop_no_action,
1616 	zone_locked_mem_usage,
1617 	zone_locked_mem_set,
1618 	zone_locked_mem_test
1619 };
1620 
1621 /*ARGSUSED*/
1622 static rctl_qty_t
1623 zone_max_swap_usage(rctl_t *rctl, struct proc *p)
1624 {
1625 	rctl_qty_t q;
1626 	zone_t *z = p->p_zone;
1627 
1628 	ASSERT(MUTEX_HELD(&p->p_lock));
1629 	mutex_enter(&z->zone_mem_lock);
1630 	q = z->zone_max_swap;
1631 	mutex_exit(&z->zone_mem_lock);
1632 	return (q);
1633 }
1634 
1635 /*ARGSUSED*/
1636 static int
1637 zone_max_swap_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1638     rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1639 {
1640 	rctl_qty_t q;
1641 	zone_t *z;
1642 
1643 	z = e->rcep_p.zone;
1644 	ASSERT(MUTEX_HELD(&p->p_lock));
1645 	ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1646 	q = z->zone_max_swap;
1647 	if (q + incr > rcntl->rcv_value)
1648 		return (1);
1649 	return (0);
1650 }
1651 
1652 /*ARGSUSED*/
1653 static int
1654 zone_max_swap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1655     rctl_qty_t nv)
1656 {
1657 	ASSERT(MUTEX_HELD(&p->p_lock));
1658 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1659 	if (e->rcep_p.zone == NULL)
1660 		return (0);
1661 	e->rcep_p.zone->zone_max_swap_ctl = nv;
1662 	return (0);
1663 }
1664 
1665 static rctl_ops_t zone_max_swap_ops = {
1666 	rcop_no_action,
1667 	zone_max_swap_usage,
1668 	zone_max_swap_set,
1669 	zone_max_swap_test
1670 };
1671 
1672 /*ARGSUSED*/
1673 static rctl_qty_t
1674 zone_max_lofi_usage(rctl_t *rctl, struct proc *p)
1675 {
1676 	rctl_qty_t q;
1677 	zone_t *z = p->p_zone;
1678 
1679 	ASSERT(MUTEX_HELD(&p->p_lock));
1680 	mutex_enter(&z->zone_rctl_lock);
1681 	q = z->zone_max_lofi;
1682 	mutex_exit(&z->zone_rctl_lock);
1683 	return (q);
1684 }
1685 
1686 /*ARGSUSED*/
1687 static int
1688 zone_max_lofi_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1689     rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1690 {
1691 	rctl_qty_t q;
1692 	zone_t *z;
1693 
1694 	z = e->rcep_p.zone;
1695 	ASSERT(MUTEX_HELD(&p->p_lock));
1696 	ASSERT(MUTEX_HELD(&z->zone_rctl_lock));
1697 	q = z->zone_max_lofi;
1698 	if (q + incr > rcntl->rcv_value)
1699 		return (1);
1700 	return (0);
1701 }
1702 
1703 /*ARGSUSED*/
1704 static int
1705 zone_max_lofi_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1706     rctl_qty_t nv)
1707 {
1708 	ASSERT(MUTEX_HELD(&p->p_lock));
1709 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1710 	if (e->rcep_p.zone == NULL)
1711 		return (0);
1712 	e->rcep_p.zone->zone_max_lofi_ctl = nv;
1713 	return (0);
1714 }
1715 
1716 static rctl_ops_t zone_max_lofi_ops = {
1717 	rcop_no_action,
1718 	zone_max_lofi_usage,
1719 	zone_max_lofi_set,
1720 	zone_max_lofi_test
1721 };
1722 
1723 /*
1724  * Helper function to brand the zone with a unique ID.
1725  */
1726 static void
1727 zone_uniqid(zone_t *zone)
1728 {
1729 	static uint64_t uniqid = 0;
1730 
1731 	ASSERT(MUTEX_HELD(&zonehash_lock));
1732 	zone->zone_uniqid = uniqid++;
1733 }
1734 
1735 /*
1736  * Returns a held pointer to the "kcred" for the specified zone.
1737  */
1738 struct cred *
1739 zone_get_kcred(zoneid_t zoneid)
1740 {
1741 	zone_t *zone;
1742 	cred_t *cr;
1743 
1744 	if ((zone = zone_find_by_id(zoneid)) == NULL)
1745 		return (NULL);
1746 	cr = zone->zone_kcred;
1747 	crhold(cr);
1748 	zone_rele(zone);
1749 	return (cr);
1750 }
1751 
1752 static int
1753 zone_lockedmem_kstat_update(kstat_t *ksp, int rw)
1754 {
1755 	zone_t *zone = ksp->ks_private;
1756 	zone_kstat_t *zk = ksp->ks_data;
1757 
1758 	if (rw == KSTAT_WRITE)
1759 		return (EACCES);
1760 
1761 	zk->zk_usage.value.ui64 = zone->zone_locked_mem;
1762 	zk->zk_value.value.ui64 = zone->zone_locked_mem_ctl;
1763 	return (0);
1764 }
1765 
1766 static int
1767 zone_nprocs_kstat_update(kstat_t *ksp, int rw)
1768 {
1769 	zone_t *zone = ksp->ks_private;
1770 	zone_kstat_t *zk = ksp->ks_data;
1771 
1772 	if (rw == KSTAT_WRITE)
1773 		return (EACCES);
1774 
1775 	zk->zk_usage.value.ui64 = zone->zone_nprocs;
1776 	zk->zk_value.value.ui64 = zone->zone_nprocs_ctl;
1777 	return (0);
1778 }
1779 
1780 static int
1781 zone_swapresv_kstat_update(kstat_t *ksp, int rw)
1782 {
1783 	zone_t *zone = ksp->ks_private;
1784 	zone_kstat_t *zk = ksp->ks_data;
1785 
1786 	if (rw == KSTAT_WRITE)
1787 		return (EACCES);
1788 
1789 	zk->zk_usage.value.ui64 = zone->zone_max_swap;
1790 	zk->zk_value.value.ui64 = zone->zone_max_swap_ctl;
1791 	return (0);
1792 }
1793 
1794 static kstat_t *
1795 zone_kstat_create_common(zone_t *zone, char *name,
1796     int (*updatefunc) (kstat_t *, int))
1797 {
1798 	kstat_t *ksp;
1799 	zone_kstat_t *zk;
1800 
1801 	ksp = rctl_kstat_create_zone(zone, name, KSTAT_TYPE_NAMED,
1802 	    sizeof (zone_kstat_t) / sizeof (kstat_named_t),
1803 	    KSTAT_FLAG_VIRTUAL);
1804 
1805 	if (ksp == NULL)
1806 		return (NULL);
1807 
1808 	zk = ksp->ks_data = kmem_alloc(sizeof (zone_kstat_t), KM_SLEEP);
1809 	ksp->ks_data_size += strlen(zone->zone_name) + 1;
1810 	kstat_named_init(&zk->zk_zonename, "zonename", KSTAT_DATA_STRING);
1811 	kstat_named_setstr(&zk->zk_zonename, zone->zone_name);
1812 	kstat_named_init(&zk->zk_usage, "usage", KSTAT_DATA_UINT64);
1813 	kstat_named_init(&zk->zk_value, "value", KSTAT_DATA_UINT64);
1814 	ksp->ks_update = updatefunc;
1815 	ksp->ks_private = zone;
1816 	kstat_install(ksp);
1817 	return (ksp);
1818 }
1819 
1820 
1821 static int
1822 zone_mcap_kstat_update(kstat_t *ksp, int rw)
1823 {
1824 	zone_t *zone = ksp->ks_private;
1825 	zone_mcap_kstat_t *zmp = ksp->ks_data;
1826 
1827 	if (rw == KSTAT_WRITE)
1828 		return (EACCES);
1829 
1830 	zmp->zm_pgpgin.value.ui64 = zone->zone_pgpgin;
1831 	zmp->zm_anonpgin.value.ui64 = zone->zone_anonpgin;
1832 	zmp->zm_execpgin.value.ui64 = zone->zone_execpgin;
1833 	zmp->zm_fspgin.value.ui64 = zone->zone_fspgin;
1834 	zmp->zm_anon_alloc_fail.value.ui64 = zone->zone_anon_alloc_fail;
1835 
1836 	return (0);
1837 }
1838 
1839 static kstat_t *
1840 zone_mcap_kstat_create(zone_t *zone)
1841 {
1842 	kstat_t *ksp;
1843 	zone_mcap_kstat_t *zmp;
1844 
1845 	if ((ksp = kstat_create_zone("memory_cap", zone->zone_id,
1846 	    zone->zone_name, "zone_memory_cap", KSTAT_TYPE_NAMED,
1847 	    sizeof (zone_mcap_kstat_t) / sizeof (kstat_named_t),
1848 	    KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
1849 		return (NULL);
1850 
1851 	if (zone->zone_id != GLOBAL_ZONEID)
1852 		kstat_zone_add(ksp, GLOBAL_ZONEID);
1853 
1854 	zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_mcap_kstat_t), KM_SLEEP);
1855 	ksp->ks_data_size += strlen(zone->zone_name) + 1;
1856 	ksp->ks_lock = &zone->zone_mcap_lock;
1857 	zone->zone_mcap_stats = zmp;
1858 
1859 	/* The kstat "name" field is not large enough for a full zonename */
1860 	kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
1861 	kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
1862 	kstat_named_init(&zmp->zm_pgpgin, "pgpgin", KSTAT_DATA_UINT64);
1863 	kstat_named_init(&zmp->zm_anonpgin, "anonpgin", KSTAT_DATA_UINT64);
1864 	kstat_named_init(&zmp->zm_execpgin, "execpgin", KSTAT_DATA_UINT64);
1865 	kstat_named_init(&zmp->zm_fspgin, "fspgin", KSTAT_DATA_UINT64);
1866 	kstat_named_init(&zmp->zm_anon_alloc_fail, "anon_alloc_fail",
1867 	    KSTAT_DATA_UINT64);
1868 
1869 	ksp->ks_update = zone_mcap_kstat_update;
1870 	ksp->ks_private = zone;
1871 
1872 	kstat_install(ksp);
1873 	return (ksp);
1874 }
1875 
1876 static int
1877 zone_misc_kstat_update(kstat_t *ksp, int rw)
1878 {
1879 	zone_t *zone = ksp->ks_private;
1880 	zone_misc_kstat_t *zmp = ksp->ks_data;
1881 	hrtime_t tmp;
1882 
1883 	if (rw == KSTAT_WRITE)
1884 		return (EACCES);
1885 
1886 	tmp = zone->zone_utime;
1887 	scalehrtime(&tmp);
1888 	zmp->zm_utime.value.ui64 = tmp;
1889 	tmp = zone->zone_stime;
1890 	scalehrtime(&tmp);
1891 	zmp->zm_stime.value.ui64 = tmp;
1892 	tmp = zone->zone_wtime;
1893 	scalehrtime(&tmp);
1894 	zmp->zm_wtime.value.ui64 = tmp;
1895 
1896 	zmp->zm_avenrun1.value.ui32 = zone->zone_avenrun[0];
1897 	zmp->zm_avenrun5.value.ui32 = zone->zone_avenrun[1];
1898 	zmp->zm_avenrun15.value.ui32 = zone->zone_avenrun[2];
1899 
1900 	zmp->zm_ffcap.value.ui32 = zone->zone_ffcap;
1901 	zmp->zm_ffnoproc.value.ui32 = zone->zone_ffnoproc;
1902 	zmp->zm_ffnomem.value.ui32 = zone->zone_ffnomem;
1903 	zmp->zm_ffmisc.value.ui32 = zone->zone_ffmisc;
1904 
1905 	zmp->zm_nested_intp.value.ui32 = zone->zone_nested_intp;
1906 
1907 	zmp->zm_init_pid.value.ui32 = zone->zone_proc_initpid;
1908 	zmp->zm_boot_time.value.ui64 = (uint64_t)zone->zone_boot_time;
1909 
1910 	return (0);
1911 }
1912 
1913 static kstat_t *
1914 zone_misc_kstat_create(zone_t *zone)
1915 {
1916 	kstat_t *ksp;
1917 	zone_misc_kstat_t *zmp;
1918 
1919 	if ((ksp = kstat_create_zone("zones", zone->zone_id,
1920 	    zone->zone_name, "zone_misc", KSTAT_TYPE_NAMED,
1921 	    sizeof (zone_misc_kstat_t) / sizeof (kstat_named_t),
1922 	    KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
1923 		return (NULL);
1924 
1925 	if (zone->zone_id != GLOBAL_ZONEID)
1926 		kstat_zone_add(ksp, GLOBAL_ZONEID);
1927 
1928 	zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_misc_kstat_t), KM_SLEEP);
1929 	ksp->ks_data_size += strlen(zone->zone_name) + 1;
1930 	ksp->ks_lock = &zone->zone_misc_lock;
1931 	zone->zone_misc_stats = zmp;
1932 
1933 	/* The kstat "name" field is not large enough for a full zonename */
1934 	kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
1935 	kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
1936 	kstat_named_init(&zmp->zm_utime, "nsec_user", KSTAT_DATA_UINT64);
1937 	kstat_named_init(&zmp->zm_stime, "nsec_sys", KSTAT_DATA_UINT64);
1938 	kstat_named_init(&zmp->zm_wtime, "nsec_waitrq", KSTAT_DATA_UINT64);
1939 	kstat_named_init(&zmp->zm_avenrun1, "avenrun_1min", KSTAT_DATA_UINT32);
1940 	kstat_named_init(&zmp->zm_avenrun5, "avenrun_5min", KSTAT_DATA_UINT32);
1941 	kstat_named_init(&zmp->zm_avenrun15, "avenrun_15min",
1942 	    KSTAT_DATA_UINT32);
1943 	kstat_named_init(&zmp->zm_ffcap, "forkfail_cap", KSTAT_DATA_UINT32);
1944 	kstat_named_init(&zmp->zm_ffnoproc, "forkfail_noproc",
1945 	    KSTAT_DATA_UINT32);
1946 	kstat_named_init(&zmp->zm_ffnomem, "forkfail_nomem", KSTAT_DATA_UINT32);
1947 	kstat_named_init(&zmp->zm_ffmisc, "forkfail_misc", KSTAT_DATA_UINT32);
1948 	kstat_named_init(&zmp->zm_nested_intp, "nested_interp",
1949 	    KSTAT_DATA_UINT32);
1950 	kstat_named_init(&zmp->zm_init_pid, "init_pid", KSTAT_DATA_UINT32);
1951 	kstat_named_init(&zmp->zm_boot_time, "boot_time", KSTAT_DATA_UINT64);
1952 
1953 	ksp->ks_update = zone_misc_kstat_update;
1954 	ksp->ks_private = zone;
1955 
1956 	kstat_install(ksp);
1957 	return (ksp);
1958 }
1959 
1960 static void
1961 zone_kstat_create(zone_t *zone)
1962 {
1963 	zone->zone_lockedmem_kstat = zone_kstat_create_common(zone,
1964 	    "lockedmem", zone_lockedmem_kstat_update);
1965 	zone->zone_swapresv_kstat = zone_kstat_create_common(zone,
1966 	    "swapresv", zone_swapresv_kstat_update);
1967 	zone->zone_nprocs_kstat = zone_kstat_create_common(zone,
1968 	    "nprocs", zone_nprocs_kstat_update);
1969 
1970 	if ((zone->zone_mcap_ksp = zone_mcap_kstat_create(zone)) == NULL) {
1971 		zone->zone_mcap_stats = kmem_zalloc(
1972 		    sizeof (zone_mcap_kstat_t), KM_SLEEP);
1973 	}
1974 
1975 	if ((zone->zone_misc_ksp = zone_misc_kstat_create(zone)) == NULL) {
1976 		zone->zone_misc_stats = kmem_zalloc(
1977 		    sizeof (zone_misc_kstat_t), KM_SLEEP);
1978 	}
1979 }
1980 
1981 static void
1982 zone_kstat_delete_common(kstat_t **pkstat, size_t datasz)
1983 {
1984 	void *data;
1985 
1986 	if (*pkstat != NULL) {
1987 		data = (*pkstat)->ks_data;
1988 		kstat_delete(*pkstat);
1989 		kmem_free(data, datasz);
1990 		*pkstat = NULL;
1991 	}
1992 }
1993 
1994 static void
1995 zone_kstat_delete(zone_t *zone)
1996 {
1997 	zone_kstat_delete_common(&zone->zone_lockedmem_kstat,
1998 	    sizeof (zone_kstat_t));
1999 	zone_kstat_delete_common(&zone->zone_swapresv_kstat,
2000 	    sizeof (zone_kstat_t));
2001 	zone_kstat_delete_common(&zone->zone_nprocs_kstat,
2002 	    sizeof (zone_kstat_t));
2003 	zone_kstat_delete_common(&zone->zone_mcap_ksp,
2004 	    sizeof (zone_mcap_kstat_t));
2005 	zone_kstat_delete_common(&zone->zone_misc_ksp,
2006 	    sizeof (zone_misc_kstat_t));
2007 }
2008 
2009 /*
2010  * Called very early on in boot to initialize the ZSD list so that
2011  * zone_key_create() can be called before zone_init().  It also initializes
2012  * portions of zone0 which may be used before zone_init() is called.  The
2013  * variable "global_zone" will be set when zone0 is fully initialized by
2014  * zone_init().
2015  */
2016 void
2017 zone_zsd_init(void)
2018 {
2019 	mutex_init(&zonehash_lock, NULL, MUTEX_DEFAULT, NULL);
2020 	mutex_init(&zsd_key_lock, NULL, MUTEX_DEFAULT, NULL);
2021 	list_create(&zsd_registered_keys, sizeof (struct zsd_entry),
2022 	    offsetof(struct zsd_entry, zsd_linkage));
2023 	list_create(&zone_active, sizeof (zone_t),
2024 	    offsetof(zone_t, zone_linkage));
2025 	list_create(&zone_deathrow, sizeof (zone_t),
2026 	    offsetof(zone_t, zone_linkage));
2027 
2028 	mutex_init(&zone0.zone_lock, NULL, MUTEX_DEFAULT, NULL);
2029 	mutex_init(&zone0.zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
2030 	mutex_init(&zone0.zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
2031 	zone0.zone_shares = 1;
2032 	zone0.zone_nlwps = 0;
2033 	zone0.zone_nlwps_ctl = INT_MAX;
2034 	zone0.zone_nprocs = 0;
2035 	zone0.zone_nprocs_ctl = INT_MAX;
2036 	zone0.zone_locked_mem = 0;
2037 	zone0.zone_locked_mem_ctl = UINT64_MAX;
2038 	ASSERT(zone0.zone_max_swap == 0);
2039 	zone0.zone_max_swap_ctl = UINT64_MAX;
2040 	zone0.zone_max_lofi = 0;
2041 	zone0.zone_max_lofi_ctl = UINT64_MAX;
2042 	zone0.zone_shmmax = 0;
2043 	zone0.zone_ipc.ipcq_shmmni = 0;
2044 	zone0.zone_ipc.ipcq_semmni = 0;
2045 	zone0.zone_ipc.ipcq_msgmni = 0;
2046 	zone0.zone_name = GLOBAL_ZONENAME;
2047 	zone0.zone_nodename = utsname.nodename;
2048 	zone0.zone_domain = srpc_domain;
2049 	zone0.zone_hostid = HW_INVALID_HOSTID;
2050 	zone0.zone_fs_allowed = NULL;
2051 	zone0.zone_ref = 1;
2052 	zone0.zone_id = GLOBAL_ZONEID;
2053 	zone0.zone_status = ZONE_IS_RUNNING;
2054 	zone0.zone_rootpath = "/";
2055 	zone0.zone_rootpathlen = 2;
2056 	zone0.zone_psetid = ZONE_PS_INVAL;
2057 	zone0.zone_ncpus = 0;
2058 	zone0.zone_ncpus_online = 0;
2059 	zone0.zone_proc_initpid = 1;
2060 	zone0.zone_initname = initname;
2061 	zone0.zone_lockedmem_kstat = NULL;
2062 	zone0.zone_swapresv_kstat = NULL;
2063 	zone0.zone_nprocs_kstat = NULL;
2064 
2065 	zone0.zone_stime = 0;
2066 	zone0.zone_utime = 0;
2067 	zone0.zone_wtime = 0;
2068 
2069 	list_create(&zone0.zone_ref_list, sizeof (zone_ref_t),
2070 	    offsetof(zone_ref_t, zref_linkage));
2071 	list_create(&zone0.zone_zsd, sizeof (struct zsd_entry),
2072 	    offsetof(struct zsd_entry, zsd_linkage));
2073 	list_insert_head(&zone_active, &zone0);
2074 
2075 	/*
2076 	 * The root filesystem is not mounted yet, so zone_rootvp cannot be set
2077 	 * to anything meaningful.  It is assigned to be 'rootdir' in
2078 	 * vfs_mountroot().
2079 	 */
2080 	zone0.zone_rootvp = NULL;
2081 	zone0.zone_vfslist = NULL;
2082 	zone0.zone_bootargs = initargs;
2083 	zone0.zone_privset = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
2084 	/*
2085 	 * The global zone has all privileges
2086 	 */
2087 	priv_fillset(zone0.zone_privset);
2088 	/*
2089 	 * Add p0 to the global zone
2090 	 */
2091 	zone0.zone_zsched = &p0;
2092 	p0.p_zone = &zone0;
2093 }
2094 
2095 /*
2096  * Compute a hash value based on the contents of the label and the DOI.  The
2097  * hash algorithm is somewhat arbitrary, but is based on the observation that
2098  * humans will likely pick labels that differ by amounts that work out to be
2099  * multiples of the number of hash chains, and thus stirring in some primes
2100  * should help.
2101  */
2102 static uint_t
2103 hash_bylabel(void *hdata, mod_hash_key_t key)
2104 {
2105 	const ts_label_t *lab = (ts_label_t *)key;
2106 	const uint32_t *up, *ue;
2107 	uint_t hash;
2108 	int i;
2109 
2110 	_NOTE(ARGUNUSED(hdata));
2111 
2112 	hash = lab->tsl_doi + (lab->tsl_doi << 1);
2113 	/* we depend on alignment of label, but not representation */
2114 	up = (const uint32_t *)&lab->tsl_label;
2115 	ue = up + sizeof (lab->tsl_label) / sizeof (*up);
2116 	i = 1;
2117 	while (up < ue) {
2118 		/* using 2^n + 1, 1 <= n <= 16 as source of many primes */
2119 		hash += *up + (*up << ((i % 16) + 1));
2120 		up++;
2121 		i++;
2122 	}
2123 	return (hash);
2124 }
2125 
2126 /*
2127  * All that mod_hash cares about here is zero (equal) versus non-zero (not
2128  * equal).  This may need to be changed if less than / greater than is ever
2129  * needed.
2130  */
2131 static int
2132 hash_labelkey_cmp(mod_hash_key_t key1, mod_hash_key_t key2)
2133 {
2134 	ts_label_t *lab1 = (ts_label_t *)key1;
2135 	ts_label_t *lab2 = (ts_label_t *)key2;
2136 
2137 	return (label_equal(lab1, lab2) ? 0 : 1);
2138 }
2139 
2140 /*
2141  * Called by main() to initialize the zones framework.
2142  */
2143 void
2144 zone_init(void)
2145 {
2146 	rctl_dict_entry_t *rde;
2147 	rctl_val_t *dval;
2148 	rctl_set_t *set;
2149 	rctl_alloc_gp_t *gp;
2150 	rctl_entity_p_t e;
2151 	int res;
2152 
2153 	ASSERT(curproc == &p0);
2154 
2155 	/*
2156 	 * Create ID space for zone IDs.  ID 0 is reserved for the
2157 	 * global zone.
2158 	 */
2159 	zoneid_space = id_space_create("zoneid_space", 1, MAX_ZONEID);
2160 
2161 	/*
2162 	 * Initialize generic zone resource controls, if any.
2163 	 */
2164 	rc_zone_cpu_shares = rctl_register("zone.cpu-shares",
2165 	    RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_NEVER |
2166 	    RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT | RCTL_GLOBAL_SYSLOG_NEVER,
2167 	    FSS_MAXSHARES, FSS_MAXSHARES, &zone_cpu_shares_ops);
2168 
2169 	rc_zone_cpu_cap = rctl_register("zone.cpu-cap",
2170 	    RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_ALWAYS |
2171 	    RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |RCTL_GLOBAL_SYSLOG_NEVER |
2172 	    RCTL_GLOBAL_INFINITE,
2173 	    MAXCAP, MAXCAP, &zone_cpu_cap_ops);
2174 
2175 	rc_zone_nlwps = rctl_register("zone.max-lwps", RCENTITY_ZONE,
2176 	    RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
2177 	    INT_MAX, INT_MAX, &zone_lwps_ops);
2178 
2179 	rc_zone_nprocs = rctl_register("zone.max-processes", RCENTITY_ZONE,
2180 	    RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
2181 	    INT_MAX, INT_MAX, &zone_procs_ops);
2182 
2183 	/*
2184 	 * System V IPC resource controls
2185 	 */
2186 	rc_zone_msgmni = rctl_register("zone.max-msg-ids",
2187 	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2188 	    RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_msgmni_ops);
2189 
2190 	rc_zone_semmni = rctl_register("zone.max-sem-ids",
2191 	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2192 	    RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_semmni_ops);
2193 
2194 	rc_zone_shmmni = rctl_register("zone.max-shm-ids",
2195 	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2196 	    RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_shmmni_ops);
2197 
2198 	rc_zone_shmmax = rctl_register("zone.max-shm-memory",
2199 	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2200 	    RCTL_GLOBAL_BYTES, UINT64_MAX, UINT64_MAX, &zone_shmmax_ops);
2201 
2202 	/*
2203 	 * Create a rctl_val with PRIVILEGED, NOACTION, value = 1.  Then attach
2204 	 * this at the head of the rctl_dict_entry for ``zone.cpu-shares''.
2205 	 */
2206 	dval = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
2207 	bzero(dval, sizeof (rctl_val_t));
2208 	dval->rcv_value = 1;
2209 	dval->rcv_privilege = RCPRIV_PRIVILEGED;
2210 	dval->rcv_flagaction = RCTL_LOCAL_NOACTION;
2211 	dval->rcv_action_recip_pid = -1;
2212 
2213 	rde = rctl_dict_lookup("zone.cpu-shares");
2214 	(void) rctl_val_list_insert(&rde->rcd_default_value, dval);
2215 
2216 	rc_zone_locked_mem = rctl_register("zone.max-locked-memory",
2217 	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
2218 	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2219 	    &zone_locked_mem_ops);
2220 
2221 	rc_zone_max_swap = rctl_register("zone.max-swap",
2222 	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
2223 	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2224 	    &zone_max_swap_ops);
2225 
2226 	rc_zone_max_lofi = rctl_register("zone.max-lofi",
2227 	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |
2228 	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2229 	    &zone_max_lofi_ops);
2230 
2231 	/*
2232 	 * Initialize the ``global zone''.
2233 	 */
2234 	set = rctl_set_create();
2235 	gp = rctl_set_init_prealloc(RCENTITY_ZONE);
2236 	mutex_enter(&p0.p_lock);
2237 	e.rcep_p.zone = &zone0;
2238 	e.rcep_t = RCENTITY_ZONE;
2239 	zone0.zone_rctls = rctl_set_init(RCENTITY_ZONE, &p0, &e, set,
2240 	    gp);
2241 
2242 	zone0.zone_nlwps = p0.p_lwpcnt;
2243 	zone0.zone_nprocs = 1;
2244 	zone0.zone_ntasks = 1;
2245 	mutex_exit(&p0.p_lock);
2246 	zone0.zone_restart_init = B_TRUE;
2247 	zone0.zone_brand = &native_brand;
2248 	rctl_prealloc_destroy(gp);
2249 	/*
2250 	 * pool_default hasn't been initialized yet, so we let pool_init()
2251 	 * take care of making sure the global zone is in the default pool.
2252 	 */
2253 
2254 	/*
2255 	 * Initialize global zone kstats
2256 	 */
2257 	zone_kstat_create(&zone0);
2258 
2259 	/*
2260 	 * Initialize zone label.
2261 	 * mlp are initialized when tnzonecfg is loaded.
2262 	 */
2263 	zone0.zone_slabel = l_admin_low;
2264 	rw_init(&zone0.zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
2265 	label_hold(l_admin_low);
2266 
2267 	/*
2268 	 * Initialise the lock for the database structure used by mntfs.
2269 	 */
2270 	rw_init(&zone0.zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL);
2271 
2272 	mutex_enter(&zonehash_lock);
2273 	zone_uniqid(&zone0);
2274 	ASSERT(zone0.zone_uniqid == GLOBAL_ZONEUNIQID);
2275 
2276 	zonehashbyid = mod_hash_create_idhash("zone_by_id", zone_hash_size,
2277 	    mod_hash_null_valdtor);
2278 	zonehashbyname = mod_hash_create_strhash("zone_by_name",
2279 	    zone_hash_size, mod_hash_null_valdtor);
2280 	/*
2281 	 * maintain zonehashbylabel only for labeled systems
2282 	 */
2283 	if (is_system_labeled())
2284 		zonehashbylabel = mod_hash_create_extended("zone_by_label",
2285 		    zone_hash_size, mod_hash_null_keydtor,
2286 		    mod_hash_null_valdtor, hash_bylabel, NULL,
2287 		    hash_labelkey_cmp, KM_SLEEP);
2288 	zonecount = 1;
2289 
2290 	(void) mod_hash_insert(zonehashbyid, (mod_hash_key_t)GLOBAL_ZONEID,
2291 	    (mod_hash_val_t)&zone0);
2292 	(void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)zone0.zone_name,
2293 	    (mod_hash_val_t)&zone0);
2294 	if (is_system_labeled()) {
2295 		zone0.zone_flags |= ZF_HASHED_LABEL;
2296 		(void) mod_hash_insert(zonehashbylabel,
2297 		    (mod_hash_key_t)zone0.zone_slabel, (mod_hash_val_t)&zone0);
2298 	}
2299 	mutex_exit(&zonehash_lock);
2300 
2301 	/*
2302 	 * We avoid setting zone_kcred until now, since kcred is initialized
2303 	 * sometime after zone_zsd_init() and before zone_init().
2304 	 */
2305 	zone0.zone_kcred = kcred;
2306 	/*
2307 	 * The global zone is fully initialized (except for zone_rootvp which
2308 	 * will be set when the root filesystem is mounted).
2309 	 */
2310 	global_zone = &zone0;
2311 
2312 	/*
2313 	 * Setup an event channel to send zone status change notifications on
2314 	 */
2315 	res = sysevent_evc_bind(ZONE_EVENT_CHANNEL, &zone_event_chan,
2316 	    EVCH_CREAT);
2317 
2318 	if (res)
2319 		panic("Sysevent_evc_bind failed during zone setup.\n");
2320 
2321 }
2322 
2323 static void
2324 zone_free(zone_t *zone)
2325 {
2326 	ASSERT(zone != global_zone);
2327 	ASSERT(zone->zone_ntasks == 0);
2328 	ASSERT(zone->zone_nlwps == 0);
2329 	ASSERT(zone->zone_nprocs == 0);
2330 	ASSERT(zone->zone_cred_ref == 0);
2331 	ASSERT(zone->zone_kcred == NULL);
2332 	ASSERT(zone_status_get(zone) == ZONE_IS_DEAD ||
2333 	    zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
2334 	ASSERT(list_is_empty(&zone->zone_ref_list));
2335 
2336 	/*
2337 	 * Remove any zone caps.
2338 	 */
2339 	cpucaps_zone_remove(zone);
2340 
2341 	ASSERT(zone->zone_cpucap == NULL);
2342 
2343 	/* remove from deathrow list */
2344 	if (zone_status_get(zone) == ZONE_IS_DEAD) {
2345 		ASSERT(zone->zone_ref == 0);
2346 		mutex_enter(&zone_deathrow_lock);
2347 		list_remove(&zone_deathrow, zone);
2348 		mutex_exit(&zone_deathrow_lock);
2349 	}
2350 
2351 	list_destroy(&zone->zone_ref_list);
2352 	zone_free_zsd(zone);
2353 	zone_free_datasets(zone);
2354 	list_destroy(&zone->zone_dl_list);
2355 
2356 	if (zone->zone_rootvp != NULL)
2357 		VN_RELE(zone->zone_rootvp);
2358 	if (zone->zone_rootpath)
2359 		kmem_free(zone->zone_rootpath, zone->zone_rootpathlen);
2360 	if (zone->zone_name != NULL)
2361 		kmem_free(zone->zone_name, ZONENAME_MAX);
2362 	if (zone->zone_slabel != NULL)
2363 		label_rele(zone->zone_slabel);
2364 	if (zone->zone_nodename != NULL)
2365 		kmem_free(zone->zone_nodename, _SYS_NMLN);
2366 	if (zone->zone_domain != NULL)
2367 		kmem_free(zone->zone_domain, _SYS_NMLN);
2368 	if (zone->zone_privset != NULL)
2369 		kmem_free(zone->zone_privset, sizeof (priv_set_t));
2370 	if (zone->zone_rctls != NULL)
2371 		rctl_set_free(zone->zone_rctls);
2372 	if (zone->zone_bootargs != NULL)
2373 		strfree(zone->zone_bootargs);
2374 	if (zone->zone_initname != NULL)
2375 		strfree(zone->zone_initname);
2376 	if (zone->zone_fs_allowed != NULL)
2377 		strfree(zone->zone_fs_allowed);
2378 	if (zone->zone_pfexecd != NULL)
2379 		klpd_freelist(&zone->zone_pfexecd);
2380 	id_free(zoneid_space, zone->zone_id);
2381 	mutex_destroy(&zone->zone_lock);
2382 	cv_destroy(&zone->zone_cv);
2383 	rw_destroy(&zone->zone_mlps.mlpl_rwlock);
2384 	rw_destroy(&zone->zone_mntfs_db_lock);
2385 	kmem_free(zone, sizeof (zone_t));
2386 }
2387 
2388 /*
2389  * See block comment at the top of this file for information about zone
2390  * status values.
2391  */
2392 /*
2393  * Convenience function for setting zone status.
2394  */
2395 static void
2396 zone_status_set(zone_t *zone, zone_status_t status)
2397 {
2398 
2399 	nvlist_t *nvl = NULL;
2400 	ASSERT(MUTEX_HELD(&zone_status_lock));
2401 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE &&
2402 	    status >= zone_status_get(zone));
2403 
2404 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) ||
2405 	    nvlist_add_string(nvl, ZONE_CB_NAME, zone->zone_name) ||
2406 	    nvlist_add_string(nvl, ZONE_CB_NEWSTATE,
2407 	    zone_status_table[status]) ||
2408 	    nvlist_add_string(nvl, ZONE_CB_OLDSTATE,
2409 	    zone_status_table[zone->zone_status]) ||
2410 	    nvlist_add_int32(nvl, ZONE_CB_ZONEID, zone->zone_id) ||
2411 	    nvlist_add_uint64(nvl, ZONE_CB_TIMESTAMP, (uint64_t)gethrtime()) ||
2412 	    sysevent_evc_publish(zone_event_chan, ZONE_EVENT_STATUS_CLASS,
2413 	    ZONE_EVENT_STATUS_SUBCLASS, "sun.com", "kernel", nvl, EVCH_SLEEP)) {
2414 #ifdef DEBUG
2415 		(void) printf(
2416 		    "Failed to allocate and send zone state change event.\n");
2417 #endif
2418 	}
2419 	nvlist_free(nvl);
2420 
2421 	zone->zone_status = status;
2422 
2423 	cv_broadcast(&zone->zone_cv);
2424 }
2425 
2426 /*
2427  * Public function to retrieve the zone status.  The zone status may
2428  * change after it is retrieved.
2429  */
2430 zone_status_t
2431 zone_status_get(zone_t *zone)
2432 {
2433 	return (zone->zone_status);
2434 }
2435 
2436 static int
2437 zone_set_bootargs(zone_t *zone, const char *zone_bootargs)
2438 {
2439 	char *buf = kmem_zalloc(BOOTARGS_MAX, KM_SLEEP);
2440 	int err = 0;
2441 
2442 	ASSERT(zone != global_zone);
2443 	if ((err = copyinstr(zone_bootargs, buf, BOOTARGS_MAX, NULL)) != 0)
2444 		goto done;	/* EFAULT or ENAMETOOLONG */
2445 
2446 	if (zone->zone_bootargs != NULL)
2447 		strfree(zone->zone_bootargs);
2448 
2449 	zone->zone_bootargs = strdup(buf);
2450 
2451 done:
2452 	kmem_free(buf, BOOTARGS_MAX);
2453 	return (err);
2454 }
2455 
2456 static int
2457 zone_set_brand(zone_t *zone, const char *brand)
2458 {
2459 	struct brand_attr *attrp;
2460 	brand_t *bp;
2461 
2462 	attrp = kmem_alloc(sizeof (struct brand_attr), KM_SLEEP);
2463 	if (copyin(brand, attrp, sizeof (struct brand_attr)) != 0) {
2464 		kmem_free(attrp, sizeof (struct brand_attr));
2465 		return (EFAULT);
2466 	}
2467 
2468 	bp = brand_register_zone(attrp);
2469 	kmem_free(attrp, sizeof (struct brand_attr));
2470 	if (bp == NULL)
2471 		return (EINVAL);
2472 
2473 	/*
2474 	 * This is the only place where a zone can change it's brand.
2475 	 * We already need to hold zone_status_lock to check the zone
2476 	 * status, so we'll just use that lock to serialize zone
2477 	 * branding requests as well.
2478 	 */
2479 	mutex_enter(&zone_status_lock);
2480 
2481 	/* Re-Branding is not allowed and the zone can't be booted yet */
2482 	if ((ZONE_IS_BRANDED(zone)) ||
2483 	    (zone_status_get(zone) >= ZONE_IS_BOOTING)) {
2484 		mutex_exit(&zone_status_lock);
2485 		brand_unregister_zone(bp);
2486 		return (EINVAL);
2487 	}
2488 
2489 	/* set up the brand specific data */
2490 	zone->zone_brand = bp;
2491 	ZBROP(zone)->b_init_brand_data(zone);
2492 
2493 	mutex_exit(&zone_status_lock);
2494 	return (0);
2495 }
2496 
2497 static int
2498 zone_set_fs_allowed(zone_t *zone, const char *zone_fs_allowed)
2499 {
2500 	char *buf = kmem_zalloc(ZONE_FS_ALLOWED_MAX, KM_SLEEP);
2501 	int err = 0;
2502 
2503 	ASSERT(zone != global_zone);
2504 	if ((err = copyinstr(zone_fs_allowed, buf,
2505 	    ZONE_FS_ALLOWED_MAX, NULL)) != 0)
2506 		goto done;
2507 
2508 	if (zone->zone_fs_allowed != NULL)
2509 		strfree(zone->zone_fs_allowed);
2510 
2511 	zone->zone_fs_allowed = strdup(buf);
2512 
2513 done:
2514 	kmem_free(buf, ZONE_FS_ALLOWED_MAX);
2515 	return (err);
2516 }
2517 
2518 static int
2519 zone_set_initname(zone_t *zone, const char *zone_initname)
2520 {
2521 	char initname[INITNAME_SZ];
2522 	size_t len;
2523 	int err = 0;
2524 
2525 	ASSERT(zone != global_zone);
2526 	if ((err = copyinstr(zone_initname, initname, INITNAME_SZ, &len)) != 0)
2527 		return (err);	/* EFAULT or ENAMETOOLONG */
2528 
2529 	if (zone->zone_initname != NULL)
2530 		strfree(zone->zone_initname);
2531 
2532 	zone->zone_initname = kmem_alloc(strlen(initname) + 1, KM_SLEEP);
2533 	(void) strcpy(zone->zone_initname, initname);
2534 	return (0);
2535 }
2536 
2537 static int
2538 zone_set_phys_mcap(zone_t *zone, const uint64_t *zone_mcap)
2539 {
2540 	uint64_t mcap;
2541 	int err = 0;
2542 
2543 	if ((err = copyin(zone_mcap, &mcap, sizeof (uint64_t))) == 0)
2544 		zone->zone_phys_mcap = mcap;
2545 
2546 	return (err);
2547 }
2548 
2549 static int
2550 zone_set_sched_class(zone_t *zone, const char *new_class)
2551 {
2552 	char sched_class[PC_CLNMSZ];
2553 	id_t classid;
2554 	int err;
2555 
2556 	ASSERT(zone != global_zone);
2557 	if ((err = copyinstr(new_class, sched_class, PC_CLNMSZ, NULL)) != 0)
2558 		return (err);	/* EFAULT or ENAMETOOLONG */
2559 
2560 	if (getcid(sched_class, &classid) != 0 || CLASS_KERNEL(classid))
2561 		return (set_errno(EINVAL));
2562 	zone->zone_defaultcid = classid;
2563 	ASSERT(zone->zone_defaultcid > 0 &&
2564 	    zone->zone_defaultcid < loaded_classes);
2565 
2566 	return (0);
2567 }
2568 
2569 /*
2570  * Block indefinitely waiting for (zone_status >= status)
2571  */
2572 void
2573 zone_status_wait(zone_t *zone, zone_status_t status)
2574 {
2575 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2576 
2577 	mutex_enter(&zone_status_lock);
2578 	while (zone->zone_status < status) {
2579 		cv_wait(&zone->zone_cv, &zone_status_lock);
2580 	}
2581 	mutex_exit(&zone_status_lock);
2582 }
2583 
2584 /*
2585  * Private CPR-safe version of zone_status_wait().
2586  */
2587 static void
2588 zone_status_wait_cpr(zone_t *zone, zone_status_t status, char *str)
2589 {
2590 	callb_cpr_t cprinfo;
2591 
2592 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2593 
2594 	CALLB_CPR_INIT(&cprinfo, &zone_status_lock, callb_generic_cpr,
2595 	    str);
2596 	mutex_enter(&zone_status_lock);
2597 	while (zone->zone_status < status) {
2598 		CALLB_CPR_SAFE_BEGIN(&cprinfo);
2599 		cv_wait(&zone->zone_cv, &zone_status_lock);
2600 		CALLB_CPR_SAFE_END(&cprinfo, &zone_status_lock);
2601 	}
2602 	/*
2603 	 * zone_status_lock is implicitly released by the following.
2604 	 */
2605 	CALLB_CPR_EXIT(&cprinfo);
2606 }
2607 
2608 /*
2609  * Block until zone enters requested state or signal is received.  Return (0)
2610  * if signaled, non-zero otherwise.
2611  */
2612 int
2613 zone_status_wait_sig(zone_t *zone, zone_status_t status)
2614 {
2615 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2616 
2617 	mutex_enter(&zone_status_lock);
2618 	while (zone->zone_status < status) {
2619 		if (!cv_wait_sig(&zone->zone_cv, &zone_status_lock)) {
2620 			mutex_exit(&zone_status_lock);
2621 			return (0);
2622 		}
2623 	}
2624 	mutex_exit(&zone_status_lock);
2625 	return (1);
2626 }
2627 
2628 /*
2629  * Block until the zone enters the requested state or the timeout expires,
2630  * whichever happens first.  Return (-1) if operation timed out, time remaining
2631  * otherwise.
2632  */
2633 clock_t
2634 zone_status_timedwait(zone_t *zone, clock_t tim, zone_status_t status)
2635 {
2636 	clock_t timeleft = 0;
2637 
2638 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2639 
2640 	mutex_enter(&zone_status_lock);
2641 	while (zone->zone_status < status && timeleft != -1) {
2642 		timeleft = cv_timedwait(&zone->zone_cv, &zone_status_lock, tim);
2643 	}
2644 	mutex_exit(&zone_status_lock);
2645 	return (timeleft);
2646 }
2647 
2648 /*
2649  * Block until the zone enters the requested state, the current process is
2650  * signaled,  or the timeout expires, whichever happens first.  Return (-1) if
2651  * operation timed out, 0 if signaled, time remaining otherwise.
2652  */
2653 clock_t
2654 zone_status_timedwait_sig(zone_t *zone, clock_t tim, zone_status_t status)
2655 {
2656 	clock_t timeleft = tim - ddi_get_lbolt();
2657 
2658 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2659 
2660 	mutex_enter(&zone_status_lock);
2661 	while (zone->zone_status < status) {
2662 		timeleft = cv_timedwait_sig(&zone->zone_cv, &zone_status_lock,
2663 		    tim);
2664 		if (timeleft <= 0)
2665 			break;
2666 	}
2667 	mutex_exit(&zone_status_lock);
2668 	return (timeleft);
2669 }
2670 
2671 /*
2672  * Zones have two reference counts: one for references from credential
2673  * structures (zone_cred_ref), and one (zone_ref) for everything else.
2674  * This is so we can allow a zone to be rebooted while there are still
2675  * outstanding cred references, since certain drivers cache dblks (which
2676  * implicitly results in cached creds).  We wait for zone_ref to drop to
2677  * 0 (actually 1), but not zone_cred_ref.  The zone structure itself is
2678  * later freed when the zone_cred_ref drops to 0, though nothing other
2679  * than the zone id and privilege set should be accessed once the zone
2680  * is "dead".
2681  *
2682  * A debugging flag, zone_wait_for_cred, can be set to a non-zero value
2683  * to force halt/reboot to block waiting for the zone_cred_ref to drop
2684  * to 0.  This can be useful to flush out other sources of cached creds
2685  * that may be less innocuous than the driver case.
2686  *
2687  * Zones also provide a tracked reference counting mechanism in which zone
2688  * references are represented by "crumbs" (zone_ref structures).  Crumbs help
2689  * debuggers determine the sources of leaked zone references.  See
2690  * zone_hold_ref() and zone_rele_ref() below for more information.
2691  */
2692 
2693 int zone_wait_for_cred = 0;
2694 
2695 static void
2696 zone_hold_locked(zone_t *z)
2697 {
2698 	ASSERT(MUTEX_HELD(&z->zone_lock));
2699 	z->zone_ref++;
2700 	ASSERT(z->zone_ref != 0);
2701 }
2702 
2703 /*
2704  * Increment the specified zone's reference count.  The zone's zone_t structure
2705  * will not be freed as long as the zone's reference count is nonzero.
2706  * Decrement the zone's reference count via zone_rele().
2707  *
2708  * NOTE: This function should only be used to hold zones for short periods of
2709  * time.  Use zone_hold_ref() if the zone must be held for a long time.
2710  */
2711 void
2712 zone_hold(zone_t *z)
2713 {
2714 	mutex_enter(&z->zone_lock);
2715 	zone_hold_locked(z);
2716 	mutex_exit(&z->zone_lock);
2717 }
2718 
2719 /*
2720  * If the non-cred ref count drops to 1 and either the cred ref count
2721  * is 0 or we aren't waiting for cred references, the zone is ready to
2722  * be destroyed.
2723  */
2724 #define	ZONE_IS_UNREF(zone)	((zone)->zone_ref == 1 && \
2725 	    (!zone_wait_for_cred || (zone)->zone_cred_ref == 0))
2726 
2727 /*
2728  * Common zone reference release function invoked by zone_rele() and
2729  * zone_rele_ref().  If subsys is ZONE_REF_NUM_SUBSYS, then the specified
2730  * zone's subsystem-specific reference counters are not affected by the
2731  * release.  If ref is not NULL, then the zone_ref_t to which it refers is
2732  * removed from the specified zone's reference list.  ref must be non-NULL iff
2733  * subsys is not ZONE_REF_NUM_SUBSYS.
2734  */
2735 static void
2736 zone_rele_common(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
2737 {
2738 	boolean_t wakeup;
2739 
2740 	mutex_enter(&z->zone_lock);
2741 	ASSERT(z->zone_ref != 0);
2742 	z->zone_ref--;
2743 	if (subsys != ZONE_REF_NUM_SUBSYS) {
2744 		ASSERT(z->zone_subsys_ref[subsys] != 0);
2745 		z->zone_subsys_ref[subsys]--;
2746 		list_remove(&z->zone_ref_list, ref);
2747 	}
2748 	if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
2749 		/* no more refs, free the structure */
2750 		mutex_exit(&z->zone_lock);
2751 		zone_free(z);
2752 		return;
2753 	}
2754 	/* signal zone_destroy so the zone can finish halting */
2755 	wakeup = (ZONE_IS_UNREF(z) && zone_status_get(z) >= ZONE_IS_DEAD);
2756 	mutex_exit(&z->zone_lock);
2757 
2758 	if (wakeup) {
2759 		/*
2760 		 * Grabbing zonehash_lock here effectively synchronizes with
2761 		 * zone_destroy() to avoid missed signals.
2762 		 */
2763 		mutex_enter(&zonehash_lock);
2764 		cv_broadcast(&zone_destroy_cv);
2765 		mutex_exit(&zonehash_lock);
2766 	}
2767 }
2768 
2769 /*
2770  * Decrement the specified zone's reference count.  The specified zone will
2771  * cease to exist after this function returns if the reference count drops to
2772  * zero.  This function should be paired with zone_hold().
2773  */
2774 void
2775 zone_rele(zone_t *z)
2776 {
2777 	zone_rele_common(z, NULL, ZONE_REF_NUM_SUBSYS);
2778 }
2779 
2780 /*
2781  * Initialize a zone reference structure.  This function must be invoked for
2782  * a reference structure before the structure is passed to zone_hold_ref().
2783  */
2784 void
2785 zone_init_ref(zone_ref_t *ref)
2786 {
2787 	ref->zref_zone = NULL;
2788 	list_link_init(&ref->zref_linkage);
2789 }
2790 
2791 /*
2792  * Acquire a reference to zone z.  The caller must specify the
2793  * zone_ref_subsys_t constant associated with its subsystem.  The specified
2794  * zone_ref_t structure will represent a reference to the specified zone.  Use
2795  * zone_rele_ref() to release the reference.
2796  *
2797  * The referenced zone_t structure will not be freed as long as the zone_t's
2798  * zone_status field is not ZONE_IS_DEAD and the zone has outstanding
2799  * references.
2800  *
2801  * NOTE: The zone_ref_t structure must be initialized before it is used.
2802  * See zone_init_ref() above.
2803  */
2804 void
2805 zone_hold_ref(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
2806 {
2807 	ASSERT(subsys >= 0 && subsys < ZONE_REF_NUM_SUBSYS);
2808 
2809 	/*
2810 	 * Prevent consumers from reusing a reference structure before
2811 	 * releasing it.
2812 	 */
2813 	VERIFY(ref->zref_zone == NULL);
2814 
2815 	ref->zref_zone = z;
2816 	mutex_enter(&z->zone_lock);
2817 	zone_hold_locked(z);
2818 	z->zone_subsys_ref[subsys]++;
2819 	ASSERT(z->zone_subsys_ref[subsys] != 0);
2820 	list_insert_head(&z->zone_ref_list, ref);
2821 	mutex_exit(&z->zone_lock);
2822 }
2823 
2824 /*
2825  * Release the zone reference represented by the specified zone_ref_t.
2826  * The reference is invalid after it's released; however, the zone_ref_t
2827  * structure can be reused without having to invoke zone_init_ref().
2828  * subsys should be the same value that was passed to zone_hold_ref()
2829  * when the reference was acquired.
2830  */
2831 void
2832 zone_rele_ref(zone_ref_t *ref, zone_ref_subsys_t subsys)
2833 {
2834 	zone_rele_common(ref->zref_zone, ref, subsys);
2835 
2836 	/*
2837 	 * Set the zone_ref_t's zref_zone field to NULL to generate panics
2838 	 * when consumers dereference the reference.  This helps us catch
2839 	 * consumers who use released references.  Furthermore, this lets
2840 	 * consumers reuse the zone_ref_t structure without having to
2841 	 * invoke zone_init_ref().
2842 	 */
2843 	ref->zref_zone = NULL;
2844 }
2845 
2846 void
2847 zone_cred_hold(zone_t *z)
2848 {
2849 	mutex_enter(&z->zone_lock);
2850 	z->zone_cred_ref++;
2851 	ASSERT(z->zone_cred_ref != 0);
2852 	mutex_exit(&z->zone_lock);
2853 }
2854 
2855 void
2856 zone_cred_rele(zone_t *z)
2857 {
2858 	boolean_t wakeup;
2859 
2860 	mutex_enter(&z->zone_lock);
2861 	ASSERT(z->zone_cred_ref != 0);
2862 	z->zone_cred_ref--;
2863 	if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
2864 		/* no more refs, free the structure */
2865 		mutex_exit(&z->zone_lock);
2866 		zone_free(z);
2867 		return;
2868 	}
2869 	/*
2870 	 * If zone_destroy is waiting for the cred references to drain
2871 	 * out, and they have, signal it.
2872 	 */
2873 	wakeup = (zone_wait_for_cred && ZONE_IS_UNREF(z) &&
2874 	    zone_status_get(z) >= ZONE_IS_DEAD);
2875 	mutex_exit(&z->zone_lock);
2876 
2877 	if (wakeup) {
2878 		/*
2879 		 * Grabbing zonehash_lock here effectively synchronizes with
2880 		 * zone_destroy() to avoid missed signals.
2881 		 */
2882 		mutex_enter(&zonehash_lock);
2883 		cv_broadcast(&zone_destroy_cv);
2884 		mutex_exit(&zonehash_lock);
2885 	}
2886 }
2887 
2888 void
2889 zone_task_hold(zone_t *z)
2890 {
2891 	mutex_enter(&z->zone_lock);
2892 	z->zone_ntasks++;
2893 	ASSERT(z->zone_ntasks != 0);
2894 	mutex_exit(&z->zone_lock);
2895 }
2896 
2897 void
2898 zone_task_rele(zone_t *zone)
2899 {
2900 	uint_t refcnt;
2901 
2902 	mutex_enter(&zone->zone_lock);
2903 	ASSERT(zone->zone_ntasks != 0);
2904 	refcnt = --zone->zone_ntasks;
2905 	if (refcnt > 1)	{	/* Common case */
2906 		mutex_exit(&zone->zone_lock);
2907 		return;
2908 	}
2909 	zone_hold_locked(zone);	/* so we can use the zone_t later */
2910 	mutex_exit(&zone->zone_lock);
2911 	if (refcnt == 1) {
2912 		/*
2913 		 * See if the zone is shutting down.
2914 		 */
2915 		mutex_enter(&zone_status_lock);
2916 		if (zone_status_get(zone) != ZONE_IS_SHUTTING_DOWN) {
2917 			goto out;
2918 		}
2919 
2920 		/*
2921 		 * Make sure the ntasks didn't change since we
2922 		 * dropped zone_lock.
2923 		 */
2924 		mutex_enter(&zone->zone_lock);
2925 		if (refcnt != zone->zone_ntasks) {
2926 			mutex_exit(&zone->zone_lock);
2927 			goto out;
2928 		}
2929 		mutex_exit(&zone->zone_lock);
2930 
2931 		/*
2932 		 * No more user processes in the zone.  The zone is empty.
2933 		 */
2934 		zone_status_set(zone, ZONE_IS_EMPTY);
2935 		goto out;
2936 	}
2937 
2938 	ASSERT(refcnt == 0);
2939 	/*
2940 	 * zsched has exited; the zone is dead.
2941 	 */
2942 	zone->zone_zsched = NULL;		/* paranoia */
2943 	mutex_enter(&zone_status_lock);
2944 	zone_status_set(zone, ZONE_IS_DEAD);
2945 out:
2946 	mutex_exit(&zone_status_lock);
2947 	zone_rele(zone);
2948 }
2949 
2950 zoneid_t
2951 getzoneid(void)
2952 {
2953 	return (curproc->p_zone->zone_id);
2954 }
2955 
2956 /*
2957  * Internal versions of zone_find_by_*().  These don't zone_hold() or
2958  * check the validity of a zone's state.
2959  */
2960 static zone_t *
2961 zone_find_all_by_id(zoneid_t zoneid)
2962 {
2963 	mod_hash_val_t hv;
2964 	zone_t *zone = NULL;
2965 
2966 	ASSERT(MUTEX_HELD(&zonehash_lock));
2967 
2968 	if (mod_hash_find(zonehashbyid,
2969 	    (mod_hash_key_t)(uintptr_t)zoneid, &hv) == 0)
2970 		zone = (zone_t *)hv;
2971 	return (zone);
2972 }
2973 
2974 static zone_t *
2975 zone_find_all_by_label(const ts_label_t *label)
2976 {
2977 	mod_hash_val_t hv;
2978 	zone_t *zone = NULL;
2979 
2980 	ASSERT(MUTEX_HELD(&zonehash_lock));
2981 
2982 	/*
2983 	 * zonehashbylabel is not maintained for unlabeled systems
2984 	 */
2985 	if (!is_system_labeled())
2986 		return (NULL);
2987 	if (mod_hash_find(zonehashbylabel, (mod_hash_key_t)label, &hv) == 0)
2988 		zone = (zone_t *)hv;
2989 	return (zone);
2990 }
2991 
2992 static zone_t *
2993 zone_find_all_by_name(char *name)
2994 {
2995 	mod_hash_val_t hv;
2996 	zone_t *zone = NULL;
2997 
2998 	ASSERT(MUTEX_HELD(&zonehash_lock));
2999 
3000 	if (mod_hash_find(zonehashbyname, (mod_hash_key_t)name, &hv) == 0)
3001 		zone = (zone_t *)hv;
3002 	return (zone);
3003 }
3004 
3005 /*
3006  * Public interface for looking up a zone by zoneid.  Only returns the zone if
3007  * it is fully initialized, and has not yet begun the zone_destroy() sequence.
3008  * Caller must call zone_rele() once it is done with the zone.
3009  *
3010  * The zone may begin the zone_destroy() sequence immediately after this
3011  * function returns, but may be safely used until zone_rele() is called.
3012  */
3013 zone_t *
3014 zone_find_by_id(zoneid_t zoneid)
3015 {
3016 	zone_t *zone;
3017 	zone_status_t status;
3018 
3019 	mutex_enter(&zonehash_lock);
3020 	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
3021 		mutex_exit(&zonehash_lock);
3022 		return (NULL);
3023 	}
3024 	status = zone_status_get(zone);
3025 	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3026 		/*
3027 		 * For all practical purposes the zone doesn't exist.
3028 		 */
3029 		mutex_exit(&zonehash_lock);
3030 		return (NULL);
3031 	}
3032 	zone_hold(zone);
3033 	mutex_exit(&zonehash_lock);
3034 	return (zone);
3035 }
3036 
3037 /*
3038  * Similar to zone_find_by_id, but using zone label as the key.
3039  */
3040 zone_t *
3041 zone_find_by_label(const ts_label_t *label)
3042 {
3043 	zone_t *zone;
3044 	zone_status_t status;
3045 
3046 	mutex_enter(&zonehash_lock);
3047 	if ((zone = zone_find_all_by_label(label)) == NULL) {
3048 		mutex_exit(&zonehash_lock);
3049 		return (NULL);
3050 	}
3051 
3052 	status = zone_status_get(zone);
3053 	if (status > ZONE_IS_DOWN) {
3054 		/*
3055 		 * For all practical purposes the zone doesn't exist.
3056 		 */
3057 		mutex_exit(&zonehash_lock);
3058 		return (NULL);
3059 	}
3060 	zone_hold(zone);
3061 	mutex_exit(&zonehash_lock);
3062 	return (zone);
3063 }
3064 
3065 /*
3066  * Similar to zone_find_by_id, but using zone name as the key.
3067  */
3068 zone_t *
3069 zone_find_by_name(char *name)
3070 {
3071 	zone_t *zone;
3072 	zone_status_t status;
3073 
3074 	mutex_enter(&zonehash_lock);
3075 	if ((zone = zone_find_all_by_name(name)) == NULL) {
3076 		mutex_exit(&zonehash_lock);
3077 		return (NULL);
3078 	}
3079 	status = zone_status_get(zone);
3080 	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3081 		/*
3082 		 * For all practical purposes the zone doesn't exist.
3083 		 */
3084 		mutex_exit(&zonehash_lock);
3085 		return (NULL);
3086 	}
3087 	zone_hold(zone);
3088 	mutex_exit(&zonehash_lock);
3089 	return (zone);
3090 }
3091 
3092 /*
3093  * Similar to zone_find_by_id(), using the path as a key.  For instance,
3094  * if there is a zone "foo" rooted at /foo/root, and the path argument
3095  * is "/foo/root/proc", it will return the held zone_t corresponding to
3096  * zone "foo".
3097  *
3098  * zone_find_by_path() always returns a non-NULL value, since at the
3099  * very least every path will be contained in the global zone.
3100  *
3101  * As with the other zone_find_by_*() functions, the caller is
3102  * responsible for zone_rele()ing the return value of this function.
3103  */
3104 zone_t *
3105 zone_find_by_path(const char *path)
3106 {
3107 	zone_t *zone;
3108 	zone_t *zret = NULL;
3109 	zone_status_t status;
3110 
3111 	if (path == NULL) {
3112 		/*
3113 		 * Call from rootconf().
3114 		 */
3115 		zone_hold(global_zone);
3116 		return (global_zone);
3117 	}
3118 	ASSERT(*path == '/');
3119 	mutex_enter(&zonehash_lock);
3120 	for (zone = list_head(&zone_active); zone != NULL;
3121 	    zone = list_next(&zone_active, zone)) {
3122 		if (ZONE_PATH_VISIBLE(path, zone))
3123 			zret = zone;
3124 	}
3125 	ASSERT(zret != NULL);
3126 	status = zone_status_get(zret);
3127 	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3128 		/*
3129 		 * Zone practically doesn't exist.
3130 		 */
3131 		zret = global_zone;
3132 	}
3133 	zone_hold(zret);
3134 	mutex_exit(&zonehash_lock);
3135 	return (zret);
3136 }
3137 
3138 /*
3139  * Public interface for updating per-zone load averages.  Called once per
3140  * second.
3141  *
3142  * Based on loadavg_update(), genloadavg() and calcloadavg() from clock.c.
3143  */
3144 void
3145 zone_loadavg_update()
3146 {
3147 	zone_t *zp;
3148 	zone_status_t status;
3149 	struct loadavg_s *lavg;
3150 	hrtime_t zone_total;
3151 	int i;
3152 	hrtime_t hr_avg;
3153 	int nrun;
3154 	static int64_t f[3] = { 135, 27, 9 };
3155 	int64_t q, r;
3156 
3157 	mutex_enter(&zonehash_lock);
3158 	for (zp = list_head(&zone_active); zp != NULL;
3159 	    zp = list_next(&zone_active, zp)) {
3160 		mutex_enter(&zp->zone_lock);
3161 
3162 		/* Skip zones that are on the way down or not yet up */
3163 		status = zone_status_get(zp);
3164 		if (status < ZONE_IS_READY || status >= ZONE_IS_DOWN) {
3165 			/* For all practical purposes the zone doesn't exist. */
3166 			mutex_exit(&zp->zone_lock);
3167 			continue;
3168 		}
3169 
3170 		/*
3171 		 * Update the 10 second moving average data in zone_loadavg.
3172 		 */
3173 		lavg = &zp->zone_loadavg;
3174 
3175 		zone_total = zp->zone_utime + zp->zone_stime + zp->zone_wtime;
3176 		scalehrtime(&zone_total);
3177 
3178 		/* The zone_total should always be increasing. */
3179 		lavg->lg_loads[lavg->lg_cur] = (zone_total > lavg->lg_total) ?
3180 		    zone_total - lavg->lg_total : 0;
3181 		lavg->lg_cur = (lavg->lg_cur + 1) % S_LOADAVG_SZ;
3182 		/* lg_total holds the prev. 1 sec. total */
3183 		lavg->lg_total = zone_total;
3184 
3185 		/*
3186 		 * To simplify the calculation, we don't calculate the load avg.
3187 		 * until the zone has been up for at least 10 seconds and our
3188 		 * moving average is thus full.
3189 		 */
3190 		if ((lavg->lg_len + 1) < S_LOADAVG_SZ) {
3191 			lavg->lg_len++;
3192 			mutex_exit(&zp->zone_lock);
3193 			continue;
3194 		}
3195 
3196 		/* Now calculate the 1min, 5min, 15 min load avg. */
3197 		hr_avg = 0;
3198 		for (i = 0; i < S_LOADAVG_SZ; i++)
3199 			hr_avg += lavg->lg_loads[i];
3200 		hr_avg = hr_avg / S_LOADAVG_SZ;
3201 		nrun = hr_avg / (NANOSEC / LGRP_LOADAVG_IN_THREAD_MAX);
3202 
3203 		/* Compute load avg. See comment in calcloadavg() */
3204 		for (i = 0; i < 3; i++) {
3205 			q = (zp->zone_hp_avenrun[i] >> 16) << 7;
3206 			r = (zp->zone_hp_avenrun[i] & 0xffff) << 7;
3207 			zp->zone_hp_avenrun[i] +=
3208 			    ((nrun - q) * f[i] - ((r * f[i]) >> 16)) >> 4;
3209 
3210 			/* avenrun[] can only hold 31 bits of load avg. */
3211 			if (zp->zone_hp_avenrun[i] <
3212 			    ((uint64_t)1<<(31+16-FSHIFT)))
3213 				zp->zone_avenrun[i] = (int32_t)
3214 				    (zp->zone_hp_avenrun[i] >> (16 - FSHIFT));
3215 			else
3216 				zp->zone_avenrun[i] = 0x7fffffff;
3217 		}
3218 
3219 		mutex_exit(&zp->zone_lock);
3220 	}
3221 	mutex_exit(&zonehash_lock);
3222 }
3223 
3224 /*
3225  * Get the number of cpus visible to this zone.  The system-wide global
3226  * 'ncpus' is returned if pools are disabled, the caller is in the
3227  * global zone, or a NULL zone argument is passed in.
3228  */
3229 int
3230 zone_ncpus_get(zone_t *zone)
3231 {
3232 	int myncpus = zone == NULL ? 0 : zone->zone_ncpus;
3233 
3234 	return (myncpus != 0 ? myncpus : ncpus);
3235 }
3236 
3237 /*
3238  * Get the number of online cpus visible to this zone.  The system-wide
3239  * global 'ncpus_online' is returned if pools are disabled, the caller
3240  * is in the global zone, or a NULL zone argument is passed in.
3241  */
3242 int
3243 zone_ncpus_online_get(zone_t *zone)
3244 {
3245 	int myncpus_online = zone == NULL ? 0 : zone->zone_ncpus_online;
3246 
3247 	return (myncpus_online != 0 ? myncpus_online : ncpus_online);
3248 }
3249 
3250 /*
3251  * Return the pool to which the zone is currently bound.
3252  */
3253 pool_t *
3254 zone_pool_get(zone_t *zone)
3255 {
3256 	ASSERT(pool_lock_held());
3257 
3258 	return (zone->zone_pool);
3259 }
3260 
3261 /*
3262  * Set the zone's pool pointer and update the zone's visibility to match
3263  * the resources in the new pool.
3264  */
3265 void
3266 zone_pool_set(zone_t *zone, pool_t *pool)
3267 {
3268 	ASSERT(pool_lock_held());
3269 	ASSERT(MUTEX_HELD(&cpu_lock));
3270 
3271 	zone->zone_pool = pool;
3272 	zone_pset_set(zone, pool->pool_pset->pset_id);
3273 }
3274 
3275 /*
3276  * Return the cached value of the id of the processor set to which the
3277  * zone is currently bound.  The value will be ZONE_PS_INVAL if the pools
3278  * facility is disabled.
3279  */
3280 psetid_t
3281 zone_pset_get(zone_t *zone)
3282 {
3283 	ASSERT(MUTEX_HELD(&cpu_lock));
3284 
3285 	return (zone->zone_psetid);
3286 }
3287 
3288 /*
3289  * Set the cached value of the id of the processor set to which the zone
3290  * is currently bound.  Also update the zone's visibility to match the
3291  * resources in the new processor set.
3292  */
3293 void
3294 zone_pset_set(zone_t *zone, psetid_t newpsetid)
3295 {
3296 	psetid_t oldpsetid;
3297 
3298 	ASSERT(MUTEX_HELD(&cpu_lock));
3299 	oldpsetid = zone_pset_get(zone);
3300 
3301 	if (oldpsetid == newpsetid)
3302 		return;
3303 	/*
3304 	 * Global zone sees all.
3305 	 */
3306 	if (zone != global_zone) {
3307 		zone->zone_psetid = newpsetid;
3308 		if (newpsetid != ZONE_PS_INVAL)
3309 			pool_pset_visibility_add(newpsetid, zone);
3310 		if (oldpsetid != ZONE_PS_INVAL)
3311 			pool_pset_visibility_remove(oldpsetid, zone);
3312 	}
3313 	/*
3314 	 * Disabling pools, so we should start using the global values
3315 	 * for ncpus and ncpus_online.
3316 	 */
3317 	if (newpsetid == ZONE_PS_INVAL) {
3318 		zone->zone_ncpus = 0;
3319 		zone->zone_ncpus_online = 0;
3320 	}
3321 }
3322 
3323 /*
3324  * Walk the list of active zones and issue the provided callback for
3325  * each of them.
3326  *
3327  * Caller must not be holding any locks that may be acquired under
3328  * zonehash_lock.  See comment at the beginning of the file for a list of
3329  * common locks and their interactions with zones.
3330  */
3331 int
3332 zone_walk(int (*cb)(zone_t *, void *), void *data)
3333 {
3334 	zone_t *zone;
3335 	int ret = 0;
3336 	zone_status_t status;
3337 
3338 	mutex_enter(&zonehash_lock);
3339 	for (zone = list_head(&zone_active); zone != NULL;
3340 	    zone = list_next(&zone_active, zone)) {
3341 		/*
3342 		 * Skip zones that shouldn't be externally visible.
3343 		 */
3344 		status = zone_status_get(zone);
3345 		if (status < ZONE_IS_READY || status > ZONE_IS_DOWN)
3346 			continue;
3347 		/*
3348 		 * Bail immediately if any callback invocation returns a
3349 		 * non-zero value.
3350 		 */
3351 		ret = (*cb)(zone, data);
3352 		if (ret != 0)
3353 			break;
3354 	}
3355 	mutex_exit(&zonehash_lock);
3356 	return (ret);
3357 }
3358 
3359 static int
3360 zone_set_root(zone_t *zone, const char *upath)
3361 {
3362 	vnode_t *vp;
3363 	int trycount;
3364 	int error = 0;
3365 	char *path;
3366 	struct pathname upn, pn;
3367 	size_t pathlen;
3368 
3369 	if ((error = pn_get((char *)upath, UIO_USERSPACE, &upn)) != 0)
3370 		return (error);
3371 
3372 	pn_alloc(&pn);
3373 
3374 	/* prevent infinite loop */
3375 	trycount = 10;
3376 	for (;;) {
3377 		if (--trycount <= 0) {
3378 			error = ESTALE;
3379 			goto out;
3380 		}
3381 
3382 		if ((error = lookuppn(&upn, &pn, FOLLOW, NULLVPP, &vp)) == 0) {
3383 			/*
3384 			 * VOP_ACCESS() may cover 'vp' with a new
3385 			 * filesystem, if 'vp' is an autoFS vnode.
3386 			 * Get the new 'vp' if so.
3387 			 */
3388 			if ((error =
3389 			    VOP_ACCESS(vp, VEXEC, 0, CRED(), NULL)) == 0 &&
3390 			    (!vn_ismntpt(vp) ||
3391 			    (error = traverse(&vp)) == 0)) {
3392 				pathlen = pn.pn_pathlen + 2;
3393 				path = kmem_alloc(pathlen, KM_SLEEP);
3394 				(void) strncpy(path, pn.pn_path,
3395 				    pn.pn_pathlen + 1);
3396 				path[pathlen - 2] = '/';
3397 				path[pathlen - 1] = '\0';
3398 				pn_free(&pn);
3399 				pn_free(&upn);
3400 
3401 				/* Success! */
3402 				break;
3403 			}
3404 			VN_RELE(vp);
3405 		}
3406 		if (error != ESTALE)
3407 			goto out;
3408 	}
3409 
3410 	ASSERT(error == 0);
3411 	zone->zone_rootvp = vp;		/* we hold a reference to vp */
3412 	zone->zone_rootpath = path;
3413 	zone->zone_rootpathlen = pathlen;
3414 	if (pathlen > 5 && strcmp(path + pathlen - 5, "/lu/") == 0)
3415 		zone->zone_flags |= ZF_IS_SCRATCH;
3416 	return (0);
3417 
3418 out:
3419 	pn_free(&pn);
3420 	pn_free(&upn);
3421 	return (error);
3422 }
3423 
3424 #define	isalnum(c)	(((c) >= '0' && (c) <= '9') || \
3425 			((c) >= 'a' && (c) <= 'z') || \
3426 			((c) >= 'A' && (c) <= 'Z'))
3427 
3428 static int
3429 zone_set_name(zone_t *zone, const char *uname)
3430 {
3431 	char *kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
3432 	size_t len;
3433 	int i, err;
3434 
3435 	if ((err = copyinstr(uname, kname, ZONENAME_MAX, &len)) != 0) {
3436 		kmem_free(kname, ZONENAME_MAX);
3437 		return (err);	/* EFAULT or ENAMETOOLONG */
3438 	}
3439 
3440 	/* must be less than ZONENAME_MAX */
3441 	if (len == ZONENAME_MAX && kname[ZONENAME_MAX - 1] != '\0') {
3442 		kmem_free(kname, ZONENAME_MAX);
3443 		return (EINVAL);
3444 	}
3445 
3446 	/*
3447 	 * Name must start with an alphanumeric and must contain only
3448 	 * alphanumerics, '-', '_' and '.'.
3449 	 */
3450 	if (!isalnum(kname[0])) {
3451 		kmem_free(kname, ZONENAME_MAX);
3452 		return (EINVAL);
3453 	}
3454 	for (i = 1; i < len - 1; i++) {
3455 		if (!isalnum(kname[i]) && kname[i] != '-' && kname[i] != '_' &&
3456 		    kname[i] != '.') {
3457 			kmem_free(kname, ZONENAME_MAX);
3458 			return (EINVAL);
3459 		}
3460 	}
3461 
3462 	zone->zone_name = kname;
3463 	return (0);
3464 }
3465 
3466 /*
3467  * Gets the 32-bit hostid of the specified zone as an unsigned int.  If 'zonep'
3468  * is NULL or it points to a zone with no hostid emulation, then the machine's
3469  * hostid (i.e., the global zone's hostid) is returned.  This function returns
3470  * zero if neither the zone nor the host machine (global zone) have hostids.  It
3471  * returns HW_INVALID_HOSTID if the function attempts to return the machine's
3472  * hostid and the machine's hostid is invalid.
3473  */
3474 uint32_t
3475 zone_get_hostid(zone_t *zonep)
3476 {
3477 	unsigned long machine_hostid;
3478 
3479 	if (zonep == NULL || zonep->zone_hostid == HW_INVALID_HOSTID) {
3480 		if (ddi_strtoul(hw_serial, NULL, 10, &machine_hostid) != 0)
3481 			return (HW_INVALID_HOSTID);
3482 		return ((uint32_t)machine_hostid);
3483 	}
3484 	return (zonep->zone_hostid);
3485 }
3486 
3487 /*
3488  * Similar to thread_create(), but makes sure the thread is in the appropriate
3489  * zone's zsched process (curproc->p_zone->zone_zsched) before returning.
3490  */
3491 /*ARGSUSED*/
3492 kthread_t *
3493 zthread_create(
3494     caddr_t stk,
3495     size_t stksize,
3496     void (*proc)(),
3497     void *arg,
3498     size_t len,
3499     pri_t pri)
3500 {
3501 	kthread_t *t;
3502 	zone_t *zone = curproc->p_zone;
3503 	proc_t *pp = zone->zone_zsched;
3504 
3505 	zone_hold(zone);	/* Reference to be dropped when thread exits */
3506 
3507 	/*
3508 	 * No-one should be trying to create threads if the zone is shutting
3509 	 * down and there aren't any kernel threads around.  See comment
3510 	 * in zthread_exit().
3511 	 */
3512 	ASSERT(!(zone->zone_kthreads == NULL &&
3513 	    zone_status_get(zone) >= ZONE_IS_EMPTY));
3514 	/*
3515 	 * Create a thread, but don't let it run until we've finished setting
3516 	 * things up.
3517 	 */
3518 	t = thread_create(stk, stksize, proc, arg, len, pp, TS_STOPPED, pri);
3519 	ASSERT(t->t_forw == NULL);
3520 	mutex_enter(&zone_status_lock);
3521 	if (zone->zone_kthreads == NULL) {
3522 		t->t_forw = t->t_back = t;
3523 	} else {
3524 		kthread_t *tx = zone->zone_kthreads;
3525 
3526 		t->t_forw = tx;
3527 		t->t_back = tx->t_back;
3528 		tx->t_back->t_forw = t;
3529 		tx->t_back = t;
3530 	}
3531 	zone->zone_kthreads = t;
3532 	mutex_exit(&zone_status_lock);
3533 
3534 	mutex_enter(&pp->p_lock);
3535 	t->t_proc_flag |= TP_ZTHREAD;
3536 	project_rele(t->t_proj);
3537 	t->t_proj = project_hold(pp->p_task->tk_proj);
3538 
3539 	/*
3540 	 * Setup complete, let it run.
3541 	 */
3542 	thread_lock(t);
3543 	t->t_schedflag |= TS_ALLSTART;
3544 	setrun_locked(t);
3545 	thread_unlock(t);
3546 
3547 	mutex_exit(&pp->p_lock);
3548 
3549 	return (t);
3550 }
3551 
3552 /*
3553  * Similar to thread_exit().  Must be called by threads created via
3554  * zthread_exit().
3555  */
3556 void
3557 zthread_exit(void)
3558 {
3559 	kthread_t *t = curthread;
3560 	proc_t *pp = curproc;
3561 	zone_t *zone = pp->p_zone;
3562 
3563 	mutex_enter(&zone_status_lock);
3564 
3565 	/*
3566 	 * Reparent to p0
3567 	 */
3568 	kpreempt_disable();
3569 	mutex_enter(&pp->p_lock);
3570 	t->t_proc_flag &= ~TP_ZTHREAD;
3571 	t->t_procp = &p0;
3572 	hat_thread_exit(t);
3573 	mutex_exit(&pp->p_lock);
3574 	kpreempt_enable();
3575 
3576 	if (t->t_back == t) {
3577 		ASSERT(t->t_forw == t);
3578 		/*
3579 		 * If the zone is empty, once the thread count
3580 		 * goes to zero no further kernel threads can be
3581 		 * created.  This is because if the creator is a process
3582 		 * in the zone, then it must have exited before the zone
3583 		 * state could be set to ZONE_IS_EMPTY.
3584 		 * Otherwise, if the creator is a kernel thread in the
3585 		 * zone, the thread count is non-zero.
3586 		 *
3587 		 * This really means that non-zone kernel threads should
3588 		 * not create zone kernel threads.
3589 		 */
3590 		zone->zone_kthreads = NULL;
3591 		if (zone_status_get(zone) == ZONE_IS_EMPTY) {
3592 			zone_status_set(zone, ZONE_IS_DOWN);
3593 			/*
3594 			 * Remove any CPU caps on this zone.
3595 			 */
3596 			cpucaps_zone_remove(zone);
3597 		}
3598 	} else {
3599 		t->t_forw->t_back = t->t_back;
3600 		t->t_back->t_forw = t->t_forw;
3601 		if (zone->zone_kthreads == t)
3602 			zone->zone_kthreads = t->t_forw;
3603 	}
3604 	mutex_exit(&zone_status_lock);
3605 	zone_rele(zone);
3606 	thread_exit();
3607 	/* NOTREACHED */
3608 }
3609 
3610 static void
3611 zone_chdir(vnode_t *vp, vnode_t **vpp, proc_t *pp)
3612 {
3613 	vnode_t *oldvp;
3614 
3615 	/* we're going to hold a reference here to the directory */
3616 	VN_HOLD(vp);
3617 
3618 	/* update abs cwd/root path see c2/audit.c */
3619 	if (AU_AUDITING())
3620 		audit_chdirec(vp, vpp);
3621 
3622 	mutex_enter(&pp->p_lock);
3623 	oldvp = *vpp;
3624 	*vpp = vp;
3625 	mutex_exit(&pp->p_lock);
3626 	if (oldvp != NULL)
3627 		VN_RELE(oldvp);
3628 }
3629 
3630 /*
3631  * Convert an rctl value represented by an nvlist_t into an rctl_val_t.
3632  */
3633 static int
3634 nvlist2rctlval(nvlist_t *nvl, rctl_val_t *rv)
3635 {
3636 	nvpair_t *nvp = NULL;
3637 	boolean_t priv_set = B_FALSE;
3638 	boolean_t limit_set = B_FALSE;
3639 	boolean_t action_set = B_FALSE;
3640 
3641 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
3642 		const char *name;
3643 		uint64_t ui64;
3644 
3645 		name = nvpair_name(nvp);
3646 		if (nvpair_type(nvp) != DATA_TYPE_UINT64)
3647 			return (EINVAL);
3648 		(void) nvpair_value_uint64(nvp, &ui64);
3649 		if (strcmp(name, "privilege") == 0) {
3650 			/*
3651 			 * Currently only privileged values are allowed, but
3652 			 * this may change in the future.
3653 			 */
3654 			if (ui64 != RCPRIV_PRIVILEGED)
3655 				return (EINVAL);
3656 			rv->rcv_privilege = ui64;
3657 			priv_set = B_TRUE;
3658 		} else if (strcmp(name, "limit") == 0) {
3659 			rv->rcv_value = ui64;
3660 			limit_set = B_TRUE;
3661 		} else if (strcmp(name, "action") == 0) {
3662 			if (ui64 != RCTL_LOCAL_NOACTION &&
3663 			    ui64 != RCTL_LOCAL_DENY)
3664 				return (EINVAL);
3665 			rv->rcv_flagaction = ui64;
3666 			action_set = B_TRUE;
3667 		} else {
3668 			return (EINVAL);
3669 		}
3670 	}
3671 
3672 	if (!(priv_set && limit_set && action_set))
3673 		return (EINVAL);
3674 	rv->rcv_action_signal = 0;
3675 	rv->rcv_action_recipient = NULL;
3676 	rv->rcv_action_recip_pid = -1;
3677 	rv->rcv_firing_time = 0;
3678 
3679 	return (0);
3680 }
3681 
3682 /*
3683  * Non-global zone version of start_init.
3684  */
3685 void
3686 zone_start_init(void)
3687 {
3688 	proc_t *p = ttoproc(curthread);
3689 	zone_t *z = p->p_zone;
3690 
3691 	ASSERT(!INGLOBALZONE(curproc));
3692 
3693 	/*
3694 	 * For all purposes (ZONE_ATTR_INITPID and restart_init),
3695 	 * storing just the pid of init is sufficient.
3696 	 */
3697 	z->zone_proc_initpid = p->p_pid;
3698 
3699 	/*
3700 	 * We maintain zone_boot_err so that we can return the cause of the
3701 	 * failure back to the caller of the zone_boot syscall.
3702 	 */
3703 	p->p_zone->zone_boot_err = start_init_common();
3704 
3705 	/*
3706 	 * We will prevent booting zones from becoming running zones if the
3707 	 * global zone is shutting down.
3708 	 */
3709 	mutex_enter(&zone_status_lock);
3710 	if (z->zone_boot_err != 0 || zone_status_get(global_zone) >=
3711 	    ZONE_IS_SHUTTING_DOWN) {
3712 		/*
3713 		 * Make sure we are still in the booting state-- we could have
3714 		 * raced and already be shutting down, or even further along.
3715 		 */
3716 		if (zone_status_get(z) == ZONE_IS_BOOTING) {
3717 			zone_status_set(z, ZONE_IS_SHUTTING_DOWN);
3718 		}
3719 		mutex_exit(&zone_status_lock);
3720 		/* It's gone bad, dispose of the process */
3721 		if (proc_exit(CLD_EXITED, z->zone_boot_err) != 0) {
3722 			mutex_enter(&p->p_lock);
3723 			ASSERT(p->p_flag & SEXITLWPS);
3724 			lwp_exit();
3725 		}
3726 	} else {
3727 		if (zone_status_get(z) == ZONE_IS_BOOTING)
3728 			zone_status_set(z, ZONE_IS_RUNNING);
3729 		mutex_exit(&zone_status_lock);
3730 		/* cause the process to return to userland. */
3731 		lwp_rtt();
3732 	}
3733 }
3734 
3735 struct zsched_arg {
3736 	zone_t *zone;
3737 	nvlist_t *nvlist;
3738 };
3739 
3740 /*
3741  * Per-zone "sched" workalike.  The similarity to "sched" doesn't have
3742  * anything to do with scheduling, but rather with the fact that
3743  * per-zone kernel threads are parented to zsched, just like regular
3744  * kernel threads are parented to sched (p0).
3745  *
3746  * zsched is also responsible for launching init for the zone.
3747  */
3748 static void
3749 zsched(void *arg)
3750 {
3751 	struct zsched_arg *za = arg;
3752 	proc_t *pp = curproc;
3753 	proc_t *initp = proc_init;
3754 	zone_t *zone = za->zone;
3755 	cred_t *cr, *oldcred;
3756 	rctl_set_t *set;
3757 	rctl_alloc_gp_t *gp;
3758 	contract_t *ct = NULL;
3759 	task_t *tk, *oldtk;
3760 	rctl_entity_p_t e;
3761 	kproject_t *pj;
3762 
3763 	nvlist_t *nvl = za->nvlist;
3764 	nvpair_t *nvp = NULL;
3765 
3766 	bcopy("zsched", PTOU(pp)->u_psargs, sizeof ("zsched"));
3767 	bcopy("zsched", PTOU(pp)->u_comm, sizeof ("zsched"));
3768 	PTOU(pp)->u_argc = 0;
3769 	PTOU(pp)->u_argv = NULL;
3770 	PTOU(pp)->u_envp = NULL;
3771 	closeall(P_FINFO(pp));
3772 
3773 	/*
3774 	 * We are this zone's "zsched" process.  As the zone isn't generally
3775 	 * visible yet we don't need to grab any locks before initializing its
3776 	 * zone_proc pointer.
3777 	 */
3778 	zone_hold(zone);  /* this hold is released by zone_destroy() */
3779 	zone->zone_zsched = pp;
3780 	mutex_enter(&pp->p_lock);
3781 	pp->p_zone = zone;
3782 	mutex_exit(&pp->p_lock);
3783 
3784 	/*
3785 	 * Disassociate process from its 'parent'; parent ourselves to init
3786 	 * (pid 1) and change other values as needed.
3787 	 */
3788 	sess_create();
3789 
3790 	mutex_enter(&pidlock);
3791 	proc_detach(pp);
3792 	pp->p_ppid = 1;
3793 	pp->p_flag |= SZONETOP;
3794 	pp->p_ancpid = 1;
3795 	pp->p_parent = initp;
3796 	pp->p_psibling = NULL;
3797 	if (initp->p_child)
3798 		initp->p_child->p_psibling = pp;
3799 	pp->p_sibling = initp->p_child;
3800 	initp->p_child = pp;
3801 
3802 	/* Decrement what newproc() incremented. */
3803 	upcount_dec(crgetruid(CRED()), GLOBAL_ZONEID);
3804 	/*
3805 	 * Our credentials are about to become kcred-like, so we don't care
3806 	 * about the caller's ruid.
3807 	 */
3808 	upcount_inc(crgetruid(kcred), zone->zone_id);
3809 	mutex_exit(&pidlock);
3810 
3811 	/*
3812 	 * getting out of global zone, so decrement lwp and process counts
3813 	 */
3814 	pj = pp->p_task->tk_proj;
3815 	mutex_enter(&global_zone->zone_nlwps_lock);
3816 	pj->kpj_nlwps -= pp->p_lwpcnt;
3817 	global_zone->zone_nlwps -= pp->p_lwpcnt;
3818 	pj->kpj_nprocs--;
3819 	global_zone->zone_nprocs--;
3820 	mutex_exit(&global_zone->zone_nlwps_lock);
3821 
3822 	/*
3823 	 * Decrement locked memory counts on old zone and project.
3824 	 */
3825 	mutex_enter(&global_zone->zone_mem_lock);
3826 	global_zone->zone_locked_mem -= pp->p_locked_mem;
3827 	pj->kpj_data.kpd_locked_mem -= pp->p_locked_mem;
3828 	mutex_exit(&global_zone->zone_mem_lock);
3829 
3830 	/*
3831 	 * Create and join a new task in project '0' of this zone.
3832 	 *
3833 	 * We don't need to call holdlwps() since we know we're the only lwp in
3834 	 * this process.
3835 	 *
3836 	 * task_join() returns with p_lock held.
3837 	 */
3838 	tk = task_create(0, zone);
3839 	mutex_enter(&cpu_lock);
3840 	oldtk = task_join(tk, 0);
3841 
3842 	pj = pp->p_task->tk_proj;
3843 
3844 	mutex_enter(&zone->zone_mem_lock);
3845 	zone->zone_locked_mem += pp->p_locked_mem;
3846 	pj->kpj_data.kpd_locked_mem += pp->p_locked_mem;
3847 	mutex_exit(&zone->zone_mem_lock);
3848 
3849 	/*
3850 	 * add lwp and process counts to zsched's zone, and increment
3851 	 * project's task and process count due to the task created in
3852 	 * the above task_create.
3853 	 */
3854 	mutex_enter(&zone->zone_nlwps_lock);
3855 	pj->kpj_nlwps += pp->p_lwpcnt;
3856 	pj->kpj_ntasks += 1;
3857 	zone->zone_nlwps += pp->p_lwpcnt;
3858 	pj->kpj_nprocs++;
3859 	zone->zone_nprocs++;
3860 	mutex_exit(&zone->zone_nlwps_lock);
3861 
3862 	mutex_exit(&curproc->p_lock);
3863 	mutex_exit(&cpu_lock);
3864 	task_rele(oldtk);
3865 
3866 	/*
3867 	 * The process was created by a process in the global zone, hence the
3868 	 * credentials are wrong.  We might as well have kcred-ish credentials.
3869 	 */
3870 	cr = zone->zone_kcred;
3871 	crhold(cr);
3872 	mutex_enter(&pp->p_crlock);
3873 	oldcred = pp->p_cred;
3874 	pp->p_cred = cr;
3875 	mutex_exit(&pp->p_crlock);
3876 	crfree(oldcred);
3877 
3878 	/*
3879 	 * Hold credentials again (for thread)
3880 	 */
3881 	crhold(cr);
3882 
3883 	/*
3884 	 * p_lwpcnt can't change since this is a kernel process.
3885 	 */
3886 	crset(pp, cr);
3887 
3888 	/*
3889 	 * Chroot
3890 	 */
3891 	zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_cdir, pp);
3892 	zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_rdir, pp);
3893 
3894 	/*
3895 	 * Initialize zone's rctl set.
3896 	 */
3897 	set = rctl_set_create();
3898 	gp = rctl_set_init_prealloc(RCENTITY_ZONE);
3899 	mutex_enter(&pp->p_lock);
3900 	e.rcep_p.zone = zone;
3901 	e.rcep_t = RCENTITY_ZONE;
3902 	zone->zone_rctls = rctl_set_init(RCENTITY_ZONE, pp, &e, set, gp);
3903 	mutex_exit(&pp->p_lock);
3904 	rctl_prealloc_destroy(gp);
3905 
3906 	/*
3907 	 * Apply the rctls passed in to zone_create().  This is basically a list
3908 	 * assignment: all of the old values are removed and the new ones
3909 	 * inserted.  That is, if an empty list is passed in, all values are
3910 	 * removed.
3911 	 */
3912 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
3913 		rctl_dict_entry_t *rde;
3914 		rctl_hndl_t hndl;
3915 		char *name;
3916 		nvlist_t **nvlarray;
3917 		uint_t i, nelem;
3918 		int error;	/* For ASSERT()s */
3919 
3920 		name = nvpair_name(nvp);
3921 		hndl = rctl_hndl_lookup(name);
3922 		ASSERT(hndl != -1);
3923 		rde = rctl_dict_lookup_hndl(hndl);
3924 		ASSERT(rde != NULL);
3925 
3926 		for (; /* ever */; ) {
3927 			rctl_val_t oval;
3928 
3929 			mutex_enter(&pp->p_lock);
3930 			error = rctl_local_get(hndl, NULL, &oval, pp);
3931 			mutex_exit(&pp->p_lock);
3932 			ASSERT(error == 0);	/* Can't fail for RCTL_FIRST */
3933 			ASSERT(oval.rcv_privilege != RCPRIV_BASIC);
3934 			if (oval.rcv_privilege == RCPRIV_SYSTEM)
3935 				break;
3936 			mutex_enter(&pp->p_lock);
3937 			error = rctl_local_delete(hndl, &oval, pp);
3938 			mutex_exit(&pp->p_lock);
3939 			ASSERT(error == 0);
3940 		}
3941 		error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
3942 		ASSERT(error == 0);
3943 		for (i = 0; i < nelem; i++) {
3944 			rctl_val_t *nvalp;
3945 
3946 			nvalp = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
3947 			error = nvlist2rctlval(nvlarray[i], nvalp);
3948 			ASSERT(error == 0);
3949 			/*
3950 			 * rctl_local_insert can fail if the value being
3951 			 * inserted is a duplicate; this is OK.
3952 			 */
3953 			mutex_enter(&pp->p_lock);
3954 			if (rctl_local_insert(hndl, nvalp, pp) != 0)
3955 				kmem_cache_free(rctl_val_cache, nvalp);
3956 			mutex_exit(&pp->p_lock);
3957 		}
3958 	}
3959 	/*
3960 	 * Tell the world that we're done setting up.
3961 	 *
3962 	 * At this point we want to set the zone status to ZONE_IS_INITIALIZED
3963 	 * and atomically set the zone's processor set visibility.  Once
3964 	 * we drop pool_lock() this zone will automatically get updated
3965 	 * to reflect any future changes to the pools configuration.
3966 	 *
3967 	 * Note that after we drop the locks below (zonehash_lock in
3968 	 * particular) other operations such as a zone_getattr call can
3969 	 * now proceed and observe the zone. That is the reason for doing a
3970 	 * state transition to the INITIALIZED state.
3971 	 */
3972 	pool_lock();
3973 	mutex_enter(&cpu_lock);
3974 	mutex_enter(&zonehash_lock);
3975 	zone_uniqid(zone);
3976 	zone_zsd_configure(zone);
3977 	if (pool_state == POOL_ENABLED)
3978 		zone_pset_set(zone, pool_default->pool_pset->pset_id);
3979 	mutex_enter(&zone_status_lock);
3980 	ASSERT(zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
3981 	zone_status_set(zone, ZONE_IS_INITIALIZED);
3982 	mutex_exit(&zone_status_lock);
3983 	mutex_exit(&zonehash_lock);
3984 	mutex_exit(&cpu_lock);
3985 	pool_unlock();
3986 
3987 	/* Now call the create callback for this key */
3988 	zsd_apply_all_keys(zsd_apply_create, zone);
3989 
3990 	/* The callbacks are complete. Mark ZONE_IS_READY */
3991 	mutex_enter(&zone_status_lock);
3992 	ASSERT(zone_status_get(zone) == ZONE_IS_INITIALIZED);
3993 	zone_status_set(zone, ZONE_IS_READY);
3994 	mutex_exit(&zone_status_lock);
3995 
3996 	/*
3997 	 * Once we see the zone transition to the ZONE_IS_BOOTING state,
3998 	 * we launch init, and set the state to running.
3999 	 */
4000 	zone_status_wait_cpr(zone, ZONE_IS_BOOTING, "zsched");
4001 
4002 	if (zone_status_get(zone) == ZONE_IS_BOOTING) {
4003 		id_t cid;
4004 
4005 		/*
4006 		 * Ok, this is a little complicated.  We need to grab the
4007 		 * zone's pool's scheduling class ID; note that by now, we
4008 		 * are already bound to a pool if we need to be (zoneadmd
4009 		 * will have done that to us while we're in the READY
4010 		 * state).  *But* the scheduling class for the zone's 'init'
4011 		 * must be explicitly passed to newproc, which doesn't
4012 		 * respect pool bindings.
4013 		 *
4014 		 * We hold the pool_lock across the call to newproc() to
4015 		 * close the obvious race: the pool's scheduling class
4016 		 * could change before we manage to create the LWP with
4017 		 * classid 'cid'.
4018 		 */
4019 		pool_lock();
4020 		if (zone->zone_defaultcid > 0)
4021 			cid = zone->zone_defaultcid;
4022 		else
4023 			cid = pool_get_class(zone->zone_pool);
4024 		if (cid == -1)
4025 			cid = defaultcid;
4026 
4027 		/*
4028 		 * If this fails, zone_boot will ultimately fail.  The
4029 		 * state of the zone will be set to SHUTTING_DOWN-- userland
4030 		 * will have to tear down the zone, and fail, or try again.
4031 		 */
4032 		if ((zone->zone_boot_err = newproc(zone_start_init, NULL, cid,
4033 		    minclsyspri - 1, &ct, 0)) != 0) {
4034 			mutex_enter(&zone_status_lock);
4035 			zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
4036 			mutex_exit(&zone_status_lock);
4037 		} else {
4038 			zone->zone_boot_time = gethrestime_sec();
4039 		}
4040 
4041 		pool_unlock();
4042 	}
4043 
4044 	/*
4045 	 * Wait for zone_destroy() to be called.  This is what we spend
4046 	 * most of our life doing.
4047 	 */
4048 	zone_status_wait_cpr(zone, ZONE_IS_DYING, "zsched");
4049 
4050 	if (ct)
4051 		/*
4052 		 * At this point the process contract should be empty.
4053 		 * (Though if it isn't, it's not the end of the world.)
4054 		 */
4055 		VERIFY(contract_abandon(ct, curproc, B_TRUE) == 0);
4056 
4057 	/*
4058 	 * Allow kcred to be freed when all referring processes
4059 	 * (including this one) go away.  We can't just do this in
4060 	 * zone_free because we need to wait for the zone_cred_ref to
4061 	 * drop to 0 before calling zone_free, and the existence of
4062 	 * zone_kcred will prevent that.  Thus, we call crfree here to
4063 	 * balance the crdup in zone_create.  The crhold calls earlier
4064 	 * in zsched will be dropped when the thread and process exit.
4065 	 */
4066 	crfree(zone->zone_kcred);
4067 	zone->zone_kcred = NULL;
4068 
4069 	exit(CLD_EXITED, 0);
4070 }
4071 
4072 /*
4073  * Helper function to determine if there are any submounts of the
4074  * provided path.  Used to make sure the zone doesn't "inherit" any
4075  * mounts from before it is created.
4076  */
4077 static uint_t
4078 zone_mount_count(const char *rootpath)
4079 {
4080 	vfs_t *vfsp;
4081 	uint_t count = 0;
4082 	size_t rootpathlen = strlen(rootpath);
4083 
4084 	/*
4085 	 * Holding zonehash_lock prevents race conditions with
4086 	 * vfs_list_add()/vfs_list_remove() since we serialize with
4087 	 * zone_find_by_path().
4088 	 */
4089 	ASSERT(MUTEX_HELD(&zonehash_lock));
4090 	/*
4091 	 * The rootpath must end with a '/'
4092 	 */
4093 	ASSERT(rootpath[rootpathlen - 1] == '/');
4094 
4095 	/*
4096 	 * This intentionally does not count the rootpath itself if that
4097 	 * happens to be a mount point.
4098 	 */
4099 	vfs_list_read_lock();
4100 	vfsp = rootvfs;
4101 	do {
4102 		if (strncmp(rootpath, refstr_value(vfsp->vfs_mntpt),
4103 		    rootpathlen) == 0)
4104 			count++;
4105 		vfsp = vfsp->vfs_next;
4106 	} while (vfsp != rootvfs);
4107 	vfs_list_unlock();
4108 	return (count);
4109 }
4110 
4111 /*
4112  * Helper function to make sure that a zone created on 'rootpath'
4113  * wouldn't end up containing other zones' rootpaths.
4114  */
4115 static boolean_t
4116 zone_is_nested(const char *rootpath)
4117 {
4118 	zone_t *zone;
4119 	size_t rootpathlen = strlen(rootpath);
4120 	size_t len;
4121 
4122 	ASSERT(MUTEX_HELD(&zonehash_lock));
4123 
4124 	/*
4125 	 * zone_set_root() appended '/' and '\0' at the end of rootpath
4126 	 */
4127 	if ((rootpathlen <= 3) && (rootpath[0] == '/') &&
4128 	    (rootpath[1] == '/') && (rootpath[2] == '\0'))
4129 		return (B_TRUE);
4130 
4131 	for (zone = list_head(&zone_active); zone != NULL;
4132 	    zone = list_next(&zone_active, zone)) {
4133 		if (zone == global_zone)
4134 			continue;
4135 		len = strlen(zone->zone_rootpath);
4136 		if (strncmp(rootpath, zone->zone_rootpath,
4137 		    MIN(rootpathlen, len)) == 0)
4138 			return (B_TRUE);
4139 	}
4140 	return (B_FALSE);
4141 }
4142 
4143 static int
4144 zone_set_privset(zone_t *zone, const priv_set_t *zone_privs,
4145     size_t zone_privssz)
4146 {
4147 	priv_set_t *privs;
4148 
4149 	if (zone_privssz < sizeof (priv_set_t))
4150 		return (ENOMEM);
4151 
4152 	privs = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
4153 
4154 	if (copyin(zone_privs, privs, sizeof (priv_set_t))) {
4155 		kmem_free(privs, sizeof (priv_set_t));
4156 		return (EFAULT);
4157 	}
4158 
4159 	zone->zone_privset = privs;
4160 	return (0);
4161 }
4162 
4163 /*
4164  * We make creative use of nvlists to pass in rctls from userland.  The list is
4165  * a list of the following structures:
4166  *
4167  * (name = rctl_name, value = nvpair_list_array)
4168  *
4169  * Where each element of the nvpair_list_array is of the form:
4170  *
4171  * [(name = "privilege", value = RCPRIV_PRIVILEGED),
4172  * 	(name = "limit", value = uint64_t),
4173  * 	(name = "action", value = (RCTL_LOCAL_NOACTION || RCTL_LOCAL_DENY))]
4174  */
4175 static int
4176 parse_rctls(caddr_t ubuf, size_t buflen, nvlist_t **nvlp)
4177 {
4178 	nvpair_t *nvp = NULL;
4179 	nvlist_t *nvl = NULL;
4180 	char *kbuf;
4181 	int error;
4182 	rctl_val_t rv;
4183 
4184 	*nvlp = NULL;
4185 
4186 	if (buflen == 0)
4187 		return (0);
4188 
4189 	if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL)
4190 		return (ENOMEM);
4191 	if (copyin(ubuf, kbuf, buflen)) {
4192 		error = EFAULT;
4193 		goto out;
4194 	}
4195 	if (nvlist_unpack(kbuf, buflen, &nvl, KM_SLEEP) != 0) {
4196 		/*
4197 		 * nvl may have been allocated/free'd, but the value set to
4198 		 * non-NULL, so we reset it here.
4199 		 */
4200 		nvl = NULL;
4201 		error = EINVAL;
4202 		goto out;
4203 	}
4204 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
4205 		rctl_dict_entry_t *rde;
4206 		rctl_hndl_t hndl;
4207 		nvlist_t **nvlarray;
4208 		uint_t i, nelem;
4209 		char *name;
4210 
4211 		error = EINVAL;
4212 		name = nvpair_name(nvp);
4213 		if (strncmp(nvpair_name(nvp), "zone.", sizeof ("zone.") - 1)
4214 		    != 0 || nvpair_type(nvp) != DATA_TYPE_NVLIST_ARRAY) {
4215 			goto out;
4216 		}
4217 		if ((hndl = rctl_hndl_lookup(name)) == -1) {
4218 			goto out;
4219 		}
4220 		rde = rctl_dict_lookup_hndl(hndl);
4221 		error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
4222 		ASSERT(error == 0);
4223 		for (i = 0; i < nelem; i++) {
4224 			if (error = nvlist2rctlval(nvlarray[i], &rv))
4225 				goto out;
4226 		}
4227 		if (rctl_invalid_value(rde, &rv)) {
4228 			error = EINVAL;
4229 			goto out;
4230 		}
4231 	}
4232 	error = 0;
4233 	*nvlp = nvl;
4234 out:
4235 	kmem_free(kbuf, buflen);
4236 	if (error && nvl != NULL)
4237 		nvlist_free(nvl);
4238 	return (error);
4239 }
4240 
4241 int
4242 zone_create_error(int er_error, int er_ext, int *er_out) {
4243 	if (er_out != NULL) {
4244 		if (copyout(&er_ext, er_out, sizeof (int))) {
4245 			return (set_errno(EFAULT));
4246 		}
4247 	}
4248 	return (set_errno(er_error));
4249 }
4250 
4251 static int
4252 zone_set_label(zone_t *zone, const bslabel_t *lab, uint32_t doi)
4253 {
4254 	ts_label_t *tsl;
4255 	bslabel_t blab;
4256 
4257 	/* Get label from user */
4258 	if (copyin(lab, &blab, sizeof (blab)) != 0)
4259 		return (EFAULT);
4260 	tsl = labelalloc(&blab, doi, KM_NOSLEEP);
4261 	if (tsl == NULL)
4262 		return (ENOMEM);
4263 
4264 	zone->zone_slabel = tsl;
4265 	return (0);
4266 }
4267 
4268 /*
4269  * Parses a comma-separated list of ZFS datasets into a per-zone dictionary.
4270  */
4271 static int
4272 parse_zfs(zone_t *zone, caddr_t ubuf, size_t buflen)
4273 {
4274 	char *kbuf;
4275 	char *dataset, *next;
4276 	zone_dataset_t *zd;
4277 	size_t len;
4278 
4279 	if (ubuf == NULL || buflen == 0)
4280 		return (0);
4281 
4282 	if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL)
4283 		return (ENOMEM);
4284 
4285 	if (copyin(ubuf, kbuf, buflen) != 0) {
4286 		kmem_free(kbuf, buflen);
4287 		return (EFAULT);
4288 	}
4289 
4290 	dataset = next = kbuf;
4291 	for (;;) {
4292 		zd = kmem_alloc(sizeof (zone_dataset_t), KM_SLEEP);
4293 
4294 		next = strchr(dataset, ',');
4295 
4296 		if (next == NULL)
4297 			len = strlen(dataset);
4298 		else
4299 			len = next - dataset;
4300 
4301 		zd->zd_dataset = kmem_alloc(len + 1, KM_SLEEP);
4302 		bcopy(dataset, zd->zd_dataset, len);
4303 		zd->zd_dataset[len] = '\0';
4304 
4305 		list_insert_head(&zone->zone_datasets, zd);
4306 
4307 		if (next == NULL)
4308 			break;
4309 
4310 		dataset = next + 1;
4311 	}
4312 
4313 	kmem_free(kbuf, buflen);
4314 	return (0);
4315 }
4316 
4317 /*
4318  * System call to create/initialize a new zone named 'zone_name', rooted
4319  * at 'zone_root', with a zone-wide privilege limit set of 'zone_privs',
4320  * and initialized with the zone-wide rctls described in 'rctlbuf', and
4321  * with labeling set by 'match', 'doi', and 'label'.
4322  *
4323  * If extended error is non-null, we may use it to return more detailed
4324  * error information.
4325  */
4326 static zoneid_t
4327 zone_create(const char *zone_name, const char *zone_root,
4328     const priv_set_t *zone_privs, size_t zone_privssz,
4329     caddr_t rctlbuf, size_t rctlbufsz,
4330     caddr_t zfsbuf, size_t zfsbufsz, int *extended_error,
4331     int match, uint32_t doi, const bslabel_t *label,
4332     int flags)
4333 {
4334 	struct zsched_arg zarg;
4335 	nvlist_t *rctls = NULL;
4336 	proc_t *pp = curproc;
4337 	zone_t *zone, *ztmp;
4338 	zoneid_t zoneid;
4339 	int error;
4340 	int error2 = 0;
4341 	char *str;
4342 	cred_t *zkcr;
4343 	boolean_t insert_label_hash;
4344 
4345 	if (secpolicy_zone_config(CRED()) != 0)
4346 		return (set_errno(EPERM));
4347 
4348 	/* can't boot zone from within chroot environment */
4349 	if (PTOU(pp)->u_rdir != NULL && PTOU(pp)->u_rdir != rootdir)
4350 		return (zone_create_error(ENOTSUP, ZE_CHROOTED,
4351 		    extended_error));
4352 
4353 	zone = kmem_zalloc(sizeof (zone_t), KM_SLEEP);
4354 	zoneid = zone->zone_id = id_alloc(zoneid_space);
4355 	zone->zone_status = ZONE_IS_UNINITIALIZED;
4356 	zone->zone_pool = pool_default;
4357 	zone->zone_pool_mod = gethrtime();
4358 	zone->zone_psetid = ZONE_PS_INVAL;
4359 	zone->zone_ncpus = 0;
4360 	zone->zone_ncpus_online = 0;
4361 	zone->zone_restart_init = B_TRUE;
4362 	zone->zone_brand = &native_brand;
4363 	zone->zone_initname = NULL;
4364 	mutex_init(&zone->zone_lock, NULL, MUTEX_DEFAULT, NULL);
4365 	mutex_init(&zone->zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
4366 	mutex_init(&zone->zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
4367 	cv_init(&zone->zone_cv, NULL, CV_DEFAULT, NULL);
4368 	list_create(&zone->zone_ref_list, sizeof (zone_ref_t),
4369 	    offsetof(zone_ref_t, zref_linkage));
4370 	list_create(&zone->zone_zsd, sizeof (struct zsd_entry),
4371 	    offsetof(struct zsd_entry, zsd_linkage));
4372 	list_create(&zone->zone_datasets, sizeof (zone_dataset_t),
4373 	    offsetof(zone_dataset_t, zd_linkage));
4374 	list_create(&zone->zone_dl_list, sizeof (zone_dl_t),
4375 	    offsetof(zone_dl_t, zdl_linkage));
4376 	rw_init(&zone->zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
4377 	rw_init(&zone->zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL);
4378 
4379 	if (flags & ZCF_NET_EXCL) {
4380 		zone->zone_flags |= ZF_NET_EXCL;
4381 	}
4382 
4383 	if ((error = zone_set_name(zone, zone_name)) != 0) {
4384 		zone_free(zone);
4385 		return (zone_create_error(error, 0, extended_error));
4386 	}
4387 
4388 	if ((error = zone_set_root(zone, zone_root)) != 0) {
4389 		zone_free(zone);
4390 		return (zone_create_error(error, 0, extended_error));
4391 	}
4392 	if ((error = zone_set_privset(zone, zone_privs, zone_privssz)) != 0) {
4393 		zone_free(zone);
4394 		return (zone_create_error(error, 0, extended_error));
4395 	}
4396 
4397 	/* initialize node name to be the same as zone name */
4398 	zone->zone_nodename = kmem_alloc(_SYS_NMLN, KM_SLEEP);
4399 	(void) strncpy(zone->zone_nodename, zone->zone_name, _SYS_NMLN);
4400 	zone->zone_nodename[_SYS_NMLN - 1] = '\0';
4401 
4402 	zone->zone_domain = kmem_alloc(_SYS_NMLN, KM_SLEEP);
4403 	zone->zone_domain[0] = '\0';
4404 	zone->zone_hostid = HW_INVALID_HOSTID;
4405 	zone->zone_shares = 1;
4406 	zone->zone_shmmax = 0;
4407 	zone->zone_ipc.ipcq_shmmni = 0;
4408 	zone->zone_ipc.ipcq_semmni = 0;
4409 	zone->zone_ipc.ipcq_msgmni = 0;
4410 	zone->zone_bootargs = NULL;
4411 	zone->zone_fs_allowed = NULL;
4412 	zone->zone_initname =
4413 	    kmem_alloc(strlen(zone_default_initname) + 1, KM_SLEEP);
4414 	(void) strcpy(zone->zone_initname, zone_default_initname);
4415 	zone->zone_nlwps = 0;
4416 	zone->zone_nlwps_ctl = INT_MAX;
4417 	zone->zone_nprocs = 0;
4418 	zone->zone_nprocs_ctl = INT_MAX;
4419 	zone->zone_locked_mem = 0;
4420 	zone->zone_locked_mem_ctl = UINT64_MAX;
4421 	zone->zone_max_swap = 0;
4422 	zone->zone_max_swap_ctl = UINT64_MAX;
4423 	zone->zone_max_lofi = 0;
4424 	zone->zone_max_lofi_ctl = UINT64_MAX;
4425 	zone0.zone_lockedmem_kstat = NULL;
4426 	zone0.zone_swapresv_kstat = NULL;
4427 
4428 	/*
4429 	 * Zsched initializes the rctls.
4430 	 */
4431 	zone->zone_rctls = NULL;
4432 
4433 	if ((error = parse_rctls(rctlbuf, rctlbufsz, &rctls)) != 0) {
4434 		zone_free(zone);
4435 		return (zone_create_error(error, 0, extended_error));
4436 	}
4437 
4438 	if ((error = parse_zfs(zone, zfsbuf, zfsbufsz)) != 0) {
4439 		zone_free(zone);
4440 		return (set_errno(error));
4441 	}
4442 
4443 	/*
4444 	 * Read in the trusted system parameters:
4445 	 * match flag and sensitivity label.
4446 	 */
4447 	zone->zone_match = match;
4448 	if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) {
4449 		/* Fail if requested to set doi to anything but system's doi */
4450 		if (doi != 0 && doi != default_doi) {
4451 			zone_free(zone);
4452 			return (set_errno(EINVAL));
4453 		}
4454 		/* Always apply system's doi to the zone */
4455 		error = zone_set_label(zone, label, default_doi);
4456 		if (error != 0) {
4457 			zone_free(zone);
4458 			return (set_errno(error));
4459 		}
4460 		insert_label_hash = B_TRUE;
4461 	} else {
4462 		/* all zones get an admin_low label if system is not labeled */
4463 		zone->zone_slabel = l_admin_low;
4464 		label_hold(l_admin_low);
4465 		insert_label_hash = B_FALSE;
4466 	}
4467 
4468 	/*
4469 	 * Stop all lwps since that's what normally happens as part of fork().
4470 	 * This needs to happen before we grab any locks to avoid deadlock
4471 	 * (another lwp in the process could be waiting for the held lock).
4472 	 */
4473 	if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK)) {
4474 		zone_free(zone);
4475 		nvlist_free(rctls);
4476 		return (zone_create_error(error, 0, extended_error));
4477 	}
4478 
4479 	if (block_mounts(zone) == 0) {
4480 		mutex_enter(&pp->p_lock);
4481 		if (curthread != pp->p_agenttp)
4482 			continuelwps(pp);
4483 		mutex_exit(&pp->p_lock);
4484 		zone_free(zone);
4485 		nvlist_free(rctls);
4486 		return (zone_create_error(error, 0, extended_error));
4487 	}
4488 
4489 	/*
4490 	 * Set up credential for kernel access.  After this, any errors
4491 	 * should go through the dance in errout rather than calling
4492 	 * zone_free directly.
4493 	 */
4494 	zone->zone_kcred = crdup(kcred);
4495 	crsetzone(zone->zone_kcred, zone);
4496 	priv_intersect(zone->zone_privset, &CR_PPRIV(zone->zone_kcred));
4497 	priv_intersect(zone->zone_privset, &CR_EPRIV(zone->zone_kcred));
4498 	priv_intersect(zone->zone_privset, &CR_IPRIV(zone->zone_kcred));
4499 	priv_intersect(zone->zone_privset, &CR_LPRIV(zone->zone_kcred));
4500 
4501 	mutex_enter(&zonehash_lock);
4502 	/*
4503 	 * Make sure zone doesn't already exist.
4504 	 *
4505 	 * If the system and zone are labeled,
4506 	 * make sure no other zone exists that has the same label.
4507 	 */
4508 	if ((ztmp = zone_find_all_by_name(zone->zone_name)) != NULL ||
4509 	    (insert_label_hash &&
4510 	    (ztmp = zone_find_all_by_label(zone->zone_slabel)) != NULL)) {
4511 		zone_status_t status;
4512 
4513 		status = zone_status_get(ztmp);
4514 		if (status == ZONE_IS_READY || status == ZONE_IS_RUNNING)
4515 			error = EEXIST;
4516 		else
4517 			error = EBUSY;
4518 
4519 		if (insert_label_hash)
4520 			error2 = ZE_LABELINUSE;
4521 
4522 		goto errout;
4523 	}
4524 
4525 	/*
4526 	 * Don't allow zone creations which would cause one zone's rootpath to
4527 	 * be accessible from that of another (non-global) zone.
4528 	 */
4529 	if (zone_is_nested(zone->zone_rootpath)) {
4530 		error = EBUSY;
4531 		goto errout;
4532 	}
4533 
4534 	ASSERT(zonecount != 0);		/* check for leaks */
4535 	if (zonecount + 1 > maxzones) {
4536 		error = ENOMEM;
4537 		goto errout;
4538 	}
4539 
4540 	if (zone_mount_count(zone->zone_rootpath) != 0) {
4541 		error = EBUSY;
4542 		error2 = ZE_AREMOUNTS;
4543 		goto errout;
4544 	}
4545 
4546 	/*
4547 	 * Zone is still incomplete, but we need to drop all locks while
4548 	 * zsched() initializes this zone's kernel process.  We
4549 	 * optimistically add the zone to the hashtable and associated
4550 	 * lists so a parallel zone_create() doesn't try to create the
4551 	 * same zone.
4552 	 */
4553 	zonecount++;
4554 	(void) mod_hash_insert(zonehashbyid,
4555 	    (mod_hash_key_t)(uintptr_t)zone->zone_id,
4556 	    (mod_hash_val_t)(uintptr_t)zone);
4557 	str = kmem_alloc(strlen(zone->zone_name) + 1, KM_SLEEP);
4558 	(void) strcpy(str, zone->zone_name);
4559 	(void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)str,
4560 	    (mod_hash_val_t)(uintptr_t)zone);
4561 	if (insert_label_hash) {
4562 		(void) mod_hash_insert(zonehashbylabel,
4563 		    (mod_hash_key_t)zone->zone_slabel, (mod_hash_val_t)zone);
4564 		zone->zone_flags |= ZF_HASHED_LABEL;
4565 	}
4566 
4567 	/*
4568 	 * Insert into active list.  At this point there are no 'hold's
4569 	 * on the zone, but everyone else knows not to use it, so we can
4570 	 * continue to use it.  zsched() will do a zone_hold() if the
4571 	 * newproc() is successful.
4572 	 */
4573 	list_insert_tail(&zone_active, zone);
4574 	mutex_exit(&zonehash_lock);
4575 
4576 	zarg.zone = zone;
4577 	zarg.nvlist = rctls;
4578 	/*
4579 	 * The process, task, and project rctls are probably wrong;
4580 	 * we need an interface to get the default values of all rctls,
4581 	 * and initialize zsched appropriately.  I'm not sure that that
4582 	 * makes much of a difference, though.
4583 	 */
4584 	error = newproc(zsched, (void *)&zarg, syscid, minclsyspri, NULL, 0);
4585 	if (error != 0) {
4586 		/*
4587 		 * We need to undo all globally visible state.
4588 		 */
4589 		mutex_enter(&zonehash_lock);
4590 		list_remove(&zone_active, zone);
4591 		if (zone->zone_flags & ZF_HASHED_LABEL) {
4592 			ASSERT(zone->zone_slabel != NULL);
4593 			(void) mod_hash_destroy(zonehashbylabel,
4594 			    (mod_hash_key_t)zone->zone_slabel);
4595 		}
4596 		(void) mod_hash_destroy(zonehashbyname,
4597 		    (mod_hash_key_t)(uintptr_t)zone->zone_name);
4598 		(void) mod_hash_destroy(zonehashbyid,
4599 		    (mod_hash_key_t)(uintptr_t)zone->zone_id);
4600 		ASSERT(zonecount > 1);
4601 		zonecount--;
4602 		goto errout;
4603 	}
4604 
4605 	/*
4606 	 * Zone creation can't fail from now on.
4607 	 */
4608 
4609 	/*
4610 	 * Create zone kstats
4611 	 */
4612 	zone_kstat_create(zone);
4613 
4614 	/*
4615 	 * Let the other lwps continue.
4616 	 */
4617 	mutex_enter(&pp->p_lock);
4618 	if (curthread != pp->p_agenttp)
4619 		continuelwps(pp);
4620 	mutex_exit(&pp->p_lock);
4621 
4622 	/*
4623 	 * Wait for zsched to finish initializing the zone.
4624 	 */
4625 	zone_status_wait(zone, ZONE_IS_READY);
4626 	/*
4627 	 * The zone is fully visible, so we can let mounts progress.
4628 	 */
4629 	resume_mounts(zone);
4630 	nvlist_free(rctls);
4631 
4632 	return (zoneid);
4633 
4634 errout:
4635 	mutex_exit(&zonehash_lock);
4636 	/*
4637 	 * Let the other lwps continue.
4638 	 */
4639 	mutex_enter(&pp->p_lock);
4640 	if (curthread != pp->p_agenttp)
4641 		continuelwps(pp);
4642 	mutex_exit(&pp->p_lock);
4643 
4644 	resume_mounts(zone);
4645 	nvlist_free(rctls);
4646 	/*
4647 	 * There is currently one reference to the zone, a cred_ref from
4648 	 * zone_kcred.  To free the zone, we call crfree, which will call
4649 	 * zone_cred_rele, which will call zone_free.
4650 	 */
4651 	ASSERT(zone->zone_cred_ref == 1);
4652 	ASSERT(zone->zone_kcred->cr_ref == 1);
4653 	ASSERT(zone->zone_ref == 0);
4654 	zkcr = zone->zone_kcred;
4655 	zone->zone_kcred = NULL;
4656 	crfree(zkcr);				/* triggers call to zone_free */
4657 	return (zone_create_error(error, error2, extended_error));
4658 }
4659 
4660 /*
4661  * Cause the zone to boot.  This is pretty simple, since we let zoneadmd do
4662  * the heavy lifting.  initname is the path to the program to launch
4663  * at the "top" of the zone; if this is NULL, we use the system default,
4664  * which is stored at zone_default_initname.
4665  */
4666 static int
4667 zone_boot(zoneid_t zoneid)
4668 {
4669 	int err;
4670 	zone_t *zone;
4671 
4672 	if (secpolicy_zone_config(CRED()) != 0)
4673 		return (set_errno(EPERM));
4674 	if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4675 		return (set_errno(EINVAL));
4676 
4677 	mutex_enter(&zonehash_lock);
4678 	/*
4679 	 * Look for zone under hash lock to prevent races with calls to
4680 	 * zone_shutdown, zone_destroy, etc.
4681 	 */
4682 	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4683 		mutex_exit(&zonehash_lock);
4684 		return (set_errno(EINVAL));
4685 	}
4686 
4687 	mutex_enter(&zone_status_lock);
4688 	if (zone_status_get(zone) != ZONE_IS_READY) {
4689 		mutex_exit(&zone_status_lock);
4690 		mutex_exit(&zonehash_lock);
4691 		return (set_errno(EINVAL));
4692 	}
4693 	zone_status_set(zone, ZONE_IS_BOOTING);
4694 	mutex_exit(&zone_status_lock);
4695 
4696 	zone_hold(zone);	/* so we can use the zone_t later */
4697 	mutex_exit(&zonehash_lock);
4698 
4699 	if (zone_status_wait_sig(zone, ZONE_IS_RUNNING) == 0) {
4700 		zone_rele(zone);
4701 		return (set_errno(EINTR));
4702 	}
4703 
4704 	/*
4705 	 * Boot (starting init) might have failed, in which case the zone
4706 	 * will go to the SHUTTING_DOWN state; an appropriate errno will
4707 	 * be placed in zone->zone_boot_err, and so we return that.
4708 	 */
4709 	err = zone->zone_boot_err;
4710 	zone_rele(zone);
4711 	return (err ? set_errno(err) : 0);
4712 }
4713 
4714 /*
4715  * Kills all user processes in the zone, waiting for them all to exit
4716  * before returning.
4717  */
4718 static int
4719 zone_empty(zone_t *zone)
4720 {
4721 	int waitstatus;
4722 
4723 	/*
4724 	 * We need to drop zonehash_lock before killing all
4725 	 * processes, otherwise we'll deadlock with zone_find_*
4726 	 * which can be called from the exit path.
4727 	 */
4728 	ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
4729 	while ((waitstatus = zone_status_timedwait_sig(zone,
4730 	    ddi_get_lbolt() + hz, ZONE_IS_EMPTY)) == -1) {
4731 		killall(zone->zone_id);
4732 	}
4733 	/*
4734 	 * return EINTR if we were signaled
4735 	 */
4736 	if (waitstatus == 0)
4737 		return (EINTR);
4738 	return (0);
4739 }
4740 
4741 /*
4742  * This function implements the policy for zone visibility.
4743  *
4744  * In standard Solaris, a non-global zone can only see itself.
4745  *
4746  * In Trusted Extensions, a labeled zone can lookup any zone whose label
4747  * it dominates. For this test, the label of the global zone is treated as
4748  * admin_high so it is special-cased instead of being checked for dominance.
4749  *
4750  * Returns true if zone attributes are viewable, false otherwise.
4751  */
4752 static boolean_t
4753 zone_list_access(zone_t *zone)
4754 {
4755 
4756 	if (curproc->p_zone == global_zone ||
4757 	    curproc->p_zone == zone) {
4758 		return (B_TRUE);
4759 	} else if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) {
4760 		bslabel_t *curproc_label;
4761 		bslabel_t *zone_label;
4762 
4763 		curproc_label = label2bslabel(curproc->p_zone->zone_slabel);
4764 		zone_label = label2bslabel(zone->zone_slabel);
4765 
4766 		if (zone->zone_id != GLOBAL_ZONEID &&
4767 		    bldominates(curproc_label, zone_label)) {
4768 			return (B_TRUE);
4769 		} else {
4770 			return (B_FALSE);
4771 		}
4772 	} else {
4773 		return (B_FALSE);
4774 	}
4775 }
4776 
4777 /*
4778  * Systemcall to start the zone's halt sequence.  By the time this
4779  * function successfully returns, all user processes and kernel threads
4780  * executing in it will have exited, ZSD shutdown callbacks executed,
4781  * and the zone status set to ZONE_IS_DOWN.
4782  *
4783  * It is possible that the call will interrupt itself if the caller is the
4784  * parent of any process running in the zone, and doesn't have SIGCHLD blocked.
4785  */
4786 static int
4787 zone_shutdown(zoneid_t zoneid)
4788 {
4789 	int error;
4790 	zone_t *zone;
4791 	zone_status_t status;
4792 
4793 	if (secpolicy_zone_config(CRED()) != 0)
4794 		return (set_errno(EPERM));
4795 	if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4796 		return (set_errno(EINVAL));
4797 
4798 	mutex_enter(&zonehash_lock);
4799 	/*
4800 	 * Look for zone under hash lock to prevent races with other
4801 	 * calls to zone_shutdown and zone_destroy.
4802 	 */
4803 	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4804 		mutex_exit(&zonehash_lock);
4805 		return (set_errno(EINVAL));
4806 	}
4807 
4808 	/*
4809 	 * We have to drop zonehash_lock before calling block_mounts.
4810 	 * Hold the zone so we can continue to use the zone_t.
4811 	 */
4812 	zone_hold(zone);
4813 	mutex_exit(&zonehash_lock);
4814 
4815 	/*
4816 	 * Block mounts so that VFS_MOUNT() can get an accurate view of
4817 	 * the zone's status with regards to ZONE_IS_SHUTTING down.
4818 	 *
4819 	 * e.g. NFS can fail the mount if it determines that the zone
4820 	 * has already begun the shutdown sequence.
4821 	 *
4822 	 */
4823 	if (block_mounts(zone) == 0) {
4824 		zone_rele(zone);
4825 		return (set_errno(EINTR));
4826 	}
4827 
4828 	mutex_enter(&zonehash_lock);
4829 	mutex_enter(&zone_status_lock);
4830 	status = zone_status_get(zone);
4831 	/*
4832 	 * Fail if the zone isn't fully initialized yet.
4833 	 */
4834 	if (status < ZONE_IS_READY) {
4835 		mutex_exit(&zone_status_lock);
4836 		mutex_exit(&zonehash_lock);
4837 		resume_mounts(zone);
4838 		zone_rele(zone);
4839 		return (set_errno(EINVAL));
4840 	}
4841 	/*
4842 	 * If conditions required for zone_shutdown() to return have been met,
4843 	 * return success.
4844 	 */
4845 	if (status >= ZONE_IS_DOWN) {
4846 		mutex_exit(&zone_status_lock);
4847 		mutex_exit(&zonehash_lock);
4848 		resume_mounts(zone);
4849 		zone_rele(zone);
4850 		return (0);
4851 	}
4852 	/*
4853 	 * If zone_shutdown() hasn't been called before, go through the motions.
4854 	 * If it has, there's nothing to do but wait for the kernel threads to
4855 	 * drain.
4856 	 */
4857 	if (status < ZONE_IS_EMPTY) {
4858 		uint_t ntasks;
4859 
4860 		mutex_enter(&zone->zone_lock);
4861 		if ((ntasks = zone->zone_ntasks) != 1) {
4862 			/*
4863 			 * There's still stuff running.
4864 			 */
4865 			zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
4866 		}
4867 		mutex_exit(&zone->zone_lock);
4868 		if (ntasks == 1) {
4869 			/*
4870 			 * The only way to create another task is through
4871 			 * zone_enter(), which will block until we drop
4872 			 * zonehash_lock.  The zone is empty.
4873 			 */
4874 			if (zone->zone_kthreads == NULL) {
4875 				/*
4876 				 * Skip ahead to ZONE_IS_DOWN
4877 				 */
4878 				zone_status_set(zone, ZONE_IS_DOWN);
4879 			} else {
4880 				zone_status_set(zone, ZONE_IS_EMPTY);
4881 			}
4882 		}
4883 	}
4884 	mutex_exit(&zone_status_lock);
4885 	mutex_exit(&zonehash_lock);
4886 	resume_mounts(zone);
4887 
4888 	if (error = zone_empty(zone)) {
4889 		zone_rele(zone);
4890 		return (set_errno(error));
4891 	}
4892 	/*
4893 	 * After the zone status goes to ZONE_IS_DOWN this zone will no
4894 	 * longer be notified of changes to the pools configuration, so
4895 	 * in order to not end up with a stale pool pointer, we point
4896 	 * ourselves at the default pool and remove all resource
4897 	 * visibility.  This is especially important as the zone_t may
4898 	 * languish on the deathrow for a very long time waiting for
4899 	 * cred's to drain out.
4900 	 *
4901 	 * This rebinding of the zone can happen multiple times
4902 	 * (presumably due to interrupted or parallel systemcalls)
4903 	 * without any adverse effects.
4904 	 */
4905 	if (pool_lock_intr() != 0) {
4906 		zone_rele(zone);
4907 		return (set_errno(EINTR));
4908 	}
4909 	if (pool_state == POOL_ENABLED) {
4910 		mutex_enter(&cpu_lock);
4911 		zone_pool_set(zone, pool_default);
4912 		/*
4913 		 * The zone no longer needs to be able to see any cpus.
4914 		 */
4915 		zone_pset_set(zone, ZONE_PS_INVAL);
4916 		mutex_exit(&cpu_lock);
4917 	}
4918 	pool_unlock();
4919 
4920 	/*
4921 	 * ZSD shutdown callbacks can be executed multiple times, hence
4922 	 * it is safe to not be holding any locks across this call.
4923 	 */
4924 	zone_zsd_callbacks(zone, ZSD_SHUTDOWN);
4925 
4926 	mutex_enter(&zone_status_lock);
4927 	if (zone->zone_kthreads == NULL && zone_status_get(zone) < ZONE_IS_DOWN)
4928 		zone_status_set(zone, ZONE_IS_DOWN);
4929 	mutex_exit(&zone_status_lock);
4930 
4931 	/*
4932 	 * Wait for kernel threads to drain.
4933 	 */
4934 	if (!zone_status_wait_sig(zone, ZONE_IS_DOWN)) {
4935 		zone_rele(zone);
4936 		return (set_errno(EINTR));
4937 	}
4938 
4939 	/*
4940 	 * Zone can be become down/destroyable even if the above wait
4941 	 * returns EINTR, so any code added here may never execute.
4942 	 * (i.e. don't add code here)
4943 	 */
4944 
4945 	zone_rele(zone);
4946 	return (0);
4947 }
4948 
4949 /*
4950  * Log the specified zone's reference counts.  The caller should not be
4951  * holding the zone's zone_lock.
4952  */
4953 static void
4954 zone_log_refcounts(zone_t *zone)
4955 {
4956 	char *buffer;
4957 	char *buffer_position;
4958 	uint32_t buffer_size;
4959 	uint32_t index;
4960 	uint_t ref;
4961 	uint_t cred_ref;
4962 
4963 	/*
4964 	 * Construct a string representing the subsystem-specific reference
4965 	 * counts.  The counts are printed in ascending order by index into the
4966 	 * zone_t::zone_subsys_ref array.  The list will be surrounded by
4967 	 * square brackets [] and will only contain nonzero reference counts.
4968 	 *
4969 	 * The buffer will hold two square bracket characters plus ten digits,
4970 	 * one colon, one space, one comma, and some characters for a
4971 	 * subsystem name per subsystem-specific reference count.  (Unsigned 32-
4972 	 * bit integers have at most ten decimal digits.)  The last
4973 	 * reference count's comma is replaced by the closing square
4974 	 * bracket and a NULL character to terminate the string.
4975 	 *
4976 	 * NOTE: We have to grab the zone's zone_lock to create a consistent
4977 	 * snapshot of the zone's reference counters.
4978 	 *
4979 	 * First, figure out how much space the string buffer will need.
4980 	 * The buffer's size is stored in buffer_size.
4981 	 */
4982 	buffer_size = 2;			/* for the square brackets */
4983 	mutex_enter(&zone->zone_lock);
4984 	zone->zone_flags |= ZF_REFCOUNTS_LOGGED;
4985 	ref = zone->zone_ref;
4986 	cred_ref = zone->zone_cred_ref;
4987 	for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index)
4988 		if (zone->zone_subsys_ref[index] != 0)
4989 			buffer_size += strlen(zone_ref_subsys_names[index]) +
4990 			    13;
4991 	if (buffer_size == 2) {
4992 		/*
4993 		 * No subsystems had nonzero reference counts.  Don't bother
4994 		 * with allocating a buffer; just log the general-purpose and
4995 		 * credential reference counts.
4996 		 */
4997 		mutex_exit(&zone->zone_lock);
4998 		(void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE,
4999 		    "Zone '%s' (ID: %d) is shutting down, but %u zone "
5000 		    "references and %u credential references are still extant",
5001 		    zone->zone_name, zone->zone_id, ref, cred_ref);
5002 		return;
5003 	}
5004 
5005 	/*
5006 	 * buffer_size contains the exact number of characters that the
5007 	 * buffer will need.  Allocate the buffer and fill it with nonzero
5008 	 * subsystem-specific reference counts.  Surround the results with
5009 	 * square brackets afterwards.
5010 	 */
5011 	buffer = kmem_alloc(buffer_size, KM_SLEEP);
5012 	buffer_position = &buffer[1];
5013 	for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index) {
5014 		/*
5015 		 * NOTE: The DDI's version of sprintf() returns a pointer to
5016 		 * the modified buffer rather than the number of bytes written
5017 		 * (as in snprintf(3C)).  This is unfortunate and annoying.
5018 		 * Therefore, we'll use snprintf() with INT_MAX to get the
5019 		 * number of bytes written.  Using INT_MAX is safe because
5020 		 * the buffer is perfectly sized for the data: we'll never
5021 		 * overrun the buffer.
5022 		 */
5023 		if (zone->zone_subsys_ref[index] != 0)
5024 			buffer_position += snprintf(buffer_position, INT_MAX,
5025 			    "%s: %u,", zone_ref_subsys_names[index],
5026 			    zone->zone_subsys_ref[index]);
5027 	}
5028 	mutex_exit(&zone->zone_lock);
5029 	buffer[0] = '[';
5030 	ASSERT((uintptr_t)(buffer_position - buffer) < buffer_size);
5031 	ASSERT(buffer_position[0] == '\0' && buffer_position[-1] == ',');
5032 	buffer_position[-1] = ']';
5033 
5034 	/*
5035 	 * Log the reference counts and free the message buffer.
5036 	 */
5037 	(void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE,
5038 	    "Zone '%s' (ID: %d) is shutting down, but %u zone references and "
5039 	    "%u credential references are still extant %s", zone->zone_name,
5040 	    zone->zone_id, ref, cred_ref, buffer);
5041 	kmem_free(buffer, buffer_size);
5042 }
5043 
5044 /*
5045  * Systemcall entry point to finalize the zone halt process.  The caller
5046  * must have already successfully called zone_shutdown().
5047  *
5048  * Upon successful completion, the zone will have been fully destroyed:
5049  * zsched will have exited, destructor callbacks executed, and the zone
5050  * removed from the list of active zones.
5051  */
5052 static int
5053 zone_destroy(zoneid_t zoneid)
5054 {
5055 	uint64_t uniqid;
5056 	zone_t *zone;
5057 	zone_status_t status;
5058 	clock_t wait_time;
5059 	boolean_t log_refcounts;
5060 
5061 	if (secpolicy_zone_config(CRED()) != 0)
5062 		return (set_errno(EPERM));
5063 	if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
5064 		return (set_errno(EINVAL));
5065 
5066 	mutex_enter(&zonehash_lock);
5067 	/*
5068 	 * Look for zone under hash lock to prevent races with other
5069 	 * calls to zone_destroy.
5070 	 */
5071 	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5072 		mutex_exit(&zonehash_lock);
5073 		return (set_errno(EINVAL));
5074 	}
5075 
5076 	if (zone_mount_count(zone->zone_rootpath) != 0) {
5077 		mutex_exit(&zonehash_lock);
5078 		return (set_errno(EBUSY));
5079 	}
5080 	mutex_enter(&zone_status_lock);
5081 	status = zone_status_get(zone);
5082 	if (status < ZONE_IS_DOWN) {
5083 		mutex_exit(&zone_status_lock);
5084 		mutex_exit(&zonehash_lock);
5085 		return (set_errno(EBUSY));
5086 	} else if (status == ZONE_IS_DOWN) {
5087 		zone_status_set(zone, ZONE_IS_DYING); /* Tell zsched to exit */
5088 	}
5089 	mutex_exit(&zone_status_lock);
5090 	zone_hold(zone);
5091 	mutex_exit(&zonehash_lock);
5092 
5093 	/*
5094 	 * wait for zsched to exit
5095 	 */
5096 	zone_status_wait(zone, ZONE_IS_DEAD);
5097 	zone_zsd_callbacks(zone, ZSD_DESTROY);
5098 	zone->zone_netstack = NULL;
5099 	uniqid = zone->zone_uniqid;
5100 	zone_rele(zone);
5101 	zone = NULL;	/* potentially free'd */
5102 
5103 	log_refcounts = B_FALSE;
5104 	wait_time = SEC_TO_TICK(ZONE_DESTROY_TIMEOUT_SECS);
5105 	mutex_enter(&zonehash_lock);
5106 	for (; /* ever */; ) {
5107 		boolean_t unref;
5108 		boolean_t refs_have_been_logged;
5109 
5110 		if ((zone = zone_find_all_by_id(zoneid)) == NULL ||
5111 		    zone->zone_uniqid != uniqid) {
5112 			/*
5113 			 * The zone has gone away.  Necessary conditions
5114 			 * are met, so we return success.
5115 			 */
5116 			mutex_exit(&zonehash_lock);
5117 			return (0);
5118 		}
5119 		mutex_enter(&zone->zone_lock);
5120 		unref = ZONE_IS_UNREF(zone);
5121 		refs_have_been_logged = (zone->zone_flags &
5122 		    ZF_REFCOUNTS_LOGGED);
5123 		mutex_exit(&zone->zone_lock);
5124 		if (unref) {
5125 			/*
5126 			 * There is only one reference to the zone -- that
5127 			 * added when the zone was added to the hashtables --
5128 			 * and things will remain this way until we drop
5129 			 * zonehash_lock... we can go ahead and cleanup the
5130 			 * zone.
5131 			 */
5132 			break;
5133 		}
5134 
5135 		/*
5136 		 * Wait for zone_rele_common() or zone_cred_rele() to signal
5137 		 * zone_destroy_cv.  zone_destroy_cv is signaled only when
5138 		 * some zone's general-purpose reference count reaches one.
5139 		 * If ZONE_DESTROY_TIMEOUT_SECS seconds elapse while waiting
5140 		 * on zone_destroy_cv, then log the zone's reference counts and
5141 		 * continue to wait for zone_rele() and zone_cred_rele().
5142 		 */
5143 		if (!refs_have_been_logged) {
5144 			if (!log_refcounts) {
5145 				/*
5146 				 * This thread hasn't timed out waiting on
5147 				 * zone_destroy_cv yet.  Wait wait_time clock
5148 				 * ticks (initially ZONE_DESTROY_TIMEOUT_SECS
5149 				 * seconds) for the zone's references to clear.
5150 				 */
5151 				ASSERT(wait_time > 0);
5152 				wait_time = cv_reltimedwait_sig(
5153 				    &zone_destroy_cv, &zonehash_lock, wait_time,
5154 				    TR_SEC);
5155 				if (wait_time > 0) {
5156 					/*
5157 					 * A thread in zone_rele() or
5158 					 * zone_cred_rele() signaled
5159 					 * zone_destroy_cv before this thread's
5160 					 * wait timed out.  The zone might have
5161 					 * only one reference left; find out!
5162 					 */
5163 					continue;
5164 				} else if (wait_time == 0) {
5165 					/* The thread's process was signaled. */
5166 					mutex_exit(&zonehash_lock);
5167 					return (set_errno(EINTR));
5168 				}
5169 
5170 				/*
5171 				 * The thread timed out while waiting on
5172 				 * zone_destroy_cv.  Even though the thread
5173 				 * timed out, it has to check whether another
5174 				 * thread woke up from zone_destroy_cv and
5175 				 * destroyed the zone.
5176 				 *
5177 				 * If the zone still exists and has more than
5178 				 * one unreleased general-purpose reference,
5179 				 * then log the zone's reference counts.
5180 				 */
5181 				log_refcounts = B_TRUE;
5182 				continue;
5183 			}
5184 
5185 			/*
5186 			 * The thread already timed out on zone_destroy_cv while
5187 			 * waiting for subsystems to release the zone's last
5188 			 * general-purpose references.  Log the zone's reference
5189 			 * counts and wait indefinitely on zone_destroy_cv.
5190 			 */
5191 			zone_log_refcounts(zone);
5192 		}
5193 		if (cv_wait_sig(&zone_destroy_cv, &zonehash_lock) == 0) {
5194 			/* The thread's process was signaled. */
5195 			mutex_exit(&zonehash_lock);
5196 			return (set_errno(EINTR));
5197 		}
5198 	}
5199 
5200 	/*
5201 	 * Remove CPU cap for this zone now since we're not going to
5202 	 * fail below this point.
5203 	 */
5204 	cpucaps_zone_remove(zone);
5205 
5206 	/* Get rid of the zone's kstats */
5207 	zone_kstat_delete(zone);
5208 
5209 	/* remove the pfexecd doors */
5210 	if (zone->zone_pfexecd != NULL) {
5211 		klpd_freelist(&zone->zone_pfexecd);
5212 		zone->zone_pfexecd = NULL;
5213 	}
5214 
5215 	/* free brand specific data */
5216 	if (ZONE_IS_BRANDED(zone))
5217 		ZBROP(zone)->b_free_brand_data(zone);
5218 
5219 	/* Say goodbye to brand framework. */
5220 	brand_unregister_zone(zone->zone_brand);
5221 
5222 	/*
5223 	 * It is now safe to let the zone be recreated; remove it from the
5224 	 * lists.  The memory will not be freed until the last cred
5225 	 * reference goes away.
5226 	 */
5227 	ASSERT(zonecount > 1);	/* must be > 1; can't destroy global zone */
5228 	zonecount--;
5229 	/* remove from active list and hash tables */
5230 	list_remove(&zone_active, zone);
5231 	(void) mod_hash_destroy(zonehashbyname,
5232 	    (mod_hash_key_t)zone->zone_name);
5233 	(void) mod_hash_destroy(zonehashbyid,
5234 	    (mod_hash_key_t)(uintptr_t)zone->zone_id);
5235 	if (zone->zone_flags & ZF_HASHED_LABEL)
5236 		(void) mod_hash_destroy(zonehashbylabel,
5237 		    (mod_hash_key_t)zone->zone_slabel);
5238 	mutex_exit(&zonehash_lock);
5239 
5240 	/*
5241 	 * Release the root vnode; we're not using it anymore.  Nor should any
5242 	 * other thread that might access it exist.
5243 	 */
5244 	if (zone->zone_rootvp != NULL) {
5245 		VN_RELE(zone->zone_rootvp);
5246 		zone->zone_rootvp = NULL;
5247 	}
5248 
5249 	/* add to deathrow list */
5250 	mutex_enter(&zone_deathrow_lock);
5251 	list_insert_tail(&zone_deathrow, zone);
5252 	mutex_exit(&zone_deathrow_lock);
5253 
5254 	/*
5255 	 * Drop last reference (which was added by zsched()), this will
5256 	 * free the zone unless there are outstanding cred references.
5257 	 */
5258 	zone_rele(zone);
5259 	return (0);
5260 }
5261 
5262 /*
5263  * Systemcall entry point for zone_getattr(2).
5264  */
5265 static ssize_t
5266 zone_getattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
5267 {
5268 	size_t size;
5269 	int error = 0, err;
5270 	zone_t *zone;
5271 	char *zonepath;
5272 	char *outstr;
5273 	zone_status_t zone_status;
5274 	pid_t initpid;
5275 	boolean_t global = (curzone == global_zone);
5276 	boolean_t inzone = (curzone->zone_id == zoneid);
5277 	ushort_t flags;
5278 	zone_net_data_t *zbuf;
5279 
5280 	mutex_enter(&zonehash_lock);
5281 	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5282 		mutex_exit(&zonehash_lock);
5283 		return (set_errno(EINVAL));
5284 	}
5285 	zone_status = zone_status_get(zone);
5286 	if (zone_status < ZONE_IS_INITIALIZED) {
5287 		mutex_exit(&zonehash_lock);
5288 		return (set_errno(EINVAL));
5289 	}
5290 	zone_hold(zone);
5291 	mutex_exit(&zonehash_lock);
5292 
5293 	/*
5294 	 * If not in the global zone, don't show information about other zones,
5295 	 * unless the system is labeled and the local zone's label dominates
5296 	 * the other zone.
5297 	 */
5298 	if (!zone_list_access(zone)) {
5299 		zone_rele(zone);
5300 		return (set_errno(EINVAL));
5301 	}
5302 
5303 	switch (attr) {
5304 	case ZONE_ATTR_ROOT:
5305 		if (global) {
5306 			/*
5307 			 * Copy the path to trim the trailing "/" (except for
5308 			 * the global zone).
5309 			 */
5310 			if (zone != global_zone)
5311 				size = zone->zone_rootpathlen - 1;
5312 			else
5313 				size = zone->zone_rootpathlen;
5314 			zonepath = kmem_alloc(size, KM_SLEEP);
5315 			bcopy(zone->zone_rootpath, zonepath, size);
5316 			zonepath[size - 1] = '\0';
5317 		} else {
5318 			if (inzone || !is_system_labeled()) {
5319 				/*
5320 				 * Caller is not in the global zone.
5321 				 * if the query is on the current zone
5322 				 * or the system is not labeled,
5323 				 * just return faked-up path for current zone.
5324 				 */
5325 				zonepath = "/";
5326 				size = 2;
5327 			} else {
5328 				/*
5329 				 * Return related path for current zone.
5330 				 */
5331 				int prefix_len = strlen(zone_prefix);
5332 				int zname_len = strlen(zone->zone_name);
5333 
5334 				size = prefix_len + zname_len + 1;
5335 				zonepath = kmem_alloc(size, KM_SLEEP);
5336 				bcopy(zone_prefix, zonepath, prefix_len);
5337 				bcopy(zone->zone_name, zonepath +
5338 				    prefix_len, zname_len);
5339 				zonepath[size - 1] = '\0';
5340 			}
5341 		}
5342 		if (bufsize > size)
5343 			bufsize = size;
5344 		if (buf != NULL) {
5345 			err = copyoutstr(zonepath, buf, bufsize, NULL);
5346 			if (err != 0 && err != ENAMETOOLONG)
5347 				error = EFAULT;
5348 		}
5349 		if (global || (is_system_labeled() && !inzone))
5350 			kmem_free(zonepath, size);
5351 		break;
5352 
5353 	case ZONE_ATTR_NAME:
5354 		size = strlen(zone->zone_name) + 1;
5355 		if (bufsize > size)
5356 			bufsize = size;
5357 		if (buf != NULL) {
5358 			err = copyoutstr(zone->zone_name, buf, bufsize, NULL);
5359 			if (err != 0 && err != ENAMETOOLONG)
5360 				error = EFAULT;
5361 		}
5362 		break;
5363 
5364 	case ZONE_ATTR_STATUS:
5365 		/*
5366 		 * Since we're not holding zonehash_lock, the zone status
5367 		 * may be anything; leave it up to userland to sort it out.
5368 		 */
5369 		size = sizeof (zone_status);
5370 		if (bufsize > size)
5371 			bufsize = size;
5372 		zone_status = zone_status_get(zone);
5373 		if (buf != NULL &&
5374 		    copyout(&zone_status, buf, bufsize) != 0)
5375 			error = EFAULT;
5376 		break;
5377 	case ZONE_ATTR_FLAGS:
5378 		size = sizeof (zone->zone_flags);
5379 		if (bufsize > size)
5380 			bufsize = size;
5381 		flags = zone->zone_flags;
5382 		if (buf != NULL &&
5383 		    copyout(&flags, buf, bufsize) != 0)
5384 			error = EFAULT;
5385 		break;
5386 	case ZONE_ATTR_PRIVSET:
5387 		size = sizeof (priv_set_t);
5388 		if (bufsize > size)
5389 			bufsize = size;
5390 		if (buf != NULL &&
5391 		    copyout(zone->zone_privset, buf, bufsize) != 0)
5392 			error = EFAULT;
5393 		break;
5394 	case ZONE_ATTR_UNIQID:
5395 		size = sizeof (zone->zone_uniqid);
5396 		if (bufsize > size)
5397 			bufsize = size;
5398 		if (buf != NULL &&
5399 		    copyout(&zone->zone_uniqid, buf, bufsize) != 0)
5400 			error = EFAULT;
5401 		break;
5402 	case ZONE_ATTR_POOLID:
5403 		{
5404 			pool_t *pool;
5405 			poolid_t poolid;
5406 
5407 			if (pool_lock_intr() != 0) {
5408 				error = EINTR;
5409 				break;
5410 			}
5411 			pool = zone_pool_get(zone);
5412 			poolid = pool->pool_id;
5413 			pool_unlock();
5414 			size = sizeof (poolid);
5415 			if (bufsize > size)
5416 				bufsize = size;
5417 			if (buf != NULL && copyout(&poolid, buf, size) != 0)
5418 				error = EFAULT;
5419 		}
5420 		break;
5421 	case ZONE_ATTR_SLBL:
5422 		size = sizeof (bslabel_t);
5423 		if (bufsize > size)
5424 			bufsize = size;
5425 		if (zone->zone_slabel == NULL)
5426 			error = EINVAL;
5427 		else if (buf != NULL &&
5428 		    copyout(label2bslabel(zone->zone_slabel), buf,
5429 		    bufsize) != 0)
5430 			error = EFAULT;
5431 		break;
5432 	case ZONE_ATTR_INITPID:
5433 		size = sizeof (initpid);
5434 		if (bufsize > size)
5435 			bufsize = size;
5436 		initpid = zone->zone_proc_initpid;
5437 		if (initpid == -1) {
5438 			error = ESRCH;
5439 			break;
5440 		}
5441 		if (buf != NULL &&
5442 		    copyout(&initpid, buf, bufsize) != 0)
5443 			error = EFAULT;
5444 		break;
5445 	case ZONE_ATTR_BRAND:
5446 		size = strlen(zone->zone_brand->b_name) + 1;
5447 
5448 		if (bufsize > size)
5449 			bufsize = size;
5450 		if (buf != NULL) {
5451 			err = copyoutstr(zone->zone_brand->b_name, buf,
5452 			    bufsize, NULL);
5453 			if (err != 0 && err != ENAMETOOLONG)
5454 				error = EFAULT;
5455 		}
5456 		break;
5457 	case ZONE_ATTR_INITNAME:
5458 		size = strlen(zone->zone_initname) + 1;
5459 		if (bufsize > size)
5460 			bufsize = size;
5461 		if (buf != NULL) {
5462 			err = copyoutstr(zone->zone_initname, buf, bufsize,
5463 			    NULL);
5464 			if (err != 0 && err != ENAMETOOLONG)
5465 				error = EFAULT;
5466 		}
5467 		break;
5468 	case ZONE_ATTR_BOOTARGS:
5469 		if (zone->zone_bootargs == NULL)
5470 			outstr = "";
5471 		else
5472 			outstr = zone->zone_bootargs;
5473 		size = strlen(outstr) + 1;
5474 		if (bufsize > size)
5475 			bufsize = size;
5476 		if (buf != NULL) {
5477 			err = copyoutstr(outstr, buf, bufsize, NULL);
5478 			if (err != 0 && err != ENAMETOOLONG)
5479 				error = EFAULT;
5480 		}
5481 		break;
5482 	case ZONE_ATTR_PHYS_MCAP:
5483 		size = sizeof (zone->zone_phys_mcap);
5484 		if (bufsize > size)
5485 			bufsize = size;
5486 		if (buf != NULL &&
5487 		    copyout(&zone->zone_phys_mcap, buf, bufsize) != 0)
5488 			error = EFAULT;
5489 		break;
5490 	case ZONE_ATTR_SCHED_CLASS:
5491 		mutex_enter(&class_lock);
5492 
5493 		if (zone->zone_defaultcid >= loaded_classes)
5494 			outstr = "";
5495 		else
5496 			outstr = sclass[zone->zone_defaultcid].cl_name;
5497 		size = strlen(outstr) + 1;
5498 		if (bufsize > size)
5499 			bufsize = size;
5500 		if (buf != NULL) {
5501 			err = copyoutstr(outstr, buf, bufsize, NULL);
5502 			if (err != 0 && err != ENAMETOOLONG)
5503 				error = EFAULT;
5504 		}
5505 
5506 		mutex_exit(&class_lock);
5507 		break;
5508 	case ZONE_ATTR_HOSTID:
5509 		if (zone->zone_hostid != HW_INVALID_HOSTID &&
5510 		    bufsize == sizeof (zone->zone_hostid)) {
5511 			size = sizeof (zone->zone_hostid);
5512 			if (buf != NULL && copyout(&zone->zone_hostid, buf,
5513 			    bufsize) != 0)
5514 				error = EFAULT;
5515 		} else {
5516 			error = EINVAL;
5517 		}
5518 		break;
5519 	case ZONE_ATTR_FS_ALLOWED:
5520 		if (zone->zone_fs_allowed == NULL)
5521 			outstr = "";
5522 		else
5523 			outstr = zone->zone_fs_allowed;
5524 		size = strlen(outstr) + 1;
5525 		if (bufsize > size)
5526 			bufsize = size;
5527 		if (buf != NULL) {
5528 			err = copyoutstr(outstr, buf, bufsize, NULL);
5529 			if (err != 0 && err != ENAMETOOLONG)
5530 				error = EFAULT;
5531 		}
5532 		break;
5533 	case ZONE_ATTR_NETWORK:
5534 		zbuf = kmem_alloc(bufsize, KM_SLEEP);
5535 		if (copyin(buf, zbuf, bufsize) != 0) {
5536 			error = EFAULT;
5537 		} else {
5538 			error = zone_get_network(zoneid, zbuf);
5539 			if (error == 0 && copyout(zbuf, buf, bufsize) != 0)
5540 				error = EFAULT;
5541 		}
5542 		kmem_free(zbuf, bufsize);
5543 		break;
5544 	default:
5545 		if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone)) {
5546 			size = bufsize;
5547 			error = ZBROP(zone)->b_getattr(zone, attr, buf, &size);
5548 		} else {
5549 			error = EINVAL;
5550 		}
5551 	}
5552 	zone_rele(zone);
5553 
5554 	if (error)
5555 		return (set_errno(error));
5556 	return ((ssize_t)size);
5557 }
5558 
5559 /*
5560  * Systemcall entry point for zone_setattr(2).
5561  */
5562 /*ARGSUSED*/
5563 static int
5564 zone_setattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
5565 {
5566 	zone_t *zone;
5567 	zone_status_t zone_status;
5568 	int err = -1;
5569 	zone_net_data_t *zbuf;
5570 
5571 	if (secpolicy_zone_config(CRED()) != 0)
5572 		return (set_errno(EPERM));
5573 
5574 	/*
5575 	 * Only the ZONE_ATTR_PHYS_MCAP attribute can be set on the
5576 	 * global zone.
5577 	 */
5578 	if (zoneid == GLOBAL_ZONEID && attr != ZONE_ATTR_PHYS_MCAP) {
5579 		return (set_errno(EINVAL));
5580 	}
5581 
5582 	mutex_enter(&zonehash_lock);
5583 	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5584 		mutex_exit(&zonehash_lock);
5585 		return (set_errno(EINVAL));
5586 	}
5587 	zone_hold(zone);
5588 	mutex_exit(&zonehash_lock);
5589 
5590 	/*
5591 	 * At present most attributes can only be set on non-running,
5592 	 * non-global zones.
5593 	 */
5594 	zone_status = zone_status_get(zone);
5595 	if (attr != ZONE_ATTR_PHYS_MCAP && zone_status > ZONE_IS_READY) {
5596 		err = EINVAL;
5597 		goto done;
5598 	}
5599 
5600 	switch (attr) {
5601 	case ZONE_ATTR_INITNAME:
5602 		err = zone_set_initname(zone, (const char *)buf);
5603 		break;
5604 	case ZONE_ATTR_INITNORESTART:
5605 		zone->zone_restart_init = B_FALSE;
5606 		err = 0;
5607 		break;
5608 	case ZONE_ATTR_BOOTARGS:
5609 		err = zone_set_bootargs(zone, (const char *)buf);
5610 		break;
5611 	case ZONE_ATTR_BRAND:
5612 		err = zone_set_brand(zone, (const char *)buf);
5613 		break;
5614 	case ZONE_ATTR_FS_ALLOWED:
5615 		err = zone_set_fs_allowed(zone, (const char *)buf);
5616 		break;
5617 	case ZONE_ATTR_PHYS_MCAP:
5618 		err = zone_set_phys_mcap(zone, (const uint64_t *)buf);
5619 		break;
5620 	case ZONE_ATTR_SCHED_CLASS:
5621 		err = zone_set_sched_class(zone, (const char *)buf);
5622 		break;
5623 	case ZONE_ATTR_HOSTID:
5624 		if (bufsize == sizeof (zone->zone_hostid)) {
5625 			if (copyin(buf, &zone->zone_hostid, bufsize) == 0)
5626 				err = 0;
5627 			else
5628 				err = EFAULT;
5629 		} else {
5630 			err = EINVAL;
5631 		}
5632 		break;
5633 	case ZONE_ATTR_NETWORK:
5634 		if (bufsize > (PIPE_BUF + sizeof (zone_net_data_t))) {
5635 			err = EINVAL;
5636 			break;
5637 		}
5638 		zbuf = kmem_alloc(bufsize, KM_SLEEP);
5639 		if (copyin(buf, zbuf, bufsize) != 0) {
5640 			kmem_free(zbuf, bufsize);
5641 			err = EFAULT;
5642 			break;
5643 		}
5644 		err = zone_set_network(zoneid, zbuf);
5645 		kmem_free(zbuf, bufsize);
5646 		break;
5647 	default:
5648 		if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone))
5649 			err = ZBROP(zone)->b_setattr(zone, attr, buf, bufsize);
5650 		else
5651 			err = EINVAL;
5652 	}
5653 
5654 done:
5655 	zone_rele(zone);
5656 	ASSERT(err != -1);
5657 	return (err != 0 ? set_errno(err) : 0);
5658 }
5659 
5660 /*
5661  * Return zero if the process has at least one vnode mapped in to its
5662  * address space which shouldn't be allowed to change zones.
5663  *
5664  * Also return zero if the process has any shared mappings which reserve
5665  * swap.  This is because the counting for zone.max-swap does not allow swap
5666  * reservation to be shared between zones.  zone swap reservation is counted
5667  * on zone->zone_max_swap.
5668  */
5669 static int
5670 as_can_change_zones(void)
5671 {
5672 	proc_t *pp = curproc;
5673 	struct seg *seg;
5674 	struct as *as = pp->p_as;
5675 	vnode_t *vp;
5676 	int allow = 1;
5677 
5678 	ASSERT(pp->p_as != &kas);
5679 	AS_LOCK_ENTER(as, RW_READER);
5680 	for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) {
5681 
5682 		/*
5683 		 * Cannot enter zone with shared anon memory which
5684 		 * reserves swap.  See comment above.
5685 		 */
5686 		if (seg_can_change_zones(seg) == B_FALSE) {
5687 			allow = 0;
5688 			break;
5689 		}
5690 		/*
5691 		 * if we can't get a backing vnode for this segment then skip
5692 		 * it.
5693 		 */
5694 		vp = NULL;
5695 		if (SEGOP_GETVP(seg, seg->s_base, &vp) != 0 || vp == NULL)
5696 			continue;
5697 		if (!vn_can_change_zones(vp)) { /* bail on first match */
5698 			allow = 0;
5699 			break;
5700 		}
5701 	}
5702 	AS_LOCK_EXIT(as);
5703 	return (allow);
5704 }
5705 
5706 /*
5707  * Count swap reserved by curproc's address space
5708  */
5709 static size_t
5710 as_swresv(void)
5711 {
5712 	proc_t *pp = curproc;
5713 	struct seg *seg;
5714 	struct as *as = pp->p_as;
5715 	size_t swap = 0;
5716 
5717 	ASSERT(pp->p_as != &kas);
5718 	ASSERT(AS_WRITE_HELD(as));
5719 	for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg))
5720 		swap += seg_swresv(seg);
5721 
5722 	return (swap);
5723 }
5724 
5725 /*
5726  * Systemcall entry point for zone_enter().
5727  *
5728  * The current process is injected into said zone.  In the process
5729  * it will change its project membership, privileges, rootdir/cwd,
5730  * zone-wide rctls, and pool association to match those of the zone.
5731  *
5732  * The first zone_enter() called while the zone is in the ZONE_IS_READY
5733  * state will transition it to ZONE_IS_RUNNING.  Processes may only
5734  * enter a zone that is "ready" or "running".
5735  */
5736 static int
5737 zone_enter(zoneid_t zoneid)
5738 {
5739 	zone_t *zone;
5740 	vnode_t *vp;
5741 	proc_t *pp = curproc;
5742 	contract_t *ct;
5743 	cont_process_t *ctp;
5744 	task_t *tk, *oldtk;
5745 	kproject_t *zone_proj0;
5746 	cred_t *cr, *newcr;
5747 	pool_t *oldpool, *newpool;
5748 	sess_t *sp;
5749 	uid_t uid;
5750 	zone_status_t status;
5751 	int err = 0;
5752 	rctl_entity_p_t e;
5753 	size_t swap;
5754 	kthread_id_t t;
5755 
5756 	if (secpolicy_zone_config(CRED()) != 0)
5757 		return (set_errno(EPERM));
5758 	if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
5759 		return (set_errno(EINVAL));
5760 
5761 	/*
5762 	 * Stop all lwps so we don't need to hold a lock to look at
5763 	 * curproc->p_zone.  This needs to happen before we grab any
5764 	 * locks to avoid deadlock (another lwp in the process could
5765 	 * be waiting for the held lock).
5766 	 */
5767 	if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK))
5768 		return (set_errno(EINTR));
5769 
5770 	/*
5771 	 * Make sure we're not changing zones with files open or mapped in
5772 	 * to our address space which shouldn't be changing zones.
5773 	 */
5774 	if (!files_can_change_zones()) {
5775 		err = EBADF;
5776 		goto out;
5777 	}
5778 	if (!as_can_change_zones()) {
5779 		err = EFAULT;
5780 		goto out;
5781 	}
5782 
5783 	mutex_enter(&zonehash_lock);
5784 	if (pp->p_zone != global_zone) {
5785 		mutex_exit(&zonehash_lock);
5786 		err = EINVAL;
5787 		goto out;
5788 	}
5789 
5790 	zone = zone_find_all_by_id(zoneid);
5791 	if (zone == NULL) {
5792 		mutex_exit(&zonehash_lock);
5793 		err = EINVAL;
5794 		goto out;
5795 	}
5796 
5797 	/*
5798 	 * To prevent processes in a zone from holding contracts on
5799 	 * extrazonal resources, and to avoid process contract
5800 	 * memberships which span zones, contract holders and processes
5801 	 * which aren't the sole members of their encapsulating process
5802 	 * contracts are not allowed to zone_enter.
5803 	 */
5804 	ctp = pp->p_ct_process;
5805 	ct = &ctp->conp_contract;
5806 	mutex_enter(&ct->ct_lock);
5807 	mutex_enter(&pp->p_lock);
5808 	if ((avl_numnodes(&pp->p_ct_held) != 0) || (ctp->conp_nmembers != 1)) {
5809 		mutex_exit(&pp->p_lock);
5810 		mutex_exit(&ct->ct_lock);
5811 		mutex_exit(&zonehash_lock);
5812 		err = EINVAL;
5813 		goto out;
5814 	}
5815 
5816 	/*
5817 	 * Moreover, we don't allow processes whose encapsulating
5818 	 * process contracts have inherited extrazonal contracts.
5819 	 * While it would be easier to eliminate all process contracts
5820 	 * with inherited contracts, we need to be able to give a
5821 	 * restarted init (or other zone-penetrating process) its
5822 	 * predecessor's contracts.
5823 	 */
5824 	if (ctp->conp_ninherited != 0) {
5825 		contract_t *next;
5826 		for (next = list_head(&ctp->conp_inherited); next;
5827 		    next = list_next(&ctp->conp_inherited, next)) {
5828 			if (contract_getzuniqid(next) != zone->zone_uniqid) {
5829 				mutex_exit(&pp->p_lock);
5830 				mutex_exit(&ct->ct_lock);
5831 				mutex_exit(&zonehash_lock);
5832 				err = EINVAL;
5833 				goto out;
5834 			}
5835 		}
5836 	}
5837 
5838 	mutex_exit(&pp->p_lock);
5839 	mutex_exit(&ct->ct_lock);
5840 
5841 	status = zone_status_get(zone);
5842 	if (status < ZONE_IS_READY || status >= ZONE_IS_SHUTTING_DOWN) {
5843 		/*
5844 		 * Can't join
5845 		 */
5846 		mutex_exit(&zonehash_lock);
5847 		err = EINVAL;
5848 		goto out;
5849 	}
5850 
5851 	/*
5852 	 * Make sure new priv set is within the permitted set for caller
5853 	 */
5854 	if (!priv_issubset(zone->zone_privset, &CR_OPPRIV(CRED()))) {
5855 		mutex_exit(&zonehash_lock);
5856 		err = EPERM;
5857 		goto out;
5858 	}
5859 	/*
5860 	 * We want to momentarily drop zonehash_lock while we optimistically
5861 	 * bind curproc to the pool it should be running in.  This is safe
5862 	 * since the zone can't disappear (we have a hold on it).
5863 	 */
5864 	zone_hold(zone);
5865 	mutex_exit(&zonehash_lock);
5866 
5867 	/*
5868 	 * Grab pool_lock to keep the pools configuration from changing
5869 	 * and to stop ourselves from getting rebound to another pool
5870 	 * until we join the zone.
5871 	 */
5872 	if (pool_lock_intr() != 0) {
5873 		zone_rele(zone);
5874 		err = EINTR;
5875 		goto out;
5876 	}
5877 	ASSERT(secpolicy_pool(CRED()) == 0);
5878 	/*
5879 	 * Bind ourselves to the pool currently associated with the zone.
5880 	 */
5881 	oldpool = curproc->p_pool;
5882 	newpool = zone_pool_get(zone);
5883 	if (pool_state == POOL_ENABLED && newpool != oldpool &&
5884 	    (err = pool_do_bind(newpool, P_PID, P_MYID,
5885 	    POOL_BIND_ALL)) != 0) {
5886 		pool_unlock();
5887 		zone_rele(zone);
5888 		goto out;
5889 	}
5890 
5891 	/*
5892 	 * Grab cpu_lock now; we'll need it later when we call
5893 	 * task_join().
5894 	 */
5895 	mutex_enter(&cpu_lock);
5896 	mutex_enter(&zonehash_lock);
5897 	/*
5898 	 * Make sure the zone hasn't moved on since we dropped zonehash_lock.
5899 	 */
5900 	if (zone_status_get(zone) >= ZONE_IS_SHUTTING_DOWN) {
5901 		/*
5902 		 * Can't join anymore.
5903 		 */
5904 		mutex_exit(&zonehash_lock);
5905 		mutex_exit(&cpu_lock);
5906 		if (pool_state == POOL_ENABLED &&
5907 		    newpool != oldpool)
5908 			(void) pool_do_bind(oldpool, P_PID, P_MYID,
5909 			    POOL_BIND_ALL);
5910 		pool_unlock();
5911 		zone_rele(zone);
5912 		err = EINVAL;
5913 		goto out;
5914 	}
5915 
5916 	/*
5917 	 * a_lock must be held while transfering locked memory and swap
5918 	 * reservation from the global zone to the non global zone because
5919 	 * asynchronous faults on the processes' address space can lock
5920 	 * memory and reserve swap via MCL_FUTURE and MAP_NORESERVE
5921 	 * segments respectively.
5922 	 */
5923 	AS_LOCK_ENTER(pp->p_as, RW_WRITER);
5924 	swap = as_swresv();
5925 	mutex_enter(&pp->p_lock);
5926 	zone_proj0 = zone->zone_zsched->p_task->tk_proj;
5927 	/* verify that we do not exceed and task or lwp limits */
5928 	mutex_enter(&zone->zone_nlwps_lock);
5929 	/* add new lwps to zone and zone's proj0 */
5930 	zone_proj0->kpj_nlwps += pp->p_lwpcnt;
5931 	zone->zone_nlwps += pp->p_lwpcnt;
5932 	/* add 1 task to zone's proj0 */
5933 	zone_proj0->kpj_ntasks += 1;
5934 
5935 	zone_proj0->kpj_nprocs++;
5936 	zone->zone_nprocs++;
5937 	mutex_exit(&zone->zone_nlwps_lock);
5938 
5939 	mutex_enter(&zone->zone_mem_lock);
5940 	zone->zone_locked_mem += pp->p_locked_mem;
5941 	zone_proj0->kpj_data.kpd_locked_mem += pp->p_locked_mem;
5942 	zone->zone_max_swap += swap;
5943 	mutex_exit(&zone->zone_mem_lock);
5944 
5945 	mutex_enter(&(zone_proj0->kpj_data.kpd_crypto_lock));
5946 	zone_proj0->kpj_data.kpd_crypto_mem += pp->p_crypto_mem;
5947 	mutex_exit(&(zone_proj0->kpj_data.kpd_crypto_lock));
5948 
5949 	/* remove lwps and process from proc's old zone and old project */
5950 	mutex_enter(&pp->p_zone->zone_nlwps_lock);
5951 	pp->p_zone->zone_nlwps -= pp->p_lwpcnt;
5952 	pp->p_task->tk_proj->kpj_nlwps -= pp->p_lwpcnt;
5953 	pp->p_task->tk_proj->kpj_nprocs--;
5954 	pp->p_zone->zone_nprocs--;
5955 	mutex_exit(&pp->p_zone->zone_nlwps_lock);
5956 
5957 	mutex_enter(&pp->p_zone->zone_mem_lock);
5958 	pp->p_zone->zone_locked_mem -= pp->p_locked_mem;
5959 	pp->p_task->tk_proj->kpj_data.kpd_locked_mem -= pp->p_locked_mem;
5960 	pp->p_zone->zone_max_swap -= swap;
5961 	mutex_exit(&pp->p_zone->zone_mem_lock);
5962 
5963 	mutex_enter(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock));
5964 	pp->p_task->tk_proj->kpj_data.kpd_crypto_mem -= pp->p_crypto_mem;
5965 	mutex_exit(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock));
5966 
5967 	pp->p_flag |= SZONETOP;
5968 	pp->p_zone = zone;
5969 	mutex_exit(&pp->p_lock);
5970 	AS_LOCK_EXIT(pp->p_as);
5971 
5972 	/*
5973 	 * Joining the zone cannot fail from now on.
5974 	 *
5975 	 * This means that a lot of the following code can be commonized and
5976 	 * shared with zsched().
5977 	 */
5978 
5979 	/*
5980 	 * If the process contract fmri was inherited, we need to
5981 	 * flag this so that any contract status will not leak
5982 	 * extra zone information, svc_fmri in this case
5983 	 */
5984 	if (ctp->conp_svc_ctid != ct->ct_id) {
5985 		mutex_enter(&ct->ct_lock);
5986 		ctp->conp_svc_zone_enter = ct->ct_id;
5987 		mutex_exit(&ct->ct_lock);
5988 	}
5989 
5990 	/*
5991 	 * Reset the encapsulating process contract's zone.
5992 	 */
5993 	ASSERT(ct->ct_mzuniqid == GLOBAL_ZONEUNIQID);
5994 	contract_setzuniqid(ct, zone->zone_uniqid);
5995 
5996 	/*
5997 	 * Create a new task and associate the process with the project keyed
5998 	 * by (projid,zoneid).
5999 	 *
6000 	 * We might as well be in project 0; the global zone's projid doesn't
6001 	 * make much sense in a zone anyhow.
6002 	 *
6003 	 * This also increments zone_ntasks, and returns with p_lock held.
6004 	 */
6005 	tk = task_create(0, zone);
6006 	oldtk = task_join(tk, 0);
6007 	mutex_exit(&cpu_lock);
6008 
6009 	/*
6010 	 * call RCTLOP_SET functions on this proc
6011 	 */
6012 	e.rcep_p.zone = zone;
6013 	e.rcep_t = RCENTITY_ZONE;
6014 	(void) rctl_set_dup(NULL, NULL, pp, &e, zone->zone_rctls, NULL,
6015 	    RCD_CALLBACK);
6016 	mutex_exit(&pp->p_lock);
6017 
6018 	/*
6019 	 * We don't need to hold any of zsched's locks here; not only do we know
6020 	 * the process and zone aren't going away, we know its session isn't
6021 	 * changing either.
6022 	 *
6023 	 * By joining zsched's session here, we mimic the behavior in the
6024 	 * global zone of init's sid being the pid of sched.  We extend this
6025 	 * to all zlogin-like zone_enter()'ing processes as well.
6026 	 */
6027 	mutex_enter(&pidlock);
6028 	sp = zone->zone_zsched->p_sessp;
6029 	sess_hold(zone->zone_zsched);
6030 	mutex_enter(&pp->p_lock);
6031 	pgexit(pp);
6032 	sess_rele(pp->p_sessp, B_TRUE);
6033 	pp->p_sessp = sp;
6034 	pgjoin(pp, zone->zone_zsched->p_pidp);
6035 
6036 	/*
6037 	 * If any threads are scheduled to be placed on zone wait queue they
6038 	 * should abandon the idea since the wait queue is changing.
6039 	 * We need to be holding pidlock & p_lock to do this.
6040 	 */
6041 	if ((t = pp->p_tlist) != NULL) {
6042 		do {
6043 			thread_lock(t);
6044 			/*
6045 			 * Kick this thread so that he doesn't sit
6046 			 * on a wrong wait queue.
6047 			 */
6048 			if (ISWAITING(t))
6049 				setrun_locked(t);
6050 
6051 			if (t->t_schedflag & TS_ANYWAITQ)
6052 				t->t_schedflag &= ~ TS_ANYWAITQ;
6053 
6054 			thread_unlock(t);
6055 		} while ((t = t->t_forw) != pp->p_tlist);
6056 	}
6057 
6058 	/*
6059 	 * If there is a default scheduling class for the zone and it is not
6060 	 * the class we are currently in, change all of the threads in the
6061 	 * process to the new class.  We need to be holding pidlock & p_lock
6062 	 * when we call parmsset so this is a good place to do it.
6063 	 */
6064 	if (zone->zone_defaultcid > 0 &&
6065 	    zone->zone_defaultcid != curthread->t_cid) {
6066 		pcparms_t pcparms;
6067 
6068 		pcparms.pc_cid = zone->zone_defaultcid;
6069 		pcparms.pc_clparms[0] = 0;
6070 
6071 		/*
6072 		 * If setting the class fails, we still want to enter the zone.
6073 		 */
6074 		if ((t = pp->p_tlist) != NULL) {
6075 			do {
6076 				(void) parmsset(&pcparms, t);
6077 			} while ((t = t->t_forw) != pp->p_tlist);
6078 		}
6079 	}
6080 
6081 	mutex_exit(&pp->p_lock);
6082 	mutex_exit(&pidlock);
6083 
6084 	mutex_exit(&zonehash_lock);
6085 	/*
6086 	 * We're firmly in the zone; let pools progress.
6087 	 */
6088 	pool_unlock();
6089 	task_rele(oldtk);
6090 	/*
6091 	 * We don't need to retain a hold on the zone since we already
6092 	 * incremented zone_ntasks, so the zone isn't going anywhere.
6093 	 */
6094 	zone_rele(zone);
6095 
6096 	/*
6097 	 * Chroot
6098 	 */
6099 	vp = zone->zone_rootvp;
6100 	zone_chdir(vp, &PTOU(pp)->u_cdir, pp);
6101 	zone_chdir(vp, &PTOU(pp)->u_rdir, pp);
6102 
6103 	/*
6104 	 * Change process credentials
6105 	 */
6106 	newcr = cralloc();
6107 	mutex_enter(&pp->p_crlock);
6108 	cr = pp->p_cred;
6109 	crcopy_to(cr, newcr);
6110 	crsetzone(newcr, zone);
6111 	pp->p_cred = newcr;
6112 
6113 	/*
6114 	 * Restrict all process privilege sets to zone limit
6115 	 */
6116 	priv_intersect(zone->zone_privset, &CR_PPRIV(newcr));
6117 	priv_intersect(zone->zone_privset, &CR_EPRIV(newcr));
6118 	priv_intersect(zone->zone_privset, &CR_IPRIV(newcr));
6119 	priv_intersect(zone->zone_privset, &CR_LPRIV(newcr));
6120 	mutex_exit(&pp->p_crlock);
6121 	crset(pp, newcr);
6122 
6123 	/*
6124 	 * Adjust upcount to reflect zone entry.
6125 	 */
6126 	uid = crgetruid(newcr);
6127 	mutex_enter(&pidlock);
6128 	upcount_dec(uid, GLOBAL_ZONEID);
6129 	upcount_inc(uid, zoneid);
6130 	mutex_exit(&pidlock);
6131 
6132 	/*
6133 	 * Set up core file path and content.
6134 	 */
6135 	set_core_defaults();
6136 
6137 out:
6138 	/*
6139 	 * Let the other lwps continue.
6140 	 */
6141 	mutex_enter(&pp->p_lock);
6142 	if (curthread != pp->p_agenttp)
6143 		continuelwps(pp);
6144 	mutex_exit(&pp->p_lock);
6145 
6146 	return (err != 0 ? set_errno(err) : 0);
6147 }
6148 
6149 /*
6150  * Systemcall entry point for zone_list(2).
6151  *
6152  * Processes running in a (non-global) zone only see themselves.
6153  * On labeled systems, they see all zones whose label they dominate.
6154  */
6155 static int
6156 zone_list(zoneid_t *zoneidlist, uint_t *numzones)
6157 {
6158 	zoneid_t *zoneids;
6159 	zone_t *zone, *myzone;
6160 	uint_t user_nzones, real_nzones;
6161 	uint_t domi_nzones;
6162 	int error;
6163 
6164 	if (copyin(numzones, &user_nzones, sizeof (uint_t)) != 0)
6165 		return (set_errno(EFAULT));
6166 
6167 	myzone = curproc->p_zone;
6168 	if (myzone != global_zone) {
6169 		bslabel_t *mybslab;
6170 
6171 		if (!is_system_labeled()) {
6172 			/* just return current zone */
6173 			real_nzones = domi_nzones = 1;
6174 			zoneids = kmem_alloc(sizeof (zoneid_t), KM_SLEEP);
6175 			zoneids[0] = myzone->zone_id;
6176 		} else {
6177 			/* return all zones that are dominated */
6178 			mutex_enter(&zonehash_lock);
6179 			real_nzones = zonecount;
6180 			domi_nzones = 0;
6181 			if (real_nzones > 0) {
6182 				zoneids = kmem_alloc(real_nzones *
6183 				    sizeof (zoneid_t), KM_SLEEP);
6184 				mybslab = label2bslabel(myzone->zone_slabel);
6185 				for (zone = list_head(&zone_active);
6186 				    zone != NULL;
6187 				    zone = list_next(&zone_active, zone)) {
6188 					if (zone->zone_id == GLOBAL_ZONEID)
6189 						continue;
6190 					if (zone != myzone &&
6191 					    (zone->zone_flags & ZF_IS_SCRATCH))
6192 						continue;
6193 					/*
6194 					 * Note that a label always dominates
6195 					 * itself, so myzone is always included
6196 					 * in the list.
6197 					 */
6198 					if (bldominates(mybslab,
6199 					    label2bslabel(zone->zone_slabel))) {
6200 						zoneids[domi_nzones++] =
6201 						    zone->zone_id;
6202 					}
6203 				}
6204 			}
6205 			mutex_exit(&zonehash_lock);
6206 		}
6207 	} else {
6208 		mutex_enter(&zonehash_lock);
6209 		real_nzones = zonecount;
6210 		domi_nzones = 0;
6211 		if (real_nzones > 0) {
6212 			zoneids = kmem_alloc(real_nzones * sizeof (zoneid_t),
6213 			    KM_SLEEP);
6214 			for (zone = list_head(&zone_active); zone != NULL;
6215 			    zone = list_next(&zone_active, zone))
6216 				zoneids[domi_nzones++] = zone->zone_id;
6217 			ASSERT(domi_nzones == real_nzones);
6218 		}
6219 		mutex_exit(&zonehash_lock);
6220 	}
6221 
6222 	/*
6223 	 * If user has allocated space for fewer entries than we found, then
6224 	 * return only up to his limit.  Either way, tell him exactly how many
6225 	 * we found.
6226 	 */
6227 	if (domi_nzones < user_nzones)
6228 		user_nzones = domi_nzones;
6229 	error = 0;
6230 	if (copyout(&domi_nzones, numzones, sizeof (uint_t)) != 0) {
6231 		error = EFAULT;
6232 	} else if (zoneidlist != NULL && user_nzones != 0) {
6233 		if (copyout(zoneids, zoneidlist,
6234 		    user_nzones * sizeof (zoneid_t)) != 0)
6235 			error = EFAULT;
6236 	}
6237 
6238 	if (real_nzones > 0)
6239 		kmem_free(zoneids, real_nzones * sizeof (zoneid_t));
6240 
6241 	if (error != 0)
6242 		return (set_errno(error));
6243 	else
6244 		return (0);
6245 }
6246 
6247 /*
6248  * Systemcall entry point for zone_lookup(2).
6249  *
6250  * Non-global zones are only able to see themselves and (on labeled systems)
6251  * the zones they dominate.
6252  */
6253 static zoneid_t
6254 zone_lookup(const char *zone_name)
6255 {
6256 	char *kname;
6257 	zone_t *zone;
6258 	zoneid_t zoneid;
6259 	int err;
6260 
6261 	if (zone_name == NULL) {
6262 		/* return caller's zone id */
6263 		return (getzoneid());
6264 	}
6265 
6266 	kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
6267 	if ((err = copyinstr(zone_name, kname, ZONENAME_MAX, NULL)) != 0) {
6268 		kmem_free(kname, ZONENAME_MAX);
6269 		return (set_errno(err));
6270 	}
6271 
6272 	mutex_enter(&zonehash_lock);
6273 	zone = zone_find_all_by_name(kname);
6274 	kmem_free(kname, ZONENAME_MAX);
6275 	/*
6276 	 * In a non-global zone, can only lookup global and own name.
6277 	 * In Trusted Extensions zone label dominance rules apply.
6278 	 */
6279 	if (zone == NULL ||
6280 	    zone_status_get(zone) < ZONE_IS_READY ||
6281 	    !zone_list_access(zone)) {
6282 		mutex_exit(&zonehash_lock);
6283 		return (set_errno(EINVAL));
6284 	} else {
6285 		zoneid = zone->zone_id;
6286 		mutex_exit(&zonehash_lock);
6287 		return (zoneid);
6288 	}
6289 }
6290 
6291 static int
6292 zone_version(int *version_arg)
6293 {
6294 	int version = ZONE_SYSCALL_API_VERSION;
6295 
6296 	if (copyout(&version, version_arg, sizeof (int)) != 0)
6297 		return (set_errno(EFAULT));
6298 	return (0);
6299 }
6300 
6301 /* ARGSUSED */
6302 long
6303 zone(int cmd, void *arg1, void *arg2, void *arg3, void *arg4)
6304 {
6305 	zone_def zs;
6306 	int err;
6307 
6308 	switch (cmd) {
6309 	case ZONE_CREATE:
6310 		if (get_udatamodel() == DATAMODEL_NATIVE) {
6311 			if (copyin(arg1, &zs, sizeof (zone_def))) {
6312 				return (set_errno(EFAULT));
6313 			}
6314 		} else {
6315 #ifdef _SYSCALL32_IMPL
6316 			zone_def32 zs32;
6317 
6318 			if (copyin(arg1, &zs32, sizeof (zone_def32))) {
6319 				return (set_errno(EFAULT));
6320 			}
6321 			zs.zone_name =
6322 			    (const char *)(unsigned long)zs32.zone_name;
6323 			zs.zone_root =
6324 			    (const char *)(unsigned long)zs32.zone_root;
6325 			zs.zone_privs =
6326 			    (const struct priv_set *)
6327 			    (unsigned long)zs32.zone_privs;
6328 			zs.zone_privssz = zs32.zone_privssz;
6329 			zs.rctlbuf = (caddr_t)(unsigned long)zs32.rctlbuf;
6330 			zs.rctlbufsz = zs32.rctlbufsz;
6331 			zs.zfsbuf = (caddr_t)(unsigned long)zs32.zfsbuf;
6332 			zs.zfsbufsz = zs32.zfsbufsz;
6333 			zs.extended_error =
6334 			    (int *)(unsigned long)zs32.extended_error;
6335 			zs.match = zs32.match;
6336 			zs.doi = zs32.doi;
6337 			zs.label = (const bslabel_t *)(uintptr_t)zs32.label;
6338 			zs.flags = zs32.flags;
6339 #else
6340 			panic("get_udatamodel() returned bogus result\n");
6341 #endif
6342 		}
6343 
6344 		return (zone_create(zs.zone_name, zs.zone_root,
6345 		    zs.zone_privs, zs.zone_privssz,
6346 		    (caddr_t)zs.rctlbuf, zs.rctlbufsz,
6347 		    (caddr_t)zs.zfsbuf, zs.zfsbufsz,
6348 		    zs.extended_error, zs.match, zs.doi,
6349 		    zs.label, zs.flags));
6350 	case ZONE_BOOT:
6351 		return (zone_boot((zoneid_t)(uintptr_t)arg1));
6352 	case ZONE_DESTROY:
6353 		return (zone_destroy((zoneid_t)(uintptr_t)arg1));
6354 	case ZONE_GETATTR:
6355 		return (zone_getattr((zoneid_t)(uintptr_t)arg1,
6356 		    (int)(uintptr_t)arg2, arg3, (size_t)arg4));
6357 	case ZONE_SETATTR:
6358 		return (zone_setattr((zoneid_t)(uintptr_t)arg1,
6359 		    (int)(uintptr_t)arg2, arg3, (size_t)arg4));
6360 	case ZONE_ENTER:
6361 		return (zone_enter((zoneid_t)(uintptr_t)arg1));
6362 	case ZONE_LIST:
6363 		return (zone_list((zoneid_t *)arg1, (uint_t *)arg2));
6364 	case ZONE_SHUTDOWN:
6365 		return (zone_shutdown((zoneid_t)(uintptr_t)arg1));
6366 	case ZONE_LOOKUP:
6367 		return (zone_lookup((const char *)arg1));
6368 	case ZONE_VERSION:
6369 		return (zone_version((int *)arg1));
6370 	case ZONE_ADD_DATALINK:
6371 		return (zone_add_datalink((zoneid_t)(uintptr_t)arg1,
6372 		    (datalink_id_t)(uintptr_t)arg2));
6373 	case ZONE_DEL_DATALINK:
6374 		return (zone_remove_datalink((zoneid_t)(uintptr_t)arg1,
6375 		    (datalink_id_t)(uintptr_t)arg2));
6376 	case ZONE_CHECK_DATALINK: {
6377 		zoneid_t	zoneid;
6378 		boolean_t	need_copyout;
6379 
6380 		if (copyin(arg1, &zoneid, sizeof (zoneid)) != 0)
6381 			return (EFAULT);
6382 		need_copyout = (zoneid == ALL_ZONES);
6383 		err = zone_check_datalink(&zoneid,
6384 		    (datalink_id_t)(uintptr_t)arg2);
6385 		if (err == 0 && need_copyout) {
6386 			if (copyout(&zoneid, arg1, sizeof (zoneid)) != 0)
6387 				err = EFAULT;
6388 		}
6389 		return (err == 0 ? 0 : set_errno(err));
6390 	}
6391 	case ZONE_LIST_DATALINK:
6392 		return (zone_list_datalink((zoneid_t)(uintptr_t)arg1,
6393 		    (int *)arg2, (datalink_id_t *)(uintptr_t)arg3));
6394 	default:
6395 		return (set_errno(EINVAL));
6396 	}
6397 }
6398 
6399 struct zarg {
6400 	zone_t *zone;
6401 	zone_cmd_arg_t arg;
6402 };
6403 
6404 static int
6405 zone_lookup_door(const char *zone_name, door_handle_t *doorp)
6406 {
6407 	char *buf;
6408 	size_t buflen;
6409 	int error;
6410 
6411 	buflen = sizeof (ZONE_DOOR_PATH) + strlen(zone_name);
6412 	buf = kmem_alloc(buflen, KM_SLEEP);
6413 	(void) snprintf(buf, buflen, ZONE_DOOR_PATH, zone_name);
6414 	error = door_ki_open(buf, doorp);
6415 	kmem_free(buf, buflen);
6416 	return (error);
6417 }
6418 
6419 static void
6420 zone_release_door(door_handle_t *doorp)
6421 {
6422 	door_ki_rele(*doorp);
6423 	*doorp = NULL;
6424 }
6425 
6426 static void
6427 zone_ki_call_zoneadmd(struct zarg *zargp)
6428 {
6429 	door_handle_t door = NULL;
6430 	door_arg_t darg, save_arg;
6431 	char *zone_name;
6432 	size_t zone_namelen;
6433 	zoneid_t zoneid;
6434 	zone_t *zone;
6435 	zone_cmd_arg_t arg;
6436 	uint64_t uniqid;
6437 	size_t size;
6438 	int error;
6439 	int retry;
6440 
6441 	zone = zargp->zone;
6442 	arg = zargp->arg;
6443 	kmem_free(zargp, sizeof (*zargp));
6444 
6445 	zone_namelen = strlen(zone->zone_name) + 1;
6446 	zone_name = kmem_alloc(zone_namelen, KM_SLEEP);
6447 	bcopy(zone->zone_name, zone_name, zone_namelen);
6448 	zoneid = zone->zone_id;
6449 	uniqid = zone->zone_uniqid;
6450 	/*
6451 	 * zoneadmd may be down, but at least we can empty out the zone.
6452 	 * We can ignore the return value of zone_empty() since we're called
6453 	 * from a kernel thread and know we won't be delivered any signals.
6454 	 */
6455 	ASSERT(curproc == &p0);
6456 	(void) zone_empty(zone);
6457 	ASSERT(zone_status_get(zone) >= ZONE_IS_EMPTY);
6458 	zone_rele(zone);
6459 
6460 	size = sizeof (arg);
6461 	darg.rbuf = (char *)&arg;
6462 	darg.data_ptr = (char *)&arg;
6463 	darg.rsize = size;
6464 	darg.data_size = size;
6465 	darg.desc_ptr = NULL;
6466 	darg.desc_num = 0;
6467 
6468 	save_arg = darg;
6469 	/*
6470 	 * Since we're not holding a reference to the zone, any number of
6471 	 * things can go wrong, including the zone disappearing before we get a
6472 	 * chance to talk to zoneadmd.
6473 	 */
6474 	for (retry = 0; /* forever */; retry++) {
6475 		if (door == NULL &&
6476 		    (error = zone_lookup_door(zone_name, &door)) != 0) {
6477 			goto next;
6478 		}
6479 		ASSERT(door != NULL);
6480 
6481 		if ((error = door_ki_upcall_limited(door, &darg, NULL,
6482 		    SIZE_MAX, 0)) == 0) {
6483 			break;
6484 		}
6485 		switch (error) {
6486 		case EINTR:
6487 			/* FALLTHROUGH */
6488 		case EAGAIN:	/* process may be forking */
6489 			/*
6490 			 * Back off for a bit
6491 			 */
6492 			break;
6493 		case EBADF:
6494 			zone_release_door(&door);
6495 			if (zone_lookup_door(zone_name, &door) != 0) {
6496 				/*
6497 				 * zoneadmd may be dead, but it may come back to
6498 				 * life later.
6499 				 */
6500 				break;
6501 			}
6502 			break;
6503 		default:
6504 			cmn_err(CE_WARN,
6505 			    "zone_ki_call_zoneadmd: door_ki_upcall error %d\n",
6506 			    error);
6507 			goto out;
6508 		}
6509 next:
6510 		/*
6511 		 * If this isn't the same zone_t that we originally had in mind,
6512 		 * then this is the same as if two kadmin requests come in at
6513 		 * the same time: the first one wins.  This means we lose, so we
6514 		 * bail.
6515 		 */
6516 		if ((zone = zone_find_by_id(zoneid)) == NULL) {
6517 			/*
6518 			 * Problem is solved.
6519 			 */
6520 			break;
6521 		}
6522 		if (zone->zone_uniqid != uniqid) {
6523 			/*
6524 			 * zoneid recycled
6525 			 */
6526 			zone_rele(zone);
6527 			break;
6528 		}
6529 		/*
6530 		 * We could zone_status_timedwait(), but there doesn't seem to
6531 		 * be much point in doing that (plus, it would mean that
6532 		 * zone_free() isn't called until this thread exits).
6533 		 */
6534 		zone_rele(zone);
6535 		delay(hz);
6536 		darg = save_arg;
6537 	}
6538 out:
6539 	if (door != NULL) {
6540 		zone_release_door(&door);
6541 	}
6542 	kmem_free(zone_name, zone_namelen);
6543 	thread_exit();
6544 }
6545 
6546 /*
6547  * Entry point for uadmin() to tell the zone to go away or reboot.  Analog to
6548  * kadmin().  The caller is a process in the zone.
6549  *
6550  * In order to shutdown the zone, we will hand off control to zoneadmd
6551  * (running in the global zone) via a door.  We do a half-hearted job at
6552  * killing all processes in the zone, create a kernel thread to contact
6553  * zoneadmd, and make note of the "uniqid" of the zone.  The uniqid is
6554  * a form of generation number used to let zoneadmd (as well as
6555  * zone_destroy()) know exactly which zone they're re talking about.
6556  */
6557 int
6558 zone_kadmin(int cmd, int fcn, const char *mdep, cred_t *credp)
6559 {
6560 	struct zarg *zargp;
6561 	zone_cmd_t zcmd;
6562 	zone_t *zone;
6563 
6564 	zone = curproc->p_zone;
6565 	ASSERT(getzoneid() != GLOBAL_ZONEID);
6566 
6567 	switch (cmd) {
6568 	case A_SHUTDOWN:
6569 		switch (fcn) {
6570 		case AD_HALT:
6571 		case AD_POWEROFF:
6572 			zcmd = Z_HALT;
6573 			break;
6574 		case AD_BOOT:
6575 			zcmd = Z_REBOOT;
6576 			break;
6577 		case AD_IBOOT:
6578 		case AD_SBOOT:
6579 		case AD_SIBOOT:
6580 		case AD_NOSYNC:
6581 			return (ENOTSUP);
6582 		default:
6583 			return (EINVAL);
6584 		}
6585 		break;
6586 	case A_REBOOT:
6587 		zcmd = Z_REBOOT;
6588 		break;
6589 	case A_FTRACE:
6590 	case A_REMOUNT:
6591 	case A_FREEZE:
6592 	case A_DUMP:
6593 	case A_CONFIG:
6594 		return (ENOTSUP);
6595 	default:
6596 		ASSERT(cmd != A_SWAPCTL);	/* handled by uadmin() */
6597 		return (EINVAL);
6598 	}
6599 
6600 	if (secpolicy_zone_admin(credp, B_FALSE))
6601 		return (EPERM);
6602 	mutex_enter(&zone_status_lock);
6603 
6604 	/*
6605 	 * zone_status can't be ZONE_IS_EMPTY or higher since curproc
6606 	 * is in the zone.
6607 	 */
6608 	ASSERT(zone_status_get(zone) < ZONE_IS_EMPTY);
6609 	if (zone_status_get(zone) > ZONE_IS_RUNNING) {
6610 		/*
6611 		 * This zone is already on its way down.
6612 		 */
6613 		mutex_exit(&zone_status_lock);
6614 		return (0);
6615 	}
6616 	/*
6617 	 * Prevent future zone_enter()s
6618 	 */
6619 	zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
6620 	mutex_exit(&zone_status_lock);
6621 
6622 	/*
6623 	 * Kill everyone now and call zoneadmd later.
6624 	 * zone_ki_call_zoneadmd() will do a more thorough job of this
6625 	 * later.
6626 	 */
6627 	killall(zone->zone_id);
6628 	/*
6629 	 * Now, create the thread to contact zoneadmd and do the rest of the
6630 	 * work.  This thread can't be created in our zone otherwise
6631 	 * zone_destroy() would deadlock.
6632 	 */
6633 	zargp = kmem_zalloc(sizeof (*zargp), KM_SLEEP);
6634 	zargp->arg.cmd = zcmd;
6635 	zargp->arg.uniqid = zone->zone_uniqid;
6636 	zargp->zone = zone;
6637 	(void) strcpy(zargp->arg.locale, "C");
6638 	/* mdep was already copied in for us by uadmin */
6639 	if (mdep != NULL)
6640 		(void) strlcpy(zargp->arg.bootbuf, mdep,
6641 		    sizeof (zargp->arg.bootbuf));
6642 	zone_hold(zone);
6643 
6644 	(void) thread_create(NULL, 0, zone_ki_call_zoneadmd, zargp, 0, &p0,
6645 	    TS_RUN, minclsyspri);
6646 	exit(CLD_EXITED, 0);
6647 
6648 	return (EINVAL);
6649 }
6650 
6651 /*
6652  * Entry point so kadmin(A_SHUTDOWN, ...) can set the global zone's
6653  * status to ZONE_IS_SHUTTING_DOWN.
6654  *
6655  * This function also shuts down all running zones to ensure that they won't
6656  * fork new processes.
6657  */
6658 void
6659 zone_shutdown_global(void)
6660 {
6661 	zone_t *current_zonep;
6662 
6663 	ASSERT(INGLOBALZONE(curproc));
6664 	mutex_enter(&zonehash_lock);
6665 	mutex_enter(&zone_status_lock);
6666 
6667 	/* Modify the global zone's status first. */
6668 	ASSERT(zone_status_get(global_zone) == ZONE_IS_RUNNING);
6669 	zone_status_set(global_zone, ZONE_IS_SHUTTING_DOWN);
6670 
6671 	/*
6672 	 * Now change the states of all running zones to ZONE_IS_SHUTTING_DOWN.
6673 	 * We don't mark all zones with ZONE_IS_SHUTTING_DOWN because doing so
6674 	 * could cause assertions to fail (e.g., assertions about a zone's
6675 	 * state during initialization, readying, or booting) or produce races.
6676 	 * We'll let threads continue to initialize and ready new zones: they'll
6677 	 * fail to boot the new zones when they see that the global zone is
6678 	 * shutting down.
6679 	 */
6680 	for (current_zonep = list_head(&zone_active); current_zonep != NULL;
6681 	    current_zonep = list_next(&zone_active, current_zonep)) {
6682 		if (zone_status_get(current_zonep) == ZONE_IS_RUNNING)
6683 			zone_status_set(current_zonep, ZONE_IS_SHUTTING_DOWN);
6684 	}
6685 	mutex_exit(&zone_status_lock);
6686 	mutex_exit(&zonehash_lock);
6687 }
6688 
6689 /*
6690  * Returns true if the named dataset is visible in the current zone.
6691  * The 'write' parameter is set to 1 if the dataset is also writable.
6692  */
6693 int
6694 zone_dataset_visible(const char *dataset, int *write)
6695 {
6696 	static int zfstype = -1;
6697 	zone_dataset_t *zd;
6698 	size_t len;
6699 	zone_t *zone = curproc->p_zone;
6700 	const char *name = NULL;
6701 	vfs_t *vfsp = NULL;
6702 
6703 	if (dataset[0] == '\0')
6704 		return (0);
6705 
6706 	/*
6707 	 * Walk the list once, looking for datasets which match exactly, or
6708 	 * specify a dataset underneath an exported dataset.  If found, return
6709 	 * true and note that it is writable.
6710 	 */
6711 	for (zd = list_head(&zone->zone_datasets); zd != NULL;
6712 	    zd = list_next(&zone->zone_datasets, zd)) {
6713 
6714 		len = strlen(zd->zd_dataset);
6715 		if (strlen(dataset) >= len &&
6716 		    bcmp(dataset, zd->zd_dataset, len) == 0 &&
6717 		    (dataset[len] == '\0' || dataset[len] == '/' ||
6718 		    dataset[len] == '@')) {
6719 			if (write)
6720 				*write = 1;
6721 			return (1);
6722 		}
6723 	}
6724 
6725 	/*
6726 	 * Walk the list a second time, searching for datasets which are parents
6727 	 * of exported datasets.  These should be visible, but read-only.
6728 	 *
6729 	 * Note that we also have to support forms such as 'pool/dataset/', with
6730 	 * a trailing slash.
6731 	 */
6732 	for (zd = list_head(&zone->zone_datasets); zd != NULL;
6733 	    zd = list_next(&zone->zone_datasets, zd)) {
6734 
6735 		len = strlen(dataset);
6736 		if (dataset[len - 1] == '/')
6737 			len--;	/* Ignore trailing slash */
6738 		if (len < strlen(zd->zd_dataset) &&
6739 		    bcmp(dataset, zd->zd_dataset, len) == 0 &&
6740 		    zd->zd_dataset[len] == '/') {
6741 			if (write)
6742 				*write = 0;
6743 			return (1);
6744 		}
6745 	}
6746 
6747 	/*
6748 	 * We reach here if the given dataset is not found in the zone_dataset
6749 	 * list. Check if this dataset was added as a filesystem (ie. "add fs")
6750 	 * instead of delegation. For this we search for the dataset in the
6751 	 * zone_vfslist of this zone. If found, return true and note that it is
6752 	 * not writable.
6753 	 */
6754 
6755 	/*
6756 	 * Initialize zfstype if it is not initialized yet.
6757 	 */
6758 	if (zfstype == -1) {
6759 		struct vfssw *vswp = vfs_getvfssw("zfs");
6760 		zfstype = vswp - vfssw;
6761 		vfs_unrefvfssw(vswp);
6762 	}
6763 
6764 	vfs_list_read_lock();
6765 	vfsp = zone->zone_vfslist;
6766 	do {
6767 		ASSERT(vfsp);
6768 		if (vfsp->vfs_fstype == zfstype) {
6769 			name = refstr_value(vfsp->vfs_resource);
6770 
6771 			/*
6772 			 * Check if we have an exact match.
6773 			 */
6774 			if (strcmp(dataset, name) == 0) {
6775 				vfs_list_unlock();
6776 				if (write)
6777 					*write = 0;
6778 				return (1);
6779 			}
6780 			/*
6781 			 * We need to check if we are looking for parents of
6782 			 * a dataset. These should be visible, but read-only.
6783 			 */
6784 			len = strlen(dataset);
6785 			if (dataset[len - 1] == '/')
6786 				len--;
6787 
6788 			if (len < strlen(name) &&
6789 			    bcmp(dataset, name, len) == 0 && name[len] == '/') {
6790 				vfs_list_unlock();
6791 				if (write)
6792 					*write = 0;
6793 				return (1);
6794 			}
6795 		}
6796 		vfsp = vfsp->vfs_zone_next;
6797 	} while (vfsp != zone->zone_vfslist);
6798 
6799 	vfs_list_unlock();
6800 	return (0);
6801 }
6802 
6803 /*
6804  * zone_find_by_any_path() -
6805  *
6806  * kernel-private routine similar to zone_find_by_path(), but which
6807  * effectively compares against zone paths rather than zonerootpath
6808  * (i.e., the last component of zonerootpaths, which should be "root/",
6809  * are not compared.)  This is done in order to accurately identify all
6810  * paths, whether zone-visible or not, including those which are parallel
6811  * to /root/, such as /dev/, /home/, etc...
6812  *
6813  * If the specified path does not fall under any zone path then global
6814  * zone is returned.
6815  *
6816  * The treat_abs parameter indicates whether the path should be treated as
6817  * an absolute path although it does not begin with "/".  (This supports
6818  * nfs mount syntax such as host:any/path.)
6819  *
6820  * The caller is responsible for zone_rele of the returned zone.
6821  */
6822 zone_t *
6823 zone_find_by_any_path(const char *path, boolean_t treat_abs)
6824 {
6825 	zone_t *zone;
6826 	int path_offset = 0;
6827 
6828 	if (path == NULL) {
6829 		zone_hold(global_zone);
6830 		return (global_zone);
6831 	}
6832 
6833 	if (*path != '/') {
6834 		ASSERT(treat_abs);
6835 		path_offset = 1;
6836 	}
6837 
6838 	mutex_enter(&zonehash_lock);
6839 	for (zone = list_head(&zone_active); zone != NULL;
6840 	    zone = list_next(&zone_active, zone)) {
6841 		char	*c;
6842 		size_t	pathlen;
6843 		char *rootpath_start;
6844 
6845 		if (zone == global_zone)	/* skip global zone */
6846 			continue;
6847 
6848 		/* scan backwards to find start of last component */
6849 		c = zone->zone_rootpath + zone->zone_rootpathlen - 2;
6850 		do {
6851 			c--;
6852 		} while (*c != '/');
6853 
6854 		pathlen = c - zone->zone_rootpath + 1 - path_offset;
6855 		rootpath_start = (zone->zone_rootpath + path_offset);
6856 		if (strncmp(path, rootpath_start, pathlen) == 0)
6857 			break;
6858 	}
6859 	if (zone == NULL)
6860 		zone = global_zone;
6861 	zone_hold(zone);
6862 	mutex_exit(&zonehash_lock);
6863 	return (zone);
6864 }
6865 
6866 /*
6867  * Finds a zone_dl_t with the given linkid in the given zone.  Returns the
6868  * zone_dl_t pointer if found, and NULL otherwise.
6869  */
6870 static zone_dl_t *
6871 zone_find_dl(zone_t *zone, datalink_id_t linkid)
6872 {
6873 	zone_dl_t *zdl;
6874 
6875 	ASSERT(mutex_owned(&zone->zone_lock));
6876 	for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
6877 	    zdl = list_next(&zone->zone_dl_list, zdl)) {
6878 		if (zdl->zdl_id == linkid)
6879 			break;
6880 	}
6881 	return (zdl);
6882 }
6883 
6884 static boolean_t
6885 zone_dl_exists(zone_t *zone, datalink_id_t linkid)
6886 {
6887 	boolean_t exists;
6888 
6889 	mutex_enter(&zone->zone_lock);
6890 	exists = (zone_find_dl(zone, linkid) != NULL);
6891 	mutex_exit(&zone->zone_lock);
6892 	return (exists);
6893 }
6894 
6895 /*
6896  * Add an data link name for the zone.
6897  */
6898 static int
6899 zone_add_datalink(zoneid_t zoneid, datalink_id_t linkid)
6900 {
6901 	zone_dl_t *zdl;
6902 	zone_t *zone;
6903 	zone_t *thiszone;
6904 
6905 	if ((thiszone = zone_find_by_id(zoneid)) == NULL)
6906 		return (set_errno(ENXIO));
6907 
6908 	/* Verify that the datalink ID doesn't already belong to a zone. */
6909 	mutex_enter(&zonehash_lock);
6910 	for (zone = list_head(&zone_active); zone != NULL;
6911 	    zone = list_next(&zone_active, zone)) {
6912 		if (zone_dl_exists(zone, linkid)) {
6913 			mutex_exit(&zonehash_lock);
6914 			zone_rele(thiszone);
6915 			return (set_errno((zone == thiszone) ? EEXIST : EPERM));
6916 		}
6917 	}
6918 
6919 	zdl = kmem_zalloc(sizeof (*zdl), KM_SLEEP);
6920 	zdl->zdl_id = linkid;
6921 	zdl->zdl_net = NULL;
6922 	mutex_enter(&thiszone->zone_lock);
6923 	list_insert_head(&thiszone->zone_dl_list, zdl);
6924 	mutex_exit(&thiszone->zone_lock);
6925 	mutex_exit(&zonehash_lock);
6926 	zone_rele(thiszone);
6927 	return (0);
6928 }
6929 
6930 static int
6931 zone_remove_datalink(zoneid_t zoneid, datalink_id_t linkid)
6932 {
6933 	zone_dl_t *zdl;
6934 	zone_t *zone;
6935 	int err = 0;
6936 
6937 	if ((zone = zone_find_by_id(zoneid)) == NULL)
6938 		return (set_errno(EINVAL));
6939 
6940 	mutex_enter(&zone->zone_lock);
6941 	if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
6942 		err = ENXIO;
6943 	} else {
6944 		list_remove(&zone->zone_dl_list, zdl);
6945 		nvlist_free(zdl->zdl_net);
6946 		kmem_free(zdl, sizeof (zone_dl_t));
6947 	}
6948 	mutex_exit(&zone->zone_lock);
6949 	zone_rele(zone);
6950 	return (err == 0 ? 0 : set_errno(err));
6951 }
6952 
6953 /*
6954  * Using the zoneidp as ALL_ZONES, we can lookup which zone has been assigned
6955  * the linkid.  Otherwise we just check if the specified zoneidp has been
6956  * assigned the supplied linkid.
6957  */
6958 int
6959 zone_check_datalink(zoneid_t *zoneidp, datalink_id_t linkid)
6960 {
6961 	zone_t *zone;
6962 	int err = ENXIO;
6963 
6964 	if (*zoneidp != ALL_ZONES) {
6965 		if ((zone = zone_find_by_id(*zoneidp)) != NULL) {
6966 			if (zone_dl_exists(zone, linkid))
6967 				err = 0;
6968 			zone_rele(zone);
6969 		}
6970 		return (err);
6971 	}
6972 
6973 	mutex_enter(&zonehash_lock);
6974 	for (zone = list_head(&zone_active); zone != NULL;
6975 	    zone = list_next(&zone_active, zone)) {
6976 		if (zone_dl_exists(zone, linkid)) {
6977 			*zoneidp = zone->zone_id;
6978 			err = 0;
6979 			break;
6980 		}
6981 	}
6982 	mutex_exit(&zonehash_lock);
6983 	return (err);
6984 }
6985 
6986 /*
6987  * Get the list of datalink IDs assigned to a zone.
6988  *
6989  * On input, *nump is the number of datalink IDs that can fit in the supplied
6990  * idarray.  Upon return, *nump is either set to the number of datalink IDs
6991  * that were placed in the array if the array was large enough, or to the
6992  * number of datalink IDs that the function needs to place in the array if the
6993  * array is too small.
6994  */
6995 static int
6996 zone_list_datalink(zoneid_t zoneid, int *nump, datalink_id_t *idarray)
6997 {
6998 	uint_t num, dlcount;
6999 	zone_t *zone;
7000 	zone_dl_t *zdl;
7001 	datalink_id_t *idptr = idarray;
7002 
7003 	if (copyin(nump, &dlcount, sizeof (dlcount)) != 0)
7004 		return (set_errno(EFAULT));
7005 	if ((zone = zone_find_by_id(zoneid)) == NULL)
7006 		return (set_errno(ENXIO));
7007 
7008 	num = 0;
7009 	mutex_enter(&zone->zone_lock);
7010 	for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7011 	    zdl = list_next(&zone->zone_dl_list, zdl)) {
7012 		/*
7013 		 * If the list is bigger than what the caller supplied, just
7014 		 * count, don't do copyout.
7015 		 */
7016 		if (++num > dlcount)
7017 			continue;
7018 		if (copyout(&zdl->zdl_id, idptr, sizeof (*idptr)) != 0) {
7019 			mutex_exit(&zone->zone_lock);
7020 			zone_rele(zone);
7021 			return (set_errno(EFAULT));
7022 		}
7023 		idptr++;
7024 	}
7025 	mutex_exit(&zone->zone_lock);
7026 	zone_rele(zone);
7027 
7028 	/* Increased or decreased, caller should be notified. */
7029 	if (num != dlcount) {
7030 		if (copyout(&num, nump, sizeof (num)) != 0)
7031 			return (set_errno(EFAULT));
7032 	}
7033 	return (0);
7034 }
7035 
7036 /*
7037  * Public interface for looking up a zone by zoneid. It's a customized version
7038  * for netstack_zone_create(). It can only be called from the zsd create
7039  * callbacks, since it doesn't have reference on the zone structure hence if
7040  * it is called elsewhere the zone could disappear after the zonehash_lock
7041  * is dropped.
7042  *
7043  * Furthermore it
7044  * 1. Doesn't check the status of the zone.
7045  * 2. It will be called even before zone_init is called, in that case the
7046  *    address of zone0 is returned directly, and netstack_zone_create()
7047  *    will only assign a value to zone0.zone_netstack, won't break anything.
7048  * 3. Returns without the zone being held.
7049  */
7050 zone_t *
7051 zone_find_by_id_nolock(zoneid_t zoneid)
7052 {
7053 	zone_t *zone;
7054 
7055 	mutex_enter(&zonehash_lock);
7056 	if (zonehashbyid == NULL)
7057 		zone = &zone0;
7058 	else
7059 		zone = zone_find_all_by_id(zoneid);
7060 	mutex_exit(&zonehash_lock);
7061 	return (zone);
7062 }
7063 
7064 /*
7065  * Walk the datalinks for a given zone
7066  */
7067 int
7068 zone_datalink_walk(zoneid_t zoneid, int (*cb)(datalink_id_t, void *),
7069     void *data)
7070 {
7071 	zone_t		*zone;
7072 	zone_dl_t	*zdl;
7073 	datalink_id_t	*idarray;
7074 	uint_t		idcount = 0;
7075 	int		i, ret = 0;
7076 
7077 	if ((zone = zone_find_by_id(zoneid)) == NULL)
7078 		return (ENOENT);
7079 
7080 	/*
7081 	 * We first build an array of linkid's so that we can walk these and
7082 	 * execute the callback with the zone_lock dropped.
7083 	 */
7084 	mutex_enter(&zone->zone_lock);
7085 	for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7086 	    zdl = list_next(&zone->zone_dl_list, zdl)) {
7087 		idcount++;
7088 	}
7089 
7090 	if (idcount == 0) {
7091 		mutex_exit(&zone->zone_lock);
7092 		zone_rele(zone);
7093 		return (0);
7094 	}
7095 
7096 	idarray = kmem_alloc(sizeof (datalink_id_t) * idcount, KM_NOSLEEP);
7097 	if (idarray == NULL) {
7098 		mutex_exit(&zone->zone_lock);
7099 		zone_rele(zone);
7100 		return (ENOMEM);
7101 	}
7102 
7103 	for (i = 0, zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7104 	    i++, zdl = list_next(&zone->zone_dl_list, zdl)) {
7105 		idarray[i] = zdl->zdl_id;
7106 	}
7107 
7108 	mutex_exit(&zone->zone_lock);
7109 
7110 	for (i = 0; i < idcount && ret == 0; i++) {
7111 		if ((ret = (*cb)(idarray[i], data)) != 0)
7112 			break;
7113 	}
7114 
7115 	zone_rele(zone);
7116 	kmem_free(idarray, sizeof (datalink_id_t) * idcount);
7117 	return (ret);
7118 }
7119 
7120 static char *
7121 zone_net_type2name(int type)
7122 {
7123 	switch (type) {
7124 	case ZONE_NETWORK_ADDRESS:
7125 		return (ZONE_NET_ADDRNAME);
7126 	case ZONE_NETWORK_DEFROUTER:
7127 		return (ZONE_NET_RTRNAME);
7128 	default:
7129 		return (NULL);
7130 	}
7131 }
7132 
7133 static int
7134 zone_set_network(zoneid_t zoneid, zone_net_data_t *znbuf)
7135 {
7136 	zone_t *zone;
7137 	zone_dl_t *zdl;
7138 	nvlist_t *nvl;
7139 	int err = 0;
7140 	uint8_t *new = NULL;
7141 	char *nvname;
7142 	int bufsize;
7143 	datalink_id_t linkid = znbuf->zn_linkid;
7144 
7145 	if (secpolicy_zone_config(CRED()) != 0)
7146 		return (set_errno(EPERM));
7147 
7148 	if (zoneid == GLOBAL_ZONEID)
7149 		return (set_errno(EINVAL));
7150 
7151 	nvname = zone_net_type2name(znbuf->zn_type);
7152 	bufsize = znbuf->zn_len;
7153 	new = znbuf->zn_val;
7154 	if (nvname == NULL)
7155 		return (set_errno(EINVAL));
7156 
7157 	if ((zone = zone_find_by_id(zoneid)) == NULL) {
7158 		return (set_errno(EINVAL));
7159 	}
7160 
7161 	mutex_enter(&zone->zone_lock);
7162 	if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
7163 		err = ENXIO;
7164 		goto done;
7165 	}
7166 	if ((nvl = zdl->zdl_net) == NULL) {
7167 		if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP)) {
7168 			err = ENOMEM;
7169 			goto done;
7170 		} else {
7171 			zdl->zdl_net = nvl;
7172 		}
7173 	}
7174 	if (nvlist_exists(nvl, nvname)) {
7175 		err = EINVAL;
7176 		goto done;
7177 	}
7178 	err = nvlist_add_uint8_array(nvl, nvname, new, bufsize);
7179 	ASSERT(err == 0);
7180 done:
7181 	mutex_exit(&zone->zone_lock);
7182 	zone_rele(zone);
7183 	if (err != 0)
7184 		return (set_errno(err));
7185 	else
7186 		return (0);
7187 }
7188 
7189 static int
7190 zone_get_network(zoneid_t zoneid, zone_net_data_t *znbuf)
7191 {
7192 	zone_t *zone;
7193 	zone_dl_t *zdl;
7194 	nvlist_t *nvl;
7195 	uint8_t *ptr;
7196 	uint_t psize;
7197 	int err = 0;
7198 	char *nvname;
7199 	int bufsize;
7200 	void *buf;
7201 	datalink_id_t linkid = znbuf->zn_linkid;
7202 
7203 	if (zoneid == GLOBAL_ZONEID)
7204 		return (set_errno(EINVAL));
7205 
7206 	nvname = zone_net_type2name(znbuf->zn_type);
7207 	bufsize = znbuf->zn_len;
7208 	buf = znbuf->zn_val;
7209 
7210 	if (nvname == NULL)
7211 		return (set_errno(EINVAL));
7212 	if ((zone = zone_find_by_id(zoneid)) == NULL)
7213 		return (set_errno(EINVAL));
7214 
7215 	mutex_enter(&zone->zone_lock);
7216 	if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
7217 		err = ENXIO;
7218 		goto done;
7219 	}
7220 	if ((nvl = zdl->zdl_net) == NULL || !nvlist_exists(nvl, nvname)) {
7221 		err = ENOENT;
7222 		goto done;
7223 	}
7224 	err = nvlist_lookup_uint8_array(nvl, nvname, &ptr, &psize);
7225 	ASSERT(err == 0);
7226 
7227 	if (psize > bufsize) {
7228 		err = ENOBUFS;
7229 		goto done;
7230 	}
7231 	znbuf->zn_len = psize;
7232 	bcopy(ptr, buf, psize);
7233 done:
7234 	mutex_exit(&zone->zone_lock);
7235 	zone_rele(zone);
7236 	if (err != 0)
7237 		return (set_errno(err));
7238 	else
7239 		return (0);
7240 }
7241