xref: /titanic_52/usr/src/uts/common/os/zone.c (revision c2e7b48d563d0e56b74d853118918af352e75cbb)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 
23 /*
24  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
25  * Use is subject to license terms.
26  */
27 
28 #pragma ident	"%Z%%M%	%I%	%E% SMI"
29 
30 /*
31  * Zones
32  *
33  *   A zone is a named collection of processes, namespace constraints,
34  *   and other system resources which comprise a secure and manageable
35  *   application containment facility.
36  *
37  *   Zones (represented by the reference counted zone_t) are tracked in
38  *   the kernel in the zonehash.  Elsewhere in the kernel, Zone IDs
39  *   (zoneid_t) are used to track zone association.  Zone IDs are
40  *   dynamically generated when the zone is created; if a persistent
41  *   identifier is needed (core files, accounting logs, audit trail,
42  *   etc.), the zone name should be used.
43  *
44  *
45  *   Global Zone:
46  *
47  *   The global zone (zoneid 0) is automatically associated with all
48  *   system resources that have not been bound to a user-created zone.
49  *   This means that even systems where zones are not in active use
50  *   have a global zone, and all processes, mounts, etc. are
51  *   associated with that zone.  The global zone is generally
52  *   unconstrained in terms of privileges and access, though the usual
53  *   credential and privilege based restrictions apply.
54  *
55  *
56  *   Zone States:
57  *
58  *   The states in which a zone may be in and the transitions are as
59  *   follows:
60  *
61  *   ZONE_IS_UNINITIALIZED: primordial state for a zone. The partially
62  *   initialized zone is added to the list of active zones on the system but
63  *   isn't accessible.
64  *
65  *   ZONE_IS_READY: zsched (the kernel dummy process for a zone) is
66  *   ready.  The zone is made visible after the ZSD constructor callbacks are
67  *   executed.  A zone remains in this state until it transitions into
68  *   the ZONE_IS_BOOTING state as a result of a call to zone_boot().
69  *
70  *   ZONE_IS_BOOTING: in this shortlived-state, zsched attempts to start
71  *   init.  Should that fail, the zone proceeds to the ZONE_IS_SHUTTING_DOWN
72  *   state.
73  *
74  *   ZONE_IS_RUNNING: The zone is open for business: zsched has
75  *   successfully started init.   A zone remains in this state until
76  *   zone_shutdown() is called.
77  *
78  *   ZONE_IS_SHUTTING_DOWN: zone_shutdown() has been called, the system is
79  *   killing all processes running in the zone. The zone remains
80  *   in this state until there are no more user processes running in the zone.
81  *   zone_create(), zone_enter(), and zone_destroy() on this zone will fail.
82  *   Since zone_shutdown() is restartable, it may be called successfully
83  *   multiple times for the same zone_t.  Setting of the zone's state to
84  *   ZONE_IS_SHUTTING_DOWN is synchronized with mounts, so VOP_MOUNT() may check
85  *   the zone's status without worrying about it being a moving target.
86  *
87  *   ZONE_IS_EMPTY: zone_shutdown() has been called, and there
88  *   are no more user processes in the zone.  The zone remains in this
89  *   state until there are no more kernel threads associated with the
90  *   zone.  zone_create(), zone_enter(), and zone_destroy() on this zone will
91  *   fail.
92  *
93  *   ZONE_IS_DOWN: All kernel threads doing work on behalf of the zone
94  *   have exited.  zone_shutdown() returns.  Henceforth it is not possible to
95  *   join the zone or create kernel threads therein.
96  *
97  *   ZONE_IS_DYING: zone_destroy() has been called on the zone; zone
98  *   remains in this state until zsched exits.  Calls to zone_find_by_*()
99  *   return NULL from now on.
100  *
101  *   ZONE_IS_DEAD: zsched has exited (zone_ntasks == 0).  There are no
102  *   processes or threads doing work on behalf of the zone.  The zone is
103  *   removed from the list of active zones.  zone_destroy() returns, and
104  *   the zone can be recreated.
105  *
106  *   ZONE_IS_FREE (internal state): zone_ref goes to 0, ZSD destructor
107  *   callbacks are executed, and all memory associated with the zone is
108  *   freed.
109  *
110  *   Threads can wait for the zone to enter a requested state by using
111  *   zone_status_wait() or zone_status_timedwait() with the desired
112  *   state passed in as an argument.  Zone state transitions are
113  *   uni-directional; it is not possible to move back to an earlier state.
114  *
115  *
116  *   Zone-Specific Data:
117  *
118  *   Subsystems needing to maintain zone-specific data can store that
119  *   data using the ZSD mechanism.  This provides a zone-specific data
120  *   store, similar to thread-specific data (see pthread_getspecific(3C)
121  *   or the TSD code in uts/common/disp/thread.c.  Also, ZSD can be used
122  *   to register callbacks to be invoked when a zone is created, shut
123  *   down, or destroyed.  This can be used to initialize zone-specific
124  *   data for new zones and to clean up when zones go away.
125  *
126  *
127  *   Data Structures:
128  *
129  *   The per-zone structure (zone_t) is reference counted, and freed
130  *   when all references are released.  zone_hold and zone_rele can be
131  *   used to adjust the reference count.  In addition, reference counts
132  *   associated with the cred_t structure are tracked separately using
133  *   zone_cred_hold and zone_cred_rele.
134  *
135  *   Pointers to active zone_t's are stored in two hash tables; one
136  *   for searching by id, the other for searching by name.  Lookups
137  *   can be performed on either basis, using zone_find_by_id and
138  *   zone_find_by_name.  Both return zone_t pointers with the zone
139  *   held, so zone_rele should be called when the pointer is no longer
140  *   needed.  Zones can also be searched by path; zone_find_by_path
141  *   returns the zone with which a path name is associated (global
142  *   zone if the path is not within some other zone's file system
143  *   hierarchy).  This currently requires iterating through each zone,
144  *   so it is slower than an id or name search via a hash table.
145  *
146  *
147  *   Locking:
148  *
149  *   zonehash_lock: This is a top-level global lock used to protect the
150  *       zone hash tables and lists.  Zones cannot be created or destroyed
151  *       while this lock is held.
152  *   zone_status_lock: This is a global lock protecting zone state.
153  *       Zones cannot change state while this lock is held.  It also
154  *       protects the list of kernel threads associated with a zone.
155  *   zone_lock: This is a per-zone lock used to protect several fields of
156  *       the zone_t (see <sys/zone.h> for details).  In addition, holding
157  *       this lock means that the zone cannot go away.
158  *   zsd_key_lock: This is a global lock protecting the key state for ZSD.
159  *   zone_deathrow_lock: This is a global lock protecting the "deathrow"
160  *       list (a list of zones in the ZONE_IS_DEAD state).
161  *
162  *   Ordering requirements:
163  *       pool_lock --> cpu_lock --> zonehash_lock --> zone_status_lock -->
164  *       	zone_lock --> zsd_key_lock --> pidlock --> p_lock
165  *
166  *   Blocking memory allocations are permitted while holding any of the
167  *   zone locks.
168  *
169  *
170  *   System Call Interface:
171  *
172  *   The zone subsystem can be managed and queried from user level with
173  *   the following system calls (all subcodes of the primary "zone"
174  *   system call):
175  *   - zone_create: creates a zone with selected attributes (name,
176  *     root path, privileges, resource controls)
177  *   - zone_enter: allows the current process to enter a zone
178  *   - zone_getattr: reports attributes of a zone
179  *   - zone_list: lists all zones active in the system
180  *   - zone_lookup: looks up zone id based on name
181  *   - zone_shutdown: initiates shutdown process (see states above)
182  *   - zone_destroy: completes shutdown process (see states above)
183  *
184  */
185 
186 #include <sys/priv_impl.h>
187 #include <sys/cred.h>
188 #include <c2/audit.h>
189 #include <sys/ddi.h>
190 #include <sys/debug.h>
191 #include <sys/file.h>
192 #include <sys/kmem.h>
193 #include <sys/mutex.h>
194 #include <sys/pathname.h>
195 #include <sys/proc.h>
196 #include <sys/project.h>
197 #include <sys/task.h>
198 #include <sys/systm.h>
199 #include <sys/types.h>
200 #include <sys/utsname.h>
201 #include <sys/vnode.h>
202 #include <sys/vfs.h>
203 #include <sys/systeminfo.h>
204 #include <sys/policy.h>
205 #include <sys/cred_impl.h>
206 #include <sys/contract_impl.h>
207 #include <sys/contract/process_impl.h>
208 #include <sys/class.h>
209 #include <sys/pool.h>
210 #include <sys/pool_pset.h>
211 #include <sys/pset.h>
212 #include <sys/log.h>
213 #include <sys/sysmacros.h>
214 #include <sys/callb.h>
215 #include <sys/vmparam.h>
216 #include <sys/corectl.h>
217 
218 #include <sys/door.h>
219 #include <sys/cpuvar.h>
220 #include <sys/fs/snode.h>
221 
222 #include <sys/uadmin.h>
223 #include <sys/session.h>
224 #include <sys/cmn_err.h>
225 #include <sys/modhash.h>
226 #include <sys/nvpair.h>
227 #include <sys/rctl.h>
228 #include <sys/fss.h>
229 #include <sys/zone.h>
230 
231 /*
232  * cv used to signal that all references to the zone have been released.  This
233  * needs to be global since there may be multiple waiters, and the first to
234  * wake up will free the zone_t, hence we cannot use zone->zone_cv.
235  */
236 static kcondvar_t zone_destroy_cv;
237 /*
238  * Lock used to serialize access to zone_cv.  This could have been per-zone,
239  * but then we'd need another lock for zone_destroy_cv, and why bother?
240  */
241 static kmutex_t zone_status_lock;
242 
243 /*
244  * ZSD-related global variables.
245  */
246 static kmutex_t zsd_key_lock;	/* protects the following two */
247 /*
248  * The next caller of zone_key_create() will be assigned a key of ++zsd_keyval.
249  */
250 static zone_key_t zsd_keyval = 0;
251 /*
252  * Global list of registered keys.  We use this when a new zone is created.
253  */
254 static list_t zsd_registered_keys;
255 
256 int zone_hash_size = 256;
257 static mod_hash_t *zonehashbyname, *zonehashbyid;
258 static kmutex_t zonehash_lock;
259 static uint_t zonecount;
260 static id_space_t *zoneid_space;
261 
262 /*
263  * The global zone (aka zone0) is the all-seeing, all-knowing zone in which the
264  * kernel proper runs, and which manages all other zones.
265  *
266  * Although not declared as static, the variable "zone0" should not be used
267  * except for by code that needs to reference the global zone early on in boot,
268  * before it is fully initialized.  All other consumers should use
269  * 'global_zone'.
270  */
271 zone_t zone0;
272 zone_t *global_zone = NULL;	/* Set when the global zone is initialized */
273 
274 /*
275  * List of active zones, protected by zonehash_lock.
276  */
277 static list_t zone_active;
278 
279 /*
280  * List of destroyed zones that still have outstanding cred references.
281  * Used for debugging.  Uses a separate lock to avoid lock ordering
282  * problems in zone_free.
283  */
284 static list_t zone_deathrow;
285 static kmutex_t zone_deathrow_lock;
286 
287 /* number of zones is limited by virtual interface limit in IP */
288 uint_t maxzones = 8192;
289 
290 /*
291  * This isn't static so lint doesn't complain.
292  */
293 rctl_hndl_t rc_zone_cpu_shares;
294 rctl_hndl_t rc_zone_nlwps;
295 /*
296  * Synchronization primitives used to synchronize between mounts and zone
297  * creation/destruction.
298  */
299 static int mounts_in_progress;
300 static kcondvar_t mount_cv;
301 static kmutex_t mount_lock;
302 
303 const char * const zone_initname = "/sbin/init";
304 
305 static int zone_shutdown(zoneid_t zoneid);
306 
307 /*
308  * Certain filesystems (such as NFS and autofs) need to know which zone
309  * the mount is being placed in.  Because of this, we need to be able to
310  * ensure that a zone isn't in the process of being created such that
311  * nfs_mount() thinks it is in the global zone, while by the time it
312  * gets added the list of mounted zones, it ends up on zoneA's mount
313  * list.
314  *
315  * The following functions: block_mounts()/resume_mounts() and
316  * mount_in_progress()/mount_completed() are used by zones and the VFS
317  * layer (respectively) to synchronize zone creation and new mounts.
318  *
319  * The semantics are like a reader-reader lock such that there may
320  * either be multiple mounts (or zone creations, if that weren't
321  * serialized by zonehash_lock) in progress at the same time, but not
322  * both.
323  *
324  * We use cv's so the user can ctrl-C out of the operation if it's
325  * taking too long.
326  *
327  * The semantics are such that there is unfair bias towards the
328  * "current" operation.  This means that zone creations may starve if
329  * there is a rapid succession of new mounts coming in to the system, or
330  * there is a remote possibility that zones will be created at such a
331  * rate that new mounts will not be able to proceed.
332  */
333 /*
334  * Prevent new mounts from progressing to the point of calling
335  * VFS_MOUNT().  If there are already mounts in this "region", wait for
336  * them to complete.
337  */
338 static int
339 block_mounts(void)
340 {
341 	int retval = 0;
342 
343 	/*
344 	 * Since it may block for a long time, block_mounts() shouldn't be
345 	 * called with zonehash_lock held.
346 	 */
347 	ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
348 	mutex_enter(&mount_lock);
349 	while (mounts_in_progress > 0) {
350 		if (cv_wait_sig(&mount_cv, &mount_lock) == 0)
351 			goto signaled;
352 	}
353 	/*
354 	 * A negative value of mounts_in_progress indicates that mounts
355 	 * have been blocked by (-mounts_in_progress) different callers.
356 	 */
357 	mounts_in_progress--;
358 	retval = 1;
359 signaled:
360 	mutex_exit(&mount_lock);
361 	return (retval);
362 }
363 
364 /*
365  * The VFS layer may progress with new mounts as far as we're concerned.
366  * Allow them to progress if we were the last obstacle.
367  */
368 static void
369 resume_mounts(void)
370 {
371 	mutex_enter(&mount_lock);
372 	if (++mounts_in_progress == 0)
373 		cv_broadcast(&mount_cv);
374 	mutex_exit(&mount_lock);
375 }
376 
377 /*
378  * The VFS layer is busy with a mount; zones should wait until all
379  * mounts are completed to progress.
380  */
381 void
382 mount_in_progress(void)
383 {
384 	mutex_enter(&mount_lock);
385 	while (mounts_in_progress < 0)
386 		cv_wait(&mount_cv, &mount_lock);
387 	mounts_in_progress++;
388 	mutex_exit(&mount_lock);
389 }
390 
391 /*
392  * VFS is done with one mount; wake up any waiting block_mounts()
393  * callers if this is the last mount.
394  */
395 void
396 mount_completed(void)
397 {
398 	mutex_enter(&mount_lock);
399 	if (--mounts_in_progress == 0)
400 		cv_broadcast(&mount_cv);
401 	mutex_exit(&mount_lock);
402 }
403 
404 /*
405  * ZSD routines.
406  *
407  * Zone Specific Data (ZSD) is modeled after Thread Specific Data as
408  * defined by the pthread_key_create() and related interfaces.
409  *
410  * Kernel subsystems may register one or more data items and/or
411  * callbacks to be executed when a zone is created, shutdown, or
412  * destroyed.
413  *
414  * Unlike the thread counterpart, destructor callbacks will be executed
415  * even if the data pointer is NULL and/or there are no constructor
416  * callbacks, so it is the responsibility of such callbacks to check for
417  * NULL data values if necessary.
418  *
419  * The locking strategy and overall picture is as follows:
420  *
421  * When someone calls zone_key_create(), a template ZSD entry is added to the
422  * global list "zsd_registered_keys", protected by zsd_key_lock.  The
423  * constructor callback is called immediately on all existing zones, and a
424  * copy of the ZSD entry added to the per-zone zone_zsd list (protected by
425  * zone_lock).  As this operation requires the list of zones, the list of
426  * registered keys, and the per-zone list of ZSD entries to remain constant
427  * throughout the entire operation, it must grab zonehash_lock, zone_lock for
428  * all existing zones, and zsd_key_lock, in that order.  Similar locking is
429  * needed when zone_key_delete() is called.  It is thus sufficient to hold
430  * zsd_key_lock *or* zone_lock to prevent additions to or removals from the
431  * per-zone zone_zsd list.
432  *
433  * Note that this implementation does not make a copy of the ZSD entry if a
434  * constructor callback is not provided.  A zone_getspecific() on such an
435  * uninitialized ZSD entry will return NULL.
436  *
437  * When new zones are created constructor callbacks for all registered ZSD
438  * entries will be called.
439  *
440  * The framework does not provide any locking around zone_getspecific() and
441  * zone_setspecific() apart from that needed for internal consistency, so
442  * callers interested in atomic "test-and-set" semantics will need to provide
443  * their own locking.
444  */
445 void
446 zone_key_create(zone_key_t *keyp, void *(*create)(zoneid_t),
447     void (*shutdown)(zoneid_t, void *), void (*destroy)(zoneid_t, void *))
448 {
449 	struct zsd_entry *zsdp;
450 	struct zsd_entry *t;
451 	struct zone *zone;
452 
453 	zsdp = kmem_alloc(sizeof (*zsdp), KM_SLEEP);
454 	zsdp->zsd_data = NULL;
455 	zsdp->zsd_create = create;
456 	zsdp->zsd_shutdown = shutdown;
457 	zsdp->zsd_destroy = destroy;
458 
459 	mutex_enter(&zonehash_lock);	/* stop the world */
460 	for (zone = list_head(&zone_active); zone != NULL;
461 	    zone = list_next(&zone_active, zone))
462 		mutex_enter(&zone->zone_lock);	/* lock all zones */
463 
464 	mutex_enter(&zsd_key_lock);
465 	*keyp = zsdp->zsd_key = ++zsd_keyval;
466 	ASSERT(zsd_keyval != 0);
467 	list_insert_tail(&zsd_registered_keys, zsdp);
468 	mutex_exit(&zsd_key_lock);
469 
470 	if (create != NULL) {
471 		for (zone = list_head(&zone_active); zone != NULL;
472 		    zone = list_next(&zone_active, zone)) {
473 			t = kmem_alloc(sizeof (*t), KM_SLEEP);
474 			t->zsd_key = *keyp;
475 			t->zsd_data = (*create)(zone->zone_id);
476 			t->zsd_create = create;
477 			t->zsd_shutdown = shutdown;
478 			t->zsd_destroy = destroy;
479 			list_insert_tail(&zone->zone_zsd, t);
480 		}
481 	}
482 	for (zone = list_head(&zone_active); zone != NULL;
483 	    zone = list_next(&zone_active, zone))
484 		mutex_exit(&zone->zone_lock);
485 	mutex_exit(&zonehash_lock);
486 }
487 
488 /*
489  * Helper function to find the zsd_entry associated with the key in the
490  * given list.
491  */
492 static struct zsd_entry *
493 zsd_find(list_t *l, zone_key_t key)
494 {
495 	struct zsd_entry *zsd;
496 
497 	for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
498 		if (zsd->zsd_key == key) {
499 			/*
500 			 * Move to head of list to keep list in MRU order.
501 			 */
502 			if (zsd != list_head(l)) {
503 				list_remove(l, zsd);
504 				list_insert_head(l, zsd);
505 			}
506 			return (zsd);
507 		}
508 	}
509 	return (NULL);
510 }
511 
512 /*
513  * Function called when a module is being unloaded, or otherwise wishes
514  * to unregister its ZSD key and callbacks.
515  */
516 int
517 zone_key_delete(zone_key_t key)
518 {
519 	struct zsd_entry *zsdp = NULL;
520 	zone_t *zone;
521 
522 	mutex_enter(&zonehash_lock);	/* Zone create/delete waits for us */
523 	for (zone = list_head(&zone_active); zone != NULL;
524 	    zone = list_next(&zone_active, zone))
525 		mutex_enter(&zone->zone_lock);	/* lock all zones */
526 
527 	mutex_enter(&zsd_key_lock);
528 	zsdp = zsd_find(&zsd_registered_keys, key);
529 	if (zsdp == NULL)
530 		goto notfound;
531 	list_remove(&zsd_registered_keys, zsdp);
532 	mutex_exit(&zsd_key_lock);
533 
534 	for (zone = list_head(&zone_active); zone != NULL;
535 	    zone = list_next(&zone_active, zone)) {
536 		struct zsd_entry *del;
537 		void *data;
538 
539 		if (!(zone->zone_flags & ZF_DESTROYED)) {
540 			del = zsd_find(&zone->zone_zsd, key);
541 			if (del != NULL) {
542 				data = del->zsd_data;
543 				ASSERT(del->zsd_shutdown == zsdp->zsd_shutdown);
544 				ASSERT(del->zsd_destroy == zsdp->zsd_destroy);
545 				list_remove(&zone->zone_zsd, del);
546 				kmem_free(del, sizeof (*del));
547 			} else {
548 				data = NULL;
549 			}
550 			if (zsdp->zsd_shutdown)
551 				zsdp->zsd_shutdown(zone->zone_id, data);
552 			if (zsdp->zsd_destroy)
553 				zsdp->zsd_destroy(zone->zone_id, data);
554 		}
555 		mutex_exit(&zone->zone_lock);
556 	}
557 	mutex_exit(&zonehash_lock);
558 	kmem_free(zsdp, sizeof (*zsdp));
559 	return (0);
560 
561 notfound:
562 	mutex_exit(&zsd_key_lock);
563 	for (zone = list_head(&zone_active); zone != NULL;
564 	    zone = list_next(&zone_active, zone))
565 		mutex_exit(&zone->zone_lock);
566 	mutex_exit(&zonehash_lock);
567 	return (-1);
568 }
569 
570 /*
571  * ZSD counterpart of pthread_setspecific().
572  */
573 int
574 zone_setspecific(zone_key_t key, zone_t *zone, const void *data)
575 {
576 	struct zsd_entry *t;
577 	struct zsd_entry *zsdp = NULL;
578 
579 	mutex_enter(&zone->zone_lock);
580 	t = zsd_find(&zone->zone_zsd, key);
581 	if (t != NULL) {
582 		/*
583 		 * Replace old value with new
584 		 */
585 		t->zsd_data = (void *)data;
586 		mutex_exit(&zone->zone_lock);
587 		return (0);
588 	}
589 	/*
590 	 * If there was no previous value, go through the list of registered
591 	 * keys.
592 	 *
593 	 * We avoid grabbing zsd_key_lock until we are sure we need it; this is
594 	 * necessary for shutdown callbacks to be able to execute without fear
595 	 * of deadlock.
596 	 */
597 	mutex_enter(&zsd_key_lock);
598 	zsdp = zsd_find(&zsd_registered_keys, key);
599 	if (zsdp == NULL) { 	/* Key was not registered */
600 		mutex_exit(&zsd_key_lock);
601 		mutex_exit(&zone->zone_lock);
602 		return (-1);
603 	}
604 
605 	/*
606 	 * Add a zsd_entry to this zone, using the template we just retrieved
607 	 * to initialize the constructor and destructor(s).
608 	 */
609 	t = kmem_alloc(sizeof (*t), KM_SLEEP);
610 	t->zsd_key = key;
611 	t->zsd_data = (void *)data;
612 	t->zsd_create = zsdp->zsd_create;
613 	t->zsd_shutdown = zsdp->zsd_shutdown;
614 	t->zsd_destroy = zsdp->zsd_destroy;
615 	list_insert_tail(&zone->zone_zsd, t);
616 	mutex_exit(&zsd_key_lock);
617 	mutex_exit(&zone->zone_lock);
618 	return (0);
619 }
620 
621 /*
622  * ZSD counterpart of pthread_getspecific().
623  */
624 void *
625 zone_getspecific(zone_key_t key, zone_t *zone)
626 {
627 	struct zsd_entry *t;
628 	void *data;
629 
630 	mutex_enter(&zone->zone_lock);
631 	t = zsd_find(&zone->zone_zsd, key);
632 	data = (t == NULL ? NULL : t->zsd_data);
633 	mutex_exit(&zone->zone_lock);
634 	return (data);
635 }
636 
637 /*
638  * Function used to initialize a zone's list of ZSD callbacks and data
639  * when the zone is being created.  The callbacks are initialized from
640  * the template list (zsd_registered_keys), and the constructor
641  * callback executed (if one exists).
642  *
643  * This is called before the zone is made publicly available, hence no
644  * need to grab zone_lock.
645  *
646  * Although we grab and release zsd_key_lock, new entries cannot be
647  * added to or removed from the zsd_registered_keys list until we
648  * release zonehash_lock, so there isn't a window for a
649  * zone_key_create() to come in after we've dropped zsd_key_lock but
650  * before the zone is added to the zone list, such that the constructor
651  * callbacks aren't executed for the new zone.
652  */
653 static void
654 zone_zsd_configure(zone_t *zone)
655 {
656 	struct zsd_entry *zsdp;
657 	struct zsd_entry *t;
658 	zoneid_t zoneid = zone->zone_id;
659 
660 	ASSERT(MUTEX_HELD(&zonehash_lock));
661 	ASSERT(list_head(&zone->zone_zsd) == NULL);
662 	mutex_enter(&zsd_key_lock);
663 	for (zsdp = list_head(&zsd_registered_keys); zsdp != NULL;
664 	    zsdp = list_next(&zsd_registered_keys, zsdp)) {
665 		if (zsdp->zsd_create != NULL) {
666 			t = kmem_alloc(sizeof (*t), KM_SLEEP);
667 			t->zsd_key = zsdp->zsd_key;
668 			t->zsd_create = zsdp->zsd_create;
669 			t->zsd_data = (*t->zsd_create)(zoneid);
670 			t->zsd_shutdown = zsdp->zsd_shutdown;
671 			t->zsd_destroy = zsdp->zsd_destroy;
672 			list_insert_tail(&zone->zone_zsd, t);
673 		}
674 	}
675 	mutex_exit(&zsd_key_lock);
676 }
677 
678 enum zsd_callback_type { ZSD_CREATE, ZSD_SHUTDOWN, ZSD_DESTROY };
679 
680 /*
681  * Helper function to execute shutdown or destructor callbacks.
682  */
683 static void
684 zone_zsd_callbacks(zone_t *zone, enum zsd_callback_type ct)
685 {
686 	struct zsd_entry *zsdp;
687 	struct zsd_entry *t;
688 	zoneid_t zoneid = zone->zone_id;
689 
690 	ASSERT(ct == ZSD_SHUTDOWN || ct == ZSD_DESTROY);
691 	ASSERT(ct != ZSD_SHUTDOWN || zone_status_get(zone) >= ZONE_IS_EMPTY);
692 	ASSERT(ct != ZSD_DESTROY || zone_status_get(zone) >= ZONE_IS_DOWN);
693 
694 	mutex_enter(&zone->zone_lock);
695 	if (ct == ZSD_DESTROY) {
696 		if (zone->zone_flags & ZF_DESTROYED) {
697 			/*
698 			 * Make sure destructors are only called once.
699 			 */
700 			mutex_exit(&zone->zone_lock);
701 			return;
702 		}
703 		zone->zone_flags |= ZF_DESTROYED;
704 	}
705 	mutex_exit(&zone->zone_lock);
706 
707 	/*
708 	 * Both zsd_key_lock and zone_lock need to be held in order to add or
709 	 * remove a ZSD key, (either globally as part of
710 	 * zone_key_create()/zone_key_delete(), or on a per-zone basis, as is
711 	 * possible through zone_setspecific()), so it's sufficient to hold
712 	 * zsd_key_lock here.
713 	 *
714 	 * This is a good thing, since we don't want to recursively try to grab
715 	 * zone_lock if a callback attempts to do something like a crfree() or
716 	 * zone_rele().
717 	 */
718 	mutex_enter(&zsd_key_lock);
719 	for (zsdp = list_head(&zsd_registered_keys); zsdp != NULL;
720 	    zsdp = list_next(&zsd_registered_keys, zsdp)) {
721 		zone_key_t key = zsdp->zsd_key;
722 
723 		/* Skip if no callbacks registered */
724 		if (ct == ZSD_SHUTDOWN && zsdp->zsd_shutdown == NULL)
725 			continue;
726 		if (ct == ZSD_DESTROY && zsdp->zsd_destroy == NULL)
727 			continue;
728 		/*
729 		 * Call the callback with the zone-specific data if we can find
730 		 * any, otherwise with NULL.
731 		 */
732 		t = zsd_find(&zone->zone_zsd, key);
733 		if (t != NULL) {
734 			if (ct == ZSD_SHUTDOWN) {
735 				t->zsd_shutdown(zoneid, t->zsd_data);
736 			} else {
737 				ASSERT(ct == ZSD_DESTROY);
738 				t->zsd_destroy(zoneid, t->zsd_data);
739 			}
740 		} else {
741 			if (ct == ZSD_SHUTDOWN) {
742 				zsdp->zsd_shutdown(zoneid, NULL);
743 			} else {
744 				ASSERT(ct == ZSD_DESTROY);
745 				zsdp->zsd_destroy(zoneid, NULL);
746 			}
747 		}
748 	}
749 	mutex_exit(&zsd_key_lock);
750 }
751 
752 /*
753  * Called when the zone is going away; free ZSD-related memory, and
754  * destroy the zone_zsd list.
755  */
756 static void
757 zone_free_zsd(zone_t *zone)
758 {
759 	struct zsd_entry *t, *next;
760 
761 	/*
762 	 * Free all the zsd_entry's we had on this zone.
763 	 */
764 	for (t = list_head(&zone->zone_zsd); t != NULL; t = next) {
765 		next = list_next(&zone->zone_zsd, t);
766 		list_remove(&zone->zone_zsd, t);
767 		kmem_free(t, sizeof (*t));
768 	}
769 	list_destroy(&zone->zone_zsd);
770 }
771 
772 /*
773  * zone.cpu-shares resource control support.
774  */
775 /*ARGSUSED*/
776 static rctl_qty_t
777 zone_cpu_shares_usage(rctl_t *rctl, struct proc *p)
778 {
779 	ASSERT(MUTEX_HELD(&p->p_lock));
780 	return (p->p_zone->zone_shares);
781 }
782 
783 /*ARGSUSED*/
784 static int
785 zone_cpu_shares_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
786     rctl_qty_t nv)
787 {
788 	ASSERT(MUTEX_HELD(&p->p_lock));
789 	ASSERT(e->rcep_t == RCENTITY_ZONE);
790 	if (e->rcep_p.zone == NULL)
791 		return (0);
792 
793 	e->rcep_p.zone->zone_shares = nv;
794 	return (0);
795 }
796 
797 static rctl_ops_t zone_cpu_shares_ops = {
798 	rcop_no_action,
799 	zone_cpu_shares_usage,
800 	zone_cpu_shares_set,
801 	rcop_no_test
802 };
803 
804 /*ARGSUSED*/
805 static rctl_qty_t
806 zone_lwps_usage(rctl_t *r, proc_t *p)
807 {
808 	rctl_qty_t nlwps;
809 	zone_t *zone = p->p_zone;
810 
811 	ASSERT(MUTEX_HELD(&p->p_lock));
812 
813 	mutex_enter(&zone->zone_nlwps_lock);
814 	nlwps = zone->zone_nlwps;
815 	mutex_exit(&zone->zone_nlwps_lock);
816 
817 	return (nlwps);
818 }
819 
820 /*ARGSUSED*/
821 static int
822 zone_lwps_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
823     rctl_qty_t incr, uint_t flags)
824 {
825 	rctl_qty_t nlwps;
826 
827 	ASSERT(MUTEX_HELD(&p->p_lock));
828 	ASSERT(e->rcep_t == RCENTITY_ZONE);
829 	if (e->rcep_p.zone == NULL)
830 		return (0);
831 	ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
832 	nlwps = e->rcep_p.zone->zone_nlwps;
833 
834 	if (nlwps + incr > rcntl->rcv_value)
835 		return (1);
836 
837 	return (0);
838 }
839 
840 /*ARGSUSED*/
841 static int
842 zone_lwps_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv) {
843 
844 	ASSERT(MUTEX_HELD(&p->p_lock));
845 	ASSERT(e->rcep_t == RCENTITY_ZONE);
846 	if (e->rcep_p.zone == NULL)
847 		return (0);
848 	e->rcep_p.zone->zone_nlwps_ctl = nv;
849 	return (0);
850 }
851 
852 static rctl_ops_t zone_lwps_ops = {
853 	rcop_no_action,
854 	zone_lwps_usage,
855 	zone_lwps_set,
856 	zone_lwps_test,
857 };
858 
859 /*
860  * Helper function to brand the zone with a unique ID.
861  */
862 static void
863 zone_uniqid(zone_t *zone)
864 {
865 	static uint64_t uniqid = 0;
866 
867 	ASSERT(MUTEX_HELD(&zonehash_lock));
868 	zone->zone_uniqid = uniqid++;
869 }
870 
871 /*
872  * Returns a held pointer to the "kcred" for the specified zone.
873  */
874 struct cred *
875 zone_get_kcred(zoneid_t zoneid)
876 {
877 	zone_t *zone;
878 	cred_t *cr;
879 
880 	if ((zone = zone_find_by_id(zoneid)) == NULL)
881 		return (NULL);
882 	cr = zone->zone_kcred;
883 	crhold(cr);
884 	zone_rele(zone);
885 	return (cr);
886 }
887 
888 /*
889  * Called very early on in boot to initialize the ZSD list so that
890  * zone_key_create() can be called before zone_init().  It also initializes
891  * portions of zone0 which may be used before zone_init() is called.  The
892  * variable "global_zone" will be set when zone0 is fully initialized by
893  * zone_init().
894  */
895 void
896 zone_zsd_init(void)
897 {
898 	mutex_init(&zonehash_lock, NULL, MUTEX_DEFAULT, NULL);
899 	mutex_init(&zsd_key_lock, NULL, MUTEX_DEFAULT, NULL);
900 	list_create(&zsd_registered_keys, sizeof (struct zsd_entry),
901 	    offsetof(struct zsd_entry, zsd_linkage));
902 	list_create(&zone_active, sizeof (zone_t),
903 	    offsetof(zone_t, zone_linkage));
904 	list_create(&zone_deathrow, sizeof (zone_t),
905 	    offsetof(zone_t, zone_linkage));
906 
907 	mutex_init(&zone0.zone_lock, NULL, MUTEX_DEFAULT, NULL);
908 	mutex_init(&zone0.zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
909 	zone0.zone_shares = 1;
910 	zone0.zone_nlwps_ctl = INT_MAX;
911 	zone0.zone_name = GLOBAL_ZONENAME;
912 	zone0.zone_nodename = utsname.nodename;
913 	zone0.zone_domain = srpc_domain;
914 	zone0.zone_ref = 1;
915 	zone0.zone_id = GLOBAL_ZONEID;
916 	zone0.zone_status = ZONE_IS_RUNNING;
917 	zone0.zone_rootpath = "/";
918 	zone0.zone_rootpathlen = 2;
919 	zone0.zone_psetid = ZONE_PS_INVAL;
920 	zone0.zone_ncpus = 0;
921 	zone0.zone_ncpus_online = 0;
922 	zone0.zone_proc_initpid = 1;
923 	list_create(&zone0.zone_zsd, sizeof (struct zsd_entry),
924 	    offsetof(struct zsd_entry, zsd_linkage));
925 	list_insert_head(&zone_active, &zone0);
926 
927 	/*
928 	 * The root filesystem is not mounted yet, so zone_rootvp cannot be set
929 	 * to anything meaningful.  It is assigned to be 'rootdir' in
930 	 * vfs_mountroot().
931 	 */
932 	zone0.zone_rootvp = NULL;
933 	zone0.zone_vfslist = NULL;
934 	zone0.zone_bootargs = NULL;
935 	zone0.zone_privset = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
936 	/*
937 	 * The global zone has all privileges
938 	 */
939 	priv_fillset(zone0.zone_privset);
940 	/*
941 	 * Add p0 to the global zone
942 	 */
943 	zone0.zone_zsched = &p0;
944 	p0.p_zone = &zone0;
945 }
946 
947 /*
948  * Called by main() to initialize the zones framework.
949  */
950 void
951 zone_init(void)
952 {
953 	rctl_dict_entry_t *rde;
954 	rctl_val_t *dval;
955 	rctl_set_t *set;
956 	rctl_alloc_gp_t *gp;
957 	rctl_entity_p_t e;
958 
959 	ASSERT(curproc == &p0);
960 
961 	/*
962 	 * Create ID space for zone IDs.  ID 0 is reserved for the
963 	 * global zone.
964 	 */
965 	zoneid_space = id_space_create("zoneid_space", 1, MAX_ZONEID);
966 
967 	/*
968 	 * Initialize generic zone resource controls, if any.
969 	 */
970 	rc_zone_cpu_shares = rctl_register("zone.cpu-shares",
971 	    RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_NEVER |
972 	    RCTL_GLOBAL_NOBASIC |
973 	    RCTL_GLOBAL_COUNT, FSS_MAXSHARES, FSS_MAXSHARES,
974 	    &zone_cpu_shares_ops);
975 
976 	rc_zone_nlwps = rctl_register("zone.max-lwps", RCENTITY_ZONE,
977 	    RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
978 	    INT_MAX, INT_MAX, &zone_lwps_ops);
979 	/*
980 	 * Create a rctl_val with PRIVILEGED, NOACTION, value = 1.  Then attach
981 	 * this at the head of the rctl_dict_entry for ``zone.cpu-shares''.
982 	 */
983 	dval = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
984 	bzero(dval, sizeof (rctl_val_t));
985 	dval->rcv_value = 1;
986 	dval->rcv_privilege = RCPRIV_PRIVILEGED;
987 	dval->rcv_flagaction = RCTL_LOCAL_NOACTION;
988 	dval->rcv_action_recip_pid = -1;
989 
990 	rde = rctl_dict_lookup("zone.cpu-shares");
991 	(void) rctl_val_list_insert(&rde->rcd_default_value, dval);
992 
993 	/*
994 	 * Initialize the ``global zone''.
995 	 */
996 	set = rctl_set_create();
997 	gp = rctl_set_init_prealloc(RCENTITY_ZONE);
998 	mutex_enter(&p0.p_lock);
999 	e.rcep_p.zone = &zone0;
1000 	e.rcep_t = RCENTITY_ZONE;
1001 	zone0.zone_rctls = rctl_set_init(RCENTITY_ZONE, &p0, &e, set,
1002 	    gp);
1003 
1004 	zone0.zone_nlwps = p0.p_lwpcnt;
1005 	zone0.zone_ntasks = 1;
1006 	mutex_exit(&p0.p_lock);
1007 	rctl_prealloc_destroy(gp);
1008 	/*
1009 	 * pool_default hasn't been initialized yet, so we let pool_init() take
1010 	 * care of making the global zone is in the default pool.
1011 	 */
1012 	mutex_enter(&zonehash_lock);
1013 	zone_uniqid(&zone0);
1014 	ASSERT(zone0.zone_uniqid == GLOBAL_ZONEUNIQID);
1015 	mutex_exit(&zonehash_lock);
1016 	zonehashbyid = mod_hash_create_idhash("zone_by_id", zone_hash_size,
1017 	    mod_hash_null_valdtor);
1018 	zonehashbyname = mod_hash_create_strhash("zone_by_name",
1019 	    zone_hash_size, mod_hash_null_valdtor);
1020 	zonecount = 1;
1021 
1022 	(void) mod_hash_insert(zonehashbyid, (mod_hash_key_t)GLOBAL_ZONEID,
1023 	    (mod_hash_val_t)&zone0);
1024 	(void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)zone0.zone_name,
1025 	    (mod_hash_val_t)&zone0);
1026 	/*
1027 	 * We avoid setting zone_kcred until now, since kcred is initialized
1028 	 * sometime after zone_zsd_init() and before zone_init().
1029 	 */
1030 	zone0.zone_kcred = kcred;
1031 	/*
1032 	 * The global zone is fully initialized (except for zone_rootvp which
1033 	 * will be set when the root filesystem is mounted).
1034 	 */
1035 	global_zone = &zone0;
1036 }
1037 
1038 static void
1039 zone_free(zone_t *zone)
1040 {
1041 	ASSERT(zone != global_zone);
1042 	ASSERT(zone->zone_ntasks == 0);
1043 	ASSERT(zone->zone_nlwps == 0);
1044 	ASSERT(zone->zone_cred_ref == 0);
1045 	ASSERT(zone->zone_kcred == NULL);
1046 	ASSERT(zone_status_get(zone) == ZONE_IS_DEAD ||
1047 	    zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
1048 
1049 	/* remove from deathrow list */
1050 	if (zone_status_get(zone) == ZONE_IS_DEAD) {
1051 		ASSERT(zone->zone_ref == 0);
1052 		mutex_enter(&zone_deathrow_lock);
1053 		list_remove(&zone_deathrow, zone);
1054 		mutex_exit(&zone_deathrow_lock);
1055 	}
1056 
1057 	zone_free_zsd(zone);
1058 
1059 	if (zone->zone_rootvp != NULL)
1060 		VN_RELE(zone->zone_rootvp);
1061 	if (zone->zone_rootpath)
1062 		kmem_free(zone->zone_rootpath, zone->zone_rootpathlen);
1063 	if (zone->zone_name != NULL)
1064 		kmem_free(zone->zone_name, ZONENAME_MAX);
1065 	if (zone->zone_nodename != NULL)
1066 		kmem_free(zone->zone_nodename, _SYS_NMLN);
1067 	if (zone->zone_domain != NULL)
1068 		kmem_free(zone->zone_domain, _SYS_NMLN);
1069 	if (zone->zone_privset != NULL)
1070 		kmem_free(zone->zone_privset, sizeof (priv_set_t));
1071 	if (zone->zone_rctls != NULL)
1072 		rctl_set_free(zone->zone_rctls);
1073 	if (zone->zone_bootargs != NULL)
1074 		kmem_free(zone->zone_bootargs, ZONEBOOTARGS_MAX);
1075 	id_free(zoneid_space, zone->zone_id);
1076 	mutex_destroy(&zone->zone_lock);
1077 	cv_destroy(&zone->zone_cv);
1078 	kmem_free(zone, sizeof (zone_t));
1079 }
1080 
1081 /*
1082  * See block comment at the top of this file for information about zone
1083  * status values.
1084  */
1085 /*
1086  * Convenience function for setting zone status.
1087  */
1088 static void
1089 zone_status_set(zone_t *zone, zone_status_t status)
1090 {
1091 	ASSERT(MUTEX_HELD(&zone_status_lock));
1092 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE &&
1093 	    status >= zone_status_get(zone));
1094 	zone->zone_status = status;
1095 	cv_broadcast(&zone->zone_cv);
1096 }
1097 
1098 /*
1099  * Public function to retrieve the zone status.  The zone status may
1100  * change after it is retrieved.
1101  */
1102 zone_status_t
1103 zone_status_get(zone_t *zone)
1104 {
1105 	return (zone->zone_status);
1106 }
1107 
1108 static int
1109 zone_set_bootargs(zone_t *zone, const char *zone_bootargs)
1110 {
1111 	char *bootargs = kmem_zalloc(ZONEBOOTARGS_MAX, KM_SLEEP);
1112 	size_t len;
1113 	int err;
1114 
1115 	err = copyinstr(zone_bootargs, bootargs, ZONEBOOTARGS_MAX - 1, &len);
1116 	if (err != 0) {
1117 		kmem_free(bootargs, ZONEBOOTARGS_MAX);
1118 		return (err);	/* EFAULT or ENAMETOOLONG */
1119 	}
1120 	bootargs[len] = '\0';
1121 
1122 	ASSERT(zone->zone_bootargs == NULL);
1123 	zone->zone_bootargs = bootargs;
1124 	return (0);
1125 }
1126 
1127 /*
1128  * Block indefinitely waiting for (zone_status >= status)
1129  */
1130 void
1131 zone_status_wait(zone_t *zone, zone_status_t status)
1132 {
1133 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
1134 
1135 	mutex_enter(&zone_status_lock);
1136 	while (zone->zone_status < status) {
1137 		cv_wait(&zone->zone_cv, &zone_status_lock);
1138 	}
1139 	mutex_exit(&zone_status_lock);
1140 }
1141 
1142 /*
1143  * Private CPR-safe version of zone_status_wait().
1144  */
1145 static void
1146 zone_status_wait_cpr(zone_t *zone, zone_status_t status, char *str)
1147 {
1148 	callb_cpr_t cprinfo;
1149 
1150 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
1151 
1152 	CALLB_CPR_INIT(&cprinfo, &zone_status_lock, callb_generic_cpr,
1153 	    str);
1154 	mutex_enter(&zone_status_lock);
1155 	while (zone->zone_status < status) {
1156 		CALLB_CPR_SAFE_BEGIN(&cprinfo);
1157 		cv_wait(&zone->zone_cv, &zone_status_lock);
1158 		CALLB_CPR_SAFE_END(&cprinfo, &zone_status_lock);
1159 	}
1160 	/*
1161 	 * zone_status_lock is implicitly released by the following.
1162 	 */
1163 	CALLB_CPR_EXIT(&cprinfo);
1164 }
1165 
1166 /*
1167  * Block until zone enters requested state or signal is received.  Return (0)
1168  * if signaled, non-zero otherwise.
1169  */
1170 int
1171 zone_status_wait_sig(zone_t *zone, zone_status_t status)
1172 {
1173 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
1174 
1175 	mutex_enter(&zone_status_lock);
1176 	while (zone->zone_status < status) {
1177 		if (!cv_wait_sig(&zone->zone_cv, &zone_status_lock)) {
1178 			mutex_exit(&zone_status_lock);
1179 			return (0);
1180 		}
1181 	}
1182 	mutex_exit(&zone_status_lock);
1183 	return (1);
1184 }
1185 
1186 /*
1187  * Block until the zone enters the requested state or the timeout expires,
1188  * whichever happens first.  Return (-1) if operation timed out, time remaining
1189  * otherwise.
1190  */
1191 clock_t
1192 zone_status_timedwait(zone_t *zone, clock_t tim, zone_status_t status)
1193 {
1194 	clock_t timeleft = 0;
1195 
1196 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
1197 
1198 	mutex_enter(&zone_status_lock);
1199 	while (zone->zone_status < status && timeleft != -1) {
1200 		timeleft = cv_timedwait(&zone->zone_cv, &zone_status_lock, tim);
1201 	}
1202 	mutex_exit(&zone_status_lock);
1203 	return (timeleft);
1204 }
1205 
1206 /*
1207  * Block until the zone enters the requested state, the current process is
1208  * signaled,  or the timeout expires, whichever happens first.  Return (-1) if
1209  * operation timed out, 0 if signaled, time remaining otherwise.
1210  */
1211 clock_t
1212 zone_status_timedwait_sig(zone_t *zone, clock_t tim, zone_status_t status)
1213 {
1214 	clock_t timeleft = tim - lbolt;
1215 
1216 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
1217 
1218 	mutex_enter(&zone_status_lock);
1219 	while (zone->zone_status < status) {
1220 		timeleft = cv_timedwait_sig(&zone->zone_cv, &zone_status_lock,
1221 		    tim);
1222 		if (timeleft <= 0)
1223 			break;
1224 	}
1225 	mutex_exit(&zone_status_lock);
1226 	return (timeleft);
1227 }
1228 
1229 /*
1230  * Zones have two reference counts: one for references from credential
1231  * structures (zone_cred_ref), and one (zone_ref) for everything else.
1232  * This is so we can allow a zone to be rebooted while there are still
1233  * outstanding cred references, since certain drivers cache dblks (which
1234  * implicitly results in cached creds).  We wait for zone_ref to drop to
1235  * 0 (actually 1), but not zone_cred_ref.  The zone structure itself is
1236  * later freed when the zone_cred_ref drops to 0, though nothing other
1237  * than the zone id and privilege set should be accessed once the zone
1238  * is "dead".
1239  *
1240  * A debugging flag, zone_wait_for_cred, can be set to a non-zero value
1241  * to force halt/reboot to block waiting for the zone_cred_ref to drop
1242  * to 0.  This can be useful to flush out other sources of cached creds
1243  * that may be less innocuous than the driver case.
1244  */
1245 
1246 int zone_wait_for_cred = 0;
1247 
1248 static void
1249 zone_hold_locked(zone_t *z)
1250 {
1251 	ASSERT(MUTEX_HELD(&z->zone_lock));
1252 	z->zone_ref++;
1253 	ASSERT(z->zone_ref != 0);
1254 }
1255 
1256 void
1257 zone_hold(zone_t *z)
1258 {
1259 	mutex_enter(&z->zone_lock);
1260 	zone_hold_locked(z);
1261 	mutex_exit(&z->zone_lock);
1262 }
1263 
1264 /*
1265  * If the non-cred ref count drops to 1 and either the cred ref count
1266  * is 0 or we aren't waiting for cred references, the zone is ready to
1267  * be destroyed.
1268  */
1269 #define	ZONE_IS_UNREF(zone)	((zone)->zone_ref == 1 && \
1270 	    (!zone_wait_for_cred || (zone)->zone_cred_ref == 0))
1271 
1272 void
1273 zone_rele(zone_t *z)
1274 {
1275 	boolean_t wakeup;
1276 
1277 	mutex_enter(&z->zone_lock);
1278 	ASSERT(z->zone_ref != 0);
1279 	z->zone_ref--;
1280 	if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
1281 		/* no more refs, free the structure */
1282 		mutex_exit(&z->zone_lock);
1283 		zone_free(z);
1284 		return;
1285 	}
1286 	/* signal zone_destroy so the zone can finish halting */
1287 	wakeup = (ZONE_IS_UNREF(z) && zone_status_get(z) >= ZONE_IS_DEAD);
1288 	mutex_exit(&z->zone_lock);
1289 
1290 	if (wakeup) {
1291 		/*
1292 		 * Grabbing zonehash_lock here effectively synchronizes with
1293 		 * zone_destroy() to avoid missed signals.
1294 		 */
1295 		mutex_enter(&zonehash_lock);
1296 		cv_broadcast(&zone_destroy_cv);
1297 		mutex_exit(&zonehash_lock);
1298 	}
1299 }
1300 
1301 void
1302 zone_cred_hold(zone_t *z)
1303 {
1304 	mutex_enter(&z->zone_lock);
1305 	z->zone_cred_ref++;
1306 	ASSERT(z->zone_cred_ref != 0);
1307 	mutex_exit(&z->zone_lock);
1308 }
1309 
1310 void
1311 zone_cred_rele(zone_t *z)
1312 {
1313 	boolean_t wakeup;
1314 
1315 	mutex_enter(&z->zone_lock);
1316 	ASSERT(z->zone_cred_ref != 0);
1317 	z->zone_cred_ref--;
1318 	if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
1319 		/* no more refs, free the structure */
1320 		mutex_exit(&z->zone_lock);
1321 		zone_free(z);
1322 		return;
1323 	}
1324 	/*
1325 	 * If zone_destroy is waiting for the cred references to drain
1326 	 * out, and they have, signal it.
1327 	 */
1328 	wakeup = (zone_wait_for_cred && ZONE_IS_UNREF(z) &&
1329 	    zone_status_get(z) >= ZONE_IS_DEAD);
1330 	mutex_exit(&z->zone_lock);
1331 
1332 	if (wakeup) {
1333 		/*
1334 		 * Grabbing zonehash_lock here effectively synchronizes with
1335 		 * zone_destroy() to avoid missed signals.
1336 		 */
1337 		mutex_enter(&zonehash_lock);
1338 		cv_broadcast(&zone_destroy_cv);
1339 		mutex_exit(&zonehash_lock);
1340 	}
1341 }
1342 
1343 void
1344 zone_task_hold(zone_t *z)
1345 {
1346 	mutex_enter(&z->zone_lock);
1347 	z->zone_ntasks++;
1348 	ASSERT(z->zone_ntasks != 0);
1349 	mutex_exit(&z->zone_lock);
1350 }
1351 
1352 void
1353 zone_task_rele(zone_t *zone)
1354 {
1355 	uint_t refcnt;
1356 
1357 	mutex_enter(&zone->zone_lock);
1358 	ASSERT(zone->zone_ntasks != 0);
1359 	refcnt = --zone->zone_ntasks;
1360 	if (refcnt > 1)	{	/* Common case */
1361 		mutex_exit(&zone->zone_lock);
1362 		return;
1363 	}
1364 	zone_hold_locked(zone);	/* so we can use the zone_t later */
1365 	mutex_exit(&zone->zone_lock);
1366 	if (refcnt == 1) {
1367 		/*
1368 		 * See if the zone is shutting down.
1369 		 */
1370 		mutex_enter(&zone_status_lock);
1371 		if (zone_status_get(zone) != ZONE_IS_SHUTTING_DOWN) {
1372 			goto out;
1373 		}
1374 
1375 		/*
1376 		 * Make sure the ntasks didn't change since we
1377 		 * dropped zone_lock.
1378 		 */
1379 		mutex_enter(&zone->zone_lock);
1380 		if (refcnt != zone->zone_ntasks) {
1381 			mutex_exit(&zone->zone_lock);
1382 			goto out;
1383 		}
1384 		mutex_exit(&zone->zone_lock);
1385 
1386 		/*
1387 		 * No more user processes in the zone.  The zone is empty.
1388 		 */
1389 		zone_status_set(zone, ZONE_IS_EMPTY);
1390 		goto out;
1391 	}
1392 
1393 	ASSERT(refcnt == 0);
1394 	/*
1395 	 * zsched has exited; the zone is dead.
1396 	 */
1397 	zone->zone_zsched = NULL;		/* paranoia */
1398 	mutex_enter(&zone_status_lock);
1399 	zone_status_set(zone, ZONE_IS_DEAD);
1400 out:
1401 	mutex_exit(&zone_status_lock);
1402 	zone_rele(zone);
1403 }
1404 
1405 zoneid_t
1406 getzoneid(void)
1407 {
1408 	return (curproc->p_zone->zone_id);
1409 }
1410 
1411 /*
1412  * Internal versions of zone_find_by_*().  These don't zone_hold() or
1413  * check the validity of a zone's state.
1414  */
1415 static zone_t *
1416 zone_find_all_by_id(zoneid_t zoneid)
1417 {
1418 	mod_hash_val_t hv;
1419 	zone_t *zone = NULL;
1420 
1421 	ASSERT(MUTEX_HELD(&zonehash_lock));
1422 
1423 	if (mod_hash_find(zonehashbyid,
1424 	    (mod_hash_key_t)(uintptr_t)zoneid, &hv) == 0)
1425 		zone = (zone_t *)hv;
1426 	return (zone);
1427 }
1428 
1429 static zone_t *
1430 zone_find_all_by_name(char *name)
1431 {
1432 	mod_hash_val_t hv;
1433 	zone_t *zone = NULL;
1434 
1435 	ASSERT(MUTEX_HELD(&zonehash_lock));
1436 
1437 	if (mod_hash_find(zonehashbyname, (mod_hash_key_t)name, &hv) == 0)
1438 		zone = (zone_t *)hv;
1439 	return (zone);
1440 }
1441 
1442 /*
1443  * Public interface for looking up a zone by zoneid.  Only returns the zone if
1444  * it is fully initialized, and has not yet begun the zone_destroy() sequence.
1445  * Caller must call zone_rele() once it is done with the zone.
1446  *
1447  * The zone may begin the zone_destroy() sequence immediately after this
1448  * function returns, but may be safely used until zone_rele() is called.
1449  */
1450 zone_t *
1451 zone_find_by_id(zoneid_t zoneid)
1452 {
1453 	zone_t *zone;
1454 	zone_status_t status;
1455 
1456 	mutex_enter(&zonehash_lock);
1457 	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
1458 		mutex_exit(&zonehash_lock);
1459 		return (NULL);
1460 	}
1461 	status = zone_status_get(zone);
1462 	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
1463 		/*
1464 		 * For all practical purposes the zone doesn't exist.
1465 		 */
1466 		mutex_exit(&zonehash_lock);
1467 		return (NULL);
1468 	}
1469 	zone_hold(zone);
1470 	mutex_exit(&zonehash_lock);
1471 	return (zone);
1472 }
1473 
1474 /*
1475  * Similar to zone_find_by_id, but using zone name as the key.
1476  */
1477 zone_t *
1478 zone_find_by_name(char *name)
1479 {
1480 	zone_t *zone;
1481 	zone_status_t status;
1482 
1483 	mutex_enter(&zonehash_lock);
1484 	if ((zone = zone_find_all_by_name(name)) == NULL) {
1485 		mutex_exit(&zonehash_lock);
1486 		return (NULL);
1487 	}
1488 	status = zone_status_get(zone);
1489 	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
1490 		/*
1491 		 * For all practical purposes the zone doesn't exist.
1492 		 */
1493 		mutex_exit(&zonehash_lock);
1494 		return (NULL);
1495 	}
1496 	zone_hold(zone);
1497 	mutex_exit(&zonehash_lock);
1498 	return (zone);
1499 }
1500 
1501 /*
1502  * Similar to zone_find_by_id(), using the path as a key.  For instance,
1503  * if there is a zone "foo" rooted at /foo/root, and the path argument
1504  * is "/foo/root/proc", it will return the held zone_t corresponding to
1505  * zone "foo".
1506  *
1507  * zone_find_by_path() always returns a non-NULL value, since at the
1508  * very least every path will be contained in the global zone.
1509  *
1510  * As with the other zone_find_by_*() functions, the caller is
1511  * responsible for zone_rele()ing the return value of this function.
1512  */
1513 zone_t *
1514 zone_find_by_path(const char *path)
1515 {
1516 	zone_t *zone;
1517 	zone_t *zret = NULL;
1518 	zone_status_t status;
1519 
1520 	if (path == NULL) {
1521 		/*
1522 		 * Call from rootconf().
1523 		 */
1524 		zone_hold(global_zone);
1525 		return (global_zone);
1526 	}
1527 	ASSERT(*path == '/');
1528 	mutex_enter(&zonehash_lock);
1529 	for (zone = list_head(&zone_active); zone != NULL;
1530 	    zone = list_next(&zone_active, zone)) {
1531 		if (ZONE_PATH_VISIBLE(path, zone))
1532 			zret = zone;
1533 	}
1534 	ASSERT(zret != NULL);
1535 	status = zone_status_get(zret);
1536 	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
1537 		/*
1538 		 * Zone practically doesn't exist.
1539 		 */
1540 		zret = global_zone;
1541 	}
1542 	zone_hold(zret);
1543 	mutex_exit(&zonehash_lock);
1544 	return (zret);
1545 }
1546 
1547 /*
1548  * Get the number of cpus visible to this zone.  The system-wide global
1549  * 'ncpus' is returned if pools are disabled, the caller is in the
1550  * global zone, or a NULL zone argument is passed in.
1551  */
1552 int
1553 zone_ncpus_get(zone_t *zone)
1554 {
1555 	int myncpus = zone == NULL ? 0 : zone->zone_ncpus;
1556 
1557 	return (myncpus != 0 ? myncpus : ncpus);
1558 }
1559 
1560 /*
1561  * Get the number of online cpus visible to this zone.  The system-wide
1562  * global 'ncpus_online' is returned if pools are disabled, the caller
1563  * is in the global zone, or a NULL zone argument is passed in.
1564  */
1565 int
1566 zone_ncpus_online_get(zone_t *zone)
1567 {
1568 	int myncpus_online = zone == NULL ? 0 : zone->zone_ncpus_online;
1569 
1570 	return (myncpus_online != 0 ? myncpus_online : ncpus_online);
1571 }
1572 
1573 /*
1574  * Return the pool to which the zone is currently bound.
1575  */
1576 pool_t *
1577 zone_pool_get(zone_t *zone)
1578 {
1579 	ASSERT(pool_lock_held());
1580 
1581 	return (zone->zone_pool);
1582 }
1583 
1584 /*
1585  * Set the zone's pool pointer and update the zone's visibility to match
1586  * the resources in the new pool.
1587  */
1588 void
1589 zone_pool_set(zone_t *zone, pool_t *pool)
1590 {
1591 	ASSERT(pool_lock_held());
1592 	ASSERT(MUTEX_HELD(&cpu_lock));
1593 
1594 	zone->zone_pool = pool;
1595 	zone_pset_set(zone, pool->pool_pset->pset_id);
1596 }
1597 
1598 /*
1599  * Return the cached value of the id of the processor set to which the
1600  * zone is currently bound.  The value will be ZONE_PS_INVAL if the pools
1601  * facility is disabled.
1602  */
1603 psetid_t
1604 zone_pset_get(zone_t *zone)
1605 {
1606 	ASSERT(MUTEX_HELD(&cpu_lock));
1607 
1608 	return (zone->zone_psetid);
1609 }
1610 
1611 /*
1612  * Set the cached value of the id of the processor set to which the zone
1613  * is currently bound.  Also update the zone's visibility to match the
1614  * resources in the new processor set.
1615  */
1616 void
1617 zone_pset_set(zone_t *zone, psetid_t newpsetid)
1618 {
1619 	psetid_t oldpsetid;
1620 
1621 	ASSERT(MUTEX_HELD(&cpu_lock));
1622 	oldpsetid = zone_pset_get(zone);
1623 
1624 	if (oldpsetid == newpsetid)
1625 		return;
1626 	/*
1627 	 * Global zone sees all.
1628 	 */
1629 	if (zone != global_zone) {
1630 		zone->zone_psetid = newpsetid;
1631 		if (newpsetid != ZONE_PS_INVAL)
1632 			pool_pset_visibility_add(newpsetid, zone);
1633 		if (oldpsetid != ZONE_PS_INVAL)
1634 			pool_pset_visibility_remove(oldpsetid, zone);
1635 	}
1636 	/*
1637 	 * Disabling pools, so we should start using the global values
1638 	 * for ncpus and ncpus_online.
1639 	 */
1640 	if (newpsetid == ZONE_PS_INVAL) {
1641 		zone->zone_ncpus = 0;
1642 		zone->zone_ncpus_online = 0;
1643 	}
1644 }
1645 
1646 /*
1647  * Walk the list of active zones and issue the provided callback for
1648  * each of them.
1649  *
1650  * Caller must not be holding any locks that may be acquired under
1651  * zonehash_lock.  See comment at the beginning of the file for a list of
1652  * common locks and their interactions with zones.
1653  */
1654 int
1655 zone_walk(int (*cb)(zone_t *, void *), void *data)
1656 {
1657 	zone_t *zone;
1658 	int ret = 0;
1659 	zone_status_t status;
1660 
1661 	mutex_enter(&zonehash_lock);
1662 	for (zone = list_head(&zone_active); zone != NULL;
1663 	    zone = list_next(&zone_active, zone)) {
1664 		/*
1665 		 * Skip zones that shouldn't be externally visible.
1666 		 */
1667 		status = zone_status_get(zone);
1668 		if (status < ZONE_IS_READY || status > ZONE_IS_DOWN)
1669 			continue;
1670 		/*
1671 		 * Bail immediately if any callback invocation returns a
1672 		 * non-zero value.
1673 		 */
1674 		ret = (*cb)(zone, data);
1675 		if (ret != 0)
1676 			break;
1677 	}
1678 	mutex_exit(&zonehash_lock);
1679 	return (ret);
1680 }
1681 
1682 static int
1683 zone_set_root(zone_t *zone, const char *upath)
1684 {
1685 	vnode_t *vp;
1686 	int trycount;
1687 	int error = 0;
1688 	char *path;
1689 	struct pathname upn, pn;
1690 	size_t pathlen;
1691 
1692 	if ((error = pn_get((char *)upath, UIO_USERSPACE, &upn)) != 0)
1693 		return (error);
1694 
1695 	pn_alloc(&pn);
1696 
1697 	/* prevent infinite loop */
1698 	trycount = 10;
1699 	for (;;) {
1700 		if (--trycount <= 0) {
1701 			error = ESTALE;
1702 			goto out;
1703 		}
1704 
1705 		if ((error = lookuppn(&upn, &pn, FOLLOW, NULLVPP, &vp)) == 0) {
1706 			/*
1707 			 * VOP_ACCESS() may cover 'vp' with a new
1708 			 * filesystem, if 'vp' is an autoFS vnode.
1709 			 * Get the new 'vp' if so.
1710 			 */
1711 			if ((error = VOP_ACCESS(vp, VEXEC, 0, CRED())) == 0 &&
1712 			    (vp->v_vfsmountedhere == NULL ||
1713 			    (error = traverse(&vp)) == 0)) {
1714 				pathlen = pn.pn_pathlen + 2;
1715 				path = kmem_alloc(pathlen, KM_SLEEP);
1716 				(void) strncpy(path, pn.pn_path,
1717 				    pn.pn_pathlen + 1);
1718 				path[pathlen - 2] = '/';
1719 				path[pathlen - 1] = '\0';
1720 				pn_free(&pn);
1721 				pn_free(&upn);
1722 
1723 				/* Success! */
1724 				break;
1725 			}
1726 			VN_RELE(vp);
1727 		}
1728 		if (error != ESTALE)
1729 			goto out;
1730 	}
1731 
1732 	ASSERT(error == 0);
1733 	zone->zone_rootvp = vp;		/* we hold a reference to vp */
1734 	zone->zone_rootpath = path;
1735 	zone->zone_rootpathlen = pathlen;
1736 	return (0);
1737 
1738 out:
1739 	pn_free(&pn);
1740 	pn_free(&upn);
1741 	return (error);
1742 }
1743 
1744 #define	isalnum(c)	(((c) >= '0' && (c) <= '9') || \
1745 			((c) >= 'a' && (c) <= 'z') || \
1746 			((c) >= 'A' && (c) <= 'Z'))
1747 
1748 static int
1749 zone_set_name(zone_t *zone, const char *uname)
1750 {
1751 	char *kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
1752 	size_t len;
1753 	int i, err;
1754 
1755 	if ((err = copyinstr(uname, kname, ZONENAME_MAX, &len)) != 0) {
1756 		kmem_free(kname, ZONENAME_MAX);
1757 		return (err);	/* EFAULT or ENAMETOOLONG */
1758 	}
1759 
1760 	/* must be less than ZONENAME_MAX */
1761 	if (len == ZONENAME_MAX && kname[ZONENAME_MAX - 1] != '\0') {
1762 		kmem_free(kname, ZONENAME_MAX);
1763 		return (EINVAL);
1764 	}
1765 
1766 	/*
1767 	 * Name must start with an alphanumeric and must contain only
1768 	 * alphanumerics, '-', '_' and '.'.
1769 	 */
1770 	if (!isalnum(kname[0])) {
1771 		kmem_free(kname, ZONENAME_MAX);
1772 		return (EINVAL);
1773 	}
1774 	for (i = 1; i < len - 1; i++) {
1775 		if (!isalnum(kname[i]) && kname[i] != '-' && kname[i] != '_' &&
1776 		    kname[i] != '.') {
1777 			kmem_free(kname, ZONENAME_MAX);
1778 			return (EINVAL);
1779 		}
1780 	}
1781 
1782 	zone->zone_name = kname;
1783 	return (0);
1784 }
1785 
1786 /*
1787  * Similar to thread_create(), but makes sure the thread is in the appropriate
1788  * zone's zsched process (curproc->p_zone->zone_zsched) before returning.
1789  */
1790 /*ARGSUSED*/
1791 kthread_t *
1792 zthread_create(
1793     caddr_t stk,
1794     size_t stksize,
1795     void (*proc)(),
1796     void *arg,
1797     size_t len,
1798     pri_t pri)
1799 {
1800 	kthread_t *t;
1801 	zone_t *zone = curproc->p_zone;
1802 	proc_t *pp = zone->zone_zsched;
1803 
1804 	zone_hold(zone);	/* Reference to be dropped when thread exits */
1805 
1806 	/*
1807 	 * No-one should be trying to create threads if the zone is shutting
1808 	 * down and there aren't any kernel threads around.  See comment
1809 	 * in zthread_exit().
1810 	 */
1811 	ASSERT(!(zone->zone_kthreads == NULL &&
1812 	    zone_status_get(zone) >= ZONE_IS_EMPTY));
1813 	/*
1814 	 * Create a thread, but don't let it run until we've finished setting
1815 	 * things up.
1816 	 */
1817 	t = thread_create(stk, stksize, proc, arg, len, pp, TS_STOPPED, pri);
1818 	ASSERT(t->t_forw == NULL);
1819 	mutex_enter(&zone_status_lock);
1820 	if (zone->zone_kthreads == NULL) {
1821 		t->t_forw = t->t_back = t;
1822 	} else {
1823 		kthread_t *tx = zone->zone_kthreads;
1824 
1825 		t->t_forw = tx;
1826 		t->t_back = tx->t_back;
1827 		tx->t_back->t_forw = t;
1828 		tx->t_back = t;
1829 	}
1830 	zone->zone_kthreads = t;
1831 	mutex_exit(&zone_status_lock);
1832 
1833 	mutex_enter(&pp->p_lock);
1834 	t->t_proc_flag |= TP_ZTHREAD;
1835 	project_rele(t->t_proj);
1836 	t->t_proj = project_hold(pp->p_task->tk_proj);
1837 
1838 	/*
1839 	 * Setup complete, let it run.
1840 	 */
1841 	thread_lock(t);
1842 	t->t_schedflag |= TS_ALLSTART;
1843 	setrun_locked(t);
1844 	thread_unlock(t);
1845 
1846 	mutex_exit(&pp->p_lock);
1847 
1848 	return (t);
1849 }
1850 
1851 /*
1852  * Similar to thread_exit().  Must be called by threads created via
1853  * zthread_exit().
1854  */
1855 void
1856 zthread_exit(void)
1857 {
1858 	kthread_t *t = curthread;
1859 	proc_t *pp = curproc;
1860 	zone_t *zone = pp->p_zone;
1861 
1862 	mutex_enter(&zone_status_lock);
1863 
1864 	/*
1865 	 * Reparent to p0
1866 	 */
1867 	mutex_enter(&pp->p_lock);
1868 	t->t_proc_flag &= ~TP_ZTHREAD;
1869 	t->t_procp = &p0;
1870 	hat_thread_exit(t);
1871 	mutex_exit(&pp->p_lock);
1872 
1873 	if (t->t_back == t) {
1874 		ASSERT(t->t_forw == t);
1875 		/*
1876 		 * If the zone is empty, once the thread count
1877 		 * goes to zero no further kernel threads can be
1878 		 * created.  This is because if the creator is a process
1879 		 * in the zone, then it must have exited before the zone
1880 		 * state could be set to ZONE_IS_EMPTY.
1881 		 * Otherwise, if the creator is a kernel thread in the
1882 		 * zone, the thread count is non-zero.
1883 		 *
1884 		 * This really means that non-zone kernel threads should
1885 		 * not create zone kernel threads.
1886 		 */
1887 		zone->zone_kthreads = NULL;
1888 		if (zone_status_get(zone) == ZONE_IS_EMPTY) {
1889 			zone_status_set(zone, ZONE_IS_DOWN);
1890 		}
1891 	} else {
1892 		t->t_forw->t_back = t->t_back;
1893 		t->t_back->t_forw = t->t_forw;
1894 		if (zone->zone_kthreads == t)
1895 			zone->zone_kthreads = t->t_forw;
1896 	}
1897 	mutex_exit(&zone_status_lock);
1898 	zone_rele(zone);
1899 	thread_exit();
1900 	/* NOTREACHED */
1901 }
1902 
1903 static void
1904 zone_chdir(vnode_t *vp, vnode_t **vpp, proc_t *pp)
1905 {
1906 	vnode_t *oldvp;
1907 
1908 	/* we're going to hold a reference here to the directory */
1909 	VN_HOLD(vp);
1910 
1911 #ifdef C2_AUDIT
1912 	if (audit_active)	/* update abs cwd/root path see c2audit.c */
1913 		audit_chdirec(vp, vpp);
1914 #endif
1915 
1916 	mutex_enter(&pp->p_lock);
1917 	oldvp = *vpp;
1918 	*vpp = vp;
1919 	mutex_exit(&pp->p_lock);
1920 	if (oldvp != NULL)
1921 		VN_RELE(oldvp);
1922 }
1923 
1924 /*
1925  * Convert an rctl value represented by an nvlist_t into an rctl_val_t.
1926  */
1927 static int
1928 nvlist2rctlval(nvlist_t *nvl, rctl_val_t *rv)
1929 {
1930 	nvpair_t *nvp = NULL;
1931 	boolean_t priv_set = B_FALSE;
1932 	boolean_t limit_set = B_FALSE;
1933 	boolean_t action_set = B_FALSE;
1934 
1935 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
1936 		const char *name;
1937 		uint64_t ui64;
1938 
1939 		name = nvpair_name(nvp);
1940 		if (nvpair_type(nvp) != DATA_TYPE_UINT64)
1941 			return (EINVAL);
1942 		(void) nvpair_value_uint64(nvp, &ui64);
1943 		if (strcmp(name, "privilege") == 0) {
1944 			/*
1945 			 * Currently only privileged values are allowed, but
1946 			 * this may change in the future.
1947 			 */
1948 			if (ui64 != RCPRIV_PRIVILEGED)
1949 				return (EINVAL);
1950 			rv->rcv_privilege = ui64;
1951 			priv_set = B_TRUE;
1952 		} else if (strcmp(name, "limit") == 0) {
1953 			rv->rcv_value = ui64;
1954 			limit_set = B_TRUE;
1955 		} else if (strcmp(name, "action") == 0) {
1956 			if (ui64 != RCTL_LOCAL_NOACTION &&
1957 			    ui64 != RCTL_LOCAL_DENY)
1958 				return (EINVAL);
1959 			rv->rcv_flagaction = ui64;
1960 			action_set = B_TRUE;
1961 		} else {
1962 			return (EINVAL);
1963 		}
1964 	}
1965 
1966 	if (!(priv_set && limit_set && action_set))
1967 		return (EINVAL);
1968 	rv->rcv_action_signal = 0;
1969 	rv->rcv_action_recipient = NULL;
1970 	rv->rcv_action_recip_pid = -1;
1971 	rv->rcv_firing_time = 0;
1972 
1973 	return (0);
1974 }
1975 
1976 void
1977 zone_icode(void)
1978 {
1979 	proc_t *p = ttoproc(curthread);
1980 	struct core_globals	*cg;
1981 
1982 	/*
1983 	 * For all purposes (ZONE_ATTR_INITPID and restart_init),
1984 	 * storing just the pid of init is sufficient.
1985 	 */
1986 	p->p_zone->zone_proc_initpid = p->p_pid;
1987 
1988 	/*
1989 	 * Allocate user address space and stack segment
1990 	 */
1991 
1992 	p->p_cstime = p->p_stime = p->p_cutime = p->p_utime = 0;
1993 	p->p_usrstack = (caddr_t)USRSTACK32;
1994 	p->p_model = DATAMODEL_ILP32;
1995 	p->p_stkprot = PROT_ZFOD & ~PROT_EXEC;
1996 	p->p_datprot = PROT_ZFOD & ~PROT_EXEC;
1997 	p->p_stk_ctl = INT32_MAX;
1998 
1999 	p->p_as = as_alloc();
2000 	p->p_as->a_userlimit = (caddr_t)USERLIMIT32;
2001 	(void) hat_setup(p->p_as->a_hat, HAT_INIT);
2002 
2003 	cg = zone_getspecific(core_zone_key, p->p_zone);
2004 	ASSERT(cg != NULL);
2005 	corectl_path_hold(cg->core_default_path);
2006 	corectl_content_hold(cg->core_default_content);
2007 	p->p_corefile = cg->core_default_path;
2008 	p->p_content = cg->core_default_content;
2009 
2010 	init_mstate(curthread, LMS_SYSTEM);
2011 
2012 	p->p_zone->zone_boot_err = exec_init(zone_initname, 0,
2013 	    p->p_zone->zone_bootargs);
2014 
2015 	mutex_enter(&zone_status_lock);
2016 	if (p->p_zone->zone_boot_err != 0) {
2017 		/*
2018 		 * Make sure we are still in the booting state-- we could have
2019 		 * raced and already be shutting down, or even further along.
2020 		 */
2021 		if (zone_status_get(p->p_zone) == ZONE_IS_BOOTING)
2022 			zone_status_set(p->p_zone, ZONE_IS_SHUTTING_DOWN);
2023 		mutex_exit(&zone_status_lock);
2024 		/* It's gone bad, dispose of the process */
2025 		if (proc_exit(CLD_EXITED, p->p_zone->zone_boot_err) != 0) {
2026 			mutex_enter(&p->p_lock);
2027 			ASSERT(p->p_flag & SEXITLWPS);
2028 			lwp_exit();
2029 		}
2030 	} else {
2031 		if (zone_status_get(p->p_zone) == ZONE_IS_BOOTING)
2032 			zone_status_set(p->p_zone, ZONE_IS_RUNNING);
2033 		mutex_exit(&zone_status_lock);
2034 		/* cause the process to return to userland. */
2035 		lwp_rtt();
2036 	}
2037 }
2038 
2039 struct zsched_arg {
2040 	zone_t *zone;
2041 	nvlist_t *nvlist;
2042 };
2043 
2044 /*
2045  * Per-zone "sched" workalike.  The similarity to "sched" doesn't have
2046  * anything to do with scheduling, but rather with the fact that
2047  * per-zone kernel threads are parented to zsched, just like regular
2048  * kernel threads are parented to sched (p0).
2049  *
2050  * zsched is also responsible for launching init for the zone.
2051  */
2052 static void
2053 zsched(void *arg)
2054 {
2055 	struct zsched_arg *za = arg;
2056 	proc_t *pp = curproc;
2057 	proc_t *initp = proc_init;
2058 	zone_t *zone = za->zone;
2059 	cred_t *cr, *oldcred;
2060 	rctl_set_t *set;
2061 	rctl_alloc_gp_t *gp;
2062 	contract_t *ct = NULL;
2063 	task_t *tk, *oldtk;
2064 	rctl_entity_p_t e;
2065 	kproject_t *pj;
2066 
2067 	nvlist_t *nvl = za->nvlist;
2068 	nvpair_t *nvp = NULL;
2069 
2070 	bcopy("zsched", u.u_psargs, sizeof ("zsched"));
2071 	bcopy("zsched", u.u_comm, sizeof ("zsched"));
2072 	u.u_argc = 0;
2073 	u.u_argv = NULL;
2074 	u.u_envp = NULL;
2075 	closeall(P_FINFO(pp));
2076 
2077 	/*
2078 	 * We are this zone's "zsched" process.  As the zone isn't generally
2079 	 * visible yet we don't need to grab any locks before initializing its
2080 	 * zone_proc pointer.
2081 	 */
2082 	zone_hold(zone);  /* this hold is released by zone_destroy() */
2083 	zone->zone_zsched = pp;
2084 	mutex_enter(&pp->p_lock);
2085 	pp->p_zone = zone;
2086 	mutex_exit(&pp->p_lock);
2087 
2088 	/*
2089 	 * Disassociate process from its 'parent'; parent ourselves to init
2090 	 * (pid 1) and change other values as needed.
2091 	 */
2092 	sess_create();
2093 
2094 	mutex_enter(&pidlock);
2095 	proc_detach(pp);
2096 	pp->p_ppid = 1;
2097 	pp->p_flag |= SZONETOP;
2098 	pp->p_ancpid = 1;
2099 	pp->p_parent = initp;
2100 	pp->p_psibling = NULL;
2101 	if (initp->p_child)
2102 		initp->p_child->p_psibling = pp;
2103 	pp->p_sibling = initp->p_child;
2104 	initp->p_child = pp;
2105 
2106 	/* Decrement what newproc() incremented. */
2107 	upcount_dec(crgetruid(CRED()), GLOBAL_ZONEID);
2108 	/*
2109 	 * Our credentials are about to become kcred-like, so we don't care
2110 	 * about the caller's ruid.
2111 	 */
2112 	upcount_inc(crgetruid(kcred), zone->zone_id);
2113 	mutex_exit(&pidlock);
2114 
2115 	/*
2116 	 * getting out of global zone, so decrement lwp counts
2117 	 */
2118 	pj = pp->p_task->tk_proj;
2119 	mutex_enter(&global_zone->zone_nlwps_lock);
2120 	pj->kpj_nlwps -= pp->p_lwpcnt;
2121 	global_zone->zone_nlwps -= pp->p_lwpcnt;
2122 	mutex_exit(&global_zone->zone_nlwps_lock);
2123 
2124 	/*
2125 	 * Create and join a new task in project '0' of this zone.
2126 	 *
2127 	 * We don't need to call holdlwps() since we know we're the only lwp in
2128 	 * this process.
2129 	 *
2130 	 * task_join() returns with p_lock held.
2131 	 */
2132 	tk = task_create(0, zone);
2133 	mutex_enter(&cpu_lock);
2134 	oldtk = task_join(tk, 0);
2135 	mutex_exit(&curproc->p_lock);
2136 	mutex_exit(&cpu_lock);
2137 	task_rele(oldtk);
2138 
2139 	/*
2140 	 * add lwp counts to zsched's zone, and increment project's task count
2141 	 * due to the task created in the above tasksys_settaskid
2142 	 */
2143 	pj = pp->p_task->tk_proj;
2144 	mutex_enter(&zone->zone_nlwps_lock);
2145 	pj->kpj_nlwps += pp->p_lwpcnt;
2146 	pj->kpj_ntasks += 1;
2147 	zone->zone_nlwps += pp->p_lwpcnt;
2148 	mutex_exit(&zone->zone_nlwps_lock);
2149 
2150 	/*
2151 	 * The process was created by a process in the global zone, hence the
2152 	 * credentials are wrong.  We might as well have kcred-ish credentials.
2153 	 */
2154 	cr = zone->zone_kcred;
2155 	crhold(cr);
2156 	mutex_enter(&pp->p_crlock);
2157 	oldcred = pp->p_cred;
2158 	pp->p_cred = cr;
2159 	mutex_exit(&pp->p_crlock);
2160 	crfree(oldcred);
2161 
2162 	/*
2163 	 * Hold credentials again (for thread)
2164 	 */
2165 	crhold(cr);
2166 
2167 	/*
2168 	 * p_lwpcnt can't change since this is a kernel process.
2169 	 */
2170 	crset(pp, cr);
2171 
2172 	/*
2173 	 * Chroot
2174 	 */
2175 	zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_cdir, pp);
2176 	zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_rdir, pp);
2177 
2178 	/*
2179 	 * Initialize zone's rctl set.
2180 	 */
2181 	set = rctl_set_create();
2182 	gp = rctl_set_init_prealloc(RCENTITY_ZONE);
2183 	mutex_enter(&pp->p_lock);
2184 	e.rcep_p.zone = zone;
2185 	e.rcep_t = RCENTITY_ZONE;
2186 	zone->zone_rctls = rctl_set_init(RCENTITY_ZONE, pp, &e, set, gp);
2187 	mutex_exit(&pp->p_lock);
2188 	rctl_prealloc_destroy(gp);
2189 
2190 	/*
2191 	 * Apply the rctls passed in to zone_create().  This is basically a list
2192 	 * assignment: all of the old values are removed and the new ones
2193 	 * inserted.  That is, if an empty list is passed in, all values are
2194 	 * removed.
2195 	 */
2196 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
2197 		rctl_dict_entry_t *rde;
2198 		rctl_hndl_t hndl;
2199 		char *name;
2200 		nvlist_t **nvlarray;
2201 		uint_t i, nelem;
2202 		int error;	/* For ASSERT()s */
2203 
2204 		name = nvpair_name(nvp);
2205 		hndl = rctl_hndl_lookup(name);
2206 		ASSERT(hndl != -1);
2207 		rde = rctl_dict_lookup_hndl(hndl);
2208 		ASSERT(rde != NULL);
2209 
2210 		for (; /* ever */; ) {
2211 			rctl_val_t oval;
2212 
2213 			mutex_enter(&pp->p_lock);
2214 			error = rctl_local_get(hndl, NULL, &oval, pp);
2215 			mutex_exit(&pp->p_lock);
2216 			ASSERT(error == 0);	/* Can't fail for RCTL_FIRST */
2217 			ASSERT(oval.rcv_privilege != RCPRIV_BASIC);
2218 			if (oval.rcv_privilege == RCPRIV_SYSTEM)
2219 				break;
2220 			mutex_enter(&pp->p_lock);
2221 			error = rctl_local_delete(hndl, &oval, pp);
2222 			mutex_exit(&pp->p_lock);
2223 			ASSERT(error == 0);
2224 		}
2225 		error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
2226 		ASSERT(error == 0);
2227 		for (i = 0; i < nelem; i++) {
2228 			rctl_val_t *nvalp;
2229 
2230 			nvalp = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
2231 			error = nvlist2rctlval(nvlarray[i], nvalp);
2232 			ASSERT(error == 0);
2233 			/*
2234 			 * rctl_local_insert can fail if the value being
2235 			 * inserted is a duplicate; this is OK.
2236 			 */
2237 			mutex_enter(&pp->p_lock);
2238 			if (rctl_local_insert(hndl, nvalp, pp) != 0)
2239 				kmem_cache_free(rctl_val_cache, nvalp);
2240 			mutex_exit(&pp->p_lock);
2241 		}
2242 	}
2243 	/*
2244 	 * Tell the world that we're done setting up.
2245 	 *
2246 	 * At this point we want to set the zone status to ZONE_IS_READY
2247 	 * and atomically set the zone's processor set visibility.  Once
2248 	 * we drop pool_lock() this zone will automatically get updated
2249 	 * to reflect any future changes to the pools configuration.
2250 	 */
2251 	pool_lock();
2252 	mutex_enter(&cpu_lock);
2253 	mutex_enter(&zonehash_lock);
2254 	zone_uniqid(zone);
2255 	zone_zsd_configure(zone);
2256 	if (pool_state == POOL_ENABLED)
2257 		zone_pset_set(zone, pool_default->pool_pset->pset_id);
2258 	mutex_enter(&zone_status_lock);
2259 	ASSERT(zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
2260 	zone_status_set(zone, ZONE_IS_READY);
2261 	mutex_exit(&zone_status_lock);
2262 	mutex_exit(&zonehash_lock);
2263 	mutex_exit(&cpu_lock);
2264 	pool_unlock();
2265 
2266 	/*
2267 	 * Once we see the zone transition to the ZONE_IS_BOOTING state,
2268 	 * we launch init, and set the state to running.
2269 	 */
2270 	zone_status_wait_cpr(zone, ZONE_IS_BOOTING, "zsched");
2271 
2272 	if (zone_status_get(zone) == ZONE_IS_BOOTING) {
2273 		id_t cid;
2274 
2275 		/*
2276 		 * Ok, this is a little complicated.  We need to grab the
2277 		 * zone's pool's scheduling class ID; note that by now, we
2278 		 * are already bound to a pool if we need to be (zoneadmd
2279 		 * will have done that to us while we're in the READY
2280 		 * state).  *But* the scheduling class for the zone's 'init'
2281 		 * must be explicitly passed to newproc, which doesn't
2282 		 * respect pool bindings.
2283 		 *
2284 		 * We hold the pool_lock across the call to newproc() to
2285 		 * close the obvious race: the pool's scheduling class
2286 		 * could change before we manage to create the LWP with
2287 		 * classid 'cid'.
2288 		 */
2289 		pool_lock();
2290 		cid = pool_get_class(zone->zone_pool);
2291 		if (cid == -1)
2292 			cid = defaultcid;
2293 
2294 		/*
2295 		 * If this fails, zone_boot will ultimately fail.  The
2296 		 * state of the zone will be set to SHUTTING_DOWN-- userland
2297 		 * will have to tear down the zone, and fail, or try again.
2298 		 */
2299 		if ((zone->zone_boot_err = newproc(zone_icode, NULL, cid,
2300 		    minclsyspri - 1, &ct)) != 0) {
2301 			mutex_enter(&zone_status_lock);
2302 			zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
2303 			mutex_exit(&zone_status_lock);
2304 		}
2305 		pool_unlock();
2306 	}
2307 
2308 	/*
2309 	 * Wait for zone_destroy() to be called.  This is what we spend
2310 	 * most of our life doing.
2311 	 */
2312 	zone_status_wait_cpr(zone, ZONE_IS_DYING, "zsched");
2313 
2314 	if (ct)
2315 		/*
2316 		 * At this point the process contract should be empty.
2317 		 * (Though if it isn't, it's not the end of the world.)
2318 		 */
2319 		VERIFY(contract_abandon(ct, curproc, B_TRUE) == 0);
2320 
2321 	/*
2322 	 * Allow kcred to be freed when all referring processes
2323 	 * (including this one) go away.  We can't just do this in
2324 	 * zone_free because we need to wait for the zone_cred_ref to
2325 	 * drop to 0 before calling zone_free, and the existence of
2326 	 * zone_kcred will prevent that.  Thus, we call crfree here to
2327 	 * balance the crdup in zone_create.  The crhold calls earlier
2328 	 * in zsched will be dropped when the thread and process exit.
2329 	 */
2330 	crfree(zone->zone_kcred);
2331 	zone->zone_kcred = NULL;
2332 
2333 	exit(CLD_EXITED, 0);
2334 }
2335 
2336 /*
2337  * Helper function to determine if there are any submounts of the
2338  * provided path.  Used to make sure the zone doesn't "inherit" any
2339  * mounts from before it is created.
2340  */
2341 static uint_t
2342 zone_mount_count(const char *rootpath)
2343 {
2344 	vfs_t *vfsp;
2345 	uint_t count = 0;
2346 	size_t rootpathlen = strlen(rootpath);
2347 
2348 	/*
2349 	 * Holding zonehash_lock prevents race conditions with
2350 	 * vfs_list_add()/vfs_list_remove() since we serialize with
2351 	 * zone_find_by_path().
2352 	 */
2353 	ASSERT(MUTEX_HELD(&zonehash_lock));
2354 	/*
2355 	 * The rootpath must end with a '/'
2356 	 */
2357 	ASSERT(rootpath[rootpathlen - 1] == '/');
2358 
2359 	/*
2360 	 * This intentionally does not count the rootpath itself if that
2361 	 * happens to be a mount point.
2362 	 */
2363 	vfs_list_read_lock();
2364 	vfsp = rootvfs;
2365 	do {
2366 		if (strncmp(rootpath, refstr_value(vfsp->vfs_mntpt),
2367 		    rootpathlen) == 0)
2368 			count++;
2369 		vfsp = vfsp->vfs_next;
2370 	} while (vfsp != rootvfs);
2371 	vfs_list_unlock();
2372 	return (count);
2373 }
2374 
2375 /*
2376  * Helper function to make sure that a zone created on 'rootpath'
2377  * wouldn't end up containing other zones' rootpaths.
2378  */
2379 static boolean_t
2380 zone_is_nested(const char *rootpath)
2381 {
2382 	zone_t *zone;
2383 	size_t rootpathlen = strlen(rootpath);
2384 	size_t len;
2385 
2386 	ASSERT(MUTEX_HELD(&zonehash_lock));
2387 
2388 	for (zone = list_head(&zone_active); zone != NULL;
2389 	    zone = list_next(&zone_active, zone)) {
2390 		if (zone == global_zone)
2391 			continue;
2392 		len = strlen(zone->zone_rootpath);
2393 		if (strncmp(rootpath, zone->zone_rootpath,
2394 		    MIN(rootpathlen, len)) == 0)
2395 			return (B_TRUE);
2396 	}
2397 	return (B_FALSE);
2398 }
2399 
2400 static int
2401 zone_set_privset(zone_t *zone, const priv_set_t *zone_privs)
2402 {
2403 	priv_set_t *privs = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
2404 
2405 	if (copyin(zone_privs, privs, sizeof (priv_set_t))) {
2406 		kmem_free(privs, sizeof (priv_set_t));
2407 		return (EFAULT);
2408 	}
2409 
2410 	zone->zone_privset = privs;
2411 	return (0);
2412 }
2413 
2414 /*
2415  * We make creative use of nvlists to pass in rctls from userland.  The list is
2416  * a list of the following structures:
2417  *
2418  * (name = rctl_name, value = nvpair_list_array)
2419  *
2420  * Where each element of the nvpair_list_array is of the form:
2421  *
2422  * [(name = "privilege", value = RCPRIV_PRIVILEGED),
2423  * 	(name = "limit", value = uint64_t),
2424  * 	(name = "action", value = (RCTL_LOCAL_NOACTION || RCTL_LOCAL_DENY))]
2425  */
2426 static int
2427 parse_rctls(caddr_t ubuf, size_t buflen, nvlist_t **nvlp)
2428 {
2429 	nvpair_t *nvp = NULL;
2430 	nvlist_t *nvl = NULL;
2431 	char *kbuf;
2432 	int error;
2433 	rctl_val_t rv;
2434 
2435 	*nvlp = NULL;
2436 
2437 	if (buflen == 0)
2438 		return (0);
2439 
2440 	if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL)
2441 		return (ENOMEM);
2442 	if (copyin(ubuf, kbuf, buflen)) {
2443 		error = EFAULT;
2444 		goto out;
2445 	}
2446 	if (nvlist_unpack(kbuf, buflen, &nvl, KM_SLEEP) != 0) {
2447 		/*
2448 		 * nvl may have been allocated/free'd, but the value set to
2449 		 * non-NULL, so we reset it here.
2450 		 */
2451 		nvl = NULL;
2452 		error = EINVAL;
2453 		goto out;
2454 	}
2455 	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
2456 		rctl_dict_entry_t *rde;
2457 		rctl_hndl_t hndl;
2458 		nvlist_t **nvlarray;
2459 		uint_t i, nelem;
2460 		char *name;
2461 
2462 		error = EINVAL;
2463 		name = nvpair_name(nvp);
2464 		if (strncmp(nvpair_name(nvp), "zone.", sizeof ("zone.") - 1)
2465 		    != 0 || nvpair_type(nvp) != DATA_TYPE_NVLIST_ARRAY) {
2466 			goto out;
2467 		}
2468 		if ((hndl = rctl_hndl_lookup(name)) == -1) {
2469 			goto out;
2470 		}
2471 		rde = rctl_dict_lookup_hndl(hndl);
2472 		error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
2473 		ASSERT(error == 0);
2474 		for (i = 0; i < nelem; i++) {
2475 			if (error = nvlist2rctlval(nvlarray[i], &rv))
2476 				goto out;
2477 		}
2478 		if (rctl_invalid_value(rde, &rv)) {
2479 			error = EINVAL;
2480 			goto out;
2481 		}
2482 	}
2483 	error = 0;
2484 	*nvlp = nvl;
2485 out:
2486 	kmem_free(kbuf, buflen);
2487 	if (error && nvl != NULL)
2488 		nvlist_free(nvl);
2489 	return (error);
2490 }
2491 
2492 int
2493 zone_create_error(int er_error, int er_ext, int *er_out) {
2494 	if (er_out != NULL) {
2495 		if (copyout(&er_ext, er_out, sizeof (int))) {
2496 			return (set_errno(EFAULT));
2497 		}
2498 	}
2499 	return (set_errno(er_error));
2500 }
2501 
2502 /*
2503  * System call to create/initialize a new zone named 'zone_name', rooted
2504  * at 'zone_root', with a zone-wide privilege limit set of 'zone_privs',
2505  * and initialized with the zone-wide rctls described in 'rctlbuf'.
2506  *
2507  * If extended error is non-null, we may use it to return more detailed
2508  * error information.
2509  */
2510 static zoneid_t
2511 zone_create(const char *zone_name, const char *zone_root,
2512     const priv_set_t *zone_privs, caddr_t rctlbuf, size_t rctlbufsz,
2513     int *extended_error)
2514 {
2515 	struct zsched_arg zarg;
2516 	nvlist_t *rctls = NULL;
2517 	proc_t *pp = curproc;
2518 	zone_t *zone, *ztmp;
2519 	zoneid_t zoneid;
2520 	int error;
2521 	int error2 = 0;
2522 	char *str;
2523 	cred_t *zkcr;
2524 
2525 	if (secpolicy_zone_config(CRED()) != 0)
2526 		return (set_errno(EPERM));
2527 
2528 	/* can't boot zone from within chroot environment */
2529 	if (PTOU(pp)->u_rdir != NULL && PTOU(pp)->u_rdir != rootdir)
2530 		return (zone_create_error(ENOTSUP, ZE_CHROOTED,
2531 			extended_error));
2532 
2533 	zone = kmem_zalloc(sizeof (zone_t), KM_SLEEP);
2534 	zoneid = zone->zone_id = id_alloc(zoneid_space);
2535 	zone->zone_status = ZONE_IS_UNINITIALIZED;
2536 	zone->zone_pool = pool_default;
2537 	zone->zone_pool_mod = gethrtime();
2538 	zone->zone_psetid = ZONE_PS_INVAL;
2539 	zone->zone_ncpus = 0;
2540 	zone->zone_ncpus_online = 0;
2541 	mutex_init(&zone->zone_lock, NULL, MUTEX_DEFAULT, NULL);
2542 	mutex_init(&zone->zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
2543 	cv_init(&zone->zone_cv, NULL, CV_DEFAULT, NULL);
2544 	list_create(&zone->zone_zsd, sizeof (struct zsd_entry),
2545 	    offsetof(struct zsd_entry, zsd_linkage));
2546 
2547 	if ((error = zone_set_name(zone, zone_name)) != 0) {
2548 		zone_free(zone);
2549 		return (zone_create_error(error, 0, extended_error));
2550 	}
2551 
2552 	if ((error = zone_set_root(zone, zone_root)) != 0) {
2553 		zone_free(zone);
2554 		return (zone_create_error(error, 0, extended_error));
2555 	}
2556 	if ((error = zone_set_privset(zone, zone_privs)) != 0) {
2557 		zone_free(zone);
2558 		return (zone_create_error(error, 0, extended_error));
2559 	}
2560 
2561 	/* initialize node name to be the same as zone name */
2562 	zone->zone_nodename = kmem_alloc(_SYS_NMLN, KM_SLEEP);
2563 	(void) strncpy(zone->zone_nodename, zone->zone_name, _SYS_NMLN);
2564 	zone->zone_nodename[_SYS_NMLN - 1] = '\0';
2565 
2566 	zone->zone_domain = kmem_alloc(_SYS_NMLN, KM_SLEEP);
2567 	zone->zone_domain[0] = '\0';
2568 	zone->zone_shares = 1;
2569 	zone->zone_bootargs = NULL;
2570 
2571 	/*
2572 	 * Zsched initializes the rctls.
2573 	 */
2574 	zone->zone_rctls = NULL;
2575 
2576 	if ((error = parse_rctls(rctlbuf, rctlbufsz, &rctls)) != 0) {
2577 		zone_free(zone);
2578 		return (zone_create_error(error, 0, extended_error));
2579 	}
2580 
2581 	/*
2582 	 * Stop all lwps since that's what normally happens as part of fork().
2583 	 * This needs to happen before we grab any locks to avoid deadlock
2584 	 * (another lwp in the process could be waiting for the held lock).
2585 	 */
2586 	if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK)) {
2587 		zone_free(zone);
2588 		if (rctls)
2589 			nvlist_free(rctls);
2590 		return (zone_create_error(error, 0, extended_error));
2591 	}
2592 
2593 	if (block_mounts() == 0) {
2594 		mutex_enter(&pp->p_lock);
2595 		if (curthread != pp->p_agenttp)
2596 			continuelwps(pp);
2597 		mutex_exit(&pp->p_lock);
2598 		zone_free(zone);
2599 		if (rctls)
2600 			nvlist_free(rctls);
2601 		return (zone_create_error(error, 0, extended_error));
2602 	}
2603 
2604 	/*
2605 	 * Set up credential for kernel access.  After this, any errors
2606 	 * should go through the dance in errout rather than calling
2607 	 * zone_free directly.
2608 	 */
2609 	zone->zone_kcred = crdup(kcred);
2610 	crsetzone(zone->zone_kcred, zone);
2611 	priv_intersect(zone->zone_privset, &CR_PPRIV(zone->zone_kcred));
2612 	priv_intersect(zone->zone_privset, &CR_EPRIV(zone->zone_kcred));
2613 	priv_intersect(zone->zone_privset, &CR_IPRIV(zone->zone_kcred));
2614 	priv_intersect(zone->zone_privset, &CR_LPRIV(zone->zone_kcred));
2615 
2616 	mutex_enter(&zonehash_lock);
2617 	/*
2618 	 * Make sure zone doesn't already exist.
2619 	 */
2620 	if ((ztmp = zone_find_all_by_name(zone->zone_name)) != NULL) {
2621 		zone_status_t status;
2622 
2623 		status = zone_status_get(ztmp);
2624 		if (status == ZONE_IS_READY || status == ZONE_IS_RUNNING)
2625 			error = EEXIST;
2626 		else
2627 			error = EBUSY;
2628 		goto errout;
2629 	}
2630 
2631 	/*
2632 	 * Don't allow zone creations which would cause one zone's rootpath to
2633 	 * be accessible from that of another (non-global) zone.
2634 	 */
2635 	if (zone_is_nested(zone->zone_rootpath)) {
2636 		error = EBUSY;
2637 		goto errout;
2638 	}
2639 
2640 	ASSERT(zonecount != 0);		/* check for leaks */
2641 	if (zonecount + 1 > maxzones) {
2642 		error = ENOMEM;
2643 		goto errout;
2644 	}
2645 
2646 	if (zone_mount_count(zone->zone_rootpath) != 0) {
2647 		error = EBUSY;
2648 		error2 = ZE_AREMOUNTS;
2649 		goto errout;
2650 	}
2651 
2652 	/*
2653 	 * Zone is still incomplete, but we need to drop all locks while
2654 	 * zsched() initializes this zone's kernel process.  We
2655 	 * optimistically add the zone to the hashtable and associated
2656 	 * lists so a parallel zone_create() doesn't try to create the
2657 	 * same zone.
2658 	 */
2659 	zonecount++;
2660 	(void) mod_hash_insert(zonehashbyid,
2661 	    (mod_hash_key_t)(uintptr_t)zone->zone_id,
2662 	    (mod_hash_val_t)(uintptr_t)zone);
2663 	str = kmem_alloc(strlen(zone->zone_name) + 1, KM_SLEEP);
2664 	(void) strcpy(str, zone->zone_name);
2665 	(void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)str,
2666 	    (mod_hash_val_t)(uintptr_t)zone);
2667 	/*
2668 	 * Insert into active list.  At this point there are no 'hold's
2669 	 * on the zone, but everyone else knows not to use it, so we can
2670 	 * continue to use it.  zsched() will do a zone_hold() if the
2671 	 * newproc() is successful.
2672 	 */
2673 	list_insert_tail(&zone_active, zone);
2674 	mutex_exit(&zonehash_lock);
2675 
2676 	zarg.zone = zone;
2677 	zarg.nvlist = rctls;
2678 	/*
2679 	 * The process, task, and project rctls are probably wrong;
2680 	 * we need an interface to get the default values of all rctls,
2681 	 * and initialize zsched appropriately.  I'm not sure that that
2682 	 * makes much of a difference, though.
2683 	 */
2684 	if (error = newproc(zsched, (void *)&zarg, syscid, minclsyspri, NULL)) {
2685 		/*
2686 		 * We need to undo all globally visible state.
2687 		 */
2688 		mutex_enter(&zonehash_lock);
2689 		list_remove(&zone_active, zone);
2690 		(void) mod_hash_destroy(zonehashbyname,
2691 		    (mod_hash_key_t)(uintptr_t)zone->zone_name);
2692 		(void) mod_hash_destroy(zonehashbyid,
2693 		    (mod_hash_key_t)(uintptr_t)zone->zone_id);
2694 		ASSERT(zonecount > 1);
2695 		zonecount--;
2696 		goto errout;
2697 	}
2698 
2699 	/*
2700 	 * Zone creation can't fail from now on.
2701 	 */
2702 
2703 	/*
2704 	 * Let the other lwps continue.
2705 	 */
2706 	mutex_enter(&pp->p_lock);
2707 	if (curthread != pp->p_agenttp)
2708 		continuelwps(pp);
2709 	mutex_exit(&pp->p_lock);
2710 
2711 	/*
2712 	 * Wait for zsched to finish initializing the zone.
2713 	 */
2714 	zone_status_wait(zone, ZONE_IS_READY);
2715 	/*
2716 	 * The zone is fully visible, so we can let mounts progress.
2717 	 */
2718 	resume_mounts();
2719 	if (rctls)
2720 		nvlist_free(rctls);
2721 
2722 	return (zoneid);
2723 
2724 errout:
2725 	mutex_exit(&zonehash_lock);
2726 	/*
2727 	 * Let the other lwps continue.
2728 	 */
2729 	mutex_enter(&pp->p_lock);
2730 	if (curthread != pp->p_agenttp)
2731 		continuelwps(pp);
2732 	mutex_exit(&pp->p_lock);
2733 
2734 	resume_mounts();
2735 	if (rctls)
2736 		nvlist_free(rctls);
2737 	/*
2738 	 * There is currently one reference to the zone, a cred_ref from
2739 	 * zone_kcred.  To free the zone, we call crfree, which will call
2740 	 * zone_cred_rele, which will call zone_free.
2741 	 */
2742 	ASSERT(zone->zone_cred_ref == 1);	/* for zone_kcred */
2743 	ASSERT(zone->zone_kcred->cr_ref == 1);
2744 	ASSERT(zone->zone_ref == 0);
2745 	zkcr = zone->zone_kcred;
2746 	zone->zone_kcred = NULL;
2747 	crfree(zkcr);				/* triggers call to zone_free */
2748 	return (zone_create_error(error, error2, extended_error));
2749 }
2750 
2751 /*
2752  * Cause the zone to boot.  This is pretty simple, since we let zoneadmd do
2753  * the heavy lifting.
2754  */
2755 static int
2756 zone_boot(zoneid_t zoneid, const char *bootargs)
2757 {
2758 	int err;
2759 	zone_t *zone;
2760 
2761 	if (secpolicy_zone_config(CRED()) != 0)
2762 		return (set_errno(EPERM));
2763 	if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
2764 		return (set_errno(EINVAL));
2765 
2766 	mutex_enter(&zonehash_lock);
2767 	/*
2768 	 * Look for zone under hash lock to prevent races with calls to
2769 	 * zone_shutdown, zone_destroy, etc.
2770 	 */
2771 	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
2772 		mutex_exit(&zonehash_lock);
2773 		return (set_errno(EINVAL));
2774 	}
2775 
2776 	if ((err = zone_set_bootargs(zone, bootargs)) != 0) {
2777 		mutex_exit(&zonehash_lock);
2778 		return (set_errno(err));
2779 	}
2780 
2781 	mutex_enter(&zone_status_lock);
2782 	if (zone_status_get(zone) != ZONE_IS_READY) {
2783 		mutex_exit(&zone_status_lock);
2784 		mutex_exit(&zonehash_lock);
2785 		return (set_errno(EINVAL));
2786 	}
2787 	zone_status_set(zone, ZONE_IS_BOOTING);
2788 	mutex_exit(&zone_status_lock);
2789 
2790 	zone_hold(zone);	/* so we can use the zone_t later */
2791 	mutex_exit(&zonehash_lock);
2792 
2793 	if (zone_status_wait_sig(zone, ZONE_IS_RUNNING) == 0) {
2794 		zone_rele(zone);
2795 		return (set_errno(EINTR));
2796 	}
2797 
2798 	/*
2799 	 * Boot (starting init) might have failed, in which case the zone
2800 	 * will go to the SHUTTING_DOWN state; an appropriate errno will
2801 	 * be placed in zone->zone_boot_err, and so we return that.
2802 	 */
2803 	err = zone->zone_boot_err;
2804 	zone_rele(zone);
2805 	return (err ? set_errno(err) : 0);
2806 }
2807 
2808 /*
2809  * Kills all user processes in the zone, waiting for them all to exit
2810  * before returning.
2811  */
2812 static int
2813 zone_empty(zone_t *zone)
2814 {
2815 	int waitstatus;
2816 
2817 	/*
2818 	 * We need to drop zonehash_lock before killing all
2819 	 * processes, otherwise we'll deadlock with zone_find_*
2820 	 * which can be called from the exit path.
2821 	 */
2822 	ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
2823 	while ((waitstatus = zone_status_timedwait_sig(zone, lbolt + hz,
2824 	    ZONE_IS_EMPTY)) == -1) {
2825 		killall(zone->zone_id);
2826 	}
2827 	/*
2828 	 * return EINTR if we were signaled
2829 	 */
2830 	if (waitstatus == 0)
2831 		return (EINTR);
2832 	return (0);
2833 }
2834 
2835 /*
2836  * Systemcall to start the zone's halt sequence.  By the time this
2837  * function successfully returns, all user processes and kernel threads
2838  * executing in it will have exited, ZSD shutdown callbacks executed,
2839  * and the zone status set to ZONE_IS_DOWN.
2840  *
2841  * It is possible that the call will interrupt itself if the caller is the
2842  * parent of any process running in the zone, and doesn't have SIGCHLD blocked.
2843  */
2844 static int
2845 zone_shutdown(zoneid_t zoneid)
2846 {
2847 	int error;
2848 	zone_t *zone;
2849 	zone_status_t status;
2850 
2851 	if (secpolicy_zone_config(CRED()) != 0)
2852 		return (set_errno(EPERM));
2853 	if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
2854 		return (set_errno(EINVAL));
2855 
2856 	/*
2857 	 * Block mounts so that VFS_MOUNT() can get an accurate view of
2858 	 * the zone's status with regards to ZONE_IS_SHUTTING down.
2859 	 *
2860 	 * e.g. NFS can fail the mount if it determines that the zone
2861 	 * has already begun the shutdown sequence.
2862 	 */
2863 	if (block_mounts() == 0)
2864 		return (set_errno(EINTR));
2865 	mutex_enter(&zonehash_lock);
2866 	/*
2867 	 * Look for zone under hash lock to prevent races with other
2868 	 * calls to zone_shutdown and zone_destroy.
2869 	 */
2870 	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
2871 		mutex_exit(&zonehash_lock);
2872 		resume_mounts();
2873 		return (set_errno(EINVAL));
2874 	}
2875 	mutex_enter(&zone_status_lock);
2876 	status = zone_status_get(zone);
2877 	/*
2878 	 * Fail if the zone isn't fully initialized yet.
2879 	 */
2880 	if (status < ZONE_IS_READY) {
2881 		mutex_exit(&zone_status_lock);
2882 		mutex_exit(&zonehash_lock);
2883 		resume_mounts();
2884 		return (set_errno(EINVAL));
2885 	}
2886 	/*
2887 	 * If conditions required for zone_shutdown() to return have been met,
2888 	 * return success.
2889 	 */
2890 	if (status >= ZONE_IS_DOWN) {
2891 		mutex_exit(&zone_status_lock);
2892 		mutex_exit(&zonehash_lock);
2893 		resume_mounts();
2894 		return (0);
2895 	}
2896 	/*
2897 	 * If zone_shutdown() hasn't been called before, go through the motions.
2898 	 * If it has, there's nothing to do but wait for the kernel threads to
2899 	 * drain.
2900 	 */
2901 	if (status < ZONE_IS_EMPTY) {
2902 		uint_t ntasks;
2903 
2904 		mutex_enter(&zone->zone_lock);
2905 		if ((ntasks = zone->zone_ntasks) != 1) {
2906 			/*
2907 			 * There's still stuff running.
2908 			 */
2909 			zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
2910 		}
2911 		mutex_exit(&zone->zone_lock);
2912 		if (ntasks == 1) {
2913 			/*
2914 			 * The only way to create another task is through
2915 			 * zone_enter(), which will block until we drop
2916 			 * zonehash_lock.  The zone is empty.
2917 			 */
2918 			if (zone->zone_kthreads == NULL) {
2919 				/*
2920 				 * Skip ahead to ZONE_IS_DOWN
2921 				 */
2922 				zone_status_set(zone, ZONE_IS_DOWN);
2923 			} else {
2924 				zone_status_set(zone, ZONE_IS_EMPTY);
2925 			}
2926 		}
2927 	}
2928 	zone_hold(zone);	/* so we can use the zone_t later */
2929 	mutex_exit(&zone_status_lock);
2930 	mutex_exit(&zonehash_lock);
2931 	resume_mounts();
2932 
2933 	if (error = zone_empty(zone)) {
2934 		zone_rele(zone);
2935 		return (set_errno(error));
2936 	}
2937 	/*
2938 	 * After the zone status goes to ZONE_IS_DOWN this zone will no
2939 	 * longer be notified of changes to the pools configuration, so
2940 	 * in order to not end up with a stale pool pointer, we point
2941 	 * ourselves at the default pool and remove all resource
2942 	 * visibility.  This is especially important as the zone_t may
2943 	 * languish on the deathrow for a very long time waiting for
2944 	 * cred's to drain out.
2945 	 *
2946 	 * This rebinding of the zone can happen multiple times
2947 	 * (presumably due to interrupted or parallel systemcalls)
2948 	 * without any adverse effects.
2949 	 */
2950 	if (pool_lock_intr() != 0) {
2951 		zone_rele(zone);
2952 		return (set_errno(EINTR));
2953 	}
2954 	if (pool_state == POOL_ENABLED) {
2955 		mutex_enter(&cpu_lock);
2956 		zone_pool_set(zone, pool_default);
2957 		/*
2958 		 * The zone no longer needs to be able to see any cpus.
2959 		 */
2960 		zone_pset_set(zone, ZONE_PS_INVAL);
2961 		mutex_exit(&cpu_lock);
2962 	}
2963 	pool_unlock();
2964 
2965 	/*
2966 	 * ZSD shutdown callbacks can be executed multiple times, hence
2967 	 * it is safe to not be holding any locks across this call.
2968 	 */
2969 	zone_zsd_callbacks(zone, ZSD_SHUTDOWN);
2970 
2971 	mutex_enter(&zone_status_lock);
2972 	if (zone->zone_kthreads == NULL && zone_status_get(zone) < ZONE_IS_DOWN)
2973 		zone_status_set(zone, ZONE_IS_DOWN);
2974 	mutex_exit(&zone_status_lock);
2975 
2976 	/*
2977 	 * Wait for kernel threads to drain.
2978 	 */
2979 	if (!zone_status_wait_sig(zone, ZONE_IS_DOWN)) {
2980 		zone_rele(zone);
2981 		return (set_errno(EINTR));
2982 	}
2983 	zone_rele(zone);
2984 	return (0);
2985 }
2986 
2987 /*
2988  * Systemcall entry point to finalize the zone halt process.  The caller
2989  * must have already successfully callefd zone_shutdown().
2990  *
2991  * Upon successful completion, the zone will have been fully destroyed:
2992  * zsched will have exited, destructor callbacks executed, and the zone
2993  * removed from the list of active zones.
2994  */
2995 static int
2996 zone_destroy(zoneid_t zoneid)
2997 {
2998 	uint64_t uniqid;
2999 	zone_t *zone;
3000 	zone_status_t status;
3001 
3002 	if (secpolicy_zone_config(CRED()) != 0)
3003 		return (set_errno(EPERM));
3004 	if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
3005 		return (set_errno(EINVAL));
3006 
3007 	mutex_enter(&zonehash_lock);
3008 	/*
3009 	 * Look for zone under hash lock to prevent races with other
3010 	 * calls to zone_destroy.
3011 	 */
3012 	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
3013 		mutex_exit(&zonehash_lock);
3014 		return (set_errno(EINVAL));
3015 	}
3016 
3017 	if (zone_mount_count(zone->zone_rootpath) != 0) {
3018 		mutex_exit(&zonehash_lock);
3019 		return (set_errno(EBUSY));
3020 	}
3021 	mutex_enter(&zone_status_lock);
3022 	status = zone_status_get(zone);
3023 	if (status < ZONE_IS_DOWN) {
3024 		mutex_exit(&zone_status_lock);
3025 		mutex_exit(&zonehash_lock);
3026 		return (set_errno(EBUSY));
3027 	} else if (status == ZONE_IS_DOWN) {
3028 		zone_status_set(zone, ZONE_IS_DYING); /* Tell zsched to exit */
3029 	}
3030 	mutex_exit(&zone_status_lock);
3031 	zone_hold(zone);
3032 	mutex_exit(&zonehash_lock);
3033 
3034 	/*
3035 	 * wait for zsched to exit
3036 	 */
3037 	zone_status_wait(zone, ZONE_IS_DEAD);
3038 	zone_zsd_callbacks(zone, ZSD_DESTROY);
3039 	uniqid = zone->zone_uniqid;
3040 	zone_rele(zone);
3041 	zone = NULL;	/* potentially free'd */
3042 
3043 	mutex_enter(&zonehash_lock);
3044 	for (; /* ever */; ) {
3045 		boolean_t unref;
3046 
3047 		if ((zone = zone_find_all_by_id(zoneid)) == NULL ||
3048 		    zone->zone_uniqid != uniqid) {
3049 			/*
3050 			 * The zone has gone away.  Necessary conditions
3051 			 * are met, so we return success.
3052 			 */
3053 			mutex_exit(&zonehash_lock);
3054 			return (0);
3055 		}
3056 		mutex_enter(&zone->zone_lock);
3057 		unref = ZONE_IS_UNREF(zone);
3058 		mutex_exit(&zone->zone_lock);
3059 		if (unref) {
3060 			/*
3061 			 * There is only one reference to the zone -- that
3062 			 * added when the zone was added to the hashtables --
3063 			 * and things will remain this way until we drop
3064 			 * zonehash_lock... we can go ahead and cleanup the
3065 			 * zone.
3066 			 */
3067 			break;
3068 		}
3069 
3070 		if (cv_wait_sig(&zone_destroy_cv, &zonehash_lock) == 0) {
3071 			/* Signaled */
3072 			mutex_exit(&zonehash_lock);
3073 			return (set_errno(EINTR));
3074 		}
3075 
3076 	}
3077 
3078 	/*
3079 	 * It is now safe to let the zone be recreated; remove it from the
3080 	 * lists.  The memory will not be freed until the last cred
3081 	 * reference goes away.
3082 	 */
3083 	ASSERT(zonecount > 1);	/* must be > 1; can't destroy global zone */
3084 	zonecount--;
3085 	/* remove from active list and hash tables */
3086 	list_remove(&zone_active, zone);
3087 	(void) mod_hash_destroy(zonehashbyname,
3088 	    (mod_hash_key_t)zone->zone_name);
3089 	(void) mod_hash_destroy(zonehashbyid,
3090 	    (mod_hash_key_t)(uintptr_t)zone->zone_id);
3091 	mutex_exit(&zonehash_lock);
3092 
3093 	/* add to deathrow list */
3094 	mutex_enter(&zone_deathrow_lock);
3095 	list_insert_tail(&zone_deathrow, zone);
3096 	mutex_exit(&zone_deathrow_lock);
3097 
3098 	/*
3099 	 * Drop last reference (which was added by zsched()), this will
3100 	 * free the zone unless there are outstanding cred references.
3101 	 */
3102 	zone_rele(zone);
3103 	return (0);
3104 }
3105 
3106 /*
3107  * Systemcall entry point for zone_getattr(2).
3108  */
3109 static ssize_t
3110 zone_getattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
3111 {
3112 	size_t size;
3113 	int error = 0, err;
3114 	zone_t *zone;
3115 	char *zonepath;
3116 	zone_status_t zone_status;
3117 	pid_t initpid;
3118 	boolean_t global = (curproc->p_zone == global_zone);
3119 
3120 	mutex_enter(&zonehash_lock);
3121 	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
3122 		mutex_exit(&zonehash_lock);
3123 		return (set_errno(EINVAL));
3124 	}
3125 	zone_status = zone_status_get(zone);
3126 	if (zone_status < ZONE_IS_READY) {
3127 		mutex_exit(&zonehash_lock);
3128 		return (set_errno(EINVAL));
3129 	}
3130 	zone_hold(zone);
3131 	mutex_exit(&zonehash_lock);
3132 
3133 	/*
3134 	 * If not in the global zone, don't show information about other zones.
3135 	 */
3136 	if (!global && curproc->p_zone != zone) {
3137 		zone_rele(zone);
3138 		return (set_errno(EINVAL));
3139 	}
3140 
3141 	switch (attr) {
3142 	case ZONE_ATTR_ROOT:
3143 		if (global) {
3144 			/*
3145 			 * Copy the path to trim the trailing "/" (except for
3146 			 * the global zone).
3147 			 */
3148 			if (zone != global_zone)
3149 				size = zone->zone_rootpathlen - 1;
3150 			else
3151 				size = zone->zone_rootpathlen;
3152 			zonepath = kmem_alloc(size, KM_SLEEP);
3153 			bcopy(zone->zone_rootpath, zonepath, size);
3154 			zonepath[size - 1] = '\0';
3155 		} else {
3156 			/*
3157 			 * Caller is not in the global zone, just return
3158 			 * faked-up path for current zone.
3159 			 */
3160 			zonepath = "/";
3161 			size = 2;
3162 		}
3163 		if (bufsize > size)
3164 			bufsize = size;
3165 		if (buf != NULL) {
3166 			err = copyoutstr(zonepath, buf, bufsize, NULL);
3167 			if (err != 0 && err != ENAMETOOLONG)
3168 				error = EFAULT;
3169 		}
3170 		if (global)
3171 			kmem_free(zonepath, size);
3172 		break;
3173 
3174 	case ZONE_ATTR_NAME:
3175 		size = strlen(zone->zone_name) + 1;
3176 		if (bufsize > size)
3177 			bufsize = size;
3178 		if (buf != NULL) {
3179 			err = copyoutstr(zone->zone_name, buf, bufsize, NULL);
3180 			if (err != 0 && err != ENAMETOOLONG)
3181 				error = EFAULT;
3182 		}
3183 		break;
3184 
3185 	case ZONE_ATTR_STATUS:
3186 		/*
3187 		 * Since we're not holding zonehash_lock, the zone status
3188 		 * may be anything; leave it up to userland to sort it out.
3189 		 */
3190 		size = sizeof (zone_status);
3191 		if (bufsize > size)
3192 			bufsize = size;
3193 		zone_status = zone_status_get(zone);
3194 		if (buf != NULL &&
3195 		    copyout(&zone_status, buf, bufsize) != 0)
3196 			error = EFAULT;
3197 		break;
3198 	case ZONE_ATTR_PRIVSET:
3199 		size = sizeof (priv_set_t);
3200 		if (bufsize > size)
3201 			bufsize = size;
3202 		if (buf != NULL &&
3203 		    copyout(zone->zone_privset, buf, bufsize) != 0)
3204 			error = EFAULT;
3205 		break;
3206 	case ZONE_ATTR_UNIQID:
3207 		size = sizeof (zone->zone_uniqid);
3208 		if (bufsize > size)
3209 			bufsize = size;
3210 		if (buf != NULL &&
3211 		    copyout(&zone->zone_uniqid, buf, bufsize) != 0)
3212 			error = EFAULT;
3213 		break;
3214 	case ZONE_ATTR_POOLID:
3215 		{
3216 			pool_t *pool;
3217 			poolid_t poolid;
3218 
3219 			if (pool_lock_intr() != 0) {
3220 				error = EINTR;
3221 				break;
3222 			}
3223 			pool = zone_pool_get(zone);
3224 			poolid = pool->pool_id;
3225 			pool_unlock();
3226 			size = sizeof (poolid);
3227 			if (bufsize > size)
3228 				bufsize = size;
3229 			if (buf != NULL && copyout(&poolid, buf, size) != 0)
3230 				error = EFAULT;
3231 		}
3232 		break;
3233 	case ZONE_ATTR_INITPID:
3234 		size = sizeof (initpid);
3235 		if (bufsize > size)
3236 			bufsize = size;
3237 		initpid = zone->zone_proc_initpid;
3238 		if (initpid == -1) {
3239 			error = ESRCH;
3240 			break;
3241 		}
3242 		if (buf != NULL &&
3243 		    copyout(&initpid, buf, bufsize) != 0)
3244 			error = EFAULT;
3245 		break;
3246 	default:
3247 		error = EINVAL;
3248 	}
3249 	zone_rele(zone);
3250 
3251 	if (error)
3252 		return (set_errno(error));
3253 	return ((ssize_t)size);
3254 }
3255 
3256 /*
3257  * Return zero if the process has at least one vnode mapped in to its
3258  * address space which shouldn't be allowed to change zones.
3259  */
3260 static int
3261 as_can_change_zones(void)
3262 {
3263 	proc_t *pp = curproc;
3264 	struct seg *seg;
3265 	struct as *as = pp->p_as;
3266 	vnode_t *vp;
3267 	int allow = 1;
3268 
3269 	ASSERT(pp->p_as != &kas);
3270 	AS_LOCK_ENTER(&as, &as->a_lock, RW_READER);
3271 	for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) {
3272 		/*
3273 		 * if we can't get a backing vnode for this segment then skip
3274 		 * it.
3275 		 */
3276 		vp = NULL;
3277 		if (SEGOP_GETVP(seg, seg->s_base, &vp) != 0 || vp == NULL)
3278 			continue;
3279 		if (!vn_can_change_zones(vp)) { /* bail on first match */
3280 			allow = 0;
3281 			break;
3282 		}
3283 	}
3284 	AS_LOCK_EXIT(&as, &as->a_lock);
3285 	return (allow);
3286 }
3287 
3288 /*
3289  * Systemcall entry point for zone_enter().
3290  *
3291  * The current process is injected into said zone.  In the process
3292  * it will change its project membership, privileges, rootdir/cwd,
3293  * zone-wide rctls, and pool association to match those of the zone.
3294  *
3295  * The first zone_enter() called while the zone is in the ZONE_IS_READY
3296  * state will transition it to ZONE_IS_RUNNING.  Processes may only
3297  * enter a zone that is "ready" or "running".
3298  */
3299 static int
3300 zone_enter(zoneid_t zoneid)
3301 {
3302 	zone_t *zone;
3303 	vnode_t *vp;
3304 	proc_t *pp = curproc;
3305 	contract_t *ct;
3306 	cont_process_t *ctp;
3307 	task_t *tk, *oldtk;
3308 	kproject_t *zone_proj0;
3309 	cred_t *cr, *newcr;
3310 	pool_t *oldpool, *newpool;
3311 	sess_t *sp;
3312 	uid_t uid;
3313 	zone_status_t status;
3314 	int err = 0;
3315 	rctl_entity_p_t e;
3316 
3317 	if (secpolicy_zone_config(CRED()) != 0)
3318 		return (set_errno(EPERM));
3319 	if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
3320 		return (set_errno(EINVAL));
3321 
3322 	/*
3323 	 * Stop all lwps so we don't need to hold a lock to look at
3324 	 * curproc->p_zone.  This needs to happen before we grab any
3325 	 * locks to avoid deadlock (another lwp in the process could
3326 	 * be waiting for the held lock).
3327 	 */
3328 	if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK))
3329 		return (set_errno(EINTR));
3330 
3331 	/*
3332 	 * Make sure we're not changing zones with files open or mapped in
3333 	 * to our address space which shouldn't be changing zones.
3334 	 */
3335 	if (!files_can_change_zones()) {
3336 		err = EBADF;
3337 		goto out;
3338 	}
3339 	if (!as_can_change_zones()) {
3340 		err = EFAULT;
3341 		goto out;
3342 	}
3343 
3344 	mutex_enter(&zonehash_lock);
3345 	if (pp->p_zone != global_zone) {
3346 		mutex_exit(&zonehash_lock);
3347 		err = EINVAL;
3348 		goto out;
3349 	}
3350 
3351 	zone = zone_find_all_by_id(zoneid);
3352 	if (zone == NULL) {
3353 		mutex_exit(&zonehash_lock);
3354 		err = EINVAL;
3355 		goto out;
3356 	}
3357 
3358 	/*
3359 	 * To prevent processes in a zone from holding contracts on
3360 	 * extrazonal resources, and to avoid process contract
3361 	 * memberships which span zones, contract holders and processes
3362 	 * which aren't the sole members of their encapsulating process
3363 	 * contracts are not allowed to zone_enter.
3364 	 */
3365 	ctp = pp->p_ct_process;
3366 	ct = &ctp->conp_contract;
3367 	mutex_enter(&ct->ct_lock);
3368 	mutex_enter(&pp->p_lock);
3369 	if ((avl_numnodes(&pp->p_ct_held) != 0) || (ctp->conp_nmembers != 1)) {
3370 		mutex_exit(&pp->p_lock);
3371 		mutex_exit(&ct->ct_lock);
3372 		mutex_exit(&zonehash_lock);
3373 		pool_unlock();
3374 		err = EINVAL;
3375 		goto out;
3376 	}
3377 
3378 	/*
3379 	 * Moreover, we don't allow processes whose encapsulating
3380 	 * process contracts have inherited extrazonal contracts.
3381 	 * While it would be easier to eliminate all process contracts
3382 	 * with inherited contracts, we need to be able to give a
3383 	 * restarted init (or other zone-penetrating process) its
3384 	 * predecessor's contracts.
3385 	 */
3386 	if (ctp->conp_ninherited != 0) {
3387 		contract_t *next;
3388 		for (next = list_head(&ctp->conp_inherited); next;
3389 		    next = list_next(&ctp->conp_inherited, next)) {
3390 			if (contract_getzuniqid(next) != zone->zone_uniqid) {
3391 				mutex_exit(&pp->p_lock);
3392 				mutex_exit(&ct->ct_lock);
3393 				mutex_exit(&zonehash_lock);
3394 				pool_unlock();
3395 				err = EINVAL;
3396 				goto out;
3397 			}
3398 		}
3399 	}
3400 	mutex_exit(&pp->p_lock);
3401 	mutex_exit(&ct->ct_lock);
3402 
3403 	status = zone_status_get(zone);
3404 	if (status < ZONE_IS_READY || status >= ZONE_IS_SHUTTING_DOWN) {
3405 		/*
3406 		 * Can't join
3407 		 */
3408 		mutex_exit(&zonehash_lock);
3409 		err = EINVAL;
3410 		goto out;
3411 	}
3412 
3413 	/*
3414 	 * Make sure new priv set is within the permitted set for caller
3415 	 */
3416 	if (!priv_issubset(zone->zone_privset, &CR_OPPRIV(CRED()))) {
3417 		mutex_exit(&zonehash_lock);
3418 		err = EPERM;
3419 		goto out;
3420 	}
3421 	/*
3422 	 * We want to momentarily drop zonehash_lock while we optimistically
3423 	 * bind curproc to the pool it should be running in.  This is safe
3424 	 * since the zone can't disappear (we have a hold on it).
3425 	 */
3426 	zone_hold(zone);
3427 	mutex_exit(&zonehash_lock);
3428 
3429 	/*
3430 	 * Grab pool_lock to keep the pools configuration from changing
3431 	 * and to stop ourselves from getting rebound to another pool
3432 	 * until we join the zone.
3433 	 */
3434 	if (pool_lock_intr() != 0) {
3435 		zone_rele(zone);
3436 		err = EINTR;
3437 		goto out;
3438 	}
3439 	ASSERT(secpolicy_pool(CRED()) == 0);
3440 	/*
3441 	 * Bind ourselves to the pool currently associated with the zone.
3442 	 */
3443 	oldpool = curproc->p_pool;
3444 	newpool = zone_pool_get(zone);
3445 	if (pool_state == POOL_ENABLED && newpool != oldpool &&
3446 	    (err = pool_do_bind(newpool, P_PID, P_MYID,
3447 	    POOL_BIND_ALL)) != 0) {
3448 		pool_unlock();
3449 		zone_rele(zone);
3450 		goto out;
3451 	}
3452 
3453 	/*
3454 	 * Grab cpu_lock now; we'll need it later when we call
3455 	 * task_join().
3456 	 */
3457 	mutex_enter(&cpu_lock);
3458 	mutex_enter(&zonehash_lock);
3459 	/*
3460 	 * Make sure the zone hasn't moved on since we dropped zonehash_lock.
3461 	 */
3462 	if (zone_status_get(zone) >= ZONE_IS_SHUTTING_DOWN) {
3463 		/*
3464 		 * Can't join anymore.
3465 		 */
3466 		mutex_exit(&zonehash_lock);
3467 		mutex_exit(&cpu_lock);
3468 		if (pool_state == POOL_ENABLED &&
3469 		    newpool != oldpool)
3470 			(void) pool_do_bind(oldpool, P_PID, P_MYID,
3471 			    POOL_BIND_ALL);
3472 		pool_unlock();
3473 		zone_rele(zone);
3474 		err = EINVAL;
3475 		goto out;
3476 	}
3477 
3478 	mutex_enter(&pp->p_lock);
3479 	zone_proj0 = zone->zone_zsched->p_task->tk_proj;
3480 	/* verify that we do not exceed and task or lwp limits */
3481 	mutex_enter(&zone->zone_nlwps_lock);
3482 	/* add new lwps to zone and zone's proj0 */
3483 	zone_proj0->kpj_nlwps += pp->p_lwpcnt;
3484 	zone->zone_nlwps += pp->p_lwpcnt;
3485 	/* add 1 task to zone's proj0 */
3486 	zone_proj0->kpj_ntasks += 1;
3487 	mutex_exit(&pp->p_lock);
3488 	mutex_exit(&zone->zone_nlwps_lock);
3489 
3490 	/* remove lwps from proc's old zone and old project */
3491 	mutex_enter(&pp->p_zone->zone_nlwps_lock);
3492 	pp->p_zone->zone_nlwps -= pp->p_lwpcnt;
3493 	pp->p_task->tk_proj->kpj_nlwps -= pp->p_lwpcnt;
3494 	mutex_exit(&pp->p_zone->zone_nlwps_lock);
3495 
3496 	/*
3497 	 * Joining the zone cannot fail from now on.
3498 	 *
3499 	 * This means that a lot of the following code can be commonized and
3500 	 * shared with zsched().
3501 	 */
3502 
3503 	/*
3504 	 * Reset the encapsulating process contract's zone.
3505 	 */
3506 	ASSERT(ct->ct_mzuniqid == GLOBAL_ZONEUNIQID);
3507 	contract_setzuniqid(ct, zone->zone_uniqid);
3508 
3509 	/*
3510 	 * Create a new task and associate the process with the project keyed
3511 	 * by (projid,zoneid).
3512 	 *
3513 	 * We might as well be in project 0; the global zone's projid doesn't
3514 	 * make much sense in a zone anyhow.
3515 	 *
3516 	 * This also increments zone_ntasks, and returns with p_lock held.
3517 	 */
3518 	tk = task_create(0, zone);
3519 	oldtk = task_join(tk, 0);
3520 	mutex_exit(&cpu_lock);
3521 
3522 	pp->p_flag |= SZONETOP;
3523 	pp->p_zone = zone;
3524 
3525 	/*
3526 	 * call RCTLOP_SET functions on this proc
3527 	 */
3528 	e.rcep_p.zone = zone;
3529 	e.rcep_t = RCENTITY_ZONE;
3530 	(void) rctl_set_dup(NULL, NULL, pp, &e, zone->zone_rctls, NULL,
3531 	    RCD_CALLBACK);
3532 	mutex_exit(&pp->p_lock);
3533 
3534 	/*
3535 	 * We don't need to hold any of zsched's locks here; not only do we know
3536 	 * the process and zone aren't going away, we know its session isn't
3537 	 * changing either.
3538 	 *
3539 	 * By joining zsched's session here, we mimic the behavior in the
3540 	 * global zone of init's sid being the pid of sched.  We extend this
3541 	 * to all zlogin-like zone_enter()'ing processes as well.
3542 	 */
3543 	mutex_enter(&pidlock);
3544 	sp = zone->zone_zsched->p_sessp;
3545 	SESS_HOLD(sp);
3546 	mutex_enter(&pp->p_lock);
3547 	pgexit(pp);
3548 	SESS_RELE(pp->p_sessp);
3549 	pp->p_sessp = sp;
3550 	pgjoin(pp, zone->zone_zsched->p_pidp);
3551 	mutex_exit(&pp->p_lock);
3552 	mutex_exit(&pidlock);
3553 
3554 	mutex_exit(&zonehash_lock);
3555 	/*
3556 	 * We're firmly in the zone; let pools progress.
3557 	 */
3558 	pool_unlock();
3559 	task_rele(oldtk);
3560 	/*
3561 	 * We don't need to retain a hold on the zone since we already
3562 	 * incremented zone_ntasks, so the zone isn't going anywhere.
3563 	 */
3564 	zone_rele(zone);
3565 
3566 	/*
3567 	 * Chroot
3568 	 */
3569 	vp = zone->zone_rootvp;
3570 	zone_chdir(vp, &PTOU(pp)->u_cdir, pp);
3571 	zone_chdir(vp, &PTOU(pp)->u_rdir, pp);
3572 
3573 	/*
3574 	 * Change process credentials
3575 	 */
3576 	newcr = cralloc();
3577 	mutex_enter(&pp->p_crlock);
3578 	cr = pp->p_cred;
3579 	crcopy_to(cr, newcr);
3580 	crsetzone(newcr, zone);
3581 	pp->p_cred = newcr;
3582 
3583 	/*
3584 	 * Restrict all process privilege sets to zone limit
3585 	 */
3586 	priv_intersect(zone->zone_privset, &CR_PPRIV(newcr));
3587 	priv_intersect(zone->zone_privset, &CR_EPRIV(newcr));
3588 	priv_intersect(zone->zone_privset, &CR_IPRIV(newcr));
3589 	priv_intersect(zone->zone_privset, &CR_LPRIV(newcr));
3590 	mutex_exit(&pp->p_crlock);
3591 	crset(pp, newcr);
3592 
3593 	/*
3594 	 * Adjust upcount to reflect zone entry.
3595 	 */
3596 	uid = crgetruid(newcr);
3597 	mutex_enter(&pidlock);
3598 	upcount_dec(uid, GLOBAL_ZONEID);
3599 	upcount_inc(uid, zoneid);
3600 	mutex_exit(&pidlock);
3601 
3602 	/*
3603 	 * Set up core file path and content.
3604 	 */
3605 	set_core_defaults();
3606 
3607 out:
3608 	/*
3609 	 * Let the other lwps continue.
3610 	 */
3611 	mutex_enter(&pp->p_lock);
3612 	if (curthread != pp->p_agenttp)
3613 		continuelwps(pp);
3614 	mutex_exit(&pp->p_lock);
3615 
3616 	return (err != 0 ? set_errno(err) : 0);
3617 }
3618 
3619 /*
3620  * Systemcall entry point for zone_list(2).
3621  *
3622  * Processes running in a (non-global) zone only see themselves.
3623  */
3624 static int
3625 zone_list(zoneid_t *zoneidlist, uint_t *numzones)
3626 {
3627 	zoneid_t *zoneids;
3628 	zone_t *zone;
3629 	uint_t user_nzones, real_nzones;
3630 	int error = 0;
3631 	uint_t i;
3632 
3633 	if (copyin(numzones, &user_nzones, sizeof (uint_t)) != 0)
3634 		return (set_errno(EFAULT));
3635 
3636 	if (curproc->p_zone != global_zone) {
3637 		/* just return current zone */
3638 		real_nzones = 1;
3639 		zoneids = kmem_alloc(sizeof (zoneid_t), KM_SLEEP);
3640 		zoneids[0] = curproc->p_zone->zone_id;
3641 	} else {
3642 		mutex_enter(&zonehash_lock);
3643 		real_nzones = zonecount;
3644 		if (real_nzones) {
3645 			zoneids = kmem_alloc(real_nzones * sizeof (zoneid_t),
3646 			    KM_SLEEP);
3647 			i = 0;
3648 			for (zone = list_head(&zone_active); zone != NULL;
3649 			    zone = list_next(&zone_active, zone))
3650 				zoneids[i++] = zone->zone_id;
3651 			ASSERT(i == real_nzones);
3652 		}
3653 		mutex_exit(&zonehash_lock);
3654 	}
3655 
3656 	if (user_nzones > real_nzones)
3657 		user_nzones = real_nzones;
3658 
3659 	if (copyout(&real_nzones, numzones, sizeof (uint_t)) != 0)
3660 		error = EFAULT;
3661 	else if (zoneidlist != NULL && user_nzones != 0) {
3662 		if (copyout(zoneids, zoneidlist,
3663 		    user_nzones * sizeof (zoneid_t)) != 0)
3664 			error = EFAULT;
3665 	}
3666 
3667 	if (real_nzones)
3668 		kmem_free(zoneids, real_nzones * sizeof (zoneid_t));
3669 
3670 	if (error)
3671 		return (set_errno(error));
3672 	else
3673 		return (0);
3674 }
3675 
3676 /*
3677  * Systemcall entry point for zone_lookup(2).
3678  *
3679  * Non-global zones are only able to see themselves.
3680  */
3681 static zoneid_t
3682 zone_lookup(const char *zone_name)
3683 {
3684 	char *kname;
3685 	zone_t *zone;
3686 	zoneid_t zoneid;
3687 	int err;
3688 
3689 	if (zone_name == NULL) {
3690 		/* return caller's zone id */
3691 		return (getzoneid());
3692 	}
3693 
3694 	kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
3695 	if ((err = copyinstr(zone_name, kname, ZONENAME_MAX, NULL)) != 0) {
3696 		kmem_free(kname, ZONENAME_MAX);
3697 		return (set_errno(err));
3698 	}
3699 
3700 	mutex_enter(&zonehash_lock);
3701 	zone = zone_find_all_by_name(kname);
3702 	kmem_free(kname, ZONENAME_MAX);
3703 	if (zone == NULL || zone_status_get(zone) < ZONE_IS_READY ||
3704 	    (curproc->p_zone != global_zone && curproc->p_zone != zone)) {
3705 		/* in non-global zone, can only lookup own name */
3706 		mutex_exit(&zonehash_lock);
3707 		return (set_errno(EINVAL));
3708 	}
3709 	zoneid = zone->zone_id;
3710 	mutex_exit(&zonehash_lock);
3711 	return (zoneid);
3712 }
3713 
3714 /* ARGSUSED */
3715 long
3716 zone(int cmd, void *arg1, void *arg2, void *arg3, void *arg4, void *arg5)
3717 {
3718 	zone_def zs;
3719 
3720 	switch (cmd) {
3721 	case ZONE_CREATE:
3722 		if (get_udatamodel() == DATAMODEL_NATIVE) {
3723 			if (copyin(arg1, &zs, sizeof (zone_def))) {
3724 				return (set_errno(EFAULT));
3725 			}
3726 		} else {
3727 #ifdef _SYSCALL32_IMPL
3728 			zone_def32 zs32;
3729 
3730 			if (copyin(arg1, &zs32, sizeof (zone_def32))) {
3731 				return (set_errno(EFAULT));
3732 			}
3733 			zs.zone_name =
3734 			    (const char *)(unsigned long)zs32.zone_name;
3735 			zs.zone_root =
3736 			    (const char *)(unsigned long)zs32.zone_root;
3737 			zs.zone_privs =
3738 			    (const struct priv_set *)
3739 			    (unsigned long)zs32.zone_privs;
3740 			zs.rctlbuf = (caddr_t)(unsigned long)zs32.rctlbuf;
3741 			zs.rctlbufsz = zs32.rctlbufsz;
3742 			zs.extended_error =
3743 			    (int *)(unsigned long)zs32.extended_error;
3744 #else
3745 			panic("get_udatamodel() returned bogus result\n");
3746 #endif
3747 		}
3748 
3749 		return (zone_create(zs.zone_name, zs.zone_root,
3750 			zs.zone_privs, (caddr_t)zs.rctlbuf, zs.rctlbufsz,
3751 			zs.extended_error));
3752 	case ZONE_BOOT:
3753 		return (zone_boot((zoneid_t)(uintptr_t)arg1,
3754 		    (const char *)arg2));
3755 	case ZONE_DESTROY:
3756 		return (zone_destroy((zoneid_t)(uintptr_t)arg1));
3757 	case ZONE_GETATTR:
3758 		return (zone_getattr((zoneid_t)(uintptr_t)arg1,
3759 		    (int)(uintptr_t)arg2, arg3, (size_t)arg4));
3760 	case ZONE_ENTER:
3761 		return (zone_enter((zoneid_t)(uintptr_t)arg1));
3762 	case ZONE_LIST:
3763 		return (zone_list((zoneid_t *)arg1, (uint_t *)arg2));
3764 	case ZONE_SHUTDOWN:
3765 		return (zone_shutdown((zoneid_t)(uintptr_t)arg1));
3766 	case ZONE_LOOKUP:
3767 		return (zone_lookup((const char *)arg1));
3768 	default:
3769 		return (set_errno(EINVAL));
3770 	}
3771 }
3772 
3773 struct zarg {
3774 	zone_t *zone;
3775 	zone_cmd_arg_t arg;
3776 };
3777 
3778 static int
3779 zone_lookup_door(const char *zone_name, door_handle_t *doorp)
3780 {
3781 	char *buf;
3782 	size_t buflen;
3783 	int error;
3784 
3785 	buflen = sizeof (ZONE_DOOR_PATH) + strlen(zone_name);
3786 	buf = kmem_alloc(buflen, KM_SLEEP);
3787 	(void) snprintf(buf, buflen, ZONE_DOOR_PATH, zone_name);
3788 	error = door_ki_open(buf, doorp);
3789 	kmem_free(buf, buflen);
3790 	return (error);
3791 }
3792 
3793 static void
3794 zone_release_door(door_handle_t *doorp)
3795 {
3796 	door_ki_rele(*doorp);
3797 	*doorp = NULL;
3798 }
3799 
3800 static void
3801 zone_ki_call_zoneadmd(struct zarg *zargp)
3802 {
3803 	door_handle_t door = NULL;
3804 	door_arg_t darg, save_arg;
3805 	char *zone_name;
3806 	size_t zone_namelen;
3807 	zoneid_t zoneid;
3808 	zone_t *zone;
3809 	zone_cmd_arg_t arg;
3810 	uint64_t uniqid;
3811 	size_t size;
3812 	int error;
3813 	int retry;
3814 
3815 	zone = zargp->zone;
3816 	arg = zargp->arg;
3817 	kmem_free(zargp, sizeof (*zargp));
3818 
3819 	zone_namelen = strlen(zone->zone_name) + 1;
3820 	zone_name = kmem_alloc(zone_namelen, KM_SLEEP);
3821 	bcopy(zone->zone_name, zone_name, zone_namelen);
3822 	zoneid = zone->zone_id;
3823 	uniqid = zone->zone_uniqid;
3824 	/*
3825 	 * zoneadmd may be down, but at least we can empty out the zone.
3826 	 * We can ignore the return value of zone_empty() since we're called
3827 	 * from a kernel thread and know we won't be delivered any signals.
3828 	 */
3829 	ASSERT(curproc == &p0);
3830 	(void) zone_empty(zone);
3831 	ASSERT(zone_status_get(zone) >= ZONE_IS_EMPTY);
3832 	zone_rele(zone);
3833 
3834 	size = sizeof (arg);
3835 	darg.rbuf = (char *)&arg;
3836 	darg.data_ptr = (char *)&arg;
3837 	darg.rsize = size;
3838 	darg.data_size = size;
3839 	darg.desc_ptr = NULL;
3840 	darg.desc_num = 0;
3841 
3842 	save_arg = darg;
3843 	/*
3844 	 * Since we're not holding a reference to the zone, any number of
3845 	 * things can go wrong, including the zone disappearing before we get a
3846 	 * chance to talk to zoneadmd.
3847 	 */
3848 	for (retry = 0; /* forever */; retry++) {
3849 		if (door == NULL &&
3850 		    (error = zone_lookup_door(zone_name, &door)) != 0) {
3851 			goto next;
3852 		}
3853 		ASSERT(door != NULL);
3854 
3855 		if ((error = door_ki_upcall(door, &darg)) == 0) {
3856 			break;
3857 		}
3858 		switch (error) {
3859 		case EINTR:
3860 			/* FALLTHROUGH */
3861 		case EAGAIN:	/* process may be forking */
3862 			/*
3863 			 * Back off for a bit
3864 			 */
3865 			break;
3866 		case EBADF:
3867 			zone_release_door(&door);
3868 			if (zone_lookup_door(zone_name, &door) != 0) {
3869 				/*
3870 				 * zoneadmd may be dead, but it may come back to
3871 				 * life later.
3872 				 */
3873 				break;
3874 			}
3875 			break;
3876 		default:
3877 			cmn_err(CE_WARN,
3878 			    "zone_ki_call_zoneadmd: door_ki_upcall error %d\n",
3879 			    error);
3880 			goto out;
3881 		}
3882 next:
3883 		/*
3884 		 * If this isn't the same zone_t that we originally had in mind,
3885 		 * then this is the same as if two kadmin requests come in at
3886 		 * the same time: the first one wins.  This means we lose, so we
3887 		 * bail.
3888 		 */
3889 		if ((zone = zone_find_by_id(zoneid)) == NULL) {
3890 			/*
3891 			 * Problem is solved.
3892 			 */
3893 			break;
3894 		}
3895 		if (zone->zone_uniqid != uniqid) {
3896 			/*
3897 			 * zoneid recycled
3898 			 */
3899 			zone_rele(zone);
3900 			break;
3901 		}
3902 		/*
3903 		 * We could zone_status_timedwait(), but there doesn't seem to
3904 		 * be much point in doing that (plus, it would mean that
3905 		 * zone_free() isn't called until this thread exits).
3906 		 */
3907 		zone_rele(zone);
3908 		delay(hz);
3909 		darg = save_arg;
3910 	}
3911 out:
3912 	if (door != NULL) {
3913 		zone_release_door(&door);
3914 	}
3915 	kmem_free(zone_name, zone_namelen);
3916 	thread_exit();
3917 }
3918 
3919 /*
3920  * Entry point for uadmin() to tell the zone to go away or reboot.  The caller
3921  * is a process in the zone to be modified.
3922  *
3923  * In order to shutdown the zone, we will hand off control to zoneadmd
3924  * (running in the global zone) via a door.  We do a half-hearted job at
3925  * killing all processes in the zone, create a kernel thread to contact
3926  * zoneadmd, and make note of the "uniqid" of the zone.  The uniqid is
3927  * a form of generation number used to let zoneadmd (as well as
3928  * zone_destroy()) know exactly which zone they're re talking about.
3929  */
3930 int
3931 zone_uadmin(int cmd, int fcn, cred_t *credp)
3932 {
3933 	struct zarg *zargp;
3934 	zone_cmd_t zcmd;
3935 	zone_t *zone;
3936 
3937 	zone = curproc->p_zone;
3938 	ASSERT(getzoneid() != GLOBAL_ZONEID);
3939 
3940 	switch (cmd) {
3941 	case A_SHUTDOWN:
3942 		switch (fcn) {
3943 		case AD_HALT:
3944 		case AD_POWEROFF:
3945 			zcmd = Z_HALT;
3946 			break;
3947 		case AD_BOOT:
3948 			zcmd = Z_REBOOT;
3949 			break;
3950 		case AD_IBOOT:
3951 		case AD_SBOOT:
3952 		case AD_SIBOOT:
3953 		case AD_NOSYNC:
3954 			return (ENOTSUP);
3955 		default:
3956 			return (EINVAL);
3957 		}
3958 		break;
3959 	case A_REBOOT:
3960 		zcmd = Z_REBOOT;
3961 		break;
3962 	case A_FTRACE:
3963 	case A_REMOUNT:
3964 	case A_FREEZE:
3965 	case A_DUMP:
3966 		return (ENOTSUP);
3967 	default:
3968 		ASSERT(cmd != A_SWAPCTL);	/* handled by uadmin() */
3969 		return (EINVAL);
3970 	}
3971 
3972 	if (secpolicy_zone_admin(credp, B_FALSE))
3973 		return (EPERM);
3974 	mutex_enter(&zone_status_lock);
3975 	/*
3976 	 * zone_status can't be ZONE_IS_EMPTY or higher since curproc
3977 	 * is in the zone.
3978 	 */
3979 	ASSERT(zone_status_get(zone) < ZONE_IS_EMPTY);
3980 	if (zone_status_get(zone) > ZONE_IS_RUNNING) {
3981 		/*
3982 		 * This zone is already on its way down.
3983 		 */
3984 		mutex_exit(&zone_status_lock);
3985 		return (0);
3986 	}
3987 	/*
3988 	 * Prevent future zone_enter()s
3989 	 */
3990 	zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
3991 	mutex_exit(&zone_status_lock);
3992 
3993 	/*
3994 	 * Kill everyone now and call zoneadmd later.
3995 	 * zone_ki_call_zoneadmd() will do a more thorough job of this
3996 	 * later.
3997 	 */
3998 	killall(zone->zone_id);
3999 	/*
4000 	 * Now, create the thread to contact zoneadmd and do the rest of the
4001 	 * work.  This thread can't be created in our zone otherwise
4002 	 * zone_destroy() would deadlock.
4003 	 */
4004 	zargp = kmem_alloc(sizeof (*zargp), KM_SLEEP);
4005 	zargp->arg.cmd = zcmd;
4006 	zargp->arg.uniqid = zone->zone_uniqid;
4007 	(void) strcpy(zargp->arg.locale, "C");
4008 	zone_hold(zargp->zone = zone);
4009 
4010 	(void) thread_create(NULL, 0, zone_ki_call_zoneadmd, zargp, 0, &p0,
4011 	    TS_RUN, minclsyspri);
4012 	exit(CLD_EXITED, 0);
4013 
4014 	return (EINVAL);
4015 }
4016 
4017 /*
4018  * Entry point so kadmin(A_SHUTDOWN, ...) can set the global zone's
4019  * status to ZONE_IS_SHUTTING_DOWN.
4020  */
4021 void
4022 zone_shutdown_global(void)
4023 {
4024 	ASSERT(curproc->p_zone == global_zone);
4025 
4026 	mutex_enter(&zone_status_lock);
4027 	ASSERT(zone_status_get(global_zone) == ZONE_IS_RUNNING);
4028 	zone_status_set(global_zone, ZONE_IS_SHUTTING_DOWN);
4029 	mutex_exit(&zone_status_lock);
4030 }
4031