zone.c revision 7b209c2cc5ea45251aba06dcc6181d3f23da807a
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
24 * Use is subject to license terms.
25 */
26
27#pragma ident	"%Z%%M%	%I%	%E% SMI"
28
29/*
30 * Zones
31 *
32 *   A zone is a named collection of processes, namespace constraints,
33 *   and other system resources which comprise a secure and manageable
34 *   application containment facility.
35 *
36 *   Zones (represented by the reference counted zone_t) are tracked in
37 *   the kernel in the zonehash.  Elsewhere in the kernel, Zone IDs
38 *   (zoneid_t) are used to track zone association.  Zone IDs are
39 *   dynamically generated when the zone is created; if a persistent
40 *   identifier is needed (core files, accounting logs, audit trail,
41 *   etc.), the zone name should be used.
42 *
43 *
44 *   Global Zone:
45 *
46 *   The global zone (zoneid 0) is automatically associated with all
47 *   system resources that have not been bound to a user-created zone.
48 *   This means that even systems where zones are not in active use
49 *   have a global zone, and all processes, mounts, etc. are
50 *   associated with that zone.  The global zone is generally
51 *   unconstrained in terms of privileges and access, though the usual
52 *   credential and privilege based restrictions apply.
53 *
54 *
55 *   Zone States:
56 *
57 *   The states in which a zone may be in and the transitions are as
58 *   follows:
59 *
60 *   ZONE_IS_UNINITIALIZED: primordial state for a zone. The partially
61 *   initialized zone is added to the list of active zones on the system but
62 *   isn't accessible.
63 *
64 *   ZONE_IS_INITIALIZED: Initialization complete except the ZSD callbacks are
65 *   not yet completed. Not possible to enter the zone, but attributes can
66 *   be retrieved.
67 *
68 *   ZONE_IS_READY: zsched (the kernel dummy process for a zone) is
69 *   ready.  The zone is made visible after the ZSD constructor callbacks are
70 *   executed.  A zone remains in this state until it transitions into
71 *   the ZONE_IS_BOOTING state as a result of a call to zone_boot().
72 *
73 *   ZONE_IS_BOOTING: in this shortlived-state, zsched attempts to start
74 *   init.  Should that fail, the zone proceeds to the ZONE_IS_SHUTTING_DOWN
75 *   state.
76 *
77 *   ZONE_IS_RUNNING: The zone is open for business: zsched has
78 *   successfully started init.   A zone remains in this state until
79 *   zone_shutdown() is called.
80 *
81 *   ZONE_IS_SHUTTING_DOWN: zone_shutdown() has been called, the system is
82 *   killing all processes running in the zone. The zone remains
83 *   in this state until there are no more user processes running in the zone.
84 *   zone_create(), zone_enter(), and zone_destroy() on this zone will fail.
85 *   Since zone_shutdown() is restartable, it may be called successfully
86 *   multiple times for the same zone_t.  Setting of the zone's state to
87 *   ZONE_IS_SHUTTING_DOWN is synchronized with mounts, so VOP_MOUNT() may check
88 *   the zone's status without worrying about it being a moving target.
89 *
90 *   ZONE_IS_EMPTY: zone_shutdown() has been called, and there
91 *   are no more user processes in the zone.  The zone remains in this
92 *   state until there are no more kernel threads associated with the
93 *   zone.  zone_create(), zone_enter(), and zone_destroy() on this zone will
94 *   fail.
95 *
96 *   ZONE_IS_DOWN: All kernel threads doing work on behalf of the zone
97 *   have exited.  zone_shutdown() returns.  Henceforth it is not possible to
98 *   join the zone or create kernel threads therein.
99 *
100 *   ZONE_IS_DYING: zone_destroy() has been called on the zone; zone
101 *   remains in this state until zsched exits.  Calls to zone_find_by_*()
102 *   return NULL from now on.
103 *
104 *   ZONE_IS_DEAD: zsched has exited (zone_ntasks == 0).  There are no
105 *   processes or threads doing work on behalf of the zone.  The zone is
106 *   removed from the list of active zones.  zone_destroy() returns, and
107 *   the zone can be recreated.
108 *
109 *   ZONE_IS_FREE (internal state): zone_ref goes to 0, ZSD destructor
110 *   callbacks are executed, and all memory associated with the zone is
111 *   freed.
112 *
113 *   Threads can wait for the zone to enter a requested state by using
114 *   zone_status_wait() or zone_status_timedwait() with the desired
115 *   state passed in as an argument.  Zone state transitions are
116 *   uni-directional; it is not possible to move back to an earlier state.
117 *
118 *
119 *   Zone-Specific Data:
120 *
121 *   Subsystems needing to maintain zone-specific data can store that
122 *   data using the ZSD mechanism.  This provides a zone-specific data
123 *   store, similar to thread-specific data (see pthread_getspecific(3C)
124 *   or the TSD code in uts/common/disp/thread.c.  Also, ZSD can be used
125 *   to register callbacks to be invoked when a zone is created, shut
126 *   down, or destroyed.  This can be used to initialize zone-specific
127 *   data for new zones and to clean up when zones go away.
128 *
129 *
130 *   Data Structures:
131 *
132 *   The per-zone structure (zone_t) is reference counted, and freed
133 *   when all references are released.  zone_hold and zone_rele can be
134 *   used to adjust the reference count.  In addition, reference counts
135 *   associated with the cred_t structure are tracked separately using
136 *   zone_cred_hold and zone_cred_rele.
137 *
138 *   Pointers to active zone_t's are stored in two hash tables; one
139 *   for searching by id, the other for searching by name.  Lookups
140 *   can be performed on either basis, using zone_find_by_id and
141 *   zone_find_by_name.  Both return zone_t pointers with the zone
142 *   held, so zone_rele should be called when the pointer is no longer
143 *   needed.  Zones can also be searched by path; zone_find_by_path
144 *   returns the zone with which a path name is associated (global
145 *   zone if the path is not within some other zone's file system
146 *   hierarchy).  This currently requires iterating through each zone,
147 *   so it is slower than an id or name search via a hash table.
148 *
149 *
150 *   Locking:
151 *
152 *   zonehash_lock: This is a top-level global lock used to protect the
153 *       zone hash tables and lists.  Zones cannot be created or destroyed
154 *       while this lock is held.
155 *   zone_status_lock: This is a global lock protecting zone state.
156 *       Zones cannot change state while this lock is held.  It also
157 *       protects the list of kernel threads associated with a zone.
158 *   zone_lock: This is a per-zone lock used to protect several fields of
159 *       the zone_t (see <sys/zone.h> for details).  In addition, holding
160 *       this lock means that the zone cannot go away.
161 *   zone_nlwps_lock: This is a per-zone lock used to protect the fields
162 *	 related to the zone.max-lwps rctl.
163 *   zone_mem_lock: This is a per-zone lock used to protect the fields
164 *	 related to the zone.max-locked-memory and zone.max-swap rctls.
165 *   zsd_key_lock: This is a global lock protecting the key state for ZSD.
166 *   zone_deathrow_lock: This is a global lock protecting the "deathrow"
167 *       list (a list of zones in the ZONE_IS_DEAD state).
168 *
169 *   Ordering requirements:
170 *       pool_lock --> cpu_lock --> zonehash_lock --> zone_status_lock -->
171 *       	zone_lock --> zsd_key_lock --> pidlock --> p_lock
172 *
173 *   When taking zone_mem_lock or zone_nlwps_lock, the lock ordering is:
174 *	zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_mem_lock
175 *	zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_mem_lock
176 *
177 *   Blocking memory allocations are permitted while holding any of the
178 *   zone locks.
179 *
180 *
181 *   System Call Interface:
182 *
183 *   The zone subsystem can be managed and queried from user level with
184 *   the following system calls (all subcodes of the primary "zone"
185 *   system call):
186 *   - zone_create: creates a zone with selected attributes (name,
187 *     root path, privileges, resource controls, ZFS datasets)
188 *   - zone_enter: allows the current process to enter a zone
189 *   - zone_getattr: reports attributes of a zone
190 *   - zone_setattr: set attributes of a zone
191 *   - zone_boot: set 'init' running for the zone
192 *   - zone_list: lists all zones active in the system
193 *   - zone_lookup: looks up zone id based on name
194 *   - zone_shutdown: initiates shutdown process (see states above)
195 *   - zone_destroy: completes shutdown process (see states above)
196 *
197 */
198
199#include <sys/priv_impl.h>
200#include <sys/cred.h>
201#include <c2/audit.h>
202#include <sys/debug.h>
203#include <sys/file.h>
204#include <sys/kmem.h>
205#include <sys/kstat.h>
206#include <sys/mutex.h>
207#include <sys/note.h>
208#include <sys/pathname.h>
209#include <sys/proc.h>
210#include <sys/project.h>
211#include <sys/sysevent.h>
212#include <sys/task.h>
213#include <sys/systm.h>
214#include <sys/types.h>
215#include <sys/utsname.h>
216#include <sys/vnode.h>
217#include <sys/vfs.h>
218#include <sys/systeminfo.h>
219#include <sys/policy.h>
220#include <sys/cred_impl.h>
221#include <sys/contract_impl.h>
222#include <sys/contract/process_impl.h>
223#include <sys/class.h>
224#include <sys/pool.h>
225#include <sys/pool_pset.h>
226#include <sys/pset.h>
227#include <sys/sysmacros.h>
228#include <sys/callb.h>
229#include <sys/vmparam.h>
230#include <sys/corectl.h>
231#include <sys/ipc_impl.h>
232
233#include <sys/door.h>
234#include <sys/cpuvar.h>
235#include <sys/sdt.h>
236
237#include <sys/uadmin.h>
238#include <sys/session.h>
239#include <sys/cmn_err.h>
240#include <sys/modhash.h>
241#include <sys/sunddi.h>
242#include <sys/nvpair.h>
243#include <sys/rctl.h>
244#include <sys/fss.h>
245#include <sys/brand.h>
246#include <sys/zone.h>
247#include <net/if.h>
248#include <sys/cpucaps.h>
249#include <vm/seg.h>
250
251/*
252 * cv used to signal that all references to the zone have been released.  This
253 * needs to be global since there may be multiple waiters, and the first to
254 * wake up will free the zone_t, hence we cannot use zone->zone_cv.
255 */
256static kcondvar_t zone_destroy_cv;
257/*
258 * Lock used to serialize access to zone_cv.  This could have been per-zone,
259 * but then we'd need another lock for zone_destroy_cv, and why bother?
260 */
261static kmutex_t zone_status_lock;
262
263/*
264 * ZSD-related global variables.
265 */
266static kmutex_t zsd_key_lock;	/* protects the following two */
267/*
268 * The next caller of zone_key_create() will be assigned a key of ++zsd_keyval.
269 */
270static zone_key_t zsd_keyval = 0;
271/*
272 * Global list of registered keys.  We use this when a new zone is created.
273 */
274static list_t zsd_registered_keys;
275
276int zone_hash_size = 256;
277static mod_hash_t *zonehashbyname, *zonehashbyid, *zonehashbylabel;
278static kmutex_t zonehash_lock;
279static uint_t zonecount;
280static id_space_t *zoneid_space;
281
282/*
283 * The global zone (aka zone0) is the all-seeing, all-knowing zone in which the
284 * kernel proper runs, and which manages all other zones.
285 *
286 * Although not declared as static, the variable "zone0" should not be used
287 * except for by code that needs to reference the global zone early on in boot,
288 * before it is fully initialized.  All other consumers should use
289 * 'global_zone'.
290 */
291zone_t zone0;
292zone_t *global_zone = NULL;	/* Set when the global zone is initialized */
293
294/*
295 * List of active zones, protected by zonehash_lock.
296 */
297static list_t zone_active;
298
299/*
300 * List of destroyed zones that still have outstanding cred references.
301 * Used for debugging.  Uses a separate lock to avoid lock ordering
302 * problems in zone_free.
303 */
304static list_t zone_deathrow;
305static kmutex_t zone_deathrow_lock;
306
307/* number of zones is limited by virtual interface limit in IP */
308uint_t maxzones = 8192;
309
310/* Event channel to sent zone state change notifications */
311evchan_t *zone_event_chan;
312
313/*
314 * This table holds the mapping from kernel zone states to
315 * states visible in the state notification API.
316 * The idea is that we only expose "obvious" states and
317 * do not expose states which are just implementation details.
318 */
319const char  *zone_status_table[] = {
320	ZONE_EVENT_UNINITIALIZED,	/* uninitialized */
321	ZONE_EVENT_INITIALIZED,		/* initialized */
322	ZONE_EVENT_READY,		/* ready */
323	ZONE_EVENT_READY,		/* booting */
324	ZONE_EVENT_RUNNING,		/* running */
325	ZONE_EVENT_SHUTTING_DOWN,	/* shutting_down */
326	ZONE_EVENT_SHUTTING_DOWN,	/* empty */
327	ZONE_EVENT_SHUTTING_DOWN,	/* down */
328	ZONE_EVENT_SHUTTING_DOWN,	/* dying */
329	ZONE_EVENT_UNINITIALIZED,	/* dead */
330};
331
332/*
333 * This isn't static so lint doesn't complain.
334 */
335rctl_hndl_t rc_zone_cpu_shares;
336rctl_hndl_t rc_zone_locked_mem;
337rctl_hndl_t rc_zone_max_swap;
338rctl_hndl_t rc_zone_cpu_cap;
339rctl_hndl_t rc_zone_nlwps;
340rctl_hndl_t rc_zone_shmmax;
341rctl_hndl_t rc_zone_shmmni;
342rctl_hndl_t rc_zone_semmni;
343rctl_hndl_t rc_zone_msgmni;
344/*
345 * Synchronization primitives used to synchronize between mounts and zone
346 * creation/destruction.
347 */
348static int mounts_in_progress;
349static kcondvar_t mount_cv;
350static kmutex_t mount_lock;
351
352const char * const zone_default_initname = "/sbin/init";
353static char * const zone_prefix = "/zone/";
354static int zone_shutdown(zoneid_t zoneid);
355static int zone_add_datalink(zoneid_t, char *);
356static int zone_remove_datalink(zoneid_t, char *);
357static int zone_check_datalink(zoneid_t *, char *);
358static int zone_list_datalink(zoneid_t, int *, char *);
359
360typedef boolean_t zsd_applyfn_t(kmutex_t *, boolean_t, zone_t *, zone_key_t);
361
362static void zsd_apply_all_zones(zsd_applyfn_t *, zone_key_t);
363static void zsd_apply_all_keys(zsd_applyfn_t *, zone_t *);
364static boolean_t zsd_apply_create(kmutex_t *, boolean_t, zone_t *, zone_key_t);
365static boolean_t zsd_apply_shutdown(kmutex_t *, boolean_t, zone_t *,
366    zone_key_t);
367static boolean_t zsd_apply_destroy(kmutex_t *, boolean_t, zone_t *, zone_key_t);
368static boolean_t zsd_wait_for_creator(zone_t *, struct zsd_entry *,
369    kmutex_t *);
370static boolean_t zsd_wait_for_inprogress(zone_t *, struct zsd_entry *,
371    kmutex_t *);
372
373/*
374 * Bump this number when you alter the zone syscall interfaces; this is
375 * because we need to have support for previous API versions in libc
376 * to support patching; libc calls into the kernel to determine this number.
377 *
378 * Version 1 of the API is the version originally shipped with Solaris 10
379 * Version 2 alters the zone_create system call in order to support more
380 *     arguments by moving the args into a structure; and to do better
381 *     error reporting when zone_create() fails.
382 * Version 3 alters the zone_create system call in order to support the
383 *     import of ZFS datasets to zones.
384 * Version 4 alters the zone_create system call in order to support
385 *     Trusted Extensions.
386 * Version 5 alters the zone_boot system call, and converts its old
387 *     bootargs parameter to be set by the zone_setattr API instead.
388 * Version 6 adds the flag argument to zone_create.
389 */
390static const int ZONE_SYSCALL_API_VERSION = 6;
391
392/*
393 * Certain filesystems (such as NFS and autofs) need to know which zone
394 * the mount is being placed in.  Because of this, we need to be able to
395 * ensure that a zone isn't in the process of being created such that
396 * nfs_mount() thinks it is in the global zone, while by the time it
397 * gets added the list of mounted zones, it ends up on zoneA's mount
398 * list.
399 *
400 * The following functions: block_mounts()/resume_mounts() and
401 * mount_in_progress()/mount_completed() are used by zones and the VFS
402 * layer (respectively) to synchronize zone creation and new mounts.
403 *
404 * The semantics are like a reader-reader lock such that there may
405 * either be multiple mounts (or zone creations, if that weren't
406 * serialized by zonehash_lock) in progress at the same time, but not
407 * both.
408 *
409 * We use cv's so the user can ctrl-C out of the operation if it's
410 * taking too long.
411 *
412 * The semantics are such that there is unfair bias towards the
413 * "current" operation.  This means that zone creations may starve if
414 * there is a rapid succession of new mounts coming in to the system, or
415 * there is a remote possibility that zones will be created at such a
416 * rate that new mounts will not be able to proceed.
417 */
418/*
419 * Prevent new mounts from progressing to the point of calling
420 * VFS_MOUNT().  If there are already mounts in this "region", wait for
421 * them to complete.
422 */
423static int
424block_mounts(void)
425{
426	int retval = 0;
427
428	/*
429	 * Since it may block for a long time, block_mounts() shouldn't be
430	 * called with zonehash_lock held.
431	 */
432	ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
433	mutex_enter(&mount_lock);
434	while (mounts_in_progress > 0) {
435		if (cv_wait_sig(&mount_cv, &mount_lock) == 0)
436			goto signaled;
437	}
438	/*
439	 * A negative value of mounts_in_progress indicates that mounts
440	 * have been blocked by (-mounts_in_progress) different callers.
441	 */
442	mounts_in_progress--;
443	retval = 1;
444signaled:
445	mutex_exit(&mount_lock);
446	return (retval);
447}
448
449/*
450 * The VFS layer may progress with new mounts as far as we're concerned.
451 * Allow them to progress if we were the last obstacle.
452 */
453static void
454resume_mounts(void)
455{
456	mutex_enter(&mount_lock);
457	if (++mounts_in_progress == 0)
458		cv_broadcast(&mount_cv);
459	mutex_exit(&mount_lock);
460}
461
462/*
463 * The VFS layer is busy with a mount; zones should wait until all
464 * mounts are completed to progress.
465 */
466void
467mount_in_progress(void)
468{
469	mutex_enter(&mount_lock);
470	while (mounts_in_progress < 0)
471		cv_wait(&mount_cv, &mount_lock);
472	mounts_in_progress++;
473	mutex_exit(&mount_lock);
474}
475
476/*
477 * VFS is done with one mount; wake up any waiting block_mounts()
478 * callers if this is the last mount.
479 */
480void
481mount_completed(void)
482{
483	mutex_enter(&mount_lock);
484	if (--mounts_in_progress == 0)
485		cv_broadcast(&mount_cv);
486	mutex_exit(&mount_lock);
487}
488
489/*
490 * ZSD routines.
491 *
492 * Zone Specific Data (ZSD) is modeled after Thread Specific Data as
493 * defined by the pthread_key_create() and related interfaces.
494 *
495 * Kernel subsystems may register one or more data items and/or
496 * callbacks to be executed when a zone is created, shutdown, or
497 * destroyed.
498 *
499 * Unlike the thread counterpart, destructor callbacks will be executed
500 * even if the data pointer is NULL and/or there are no constructor
501 * callbacks, so it is the responsibility of such callbacks to check for
502 * NULL data values if necessary.
503 *
504 * The locking strategy and overall picture is as follows:
505 *
506 * When someone calls zone_key_create(), a template ZSD entry is added to the
507 * global list "zsd_registered_keys", protected by zsd_key_lock.  While
508 * holding that lock all the existing zones are marked as
509 * ZSD_CREATE_NEEDED and a copy of the ZSD entry added to the per-zone
510 * zone_zsd list (protected by zone_lock). The global list is updated first
511 * (under zone_key_lock) to make sure that newly created zones use the
512 * most recent list of keys. Then under zonehash_lock we walk the zones
513 * and mark them.  Similar locking is used in zone_key_delete().
514 *
515 * The actual create, shutdown, and destroy callbacks are done without
516 * holding any lock. And zsd_flags are used to ensure that the operations
517 * completed so that when zone_key_create (and zone_create) is done, as well as
518 * zone_key_delete (and zone_destroy) is done, all the necessary callbacks
519 * are completed.
520 *
521 * When new zones are created constructor callbacks for all registered ZSD
522 * entries will be called. That also uses the above two phases of marking
523 * what needs to be done, and then running the callbacks without holding
524 * any locks.
525 *
526 * The framework does not provide any locking around zone_getspecific() and
527 * zone_setspecific() apart from that needed for internal consistency, so
528 * callers interested in atomic "test-and-set" semantics will need to provide
529 * their own locking.
530 */
531
532/*
533 * Helper function to find the zsd_entry associated with the key in the
534 * given list.
535 */
536static struct zsd_entry *
537zsd_find(list_t *l, zone_key_t key)
538{
539	struct zsd_entry *zsd;
540
541	for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
542		if (zsd->zsd_key == key) {
543			return (zsd);
544		}
545	}
546	return (NULL);
547}
548
549/*
550 * Helper function to find the zsd_entry associated with the key in the
551 * given list. Move it to the front of the list.
552 */
553static struct zsd_entry *
554zsd_find_mru(list_t *l, zone_key_t key)
555{
556	struct zsd_entry *zsd;
557
558	for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
559		if (zsd->zsd_key == key) {
560			/*
561			 * Move to head of list to keep list in MRU order.
562			 */
563			if (zsd != list_head(l)) {
564				list_remove(l, zsd);
565				list_insert_head(l, zsd);
566			}
567			return (zsd);
568		}
569	}
570	return (NULL);
571}
572
573void
574zone_key_create(zone_key_t *keyp, void *(*create)(zoneid_t),
575    void (*shutdown)(zoneid_t, void *), void (*destroy)(zoneid_t, void *))
576{
577	struct zsd_entry *zsdp;
578	struct zsd_entry *t;
579	struct zone *zone;
580	zone_key_t  key;
581
582	zsdp = kmem_zalloc(sizeof (*zsdp), KM_SLEEP);
583	zsdp->zsd_data = NULL;
584	zsdp->zsd_create = create;
585	zsdp->zsd_shutdown = shutdown;
586	zsdp->zsd_destroy = destroy;
587
588	/*
589	 * Insert in global list of callbacks. Makes future zone creations
590	 * see it.
591	 */
592	mutex_enter(&zsd_key_lock);
593	*keyp = key = zsdp->zsd_key = ++zsd_keyval;
594	ASSERT(zsd_keyval != 0);
595	list_insert_tail(&zsd_registered_keys, zsdp);
596	mutex_exit(&zsd_key_lock);
597
598	/*
599	 * Insert for all existing zones and mark them as needing
600	 * a create callback.
601	 */
602	mutex_enter(&zonehash_lock);	/* stop the world */
603	for (zone = list_head(&zone_active); zone != NULL;
604	    zone = list_next(&zone_active, zone)) {
605		zone_status_t status;
606
607		mutex_enter(&zone->zone_lock);
608
609		/* Skip zones that are on the way down or not yet up */
610		status = zone_status_get(zone);
611		if (status >= ZONE_IS_DOWN ||
612		    status == ZONE_IS_UNINITIALIZED) {
613			mutex_exit(&zone->zone_lock);
614			continue;
615		}
616
617		t = zsd_find_mru(&zone->zone_zsd, key);
618		if (t != NULL) {
619			/*
620			 * A zsd_configure already inserted it after
621			 * we dropped zsd_key_lock above.
622			 */
623			mutex_exit(&zone->zone_lock);
624			continue;
625		}
626		t = kmem_zalloc(sizeof (*t), KM_SLEEP);
627		t->zsd_key = key;
628		t->zsd_create = create;
629		t->zsd_shutdown = shutdown;
630		t->zsd_destroy = destroy;
631		if (create != NULL) {
632			t->zsd_flags = ZSD_CREATE_NEEDED;
633			DTRACE_PROBE2(zsd__create__needed,
634			    zone_t *, zone, zone_key_t, key);
635		}
636		list_insert_tail(&zone->zone_zsd, t);
637		mutex_exit(&zone->zone_lock);
638	}
639	mutex_exit(&zonehash_lock);
640
641	if (create != NULL) {
642		/* Now call the create callback for this key */
643		zsd_apply_all_zones(zsd_apply_create, key);
644	}
645}
646
647/*
648 * Function called when a module is being unloaded, or otherwise wishes
649 * to unregister its ZSD key and callbacks.
650 *
651 * Remove from the global list and determine the functions that need to
652 * be called under a global lock. Then call the functions without
653 * holding any locks. Finally free up the zone_zsd entries. (The apply
654 * functions need to access the zone_zsd entries to find zsd_data etc.)
655 */
656int
657zone_key_delete(zone_key_t key)
658{
659	struct zsd_entry *zsdp = NULL;
660	zone_t *zone;
661
662	mutex_enter(&zsd_key_lock);
663	zsdp = zsd_find_mru(&zsd_registered_keys, key);
664	if (zsdp == NULL) {
665		mutex_exit(&zsd_key_lock);
666		return (-1);
667	}
668	list_remove(&zsd_registered_keys, zsdp);
669	mutex_exit(&zsd_key_lock);
670
671	mutex_enter(&zonehash_lock);
672	for (zone = list_head(&zone_active); zone != NULL;
673	    zone = list_next(&zone_active, zone)) {
674		struct zsd_entry *del;
675
676		mutex_enter(&zone->zone_lock);
677		del = zsd_find_mru(&zone->zone_zsd, key);
678		if (del == NULL) {
679			/*
680			 * Somebody else got here first e.g the zone going
681			 * away.
682			 */
683			mutex_exit(&zone->zone_lock);
684			continue;
685		}
686		ASSERT(del->zsd_shutdown == zsdp->zsd_shutdown);
687		ASSERT(del->zsd_destroy == zsdp->zsd_destroy);
688		if (del->zsd_shutdown != NULL &&
689		    (del->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
690			del->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
691			DTRACE_PROBE2(zsd__shutdown__needed,
692			    zone_t *, zone, zone_key_t, key);
693		}
694		if (del->zsd_destroy != NULL &&
695		    (del->zsd_flags & ZSD_DESTROY_ALL) == 0) {
696			del->zsd_flags |= ZSD_DESTROY_NEEDED;
697			DTRACE_PROBE2(zsd__destroy__needed,
698			    zone_t *, zone, zone_key_t, key);
699		}
700		mutex_exit(&zone->zone_lock);
701	}
702	mutex_exit(&zonehash_lock);
703	kmem_free(zsdp, sizeof (*zsdp));
704
705	/* Now call the shutdown and destroy callback for this key */
706	zsd_apply_all_zones(zsd_apply_shutdown, key);
707	zsd_apply_all_zones(zsd_apply_destroy, key);
708
709	/* Now we can free up the zsdp structures in each zone */
710	mutex_enter(&zonehash_lock);
711	for (zone = list_head(&zone_active); zone != NULL;
712	    zone = list_next(&zone_active, zone)) {
713		struct zsd_entry *del;
714
715		mutex_enter(&zone->zone_lock);
716		del = zsd_find(&zone->zone_zsd, key);
717		if (del != NULL) {
718			list_remove(&zone->zone_zsd, del);
719			ASSERT(!(del->zsd_flags & ZSD_ALL_INPROGRESS));
720			kmem_free(del, sizeof (*del));
721		}
722		mutex_exit(&zone->zone_lock);
723	}
724	mutex_exit(&zonehash_lock);
725
726	return (0);
727}
728
729/*
730 * ZSD counterpart of pthread_setspecific().
731 *
732 * Since all zsd callbacks, including those with no create function,
733 * have an entry in zone_zsd, if the key is registered it is part of
734 * the zone_zsd list.
735 * Return an error if the key wasn't registerd.
736 */
737int
738zone_setspecific(zone_key_t key, zone_t *zone, const void *data)
739{
740	struct zsd_entry *t;
741
742	mutex_enter(&zone->zone_lock);
743	t = zsd_find_mru(&zone->zone_zsd, key);
744	if (t != NULL) {
745		/*
746		 * Replace old value with new
747		 */
748		t->zsd_data = (void *)data;
749		mutex_exit(&zone->zone_lock);
750		return (0);
751	}
752	mutex_exit(&zone->zone_lock);
753	return (-1);
754}
755
756/*
757 * ZSD counterpart of pthread_getspecific().
758 */
759void *
760zone_getspecific(zone_key_t key, zone_t *zone)
761{
762	struct zsd_entry *t;
763	void *data;
764
765	mutex_enter(&zone->zone_lock);
766	t = zsd_find_mru(&zone->zone_zsd, key);
767	data = (t == NULL ? NULL : t->zsd_data);
768	mutex_exit(&zone->zone_lock);
769	return (data);
770}
771
772/*
773 * Function used to initialize a zone's list of ZSD callbacks and data
774 * when the zone is being created.  The callbacks are initialized from
775 * the template list (zsd_registered_keys). The constructor callback is
776 * executed later (once the zone exists and with locks dropped).
777 */
778static void
779zone_zsd_configure(zone_t *zone)
780{
781	struct zsd_entry *zsdp;
782	struct zsd_entry *t;
783
784	ASSERT(MUTEX_HELD(&zonehash_lock));
785	ASSERT(list_head(&zone->zone_zsd) == NULL);
786	mutex_enter(&zone->zone_lock);
787	mutex_enter(&zsd_key_lock);
788	for (zsdp = list_head(&zsd_registered_keys); zsdp != NULL;
789	    zsdp = list_next(&zsd_registered_keys, zsdp)) {
790		/*
791		 * Since this zone is ZONE_IS_UNCONFIGURED, zone_key_create
792		 * should not have added anything to it.
793		 */
794		ASSERT(zsd_find(&zone->zone_zsd, zsdp->zsd_key) == NULL);
795
796		t = kmem_zalloc(sizeof (*t), KM_SLEEP);
797		t->zsd_key = zsdp->zsd_key;
798		t->zsd_create = zsdp->zsd_create;
799		t->zsd_shutdown = zsdp->zsd_shutdown;
800		t->zsd_destroy = zsdp->zsd_destroy;
801		if (zsdp->zsd_create != NULL) {
802			t->zsd_flags = ZSD_CREATE_NEEDED;
803			DTRACE_PROBE2(zsd__create__needed,
804			    zone_t *, zone, zone_key_t, zsdp->zsd_key);
805		}
806		list_insert_tail(&zone->zone_zsd, t);
807	}
808	mutex_exit(&zsd_key_lock);
809	mutex_exit(&zone->zone_lock);
810}
811
812enum zsd_callback_type { ZSD_CREATE, ZSD_SHUTDOWN, ZSD_DESTROY };
813
814/*
815 * Helper function to execute shutdown or destructor callbacks.
816 */
817static void
818zone_zsd_callbacks(zone_t *zone, enum zsd_callback_type ct)
819{
820	struct zsd_entry *t;
821
822	ASSERT(ct == ZSD_SHUTDOWN || ct == ZSD_DESTROY);
823	ASSERT(ct != ZSD_SHUTDOWN || zone_status_get(zone) >= ZONE_IS_EMPTY);
824	ASSERT(ct != ZSD_DESTROY || zone_status_get(zone) >= ZONE_IS_DOWN);
825
826	/*
827	 * Run the callback solely based on what is registered for the zone
828	 * in zone_zsd. The global list can change independently of this
829	 * as keys are registered and unregistered and we don't register new
830	 * callbacks for a zone that is in the process of going away.
831	 */
832	mutex_enter(&zone->zone_lock);
833	for (t = list_head(&zone->zone_zsd); t != NULL;
834	    t = list_next(&zone->zone_zsd, t)) {
835		zone_key_t key = t->zsd_key;
836
837		/* Skip if no callbacks registered */
838
839		if (ct == ZSD_SHUTDOWN) {
840			if (t->zsd_shutdown != NULL &&
841			    (t->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
842				t->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
843				DTRACE_PROBE2(zsd__shutdown__needed,
844				    zone_t *, zone, zone_key_t, key);
845			}
846		} else {
847			if (t->zsd_destroy != NULL &&
848			    (t->zsd_flags & ZSD_DESTROY_ALL) == 0) {
849				t->zsd_flags |= ZSD_DESTROY_NEEDED;
850				DTRACE_PROBE2(zsd__destroy__needed,
851				    zone_t *, zone, zone_key_t, key);
852			}
853		}
854	}
855	mutex_exit(&zone->zone_lock);
856
857	/* Now call the shutdown and destroy callback for this key */
858	zsd_apply_all_keys(zsd_apply_shutdown, zone);
859	zsd_apply_all_keys(zsd_apply_destroy, zone);
860
861}
862
863/*
864 * Called when the zone is going away; free ZSD-related memory, and
865 * destroy the zone_zsd list.
866 */
867static void
868zone_free_zsd(zone_t *zone)
869{
870	struct zsd_entry *t, *next;
871
872	/*
873	 * Free all the zsd_entry's we had on this zone.
874	 */
875	mutex_enter(&zone->zone_lock);
876	for (t = list_head(&zone->zone_zsd); t != NULL; t = next) {
877		next = list_next(&zone->zone_zsd, t);
878		list_remove(&zone->zone_zsd, t);
879		ASSERT(!(t->zsd_flags & ZSD_ALL_INPROGRESS));
880		kmem_free(t, sizeof (*t));
881	}
882	list_destroy(&zone->zone_zsd);
883	mutex_exit(&zone->zone_lock);
884
885}
886
887/*
888 * Apply a function to all zones for particular key value.
889 *
890 * The applyfn has to drop zonehash_lock if it does some work, and
891 * then reacquire it before it returns.
892 * When the lock is dropped we don't follow list_next even
893 * if it is possible to do so without any hazards. This is
894 * because we want the design to allow for the list of zones
895 * to change in any arbitrary way during the time the
896 * lock was dropped.
897 *
898 * It is safe to restart the loop at list_head since the applyfn
899 * changes the zsd_flags as it does work, so a subsequent
900 * pass through will have no effect in applyfn, hence the loop will terminate
901 * in at worst O(N^2).
902 */
903static void
904zsd_apply_all_zones(zsd_applyfn_t *applyfn, zone_key_t key)
905{
906	zone_t *zone;
907
908	mutex_enter(&zonehash_lock);
909	zone = list_head(&zone_active);
910	while (zone != NULL) {
911		if ((applyfn)(&zonehash_lock, B_FALSE, zone, key)) {
912			/* Lock dropped - restart at head */
913			zone = list_head(&zone_active);
914		} else {
915			zone = list_next(&zone_active, zone);
916		}
917	}
918	mutex_exit(&zonehash_lock);
919}
920
921/*
922 * Apply a function to all keys for a particular zone.
923 *
924 * The applyfn has to drop zonehash_lock if it does some work, and
925 * then reacquire it before it returns.
926 * When the lock is dropped we don't follow list_next even
927 * if it is possible to do so without any hazards. This is
928 * because we want the design to allow for the list of zsd callbacks
929 * to change in any arbitrary way during the time the
930 * lock was dropped.
931 *
932 * It is safe to restart the loop at list_head since the applyfn
933 * changes the zsd_flags as it does work, so a subsequent
934 * pass through will have no effect in applyfn, hence the loop will terminate
935 * in at worst O(N^2).
936 */
937static void
938zsd_apply_all_keys(zsd_applyfn_t *applyfn, zone_t *zone)
939{
940	struct zsd_entry *t;
941
942	mutex_enter(&zone->zone_lock);
943	t = list_head(&zone->zone_zsd);
944	while (t != NULL) {
945		if ((applyfn)(NULL, B_TRUE, zone, t->zsd_key)) {
946			/* Lock dropped - restart at head */
947			t = list_head(&zone->zone_zsd);
948		} else {
949			t = list_next(&zone->zone_zsd, t);
950		}
951	}
952	mutex_exit(&zone->zone_lock);
953}
954
955/*
956 * Call the create function for the zone and key if CREATE_NEEDED
957 * is set.
958 * If some other thread gets here first and sets CREATE_INPROGRESS, then
959 * we wait for that thread to complete so that we can ensure that
960 * all the callbacks are done when we've looped over all zones/keys.
961 *
962 * When we call the create function, we drop the global held by the
963 * caller, and return true to tell the caller it needs to re-evalute the
964 * state.
965 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
966 * remains held on exit.
967 */
968static boolean_t
969zsd_apply_create(kmutex_t *lockp, boolean_t zone_lock_held,
970    zone_t *zone, zone_key_t key)
971{
972	void *result;
973	struct zsd_entry *t;
974	boolean_t dropped;
975
976	if (lockp != NULL) {
977		ASSERT(MUTEX_HELD(lockp));
978	}
979	if (zone_lock_held) {
980		ASSERT(MUTEX_HELD(&zone->zone_lock));
981	} else {
982		mutex_enter(&zone->zone_lock);
983	}
984
985	t = zsd_find(&zone->zone_zsd, key);
986	if (t == NULL) {
987		/*
988		 * Somebody else got here first e.g the zone going
989		 * away.
990		 */
991		if (!zone_lock_held)
992			mutex_exit(&zone->zone_lock);
993		return (B_FALSE);
994	}
995	dropped = B_FALSE;
996	if (zsd_wait_for_inprogress(zone, t, lockp))
997		dropped = B_TRUE;
998
999	if (t->zsd_flags & ZSD_CREATE_NEEDED) {
1000		t->zsd_flags &= ~ZSD_CREATE_NEEDED;
1001		t->zsd_flags |= ZSD_CREATE_INPROGRESS;
1002		DTRACE_PROBE2(zsd__create__inprogress,
1003		    zone_t *, zone, zone_key_t, key);
1004		mutex_exit(&zone->zone_lock);
1005		if (lockp != NULL)
1006			mutex_exit(lockp);
1007
1008		dropped = B_TRUE;
1009		ASSERT(t->zsd_create != NULL);
1010		DTRACE_PROBE2(zsd__create__start,
1011		    zone_t *, zone, zone_key_t, key);
1012
1013		result = (*t->zsd_create)(zone->zone_id);
1014
1015		DTRACE_PROBE2(zsd__create__end,
1016		    zone_t *, zone, voidn *, result);
1017
1018		ASSERT(result != NULL);
1019		if (lockp != NULL)
1020			mutex_enter(lockp);
1021		mutex_enter(&zone->zone_lock);
1022		t->zsd_data = result;
1023		t->zsd_flags &= ~ZSD_CREATE_INPROGRESS;
1024		t->zsd_flags |= ZSD_CREATE_COMPLETED;
1025		cv_broadcast(&t->zsd_cv);
1026		DTRACE_PROBE2(zsd__create__completed,
1027		    zone_t *, zone, zone_key_t, key);
1028	}
1029	if (!zone_lock_held)
1030		mutex_exit(&zone->zone_lock);
1031	return (dropped);
1032}
1033
1034/*
1035 * Call the shutdown function for the zone and key if SHUTDOWN_NEEDED
1036 * is set.
1037 * If some other thread gets here first and sets *_INPROGRESS, then
1038 * we wait for that thread to complete so that we can ensure that
1039 * all the callbacks are done when we've looped over all zones/keys.
1040 *
1041 * When we call the shutdown function, we drop the global held by the
1042 * caller, and return true to tell the caller it needs to re-evalute the
1043 * state.
1044 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1045 * remains held on exit.
1046 */
1047static boolean_t
1048zsd_apply_shutdown(kmutex_t *lockp, boolean_t zone_lock_held,
1049    zone_t *zone, zone_key_t key)
1050{
1051	struct zsd_entry *t;
1052	void *data;
1053	boolean_t dropped;
1054
1055	if (lockp != NULL) {
1056		ASSERT(MUTEX_HELD(lockp));
1057	}
1058	if (zone_lock_held) {
1059		ASSERT(MUTEX_HELD(&zone->zone_lock));
1060	} else {
1061		mutex_enter(&zone->zone_lock);
1062	}
1063
1064	t = zsd_find(&zone->zone_zsd, key);
1065	if (t == NULL) {
1066		/*
1067		 * Somebody else got here first e.g the zone going
1068		 * away.
1069		 */
1070		if (!zone_lock_held)
1071			mutex_exit(&zone->zone_lock);
1072		return (B_FALSE);
1073	}
1074	dropped = B_FALSE;
1075	if (zsd_wait_for_creator(zone, t, lockp))
1076		dropped = B_TRUE;
1077
1078	if (zsd_wait_for_inprogress(zone, t, lockp))
1079		dropped = B_TRUE;
1080
1081	if (t->zsd_flags & ZSD_SHUTDOWN_NEEDED) {
1082		t->zsd_flags &= ~ZSD_SHUTDOWN_NEEDED;
1083		t->zsd_flags |= ZSD_SHUTDOWN_INPROGRESS;
1084		DTRACE_PROBE2(zsd__shutdown__inprogress,
1085		    zone_t *, zone, zone_key_t, key);
1086		mutex_exit(&zone->zone_lock);
1087		if (lockp != NULL)
1088			mutex_exit(lockp);
1089		dropped = B_TRUE;
1090
1091		ASSERT(t->zsd_shutdown != NULL);
1092		data = t->zsd_data;
1093
1094		DTRACE_PROBE2(zsd__shutdown__start,
1095		    zone_t *, zone, zone_key_t, key);
1096
1097		(t->zsd_shutdown)(zone->zone_id, data);
1098		DTRACE_PROBE2(zsd__shutdown__end,
1099		    zone_t *, zone, zone_key_t, key);
1100
1101		if (lockp != NULL)
1102			mutex_enter(lockp);
1103		mutex_enter(&zone->zone_lock);
1104		t->zsd_flags &= ~ZSD_SHUTDOWN_INPROGRESS;
1105		t->zsd_flags |= ZSD_SHUTDOWN_COMPLETED;
1106		cv_broadcast(&t->zsd_cv);
1107		DTRACE_PROBE2(zsd__shutdown__completed,
1108		    zone_t *, zone, zone_key_t, key);
1109	}
1110	if (!zone_lock_held)
1111		mutex_exit(&zone->zone_lock);
1112	return (dropped);
1113}
1114
1115/*
1116 * Call the destroy function for the zone and key if DESTROY_NEEDED
1117 * is set.
1118 * If some other thread gets here first and sets *_INPROGRESS, then
1119 * we wait for that thread to complete so that we can ensure that
1120 * all the callbacks are done when we've looped over all zones/keys.
1121 *
1122 * When we call the destroy function, we drop the global held by the
1123 * caller, and return true to tell the caller it needs to re-evalute the
1124 * state.
1125 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1126 * remains held on exit.
1127 */
1128static boolean_t
1129zsd_apply_destroy(kmutex_t *lockp, boolean_t zone_lock_held,
1130    zone_t *zone, zone_key_t key)
1131{
1132	struct zsd_entry *t;
1133	void *data;
1134	boolean_t dropped;
1135
1136	if (lockp != NULL) {
1137		ASSERT(MUTEX_HELD(lockp));
1138	}
1139	if (zone_lock_held) {
1140		ASSERT(MUTEX_HELD(&zone->zone_lock));
1141	} else {
1142		mutex_enter(&zone->zone_lock);
1143	}
1144
1145	t = zsd_find(&zone->zone_zsd, key);
1146	if (t == NULL) {
1147		/*
1148		 * Somebody else got here first e.g the zone going
1149		 * away.
1150		 */
1151		if (!zone_lock_held)
1152			mutex_exit(&zone->zone_lock);
1153		return (B_FALSE);
1154	}
1155	dropped = B_FALSE;
1156	if (zsd_wait_for_creator(zone, t, lockp))
1157		dropped = B_TRUE;
1158
1159	if (zsd_wait_for_inprogress(zone, t, lockp))
1160		dropped = B_TRUE;
1161
1162	if (t->zsd_flags & ZSD_DESTROY_NEEDED) {
1163		t->zsd_flags &= ~ZSD_DESTROY_NEEDED;
1164		t->zsd_flags |= ZSD_DESTROY_INPROGRESS;
1165		DTRACE_PROBE2(zsd__destroy__inprogress,
1166		    zone_t *, zone, zone_key_t, key);
1167		mutex_exit(&zone->zone_lock);
1168		if (lockp != NULL)
1169			mutex_exit(lockp);
1170		dropped = B_TRUE;
1171
1172		ASSERT(t->zsd_destroy != NULL);
1173		data = t->zsd_data;
1174		DTRACE_PROBE2(zsd__destroy__start,
1175		    zone_t *, zone, zone_key_t, key);
1176
1177		(t->zsd_destroy)(zone->zone_id, data);
1178		DTRACE_PROBE2(zsd__destroy__end,
1179		    zone_t *, zone, zone_key_t, key);
1180
1181		if (lockp != NULL)
1182			mutex_enter(lockp);
1183		mutex_enter(&zone->zone_lock);
1184		t->zsd_data = NULL;
1185		t->zsd_flags &= ~ZSD_DESTROY_INPROGRESS;
1186		t->zsd_flags |= ZSD_DESTROY_COMPLETED;
1187		cv_broadcast(&t->zsd_cv);
1188		DTRACE_PROBE2(zsd__destroy__completed,
1189		    zone_t *, zone, zone_key_t, key);
1190	}
1191	if (!zone_lock_held)
1192		mutex_exit(&zone->zone_lock);
1193	return (dropped);
1194}
1195
1196/*
1197 * Wait for any CREATE_NEEDED flag to be cleared.
1198 * Returns true if lockp was temporarily dropped while waiting.
1199 */
1200static boolean_t
1201zsd_wait_for_creator(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
1202{
1203	boolean_t dropped = B_FALSE;
1204
1205	while (t->zsd_flags & ZSD_CREATE_NEEDED) {
1206		DTRACE_PROBE2(zsd__wait__for__creator,
1207		    zone_t *, zone, struct zsd_entry *, t);
1208		if (lockp != NULL) {
1209			dropped = B_TRUE;
1210			mutex_exit(lockp);
1211		}
1212		cv_wait(&t->zsd_cv, &zone->zone_lock);
1213		if (lockp != NULL) {
1214			/* First drop zone_lock to preserve order */
1215			mutex_exit(&zone->zone_lock);
1216			mutex_enter(lockp);
1217			mutex_enter(&zone->zone_lock);
1218		}
1219	}
1220	return (dropped);
1221}
1222
1223/*
1224 * Wait for any INPROGRESS flag to be cleared.
1225 * Returns true if lockp was temporarily dropped while waiting.
1226 */
1227static boolean_t
1228zsd_wait_for_inprogress(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
1229{
1230	boolean_t dropped = B_FALSE;
1231
1232	while (t->zsd_flags & ZSD_ALL_INPROGRESS) {
1233		DTRACE_PROBE2(zsd__wait__for__inprogress,
1234		    zone_t *, zone, struct zsd_entry *, t);
1235		if (lockp != NULL) {
1236			dropped = B_TRUE;
1237			mutex_exit(lockp);
1238		}
1239		cv_wait(&t->zsd_cv, &zone->zone_lock);
1240		if (lockp != NULL) {
1241			/* First drop zone_lock to preserve order */
1242			mutex_exit(&zone->zone_lock);
1243			mutex_enter(lockp);
1244			mutex_enter(&zone->zone_lock);
1245		}
1246	}
1247	return (dropped);
1248}
1249
1250/*
1251 * Frees memory associated with the zone dataset list.
1252 */
1253static void
1254zone_free_datasets(zone_t *zone)
1255{
1256	zone_dataset_t *t, *next;
1257
1258	for (t = list_head(&zone->zone_datasets); t != NULL; t = next) {
1259		next = list_next(&zone->zone_datasets, t);
1260		list_remove(&zone->zone_datasets, t);
1261		kmem_free(t->zd_dataset, strlen(t->zd_dataset) + 1);
1262		kmem_free(t, sizeof (*t));
1263	}
1264	list_destroy(&zone->zone_datasets);
1265}
1266
1267/*
1268 * zone.cpu-shares resource control support.
1269 */
1270/*ARGSUSED*/
1271static rctl_qty_t
1272zone_cpu_shares_usage(rctl_t *rctl, struct proc *p)
1273{
1274	ASSERT(MUTEX_HELD(&p->p_lock));
1275	return (p->p_zone->zone_shares);
1276}
1277
1278/*ARGSUSED*/
1279static int
1280zone_cpu_shares_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1281    rctl_qty_t nv)
1282{
1283	ASSERT(MUTEX_HELD(&p->p_lock));
1284	ASSERT(e->rcep_t == RCENTITY_ZONE);
1285	if (e->rcep_p.zone == NULL)
1286		return (0);
1287
1288	e->rcep_p.zone->zone_shares = nv;
1289	return (0);
1290}
1291
1292static rctl_ops_t zone_cpu_shares_ops = {
1293	rcop_no_action,
1294	zone_cpu_shares_usage,
1295	zone_cpu_shares_set,
1296	rcop_no_test
1297};
1298
1299/*
1300 * zone.cpu-cap resource control support.
1301 */
1302/*ARGSUSED*/
1303static rctl_qty_t
1304zone_cpu_cap_get(rctl_t *rctl, struct proc *p)
1305{
1306	ASSERT(MUTEX_HELD(&p->p_lock));
1307	return (cpucaps_zone_get(p->p_zone));
1308}
1309
1310/*ARGSUSED*/
1311static int
1312zone_cpu_cap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1313    rctl_qty_t nv)
1314{
1315	zone_t *zone = e->rcep_p.zone;
1316
1317	ASSERT(MUTEX_HELD(&p->p_lock));
1318	ASSERT(e->rcep_t == RCENTITY_ZONE);
1319
1320	if (zone == NULL)
1321		return (0);
1322
1323	/*
1324	 * set cap to the new value.
1325	 */
1326	return (cpucaps_zone_set(zone, nv));
1327}
1328
1329static rctl_ops_t zone_cpu_cap_ops = {
1330	rcop_no_action,
1331	zone_cpu_cap_get,
1332	zone_cpu_cap_set,
1333	rcop_no_test
1334};
1335
1336/*ARGSUSED*/
1337static rctl_qty_t
1338zone_lwps_usage(rctl_t *r, proc_t *p)
1339{
1340	rctl_qty_t nlwps;
1341	zone_t *zone = p->p_zone;
1342
1343	ASSERT(MUTEX_HELD(&p->p_lock));
1344
1345	mutex_enter(&zone->zone_nlwps_lock);
1346	nlwps = zone->zone_nlwps;
1347	mutex_exit(&zone->zone_nlwps_lock);
1348
1349	return (nlwps);
1350}
1351
1352/*ARGSUSED*/
1353static int
1354zone_lwps_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
1355    rctl_qty_t incr, uint_t flags)
1356{
1357	rctl_qty_t nlwps;
1358
1359	ASSERT(MUTEX_HELD(&p->p_lock));
1360	ASSERT(e->rcep_t == RCENTITY_ZONE);
1361	if (e->rcep_p.zone == NULL)
1362		return (0);
1363	ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
1364	nlwps = e->rcep_p.zone->zone_nlwps;
1365
1366	if (nlwps + incr > rcntl->rcv_value)
1367		return (1);
1368
1369	return (0);
1370}
1371
1372/*ARGSUSED*/
1373static int
1374zone_lwps_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv)
1375{
1376	ASSERT(MUTEX_HELD(&p->p_lock));
1377	ASSERT(e->rcep_t == RCENTITY_ZONE);
1378	if (e->rcep_p.zone == NULL)
1379		return (0);
1380	e->rcep_p.zone->zone_nlwps_ctl = nv;
1381	return (0);
1382}
1383
1384static rctl_ops_t zone_lwps_ops = {
1385	rcop_no_action,
1386	zone_lwps_usage,
1387	zone_lwps_set,
1388	zone_lwps_test,
1389};
1390
1391/*ARGSUSED*/
1392static int
1393zone_shmmax_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1394    rctl_qty_t incr, uint_t flags)
1395{
1396	rctl_qty_t v;
1397	ASSERT(MUTEX_HELD(&p->p_lock));
1398	ASSERT(e->rcep_t == RCENTITY_ZONE);
1399	v = e->rcep_p.zone->zone_shmmax + incr;
1400	if (v > rval->rcv_value)
1401		return (1);
1402	return (0);
1403}
1404
1405static rctl_ops_t zone_shmmax_ops = {
1406	rcop_no_action,
1407	rcop_no_usage,
1408	rcop_no_set,
1409	zone_shmmax_test
1410};
1411
1412/*ARGSUSED*/
1413static int
1414zone_shmmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1415    rctl_qty_t incr, uint_t flags)
1416{
1417	rctl_qty_t v;
1418	ASSERT(MUTEX_HELD(&p->p_lock));
1419	ASSERT(e->rcep_t == RCENTITY_ZONE);
1420	v = e->rcep_p.zone->zone_ipc.ipcq_shmmni + incr;
1421	if (v > rval->rcv_value)
1422		return (1);
1423	return (0);
1424}
1425
1426static rctl_ops_t zone_shmmni_ops = {
1427	rcop_no_action,
1428	rcop_no_usage,
1429	rcop_no_set,
1430	zone_shmmni_test
1431};
1432
1433/*ARGSUSED*/
1434static int
1435zone_semmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1436    rctl_qty_t incr, uint_t flags)
1437{
1438	rctl_qty_t v;
1439	ASSERT(MUTEX_HELD(&p->p_lock));
1440	ASSERT(e->rcep_t == RCENTITY_ZONE);
1441	v = e->rcep_p.zone->zone_ipc.ipcq_semmni + incr;
1442	if (v > rval->rcv_value)
1443		return (1);
1444	return (0);
1445}
1446
1447static rctl_ops_t zone_semmni_ops = {
1448	rcop_no_action,
1449	rcop_no_usage,
1450	rcop_no_set,
1451	zone_semmni_test
1452};
1453
1454/*ARGSUSED*/
1455static int
1456zone_msgmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1457    rctl_qty_t incr, uint_t flags)
1458{
1459	rctl_qty_t v;
1460	ASSERT(MUTEX_HELD(&p->p_lock));
1461	ASSERT(e->rcep_t == RCENTITY_ZONE);
1462	v = e->rcep_p.zone->zone_ipc.ipcq_msgmni + incr;
1463	if (v > rval->rcv_value)
1464		return (1);
1465	return (0);
1466}
1467
1468static rctl_ops_t zone_msgmni_ops = {
1469	rcop_no_action,
1470	rcop_no_usage,
1471	rcop_no_set,
1472	zone_msgmni_test
1473};
1474
1475/*ARGSUSED*/
1476static rctl_qty_t
1477zone_locked_mem_usage(rctl_t *rctl, struct proc *p)
1478{
1479	rctl_qty_t q;
1480	ASSERT(MUTEX_HELD(&p->p_lock));
1481	mutex_enter(&p->p_zone->zone_mem_lock);
1482	q = p->p_zone->zone_locked_mem;
1483	mutex_exit(&p->p_zone->zone_mem_lock);
1484	return (q);
1485}
1486
1487/*ARGSUSED*/
1488static int
1489zone_locked_mem_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1490    rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1491{
1492	rctl_qty_t q;
1493	zone_t *z;
1494
1495	z = e->rcep_p.zone;
1496	ASSERT(MUTEX_HELD(&p->p_lock));
1497	ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1498	q = z->zone_locked_mem;
1499	if (q + incr > rcntl->rcv_value)
1500		return (1);
1501	return (0);
1502}
1503
1504/*ARGSUSED*/
1505static int
1506zone_locked_mem_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1507    rctl_qty_t nv)
1508{
1509	ASSERT(MUTEX_HELD(&p->p_lock));
1510	ASSERT(e->rcep_t == RCENTITY_ZONE);
1511	if (e->rcep_p.zone == NULL)
1512		return (0);
1513	e->rcep_p.zone->zone_locked_mem_ctl = nv;
1514	return (0);
1515}
1516
1517static rctl_ops_t zone_locked_mem_ops = {
1518	rcop_no_action,
1519	zone_locked_mem_usage,
1520	zone_locked_mem_set,
1521	zone_locked_mem_test
1522};
1523
1524/*ARGSUSED*/
1525static rctl_qty_t
1526zone_max_swap_usage(rctl_t *rctl, struct proc *p)
1527{
1528	rctl_qty_t q;
1529	zone_t *z = p->p_zone;
1530
1531	ASSERT(MUTEX_HELD(&p->p_lock));
1532	mutex_enter(&z->zone_mem_lock);
1533	q = z->zone_max_swap;
1534	mutex_exit(&z->zone_mem_lock);
1535	return (q);
1536}
1537
1538/*ARGSUSED*/
1539static int
1540zone_max_swap_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1541    rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1542{
1543	rctl_qty_t q;
1544	zone_t *z;
1545
1546	z = e->rcep_p.zone;
1547	ASSERT(MUTEX_HELD(&p->p_lock));
1548	ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1549	q = z->zone_max_swap;
1550	if (q + incr > rcntl->rcv_value)
1551		return (1);
1552	return (0);
1553}
1554
1555/*ARGSUSED*/
1556static int
1557zone_max_swap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1558    rctl_qty_t nv)
1559{
1560	ASSERT(MUTEX_HELD(&p->p_lock));
1561	ASSERT(e->rcep_t == RCENTITY_ZONE);
1562	if (e->rcep_p.zone == NULL)
1563		return (0);
1564	e->rcep_p.zone->zone_max_swap_ctl = nv;
1565	return (0);
1566}
1567
1568static rctl_ops_t zone_max_swap_ops = {
1569	rcop_no_action,
1570	zone_max_swap_usage,
1571	zone_max_swap_set,
1572	zone_max_swap_test
1573};
1574
1575/*
1576 * Helper function to brand the zone with a unique ID.
1577 */
1578static void
1579zone_uniqid(zone_t *zone)
1580{
1581	static uint64_t uniqid = 0;
1582
1583	ASSERT(MUTEX_HELD(&zonehash_lock));
1584	zone->zone_uniqid = uniqid++;
1585}
1586
1587/*
1588 * Returns a held pointer to the "kcred" for the specified zone.
1589 */
1590struct cred *
1591zone_get_kcred(zoneid_t zoneid)
1592{
1593	zone_t *zone;
1594	cred_t *cr;
1595
1596	if ((zone = zone_find_by_id(zoneid)) == NULL)
1597		return (NULL);
1598	cr = zone->zone_kcred;
1599	crhold(cr);
1600	zone_rele(zone);
1601	return (cr);
1602}
1603
1604static int
1605zone_lockedmem_kstat_update(kstat_t *ksp, int rw)
1606{
1607	zone_t *zone = ksp->ks_private;
1608	zone_kstat_t *zk = ksp->ks_data;
1609
1610	if (rw == KSTAT_WRITE)
1611		return (EACCES);
1612
1613	zk->zk_usage.value.ui64 = zone->zone_locked_mem;
1614	zk->zk_value.value.ui64 = zone->zone_locked_mem_ctl;
1615	return (0);
1616}
1617
1618static int
1619zone_swapresv_kstat_update(kstat_t *ksp, int rw)
1620{
1621	zone_t *zone = ksp->ks_private;
1622	zone_kstat_t *zk = ksp->ks_data;
1623
1624	if (rw == KSTAT_WRITE)
1625		return (EACCES);
1626
1627	zk->zk_usage.value.ui64 = zone->zone_max_swap;
1628	zk->zk_value.value.ui64 = zone->zone_max_swap_ctl;
1629	return (0);
1630}
1631
1632static void
1633zone_kstat_create(zone_t *zone)
1634{
1635	kstat_t *ksp;
1636	zone_kstat_t *zk;
1637
1638	ksp = rctl_kstat_create_zone(zone, "lockedmem", KSTAT_TYPE_NAMED,
1639	    sizeof (zone_kstat_t) / sizeof (kstat_named_t),
1640	    KSTAT_FLAG_VIRTUAL);
1641
1642	if (ksp == NULL)
1643		return;
1644
1645	zk = ksp->ks_data = kmem_alloc(sizeof (zone_kstat_t), KM_SLEEP);
1646	ksp->ks_data_size += strlen(zone->zone_name) + 1;
1647	kstat_named_init(&zk->zk_zonename, "zonename", KSTAT_DATA_STRING);
1648	kstat_named_setstr(&zk->zk_zonename, zone->zone_name);
1649	kstat_named_init(&zk->zk_usage, "usage", KSTAT_DATA_UINT64);
1650	kstat_named_init(&zk->zk_value, "value", KSTAT_DATA_UINT64);
1651	ksp->ks_update = zone_lockedmem_kstat_update;
1652	ksp->ks_private = zone;
1653	kstat_install(ksp);
1654
1655	zone->zone_lockedmem_kstat = ksp;
1656
1657	ksp = rctl_kstat_create_zone(zone, "swapresv", KSTAT_TYPE_NAMED,
1658	    sizeof (zone_kstat_t) / sizeof (kstat_named_t),
1659	    KSTAT_FLAG_VIRTUAL);
1660
1661	if (ksp == NULL)
1662		return;
1663
1664	zk = ksp->ks_data = kmem_alloc(sizeof (zone_kstat_t), KM_SLEEP);
1665	ksp->ks_data_size += strlen(zone->zone_name) + 1;
1666	kstat_named_init(&zk->zk_zonename, "zonename", KSTAT_DATA_STRING);
1667	kstat_named_setstr(&zk->zk_zonename, zone->zone_name);
1668	kstat_named_init(&zk->zk_usage, "usage", KSTAT_DATA_UINT64);
1669	kstat_named_init(&zk->zk_value, "value", KSTAT_DATA_UINT64);
1670	ksp->ks_update = zone_swapresv_kstat_update;
1671	ksp->ks_private = zone;
1672	kstat_install(ksp);
1673
1674	zone->zone_swapresv_kstat = ksp;
1675}
1676
1677static void
1678zone_kstat_delete(zone_t *zone)
1679{
1680	void *data;
1681
1682	if (zone->zone_lockedmem_kstat != NULL) {
1683		data = zone->zone_lockedmem_kstat->ks_data;
1684		kstat_delete(zone->zone_lockedmem_kstat);
1685		kmem_free(data, sizeof (zone_kstat_t));
1686	}
1687	if (zone->zone_swapresv_kstat != NULL) {
1688		data = zone->zone_swapresv_kstat->ks_data;
1689		kstat_delete(zone->zone_swapresv_kstat);
1690		kmem_free(data, sizeof (zone_kstat_t));
1691	}
1692}
1693
1694/*
1695 * Called very early on in boot to initialize the ZSD list so that
1696 * zone_key_create() can be called before zone_init().  It also initializes
1697 * portions of zone0 which may be used before zone_init() is called.  The
1698 * variable "global_zone" will be set when zone0 is fully initialized by
1699 * zone_init().
1700 */
1701void
1702zone_zsd_init(void)
1703{
1704	mutex_init(&zonehash_lock, NULL, MUTEX_DEFAULT, NULL);
1705	mutex_init(&zsd_key_lock, NULL, MUTEX_DEFAULT, NULL);
1706	list_create(&zsd_registered_keys, sizeof (struct zsd_entry),
1707	    offsetof(struct zsd_entry, zsd_linkage));
1708	list_create(&zone_active, sizeof (zone_t),
1709	    offsetof(zone_t, zone_linkage));
1710	list_create(&zone_deathrow, sizeof (zone_t),
1711	    offsetof(zone_t, zone_linkage));
1712
1713	mutex_init(&zone0.zone_lock, NULL, MUTEX_DEFAULT, NULL);
1714	mutex_init(&zone0.zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
1715	mutex_init(&zone0.zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
1716	zone0.zone_shares = 1;
1717	zone0.zone_nlwps = 0;
1718	zone0.zone_nlwps_ctl = INT_MAX;
1719	zone0.zone_locked_mem = 0;
1720	zone0.zone_locked_mem_ctl = UINT64_MAX;
1721	ASSERT(zone0.zone_max_swap == 0);
1722	zone0.zone_max_swap_ctl = UINT64_MAX;
1723	zone0.zone_shmmax = 0;
1724	zone0.zone_ipc.ipcq_shmmni = 0;
1725	zone0.zone_ipc.ipcq_semmni = 0;
1726	zone0.zone_ipc.ipcq_msgmni = 0;
1727	zone0.zone_name = GLOBAL_ZONENAME;
1728	zone0.zone_nodename = utsname.nodename;
1729	zone0.zone_domain = srpc_domain;
1730	zone0.zone_ref = 1;
1731	zone0.zone_id = GLOBAL_ZONEID;
1732	zone0.zone_status = ZONE_IS_RUNNING;
1733	zone0.zone_rootpath = "/";
1734	zone0.zone_rootpathlen = 2;
1735	zone0.zone_psetid = ZONE_PS_INVAL;
1736	zone0.zone_ncpus = 0;
1737	zone0.zone_ncpus_online = 0;
1738	zone0.zone_proc_initpid = 1;
1739	zone0.zone_initname = initname;
1740	zone0.zone_lockedmem_kstat = NULL;
1741	zone0.zone_swapresv_kstat = NULL;
1742	list_create(&zone0.zone_zsd, sizeof (struct zsd_entry),
1743	    offsetof(struct zsd_entry, zsd_linkage));
1744	list_insert_head(&zone_active, &zone0);
1745
1746	/*
1747	 * The root filesystem is not mounted yet, so zone_rootvp cannot be set
1748	 * to anything meaningful.  It is assigned to be 'rootdir' in
1749	 * vfs_mountroot().
1750	 */
1751	zone0.zone_rootvp = NULL;
1752	zone0.zone_vfslist = NULL;
1753	zone0.zone_bootargs = initargs;
1754	zone0.zone_privset = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
1755	/*
1756	 * The global zone has all privileges
1757	 */
1758	priv_fillset(zone0.zone_privset);
1759	/*
1760	 * Add p0 to the global zone
1761	 */
1762	zone0.zone_zsched = &p0;
1763	p0.p_zone = &zone0;
1764}
1765
1766/*
1767 * Compute a hash value based on the contents of the label and the DOI.  The
1768 * hash algorithm is somewhat arbitrary, but is based on the observation that
1769 * humans will likely pick labels that differ by amounts that work out to be
1770 * multiples of the number of hash chains, and thus stirring in some primes
1771 * should help.
1772 */
1773static uint_t
1774hash_bylabel(void *hdata, mod_hash_key_t key)
1775{
1776	const ts_label_t *lab = (ts_label_t *)key;
1777	const uint32_t *up, *ue;
1778	uint_t hash;
1779	int i;
1780
1781	_NOTE(ARGUNUSED(hdata));
1782
1783	hash = lab->tsl_doi + (lab->tsl_doi << 1);
1784	/* we depend on alignment of label, but not representation */
1785	up = (const uint32_t *)&lab->tsl_label;
1786	ue = up + sizeof (lab->tsl_label) / sizeof (*up);
1787	i = 1;
1788	while (up < ue) {
1789		/* using 2^n + 1, 1 <= n <= 16 as source of many primes */
1790		hash += *up + (*up << ((i % 16) + 1));
1791		up++;
1792		i++;
1793	}
1794	return (hash);
1795}
1796
1797/*
1798 * All that mod_hash cares about here is zero (equal) versus non-zero (not
1799 * equal).  This may need to be changed if less than / greater than is ever
1800 * needed.
1801 */
1802static int
1803hash_labelkey_cmp(mod_hash_key_t key1, mod_hash_key_t key2)
1804{
1805	ts_label_t *lab1 = (ts_label_t *)key1;
1806	ts_label_t *lab2 = (ts_label_t *)key2;
1807
1808	return (label_equal(lab1, lab2) ? 0 : 1);
1809}
1810
1811/*
1812 * Called by main() to initialize the zones framework.
1813 */
1814void
1815zone_init(void)
1816{
1817	rctl_dict_entry_t *rde;
1818	rctl_val_t *dval;
1819	rctl_set_t *set;
1820	rctl_alloc_gp_t *gp;
1821	rctl_entity_p_t e;
1822	int res;
1823
1824	ASSERT(curproc == &p0);
1825
1826	/*
1827	 * Create ID space for zone IDs.  ID 0 is reserved for the
1828	 * global zone.
1829	 */
1830	zoneid_space = id_space_create("zoneid_space", 1, MAX_ZONEID);
1831
1832	/*
1833	 * Initialize generic zone resource controls, if any.
1834	 */
1835	rc_zone_cpu_shares = rctl_register("zone.cpu-shares",
1836	    RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_NEVER |
1837	    RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT | RCTL_GLOBAL_SYSLOG_NEVER,
1838	    FSS_MAXSHARES, FSS_MAXSHARES, &zone_cpu_shares_ops);
1839
1840	rc_zone_cpu_cap = rctl_register("zone.cpu-cap",
1841	    RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_ALWAYS |
1842	    RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |RCTL_GLOBAL_SYSLOG_NEVER |
1843	    RCTL_GLOBAL_INFINITE,
1844	    MAXCAP, MAXCAP, &zone_cpu_cap_ops);
1845
1846	rc_zone_nlwps = rctl_register("zone.max-lwps", RCENTITY_ZONE,
1847	    RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
1848	    INT_MAX, INT_MAX, &zone_lwps_ops);
1849	/*
1850	 * System V IPC resource controls
1851	 */
1852	rc_zone_msgmni = rctl_register("zone.max-msg-ids",
1853	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
1854	    RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_msgmni_ops);
1855
1856	rc_zone_semmni = rctl_register("zone.max-sem-ids",
1857	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
1858	    RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_semmni_ops);
1859
1860	rc_zone_shmmni = rctl_register("zone.max-shm-ids",
1861	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
1862	    RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_shmmni_ops);
1863
1864	rc_zone_shmmax = rctl_register("zone.max-shm-memory",
1865	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
1866	    RCTL_GLOBAL_BYTES, UINT64_MAX, UINT64_MAX, &zone_shmmax_ops);
1867
1868	/*
1869	 * Create a rctl_val with PRIVILEGED, NOACTION, value = 1.  Then attach
1870	 * this at the head of the rctl_dict_entry for ``zone.cpu-shares''.
1871	 */
1872	dval = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
1873	bzero(dval, sizeof (rctl_val_t));
1874	dval->rcv_value = 1;
1875	dval->rcv_privilege = RCPRIV_PRIVILEGED;
1876	dval->rcv_flagaction = RCTL_LOCAL_NOACTION;
1877	dval->rcv_action_recip_pid = -1;
1878
1879	rde = rctl_dict_lookup("zone.cpu-shares");
1880	(void) rctl_val_list_insert(&rde->rcd_default_value, dval);
1881
1882	rc_zone_locked_mem = rctl_register("zone.max-locked-memory",
1883	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
1884	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
1885	    &zone_locked_mem_ops);
1886
1887	rc_zone_max_swap = rctl_register("zone.max-swap",
1888	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
1889	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
1890	    &zone_max_swap_ops);
1891
1892	/*
1893	 * Initialize the ``global zone''.
1894	 */
1895	set = rctl_set_create();
1896	gp = rctl_set_init_prealloc(RCENTITY_ZONE);
1897	mutex_enter(&p0.p_lock);
1898	e.rcep_p.zone = &zone0;
1899	e.rcep_t = RCENTITY_ZONE;
1900	zone0.zone_rctls = rctl_set_init(RCENTITY_ZONE, &p0, &e, set,
1901	    gp);
1902
1903	zone0.zone_nlwps = p0.p_lwpcnt;
1904	zone0.zone_ntasks = 1;
1905	mutex_exit(&p0.p_lock);
1906	zone0.zone_restart_init = B_TRUE;
1907	zone0.zone_brand = &native_brand;
1908	rctl_prealloc_destroy(gp);
1909	/*
1910	 * pool_default hasn't been initialized yet, so we let pool_init()
1911	 * take care of making sure the global zone is in the default pool.
1912	 */
1913
1914	/*
1915	 * Initialize global zone kstats
1916	 */
1917	zone_kstat_create(&zone0);
1918
1919	/*
1920	 * Initialize zone label.
1921	 * mlp are initialized when tnzonecfg is loaded.
1922	 */
1923	zone0.zone_slabel = l_admin_low;
1924	rw_init(&zone0.zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
1925	label_hold(l_admin_low);
1926
1927	mutex_enter(&zonehash_lock);
1928	zone_uniqid(&zone0);
1929	ASSERT(zone0.zone_uniqid == GLOBAL_ZONEUNIQID);
1930
1931	zonehashbyid = mod_hash_create_idhash("zone_by_id", zone_hash_size,
1932	    mod_hash_null_valdtor);
1933	zonehashbyname = mod_hash_create_strhash("zone_by_name",
1934	    zone_hash_size, mod_hash_null_valdtor);
1935	/*
1936	 * maintain zonehashbylabel only for labeled systems
1937	 */
1938	if (is_system_labeled())
1939		zonehashbylabel = mod_hash_create_extended("zone_by_label",
1940		    zone_hash_size, mod_hash_null_keydtor,
1941		    mod_hash_null_valdtor, hash_bylabel, NULL,
1942		    hash_labelkey_cmp, KM_SLEEP);
1943	zonecount = 1;
1944
1945	(void) mod_hash_insert(zonehashbyid, (mod_hash_key_t)GLOBAL_ZONEID,
1946	    (mod_hash_val_t)&zone0);
1947	(void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)zone0.zone_name,
1948	    (mod_hash_val_t)&zone0);
1949	if (is_system_labeled()) {
1950		zone0.zone_flags |= ZF_HASHED_LABEL;
1951		(void) mod_hash_insert(zonehashbylabel,
1952		    (mod_hash_key_t)zone0.zone_slabel, (mod_hash_val_t)&zone0);
1953	}
1954	mutex_exit(&zonehash_lock);
1955
1956	/*
1957	 * We avoid setting zone_kcred until now, since kcred is initialized
1958	 * sometime after zone_zsd_init() and before zone_init().
1959	 */
1960	zone0.zone_kcred = kcred;
1961	/*
1962	 * The global zone is fully initialized (except for zone_rootvp which
1963	 * will be set when the root filesystem is mounted).
1964	 */
1965	global_zone = &zone0;
1966
1967	/*
1968	 * Setup an event channel to send zone status change notifications on
1969	 */
1970	res = sysevent_evc_bind(ZONE_EVENT_CHANNEL, &zone_event_chan,
1971	    EVCH_CREAT);
1972
1973	if (res)
1974		panic("Sysevent_evc_bind failed during zone setup.\n");
1975
1976}
1977
1978static void
1979zone_free(zone_t *zone)
1980{
1981	ASSERT(zone != global_zone);
1982	ASSERT(zone->zone_ntasks == 0);
1983	ASSERT(zone->zone_nlwps == 0);
1984	ASSERT(zone->zone_cred_ref == 0);
1985	ASSERT(zone->zone_kcred == NULL);
1986	ASSERT(zone_status_get(zone) == ZONE_IS_DEAD ||
1987	    zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
1988
1989	/*
1990	 * Remove any zone caps.
1991	 */
1992	cpucaps_zone_remove(zone);
1993
1994	ASSERT(zone->zone_cpucap == NULL);
1995
1996	/* remove from deathrow list */
1997	if (zone_status_get(zone) == ZONE_IS_DEAD) {
1998		ASSERT(zone->zone_ref == 0);
1999		mutex_enter(&zone_deathrow_lock);
2000		list_remove(&zone_deathrow, zone);
2001		mutex_exit(&zone_deathrow_lock);
2002	}
2003
2004	zone_free_zsd(zone);
2005	zone_free_datasets(zone);
2006
2007	if (zone->zone_rootvp != NULL)
2008		VN_RELE(zone->zone_rootvp);
2009	if (zone->zone_rootpath)
2010		kmem_free(zone->zone_rootpath, zone->zone_rootpathlen);
2011	if (zone->zone_name != NULL)
2012		kmem_free(zone->zone_name, ZONENAME_MAX);
2013	if (zone->zone_slabel != NULL)
2014		label_rele(zone->zone_slabel);
2015	if (zone->zone_nodename != NULL)
2016		kmem_free(zone->zone_nodename, _SYS_NMLN);
2017	if (zone->zone_domain != NULL)
2018		kmem_free(zone->zone_domain, _SYS_NMLN);
2019	if (zone->zone_privset != NULL)
2020		kmem_free(zone->zone_privset, sizeof (priv_set_t));
2021	if (zone->zone_rctls != NULL)
2022		rctl_set_free(zone->zone_rctls);
2023	if (zone->zone_bootargs != NULL)
2024		kmem_free(zone->zone_bootargs, strlen(zone->zone_bootargs) + 1);
2025	if (zone->zone_initname != NULL)
2026		kmem_free(zone->zone_initname, strlen(zone->zone_initname) + 1);
2027	id_free(zoneid_space, zone->zone_id);
2028	mutex_destroy(&zone->zone_lock);
2029	cv_destroy(&zone->zone_cv);
2030	rw_destroy(&zone->zone_mlps.mlpl_rwlock);
2031	kmem_free(zone, sizeof (zone_t));
2032}
2033
2034/*
2035 * See block comment at the top of this file for information about zone
2036 * status values.
2037 */
2038/*
2039 * Convenience function for setting zone status.
2040 */
2041static void
2042zone_status_set(zone_t *zone, zone_status_t status)
2043{
2044
2045	nvlist_t *nvl = NULL;
2046	ASSERT(MUTEX_HELD(&zone_status_lock));
2047	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE &&
2048	    status >= zone_status_get(zone));
2049
2050	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) ||
2051	    nvlist_add_string(nvl, ZONE_CB_NAME, zone->zone_name) ||
2052	    nvlist_add_string(nvl, ZONE_CB_NEWSTATE,
2053	    zone_status_table[status]) ||
2054	    nvlist_add_string(nvl, ZONE_CB_OLDSTATE,
2055	    zone_status_table[zone->zone_status]) ||
2056	    nvlist_add_int32(nvl, ZONE_CB_ZONEID, zone->zone_id) ||
2057	    nvlist_add_uint64(nvl, ZONE_CB_TIMESTAMP, (uint64_t)gethrtime()) ||
2058	    sysevent_evc_publish(zone_event_chan, ZONE_EVENT_STATUS_CLASS,
2059	    ZONE_EVENT_STATUS_SUBCLASS, "sun.com", "kernel", nvl, EVCH_SLEEP)) {
2060#ifdef DEBUG
2061		(void) printf(
2062		    "Failed to allocate and send zone state change event.\n");
2063#endif
2064	}
2065	nvlist_free(nvl);
2066
2067	zone->zone_status = status;
2068
2069	cv_broadcast(&zone->zone_cv);
2070}
2071
2072/*
2073 * Public function to retrieve the zone status.  The zone status may
2074 * change after it is retrieved.
2075 */
2076zone_status_t
2077zone_status_get(zone_t *zone)
2078{
2079	return (zone->zone_status);
2080}
2081
2082static int
2083zone_set_bootargs(zone_t *zone, const char *zone_bootargs)
2084{
2085	char *bootargs = kmem_zalloc(BOOTARGS_MAX, KM_SLEEP);
2086	int err = 0;
2087
2088	ASSERT(zone != global_zone);
2089	if ((err = copyinstr(zone_bootargs, bootargs, BOOTARGS_MAX, NULL)) != 0)
2090		goto done;	/* EFAULT or ENAMETOOLONG */
2091
2092	if (zone->zone_bootargs != NULL)
2093		kmem_free(zone->zone_bootargs, strlen(zone->zone_bootargs) + 1);
2094
2095	zone->zone_bootargs = kmem_alloc(strlen(bootargs) + 1, KM_SLEEP);
2096	(void) strcpy(zone->zone_bootargs, bootargs);
2097
2098done:
2099	kmem_free(bootargs, BOOTARGS_MAX);
2100	return (err);
2101}
2102
2103static int
2104zone_set_brand(zone_t *zone, const char *brand)
2105{
2106	struct brand_attr *attrp;
2107	brand_t *bp;
2108
2109	attrp = kmem_alloc(sizeof (struct brand_attr), KM_SLEEP);
2110	if (copyin(brand, attrp, sizeof (struct brand_attr)) != 0) {
2111		kmem_free(attrp, sizeof (struct brand_attr));
2112		return (EFAULT);
2113	}
2114
2115	bp = brand_register_zone(attrp);
2116	kmem_free(attrp, sizeof (struct brand_attr));
2117	if (bp == NULL)
2118		return (EINVAL);
2119
2120	/*
2121	 * This is the only place where a zone can change it's brand.
2122	 * We already need to hold zone_status_lock to check the zone
2123	 * status, so we'll just use that lock to serialize zone
2124	 * branding requests as well.
2125	 */
2126	mutex_enter(&zone_status_lock);
2127
2128	/* Re-Branding is not allowed and the zone can't be booted yet */
2129	if ((ZONE_IS_BRANDED(zone)) ||
2130	    (zone_status_get(zone) >= ZONE_IS_BOOTING)) {
2131		mutex_exit(&zone_status_lock);
2132		brand_unregister_zone(bp);
2133		return (EINVAL);
2134	}
2135
2136	if (is_system_labeled() &&
2137	    strncmp(attrp->ba_brandname, NATIVE_BRAND_NAME, MAXNAMELEN) != 0) {
2138		mutex_exit(&zone_status_lock);
2139		brand_unregister_zone(bp);
2140		return (EPERM);
2141	}
2142
2143	/* set up the brand specific data */
2144	zone->zone_brand = bp;
2145	ZBROP(zone)->b_init_brand_data(zone);
2146
2147	mutex_exit(&zone_status_lock);
2148	return (0);
2149}
2150
2151static int
2152zone_set_initname(zone_t *zone, const char *zone_initname)
2153{
2154	char initname[INITNAME_SZ];
2155	size_t len;
2156	int err = 0;
2157
2158	ASSERT(zone != global_zone);
2159	if ((err = copyinstr(zone_initname, initname, INITNAME_SZ, &len)) != 0)
2160		return (err);	/* EFAULT or ENAMETOOLONG */
2161
2162	if (zone->zone_initname != NULL)
2163		kmem_free(zone->zone_initname, strlen(zone->zone_initname) + 1);
2164
2165	zone->zone_initname = kmem_alloc(strlen(initname) + 1, KM_SLEEP);
2166	(void) strcpy(zone->zone_initname, initname);
2167	return (0);
2168}
2169
2170static int
2171zone_set_phys_mcap(zone_t *zone, const uint64_t *zone_mcap)
2172{
2173	uint64_t mcap;
2174	int err = 0;
2175
2176	if ((err = copyin(zone_mcap, &mcap, sizeof (uint64_t))) == 0)
2177		zone->zone_phys_mcap = mcap;
2178
2179	return (err);
2180}
2181
2182static int
2183zone_set_sched_class(zone_t *zone, const char *new_class)
2184{
2185	char sched_class[PC_CLNMSZ];
2186	id_t classid;
2187	int err;
2188
2189	ASSERT(zone != global_zone);
2190	if ((err = copyinstr(new_class, sched_class, PC_CLNMSZ, NULL)) != 0)
2191		return (err);	/* EFAULT or ENAMETOOLONG */
2192
2193	if (getcid(sched_class, &classid) != 0 || classid == syscid)
2194		return (set_errno(EINVAL));
2195	zone->zone_defaultcid = classid;
2196	ASSERT(zone->zone_defaultcid > 0 &&
2197	    zone->zone_defaultcid < loaded_classes);
2198
2199	return (0);
2200}
2201
2202/*
2203 * Block indefinitely waiting for (zone_status >= status)
2204 */
2205void
2206zone_status_wait(zone_t *zone, zone_status_t status)
2207{
2208	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2209
2210	mutex_enter(&zone_status_lock);
2211	while (zone->zone_status < status) {
2212		cv_wait(&zone->zone_cv, &zone_status_lock);
2213	}
2214	mutex_exit(&zone_status_lock);
2215}
2216
2217/*
2218 * Private CPR-safe version of zone_status_wait().
2219 */
2220static void
2221zone_status_wait_cpr(zone_t *zone, zone_status_t status, char *str)
2222{
2223	callb_cpr_t cprinfo;
2224
2225	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2226
2227	CALLB_CPR_INIT(&cprinfo, &zone_status_lock, callb_generic_cpr,
2228	    str);
2229	mutex_enter(&zone_status_lock);
2230	while (zone->zone_status < status) {
2231		CALLB_CPR_SAFE_BEGIN(&cprinfo);
2232		cv_wait(&zone->zone_cv, &zone_status_lock);
2233		CALLB_CPR_SAFE_END(&cprinfo, &zone_status_lock);
2234	}
2235	/*
2236	 * zone_status_lock is implicitly released by the following.
2237	 */
2238	CALLB_CPR_EXIT(&cprinfo);
2239}
2240
2241/*
2242 * Block until zone enters requested state or signal is received.  Return (0)
2243 * if signaled, non-zero otherwise.
2244 */
2245int
2246zone_status_wait_sig(zone_t *zone, zone_status_t status)
2247{
2248	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2249
2250	mutex_enter(&zone_status_lock);
2251	while (zone->zone_status < status) {
2252		if (!cv_wait_sig(&zone->zone_cv, &zone_status_lock)) {
2253			mutex_exit(&zone_status_lock);
2254			return (0);
2255		}
2256	}
2257	mutex_exit(&zone_status_lock);
2258	return (1);
2259}
2260
2261/*
2262 * Block until the zone enters the requested state or the timeout expires,
2263 * whichever happens first.  Return (-1) if operation timed out, time remaining
2264 * otherwise.
2265 */
2266clock_t
2267zone_status_timedwait(zone_t *zone, clock_t tim, zone_status_t status)
2268{
2269	clock_t timeleft = 0;
2270
2271	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2272
2273	mutex_enter(&zone_status_lock);
2274	while (zone->zone_status < status && timeleft != -1) {
2275		timeleft = cv_timedwait(&zone->zone_cv, &zone_status_lock, tim);
2276	}
2277	mutex_exit(&zone_status_lock);
2278	return (timeleft);
2279}
2280
2281/*
2282 * Block until the zone enters the requested state, the current process is
2283 * signaled,  or the timeout expires, whichever happens first.  Return (-1) if
2284 * operation timed out, 0 if signaled, time remaining otherwise.
2285 */
2286clock_t
2287zone_status_timedwait_sig(zone_t *zone, clock_t tim, zone_status_t status)
2288{
2289	clock_t timeleft = tim - lbolt;
2290
2291	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2292
2293	mutex_enter(&zone_status_lock);
2294	while (zone->zone_status < status) {
2295		timeleft = cv_timedwait_sig(&zone->zone_cv, &zone_status_lock,
2296		    tim);
2297		if (timeleft <= 0)
2298			break;
2299	}
2300	mutex_exit(&zone_status_lock);
2301	return (timeleft);
2302}
2303
2304/*
2305 * Zones have two reference counts: one for references from credential
2306 * structures (zone_cred_ref), and one (zone_ref) for everything else.
2307 * This is so we can allow a zone to be rebooted while there are still
2308 * outstanding cred references, since certain drivers cache dblks (which
2309 * implicitly results in cached creds).  We wait for zone_ref to drop to
2310 * 0 (actually 1), but not zone_cred_ref.  The zone structure itself is
2311 * later freed when the zone_cred_ref drops to 0, though nothing other
2312 * than the zone id and privilege set should be accessed once the zone
2313 * is "dead".
2314 *
2315 * A debugging flag, zone_wait_for_cred, can be set to a non-zero value
2316 * to force halt/reboot to block waiting for the zone_cred_ref to drop
2317 * to 0.  This can be useful to flush out other sources of cached creds
2318 * that may be less innocuous than the driver case.
2319 */
2320
2321int zone_wait_for_cred = 0;
2322
2323static void
2324zone_hold_locked(zone_t *z)
2325{
2326	ASSERT(MUTEX_HELD(&z->zone_lock));
2327	z->zone_ref++;
2328	ASSERT(z->zone_ref != 0);
2329}
2330
2331void
2332zone_hold(zone_t *z)
2333{
2334	mutex_enter(&z->zone_lock);
2335	zone_hold_locked(z);
2336	mutex_exit(&z->zone_lock);
2337}
2338
2339/*
2340 * If the non-cred ref count drops to 1 and either the cred ref count
2341 * is 0 or we aren't waiting for cred references, the zone is ready to
2342 * be destroyed.
2343 */
2344#define	ZONE_IS_UNREF(zone)	((zone)->zone_ref == 1 && \
2345	    (!zone_wait_for_cred || (zone)->zone_cred_ref == 0))
2346
2347void
2348zone_rele(zone_t *z)
2349{
2350	boolean_t wakeup;
2351
2352	mutex_enter(&z->zone_lock);
2353	ASSERT(z->zone_ref != 0);
2354	z->zone_ref--;
2355	if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
2356		/* no more refs, free the structure */
2357		mutex_exit(&z->zone_lock);
2358		zone_free(z);
2359		return;
2360	}
2361	/* signal zone_destroy so the zone can finish halting */
2362	wakeup = (ZONE_IS_UNREF(z) && zone_status_get(z) >= ZONE_IS_DEAD);
2363	mutex_exit(&z->zone_lock);
2364
2365	if (wakeup) {
2366		/*
2367		 * Grabbing zonehash_lock here effectively synchronizes with
2368		 * zone_destroy() to avoid missed signals.
2369		 */
2370		mutex_enter(&zonehash_lock);
2371		cv_broadcast(&zone_destroy_cv);
2372		mutex_exit(&zonehash_lock);
2373	}
2374}
2375
2376void
2377zone_cred_hold(zone_t *z)
2378{
2379	mutex_enter(&z->zone_lock);
2380	z->zone_cred_ref++;
2381	ASSERT(z->zone_cred_ref != 0);
2382	mutex_exit(&z->zone_lock);
2383}
2384
2385void
2386zone_cred_rele(zone_t *z)
2387{
2388	boolean_t wakeup;
2389
2390	mutex_enter(&z->zone_lock);
2391	ASSERT(z->zone_cred_ref != 0);
2392	z->zone_cred_ref--;
2393	if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
2394		/* no more refs, free the structure */
2395		mutex_exit(&z->zone_lock);
2396		zone_free(z);
2397		return;
2398	}
2399	/*
2400	 * If zone_destroy is waiting for the cred references to drain
2401	 * out, and they have, signal it.
2402	 */
2403	wakeup = (zone_wait_for_cred && ZONE_IS_UNREF(z) &&
2404	    zone_status_get(z) >= ZONE_IS_DEAD);
2405	mutex_exit(&z->zone_lock);
2406
2407	if (wakeup) {
2408		/*
2409		 * Grabbing zonehash_lock here effectively synchronizes with
2410		 * zone_destroy() to avoid missed signals.
2411		 */
2412		mutex_enter(&zonehash_lock);
2413		cv_broadcast(&zone_destroy_cv);
2414		mutex_exit(&zonehash_lock);
2415	}
2416}
2417
2418void
2419zone_task_hold(zone_t *z)
2420{
2421	mutex_enter(&z->zone_lock);
2422	z->zone_ntasks++;
2423	ASSERT(z->zone_ntasks != 0);
2424	mutex_exit(&z->zone_lock);
2425}
2426
2427void
2428zone_task_rele(zone_t *zone)
2429{
2430	uint_t refcnt;
2431
2432	mutex_enter(&zone->zone_lock);
2433	ASSERT(zone->zone_ntasks != 0);
2434	refcnt = --zone->zone_ntasks;
2435	if (refcnt > 1)	{	/* Common case */
2436		mutex_exit(&zone->zone_lock);
2437		return;
2438	}
2439	zone_hold_locked(zone);	/* so we can use the zone_t later */
2440	mutex_exit(&zone->zone_lock);
2441	if (refcnt == 1) {
2442		/*
2443		 * See if the zone is shutting down.
2444		 */
2445		mutex_enter(&zone_status_lock);
2446		if (zone_status_get(zone) != ZONE_IS_SHUTTING_DOWN) {
2447			goto out;
2448		}
2449
2450		/*
2451		 * Make sure the ntasks didn't change since we
2452		 * dropped zone_lock.
2453		 */
2454		mutex_enter(&zone->zone_lock);
2455		if (refcnt != zone->zone_ntasks) {
2456			mutex_exit(&zone->zone_lock);
2457			goto out;
2458		}
2459		mutex_exit(&zone->zone_lock);
2460
2461		/*
2462		 * No more user processes in the zone.  The zone is empty.
2463		 */
2464		zone_status_set(zone, ZONE_IS_EMPTY);
2465		goto out;
2466	}
2467
2468	ASSERT(refcnt == 0);
2469	/*
2470	 * zsched has exited; the zone is dead.
2471	 */
2472	zone->zone_zsched = NULL;		/* paranoia */
2473	mutex_enter(&zone_status_lock);
2474	zone_status_set(zone, ZONE_IS_DEAD);
2475out:
2476	mutex_exit(&zone_status_lock);
2477	zone_rele(zone);
2478}
2479
2480zoneid_t
2481getzoneid(void)
2482{
2483	return (curproc->p_zone->zone_id);
2484}
2485
2486/*
2487 * Internal versions of zone_find_by_*().  These don't zone_hold() or
2488 * check the validity of a zone's state.
2489 */
2490static zone_t *
2491zone_find_all_by_id(zoneid_t zoneid)
2492{
2493	mod_hash_val_t hv;
2494	zone_t *zone = NULL;
2495
2496	ASSERT(MUTEX_HELD(&zonehash_lock));
2497
2498	if (mod_hash_find(zonehashbyid,
2499	    (mod_hash_key_t)(uintptr_t)zoneid, &hv) == 0)
2500		zone = (zone_t *)hv;
2501	return (zone);
2502}
2503
2504static zone_t *
2505zone_find_all_by_label(const ts_label_t *label)
2506{
2507	mod_hash_val_t hv;
2508	zone_t *zone = NULL;
2509
2510	ASSERT(MUTEX_HELD(&zonehash_lock));
2511
2512	/*
2513	 * zonehashbylabel is not maintained for unlabeled systems
2514	 */
2515	if (!is_system_labeled())
2516		return (NULL);
2517	if (mod_hash_find(zonehashbylabel, (mod_hash_key_t)label, &hv) == 0)
2518		zone = (zone_t *)hv;
2519	return (zone);
2520}
2521
2522static zone_t *
2523zone_find_all_by_name(char *name)
2524{
2525	mod_hash_val_t hv;
2526	zone_t *zone = NULL;
2527
2528	ASSERT(MUTEX_HELD(&zonehash_lock));
2529
2530	if (mod_hash_find(zonehashbyname, (mod_hash_key_t)name, &hv) == 0)
2531		zone = (zone_t *)hv;
2532	return (zone);
2533}
2534
2535/*
2536 * Public interface for looking up a zone by zoneid.  Only returns the zone if
2537 * it is fully initialized, and has not yet begun the zone_destroy() sequence.
2538 * Caller must call zone_rele() once it is done with the zone.
2539 *
2540 * The zone may begin the zone_destroy() sequence immediately after this
2541 * function returns, but may be safely used until zone_rele() is called.
2542 */
2543zone_t *
2544zone_find_by_id(zoneid_t zoneid)
2545{
2546	zone_t *zone;
2547	zone_status_t status;
2548
2549	mutex_enter(&zonehash_lock);
2550	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
2551		mutex_exit(&zonehash_lock);
2552		return (NULL);
2553	}
2554	status = zone_status_get(zone);
2555	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
2556		/*
2557		 * For all practical purposes the zone doesn't exist.
2558		 */
2559		mutex_exit(&zonehash_lock);
2560		return (NULL);
2561	}
2562	zone_hold(zone);
2563	mutex_exit(&zonehash_lock);
2564	return (zone);
2565}
2566
2567/*
2568 * Similar to zone_find_by_id, but using zone label as the key.
2569 */
2570zone_t *
2571zone_find_by_label(const ts_label_t *label)
2572{
2573	zone_t *zone;
2574	zone_status_t status;
2575
2576	mutex_enter(&zonehash_lock);
2577	if ((zone = zone_find_all_by_label(label)) == NULL) {
2578		mutex_exit(&zonehash_lock);
2579		return (NULL);
2580	}
2581
2582	status = zone_status_get(zone);
2583	if (status > ZONE_IS_DOWN) {
2584		/*
2585		 * For all practical purposes the zone doesn't exist.
2586		 */
2587		mutex_exit(&zonehash_lock);
2588		return (NULL);
2589	}
2590	zone_hold(zone);
2591	mutex_exit(&zonehash_lock);
2592	return (zone);
2593}
2594
2595/*
2596 * Similar to zone_find_by_id, but using zone name as the key.
2597 */
2598zone_t *
2599zone_find_by_name(char *name)
2600{
2601	zone_t *zone;
2602	zone_status_t status;
2603
2604	mutex_enter(&zonehash_lock);
2605	if ((zone = zone_find_all_by_name(name)) == NULL) {
2606		mutex_exit(&zonehash_lock);
2607		return (NULL);
2608	}
2609	status = zone_status_get(zone);
2610	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
2611		/*
2612		 * For all practical purposes the zone doesn't exist.
2613		 */
2614		mutex_exit(&zonehash_lock);
2615		return (NULL);
2616	}
2617	zone_hold(zone);
2618	mutex_exit(&zonehash_lock);
2619	return (zone);
2620}
2621
2622/*
2623 * Similar to zone_find_by_id(), using the path as a key.  For instance,
2624 * if there is a zone "foo" rooted at /foo/root, and the path argument
2625 * is "/foo/root/proc", it will return the held zone_t corresponding to
2626 * zone "foo".
2627 *
2628 * zone_find_by_path() always returns a non-NULL value, since at the
2629 * very least every path will be contained in the global zone.
2630 *
2631 * As with the other zone_find_by_*() functions, the caller is
2632 * responsible for zone_rele()ing the return value of this function.
2633 */
2634zone_t *
2635zone_find_by_path(const char *path)
2636{
2637	zone_t *zone;
2638	zone_t *zret = NULL;
2639	zone_status_t status;
2640
2641	if (path == NULL) {
2642		/*
2643		 * Call from rootconf().
2644		 */
2645		zone_hold(global_zone);
2646		return (global_zone);
2647	}
2648	ASSERT(*path == '/');
2649	mutex_enter(&zonehash_lock);
2650	for (zone = list_head(&zone_active); zone != NULL;
2651	    zone = list_next(&zone_active, zone)) {
2652		if (ZONE_PATH_VISIBLE(path, zone))
2653			zret = zone;
2654	}
2655	ASSERT(zret != NULL);
2656	status = zone_status_get(zret);
2657	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
2658		/*
2659		 * Zone practically doesn't exist.
2660		 */
2661		zret = global_zone;
2662	}
2663	zone_hold(zret);
2664	mutex_exit(&zonehash_lock);
2665	return (zret);
2666}
2667
2668/*
2669 * Get the number of cpus visible to this zone.  The system-wide global
2670 * 'ncpus' is returned if pools are disabled, the caller is in the
2671 * global zone, or a NULL zone argument is passed in.
2672 */
2673int
2674zone_ncpus_get(zone_t *zone)
2675{
2676	int myncpus = zone == NULL ? 0 : zone->zone_ncpus;
2677
2678	return (myncpus != 0 ? myncpus : ncpus);
2679}
2680
2681/*
2682 * Get the number of online cpus visible to this zone.  The system-wide
2683 * global 'ncpus_online' is returned if pools are disabled, the caller
2684 * is in the global zone, or a NULL zone argument is passed in.
2685 */
2686int
2687zone_ncpus_online_get(zone_t *zone)
2688{
2689	int myncpus_online = zone == NULL ? 0 : zone->zone_ncpus_online;
2690
2691	return (myncpus_online != 0 ? myncpus_online : ncpus_online);
2692}
2693
2694/*
2695 * Return the pool to which the zone is currently bound.
2696 */
2697pool_t *
2698zone_pool_get(zone_t *zone)
2699{
2700	ASSERT(pool_lock_held());
2701
2702	return (zone->zone_pool);
2703}
2704
2705/*
2706 * Set the zone's pool pointer and update the zone's visibility to match
2707 * the resources in the new pool.
2708 */
2709void
2710zone_pool_set(zone_t *zone, pool_t *pool)
2711{
2712	ASSERT(pool_lock_held());
2713	ASSERT(MUTEX_HELD(&cpu_lock));
2714
2715	zone->zone_pool = pool;
2716	zone_pset_set(zone, pool->pool_pset->pset_id);
2717}
2718
2719/*
2720 * Return the cached value of the id of the processor set to which the
2721 * zone is currently bound.  The value will be ZONE_PS_INVAL if the pools
2722 * facility is disabled.
2723 */
2724psetid_t
2725zone_pset_get(zone_t *zone)
2726{
2727	ASSERT(MUTEX_HELD(&cpu_lock));
2728
2729	return (zone->zone_psetid);
2730}
2731
2732/*
2733 * Set the cached value of the id of the processor set to which the zone
2734 * is currently bound.  Also update the zone's visibility to match the
2735 * resources in the new processor set.
2736 */
2737void
2738zone_pset_set(zone_t *zone, psetid_t newpsetid)
2739{
2740	psetid_t oldpsetid;
2741
2742	ASSERT(MUTEX_HELD(&cpu_lock));
2743	oldpsetid = zone_pset_get(zone);
2744
2745	if (oldpsetid == newpsetid)
2746		return;
2747	/*
2748	 * Global zone sees all.
2749	 */
2750	if (zone != global_zone) {
2751		zone->zone_psetid = newpsetid;
2752		if (newpsetid != ZONE_PS_INVAL)
2753			pool_pset_visibility_add(newpsetid, zone);
2754		if (oldpsetid != ZONE_PS_INVAL)
2755			pool_pset_visibility_remove(oldpsetid, zone);
2756	}
2757	/*
2758	 * Disabling pools, so we should start using the global values
2759	 * for ncpus and ncpus_online.
2760	 */
2761	if (newpsetid == ZONE_PS_INVAL) {
2762		zone->zone_ncpus = 0;
2763		zone->zone_ncpus_online = 0;
2764	}
2765}
2766
2767/*
2768 * Walk the list of active zones and issue the provided callback for
2769 * each of them.
2770 *
2771 * Caller must not be holding any locks that may be acquired under
2772 * zonehash_lock.  See comment at the beginning of the file for a list of
2773 * common locks and their interactions with zones.
2774 */
2775int
2776zone_walk(int (*cb)(zone_t *, void *), void *data)
2777{
2778	zone_t *zone;
2779	int ret = 0;
2780	zone_status_t status;
2781
2782	mutex_enter(&zonehash_lock);
2783	for (zone = list_head(&zone_active); zone != NULL;
2784	    zone = list_next(&zone_active, zone)) {
2785		/*
2786		 * Skip zones that shouldn't be externally visible.
2787		 */
2788		status = zone_status_get(zone);
2789		if (status < ZONE_IS_READY || status > ZONE_IS_DOWN)
2790			continue;
2791		/*
2792		 * Bail immediately if any callback invocation returns a
2793		 * non-zero value.
2794		 */
2795		ret = (*cb)(zone, data);
2796		if (ret != 0)
2797			break;
2798	}
2799	mutex_exit(&zonehash_lock);
2800	return (ret);
2801}
2802
2803static int
2804zone_set_root(zone_t *zone, const char *upath)
2805{
2806	vnode_t *vp;
2807	int trycount;
2808	int error = 0;
2809	char *path;
2810	struct pathname upn, pn;
2811	size_t pathlen;
2812
2813	if ((error = pn_get((char *)upath, UIO_USERSPACE, &upn)) != 0)
2814		return (error);
2815
2816	pn_alloc(&pn);
2817
2818	/* prevent infinite loop */
2819	trycount = 10;
2820	for (;;) {
2821		if (--trycount <= 0) {
2822			error = ESTALE;
2823			goto out;
2824		}
2825
2826		if ((error = lookuppn(&upn, &pn, FOLLOW, NULLVPP, &vp)) == 0) {
2827			/*
2828			 * VOP_ACCESS() may cover 'vp' with a new
2829			 * filesystem, if 'vp' is an autoFS vnode.
2830			 * Get the new 'vp' if so.
2831			 */
2832			if ((error =
2833			    VOP_ACCESS(vp, VEXEC, 0, CRED(), NULL)) == 0 &&
2834			    (!vn_ismntpt(vp) ||
2835			    (error = traverse(&vp)) == 0)) {
2836				pathlen = pn.pn_pathlen + 2;
2837				path = kmem_alloc(pathlen, KM_SLEEP);
2838				(void) strncpy(path, pn.pn_path,
2839				    pn.pn_pathlen + 1);
2840				path[pathlen - 2] = '/';
2841				path[pathlen - 1] = '\0';
2842				pn_free(&pn);
2843				pn_free(&upn);
2844
2845				/* Success! */
2846				break;
2847			}
2848			VN_RELE(vp);
2849		}
2850		if (error != ESTALE)
2851			goto out;
2852	}
2853
2854	ASSERT(error == 0);
2855	zone->zone_rootvp = vp;		/* we hold a reference to vp */
2856	zone->zone_rootpath = path;
2857	zone->zone_rootpathlen = pathlen;
2858	if (pathlen > 5 && strcmp(path + pathlen - 5, "/lu/") == 0)
2859		zone->zone_flags |= ZF_IS_SCRATCH;
2860	return (0);
2861
2862out:
2863	pn_free(&pn);
2864	pn_free(&upn);
2865	return (error);
2866}
2867
2868#define	isalnum(c)	(((c) >= '0' && (c) <= '9') || \
2869			((c) >= 'a' && (c) <= 'z') || \
2870			((c) >= 'A' && (c) <= 'Z'))
2871
2872static int
2873zone_set_name(zone_t *zone, const char *uname)
2874{
2875	char *kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
2876	size_t len;
2877	int i, err;
2878
2879	if ((err = copyinstr(uname, kname, ZONENAME_MAX, &len)) != 0) {
2880		kmem_free(kname, ZONENAME_MAX);
2881		return (err);	/* EFAULT or ENAMETOOLONG */
2882	}
2883
2884	/* must be less than ZONENAME_MAX */
2885	if (len == ZONENAME_MAX && kname[ZONENAME_MAX - 1] != '\0') {
2886		kmem_free(kname, ZONENAME_MAX);
2887		return (EINVAL);
2888	}
2889
2890	/*
2891	 * Name must start with an alphanumeric and must contain only
2892	 * alphanumerics, '-', '_' and '.'.
2893	 */
2894	if (!isalnum(kname[0])) {
2895		kmem_free(kname, ZONENAME_MAX);
2896		return (EINVAL);
2897	}
2898	for (i = 1; i < len - 1; i++) {
2899		if (!isalnum(kname[i]) && kname[i] != '-' && kname[i] != '_' &&
2900		    kname[i] != '.') {
2901			kmem_free(kname, ZONENAME_MAX);
2902			return (EINVAL);
2903		}
2904	}
2905
2906	zone->zone_name = kname;
2907	return (0);
2908}
2909
2910/*
2911 * Similar to thread_create(), but makes sure the thread is in the appropriate
2912 * zone's zsched process (curproc->p_zone->zone_zsched) before returning.
2913 */
2914/*ARGSUSED*/
2915kthread_t *
2916zthread_create(
2917    caddr_t stk,
2918    size_t stksize,
2919    void (*proc)(),
2920    void *arg,
2921    size_t len,
2922    pri_t pri)
2923{
2924	kthread_t *t;
2925	zone_t *zone = curproc->p_zone;
2926	proc_t *pp = zone->zone_zsched;
2927
2928	zone_hold(zone);	/* Reference to be dropped when thread exits */
2929
2930	/*
2931	 * No-one should be trying to create threads if the zone is shutting
2932	 * down and there aren't any kernel threads around.  See comment
2933	 * in zthread_exit().
2934	 */
2935	ASSERT(!(zone->zone_kthreads == NULL &&
2936	    zone_status_get(zone) >= ZONE_IS_EMPTY));
2937	/*
2938	 * Create a thread, but don't let it run until we've finished setting
2939	 * things up.
2940	 */
2941	t = thread_create(stk, stksize, proc, arg, len, pp, TS_STOPPED, pri);
2942	ASSERT(t->t_forw == NULL);
2943	mutex_enter(&zone_status_lock);
2944	if (zone->zone_kthreads == NULL) {
2945		t->t_forw = t->t_back = t;
2946	} else {
2947		kthread_t *tx = zone->zone_kthreads;
2948
2949		t->t_forw = tx;
2950		t->t_back = tx->t_back;
2951		tx->t_back->t_forw = t;
2952		tx->t_back = t;
2953	}
2954	zone->zone_kthreads = t;
2955	mutex_exit(&zone_status_lock);
2956
2957	mutex_enter(&pp->p_lock);
2958	t->t_proc_flag |= TP_ZTHREAD;
2959	project_rele(t->t_proj);
2960	t->t_proj = project_hold(pp->p_task->tk_proj);
2961
2962	/*
2963	 * Setup complete, let it run.
2964	 */
2965	thread_lock(t);
2966	t->t_schedflag |= TS_ALLSTART;
2967	setrun_locked(t);
2968	thread_unlock(t);
2969
2970	mutex_exit(&pp->p_lock);
2971
2972	return (t);
2973}
2974
2975/*
2976 * Similar to thread_exit().  Must be called by threads created via
2977 * zthread_exit().
2978 */
2979void
2980zthread_exit(void)
2981{
2982	kthread_t *t = curthread;
2983	proc_t *pp = curproc;
2984	zone_t *zone = pp->p_zone;
2985
2986	mutex_enter(&zone_status_lock);
2987
2988	/*
2989	 * Reparent to p0
2990	 */
2991	kpreempt_disable();
2992	mutex_enter(&pp->p_lock);
2993	t->t_proc_flag &= ~TP_ZTHREAD;
2994	t->t_procp = &p0;
2995	hat_thread_exit(t);
2996	mutex_exit(&pp->p_lock);
2997	kpreempt_enable();
2998
2999	if (t->t_back == t) {
3000		ASSERT(t->t_forw == t);
3001		/*
3002		 * If the zone is empty, once the thread count
3003		 * goes to zero no further kernel threads can be
3004		 * created.  This is because if the creator is a process
3005		 * in the zone, then it must have exited before the zone
3006		 * state could be set to ZONE_IS_EMPTY.
3007		 * Otherwise, if the creator is a kernel thread in the
3008		 * zone, the thread count is non-zero.
3009		 *
3010		 * This really means that non-zone kernel threads should
3011		 * not create zone kernel threads.
3012		 */
3013		zone->zone_kthreads = NULL;
3014		if (zone_status_get(zone) == ZONE_IS_EMPTY) {
3015			zone_status_set(zone, ZONE_IS_DOWN);
3016			/*
3017			 * Remove any CPU caps on this zone.
3018			 */
3019			cpucaps_zone_remove(zone);
3020		}
3021	} else {
3022		t->t_forw->t_back = t->t_back;
3023		t->t_back->t_forw = t->t_forw;
3024		if (zone->zone_kthreads == t)
3025			zone->zone_kthreads = t->t_forw;
3026	}
3027	mutex_exit(&zone_status_lock);
3028	zone_rele(zone);
3029	thread_exit();
3030	/* NOTREACHED */
3031}
3032
3033static void
3034zone_chdir(vnode_t *vp, vnode_t **vpp, proc_t *pp)
3035{
3036	vnode_t *oldvp;
3037
3038	/* we're going to hold a reference here to the directory */
3039	VN_HOLD(vp);
3040
3041	if (audit_active)	/* update abs cwd/root path see c2audit.c */
3042		audit_chdirec(vp, vpp);
3043
3044	mutex_enter(&pp->p_lock);
3045	oldvp = *vpp;
3046	*vpp = vp;
3047	mutex_exit(&pp->p_lock);
3048	if (oldvp != NULL)
3049		VN_RELE(oldvp);
3050}
3051
3052/*
3053 * Convert an rctl value represented by an nvlist_t into an rctl_val_t.
3054 */
3055static int
3056nvlist2rctlval(nvlist_t *nvl, rctl_val_t *rv)
3057{
3058	nvpair_t *nvp = NULL;
3059	boolean_t priv_set = B_FALSE;
3060	boolean_t limit_set = B_FALSE;
3061	boolean_t action_set = B_FALSE;
3062
3063	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
3064		const char *name;
3065		uint64_t ui64;
3066
3067		name = nvpair_name(nvp);
3068		if (nvpair_type(nvp) != DATA_TYPE_UINT64)
3069			return (EINVAL);
3070		(void) nvpair_value_uint64(nvp, &ui64);
3071		if (strcmp(name, "privilege") == 0) {
3072			/*
3073			 * Currently only privileged values are allowed, but
3074			 * this may change in the future.
3075			 */
3076			if (ui64 != RCPRIV_PRIVILEGED)
3077				return (EINVAL);
3078			rv->rcv_privilege = ui64;
3079			priv_set = B_TRUE;
3080		} else if (strcmp(name, "limit") == 0) {
3081			rv->rcv_value = ui64;
3082			limit_set = B_TRUE;
3083		} else if (strcmp(name, "action") == 0) {
3084			if (ui64 != RCTL_LOCAL_NOACTION &&
3085			    ui64 != RCTL_LOCAL_DENY)
3086				return (EINVAL);
3087			rv->rcv_flagaction = ui64;
3088			action_set = B_TRUE;
3089		} else {
3090			return (EINVAL);
3091		}
3092	}
3093
3094	if (!(priv_set && limit_set && action_set))
3095		return (EINVAL);
3096	rv->rcv_action_signal = 0;
3097	rv->rcv_action_recipient = NULL;
3098	rv->rcv_action_recip_pid = -1;
3099	rv->rcv_firing_time = 0;
3100
3101	return (0);
3102}
3103
3104/*
3105 * Non-global zone version of start_init.
3106 */
3107void
3108zone_start_init(void)
3109{
3110	proc_t *p = ttoproc(curthread);
3111	zone_t *z = p->p_zone;
3112
3113	ASSERT(!INGLOBALZONE(curproc));
3114
3115	/*
3116	 * For all purposes (ZONE_ATTR_INITPID and restart_init),
3117	 * storing just the pid of init is sufficient.
3118	 */
3119	z->zone_proc_initpid = p->p_pid;
3120
3121	/*
3122	 * We maintain zone_boot_err so that we can return the cause of the
3123	 * failure back to the caller of the zone_boot syscall.
3124	 */
3125	p->p_zone->zone_boot_err = start_init_common();
3126
3127	mutex_enter(&zone_status_lock);
3128	if (z->zone_boot_err != 0) {
3129		/*
3130		 * Make sure we are still in the booting state-- we could have
3131		 * raced and already be shutting down, or even further along.
3132		 */
3133		if (zone_status_get(z) == ZONE_IS_BOOTING) {
3134			zone_status_set(z, ZONE_IS_SHUTTING_DOWN);
3135		}
3136		mutex_exit(&zone_status_lock);
3137		/* It's gone bad, dispose of the process */
3138		if (proc_exit(CLD_EXITED, z->zone_boot_err) != 0) {
3139			mutex_enter(&p->p_lock);
3140			ASSERT(p->p_flag & SEXITLWPS);
3141			lwp_exit();
3142		}
3143	} else {
3144		if (zone_status_get(z) == ZONE_IS_BOOTING)
3145			zone_status_set(z, ZONE_IS_RUNNING);
3146		mutex_exit(&zone_status_lock);
3147		/* cause the process to return to userland. */
3148		lwp_rtt();
3149	}
3150}
3151
3152struct zsched_arg {
3153	zone_t *zone;
3154	nvlist_t *nvlist;
3155};
3156
3157/*
3158 * Per-zone "sched" workalike.  The similarity to "sched" doesn't have
3159 * anything to do with scheduling, but rather with the fact that
3160 * per-zone kernel threads are parented to zsched, just like regular
3161 * kernel threads are parented to sched (p0).
3162 *
3163 * zsched is also responsible for launching init for the zone.
3164 */
3165static void
3166zsched(void *arg)
3167{
3168	struct zsched_arg *za = arg;
3169	proc_t *pp = curproc;
3170	proc_t *initp = proc_init;
3171	zone_t *zone = za->zone;
3172	cred_t *cr, *oldcred;
3173	rctl_set_t *set;
3174	rctl_alloc_gp_t *gp;
3175	contract_t *ct = NULL;
3176	task_t *tk, *oldtk;
3177	rctl_entity_p_t e;
3178	kproject_t *pj;
3179
3180	nvlist_t *nvl = za->nvlist;
3181	nvpair_t *nvp = NULL;
3182
3183	bcopy("zsched", PTOU(pp)->u_psargs, sizeof ("zsched"));
3184	bcopy("zsched", PTOU(pp)->u_comm, sizeof ("zsched"));
3185	PTOU(pp)->u_argc = 0;
3186	PTOU(pp)->u_argv = NULL;
3187	PTOU(pp)->u_envp = NULL;
3188	closeall(P_FINFO(pp));
3189
3190	/*
3191	 * We are this zone's "zsched" process.  As the zone isn't generally
3192	 * visible yet we don't need to grab any locks before initializing its
3193	 * zone_proc pointer.
3194	 */
3195	zone_hold(zone);  /* this hold is released by zone_destroy() */
3196	zone->zone_zsched = pp;
3197	mutex_enter(&pp->p_lock);
3198	pp->p_zone = zone;
3199	mutex_exit(&pp->p_lock);
3200
3201	/*
3202	 * Disassociate process from its 'parent'; parent ourselves to init
3203	 * (pid 1) and change other values as needed.
3204	 */
3205	sess_create();
3206
3207	mutex_enter(&pidlock);
3208	proc_detach(pp);
3209	pp->p_ppid = 1;
3210	pp->p_flag |= SZONETOP;
3211	pp->p_ancpid = 1;
3212	pp->p_parent = initp;
3213	pp->p_psibling = NULL;
3214	if (initp->p_child)
3215		initp->p_child->p_psibling = pp;
3216	pp->p_sibling = initp->p_child;
3217	initp->p_child = pp;
3218
3219	/* Decrement what newproc() incremented. */
3220	upcount_dec(crgetruid(CRED()), GLOBAL_ZONEID);
3221	/*
3222	 * Our credentials are about to become kcred-like, so we don't care
3223	 * about the caller's ruid.
3224	 */
3225	upcount_inc(crgetruid(kcred), zone->zone_id);
3226	mutex_exit(&pidlock);
3227
3228	/*
3229	 * getting out of global zone, so decrement lwp counts
3230	 */
3231	pj = pp->p_task->tk_proj;
3232	mutex_enter(&global_zone->zone_nlwps_lock);
3233	pj->kpj_nlwps -= pp->p_lwpcnt;
3234	global_zone->zone_nlwps -= pp->p_lwpcnt;
3235	mutex_exit(&global_zone->zone_nlwps_lock);
3236
3237	/*
3238	 * Decrement locked memory counts on old zone and project.
3239	 */
3240	mutex_enter(&global_zone->zone_mem_lock);
3241	global_zone->zone_locked_mem -= pp->p_locked_mem;
3242	pj->kpj_data.kpd_locked_mem -= pp->p_locked_mem;
3243	mutex_exit(&global_zone->zone_mem_lock);
3244
3245	/*
3246	 * Create and join a new task in project '0' of this zone.
3247	 *
3248	 * We don't need to call holdlwps() since we know we're the only lwp in
3249	 * this process.
3250	 *
3251	 * task_join() returns with p_lock held.
3252	 */
3253	tk = task_create(0, zone);
3254	mutex_enter(&cpu_lock);
3255	oldtk = task_join(tk, 0);
3256
3257	pj = pp->p_task->tk_proj;
3258
3259	mutex_enter(&zone->zone_mem_lock);
3260	zone->zone_locked_mem += pp->p_locked_mem;
3261	pj->kpj_data.kpd_locked_mem += pp->p_locked_mem;
3262	mutex_exit(&zone->zone_mem_lock);
3263
3264	/*
3265	 * add lwp counts to zsched's zone, and increment project's task count
3266	 * due to the task created in the above tasksys_settaskid
3267	 */
3268
3269	mutex_enter(&zone->zone_nlwps_lock);
3270	pj->kpj_nlwps += pp->p_lwpcnt;
3271	pj->kpj_ntasks += 1;
3272	zone->zone_nlwps += pp->p_lwpcnt;
3273	mutex_exit(&zone->zone_nlwps_lock);
3274
3275	mutex_exit(&curproc->p_lock);
3276	mutex_exit(&cpu_lock);
3277	task_rele(oldtk);
3278
3279	/*
3280	 * The process was created by a process in the global zone, hence the
3281	 * credentials are wrong.  We might as well have kcred-ish credentials.
3282	 */
3283	cr = zone->zone_kcred;
3284	crhold(cr);
3285	mutex_enter(&pp->p_crlock);
3286	oldcred = pp->p_cred;
3287	pp->p_cred = cr;
3288	mutex_exit(&pp->p_crlock);
3289	crfree(oldcred);
3290
3291	/*
3292	 * Hold credentials again (for thread)
3293	 */
3294	crhold(cr);
3295
3296	/*
3297	 * p_lwpcnt can't change since this is a kernel process.
3298	 */
3299	crset(pp, cr);
3300
3301	/*
3302	 * Chroot
3303	 */
3304	zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_cdir, pp);
3305	zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_rdir, pp);
3306
3307	/*
3308	 * Initialize zone's rctl set.
3309	 */
3310	set = rctl_set_create();
3311	gp = rctl_set_init_prealloc(RCENTITY_ZONE);
3312	mutex_enter(&pp->p_lock);
3313	e.rcep_p.zone = zone;
3314	e.rcep_t = RCENTITY_ZONE;
3315	zone->zone_rctls = rctl_set_init(RCENTITY_ZONE, pp, &e, set, gp);
3316	mutex_exit(&pp->p_lock);
3317	rctl_prealloc_destroy(gp);
3318
3319	/*
3320	 * Apply the rctls passed in to zone_create().  This is basically a list
3321	 * assignment: all of the old values are removed and the new ones
3322	 * inserted.  That is, if an empty list is passed in, all values are
3323	 * removed.
3324	 */
3325	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
3326		rctl_dict_entry_t *rde;
3327		rctl_hndl_t hndl;
3328		char *name;
3329		nvlist_t **nvlarray;
3330		uint_t i, nelem;
3331		int error;	/* For ASSERT()s */
3332
3333		name = nvpair_name(nvp);
3334		hndl = rctl_hndl_lookup(name);
3335		ASSERT(hndl != -1);
3336		rde = rctl_dict_lookup_hndl(hndl);
3337		ASSERT(rde != NULL);
3338
3339		for (; /* ever */; ) {
3340			rctl_val_t oval;
3341
3342			mutex_enter(&pp->p_lock);
3343			error = rctl_local_get(hndl, NULL, &oval, pp);
3344			mutex_exit(&pp->p_lock);
3345			ASSERT(error == 0);	/* Can't fail for RCTL_FIRST */
3346			ASSERT(oval.rcv_privilege != RCPRIV_BASIC);
3347			if (oval.rcv_privilege == RCPRIV_SYSTEM)
3348				break;
3349			mutex_enter(&pp->p_lock);
3350			error = rctl_local_delete(hndl, &oval, pp);
3351			mutex_exit(&pp->p_lock);
3352			ASSERT(error == 0);
3353		}
3354		error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
3355		ASSERT(error == 0);
3356		for (i = 0; i < nelem; i++) {
3357			rctl_val_t *nvalp;
3358
3359			nvalp = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
3360			error = nvlist2rctlval(nvlarray[i], nvalp);
3361			ASSERT(error == 0);
3362			/*
3363			 * rctl_local_insert can fail if the value being
3364			 * inserted is a duplicate; this is OK.
3365			 */
3366			mutex_enter(&pp->p_lock);
3367			if (rctl_local_insert(hndl, nvalp, pp) != 0)
3368				kmem_cache_free(rctl_val_cache, nvalp);
3369			mutex_exit(&pp->p_lock);
3370		}
3371	}
3372	/*
3373	 * Tell the world that we're done setting up.
3374	 *
3375	 * At this point we want to set the zone status to ZONE_IS_INITIALIZED
3376	 * and atomically set the zone's processor set visibility.  Once
3377	 * we drop pool_lock() this zone will automatically get updated
3378	 * to reflect any future changes to the pools configuration.
3379	 *
3380	 * Note that after we drop the locks below (zonehash_lock in
3381	 * particular) other operations such as a zone_getattr call can
3382	 * now proceed and observe the zone. That is the reason for doing a
3383	 * state transition to the INITIALIZED state.
3384	 */
3385	pool_lock();
3386	mutex_enter(&cpu_lock);
3387	mutex_enter(&zonehash_lock);
3388	zone_uniqid(zone);
3389	zone_zsd_configure(zone);
3390	if (pool_state == POOL_ENABLED)
3391		zone_pset_set(zone, pool_default->pool_pset->pset_id);
3392	mutex_enter(&zone_status_lock);
3393	ASSERT(zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
3394	zone_status_set(zone, ZONE_IS_INITIALIZED);
3395	mutex_exit(&zone_status_lock);
3396	mutex_exit(&zonehash_lock);
3397	mutex_exit(&cpu_lock);
3398	pool_unlock();
3399
3400	/* Now call the create callback for this key */
3401	zsd_apply_all_keys(zsd_apply_create, zone);
3402
3403	/* The callbacks are complete. Mark ZONE_IS_READY */
3404	mutex_enter(&zone_status_lock);
3405	ASSERT(zone_status_get(zone) == ZONE_IS_INITIALIZED);
3406	zone_status_set(zone, ZONE_IS_READY);
3407	mutex_exit(&zone_status_lock);
3408
3409	/*
3410	 * Once we see the zone transition to the ZONE_IS_BOOTING state,
3411	 * we launch init, and set the state to running.
3412	 */
3413	zone_status_wait_cpr(zone, ZONE_IS_BOOTING, "zsched");
3414
3415	if (zone_status_get(zone) == ZONE_IS_BOOTING) {
3416		id_t cid;
3417
3418		/*
3419		 * Ok, this is a little complicated.  We need to grab the
3420		 * zone's pool's scheduling class ID; note that by now, we
3421		 * are already bound to a pool if we need to be (zoneadmd
3422		 * will have done that to us while we're in the READY
3423		 * state).  *But* the scheduling class for the zone's 'init'
3424		 * must be explicitly passed to newproc, which doesn't
3425		 * respect pool bindings.
3426		 *
3427		 * We hold the pool_lock across the call to newproc() to
3428		 * close the obvious race: the pool's scheduling class
3429		 * could change before we manage to create the LWP with
3430		 * classid 'cid'.
3431		 */
3432		pool_lock();
3433		if (zone->zone_defaultcid > 0)
3434			cid = zone->zone_defaultcid;
3435		else
3436			cid = pool_get_class(zone->zone_pool);
3437		if (cid == -1)
3438			cid = defaultcid;
3439
3440		/*
3441		 * If this fails, zone_boot will ultimately fail.  The
3442		 * state of the zone will be set to SHUTTING_DOWN-- userland
3443		 * will have to tear down the zone, and fail, or try again.
3444		 */
3445		if ((zone->zone_boot_err = newproc(zone_start_init, NULL, cid,
3446		    minclsyspri - 1, &ct)) != 0) {
3447			mutex_enter(&zone_status_lock);
3448			zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
3449			mutex_exit(&zone_status_lock);
3450		}
3451		pool_unlock();
3452	}
3453
3454	/*
3455	 * Wait for zone_destroy() to be called.  This is what we spend
3456	 * most of our life doing.
3457	 */
3458	zone_status_wait_cpr(zone, ZONE_IS_DYING, "zsched");
3459
3460	if (ct)
3461		/*
3462		 * At this point the process contract should be empty.
3463		 * (Though if it isn't, it's not the end of the world.)
3464		 */
3465		VERIFY(contract_abandon(ct, curproc, B_TRUE) == 0);
3466
3467	/*
3468	 * Allow kcred to be freed when all referring processes
3469	 * (including this one) go away.  We can't just do this in
3470	 * zone_free because we need to wait for the zone_cred_ref to
3471	 * drop to 0 before calling zone_free, and the existence of
3472	 * zone_kcred will prevent that.  Thus, we call crfree here to
3473	 * balance the crdup in zone_create.  The crhold calls earlier
3474	 * in zsched will be dropped when the thread and process exit.
3475	 */
3476	crfree(zone->zone_kcred);
3477	zone->zone_kcred = NULL;
3478
3479	exit(CLD_EXITED, 0);
3480}
3481
3482/*
3483 * Helper function to determine if there are any submounts of the
3484 * provided path.  Used to make sure the zone doesn't "inherit" any
3485 * mounts from before it is created.
3486 */
3487static uint_t
3488zone_mount_count(const char *rootpath)
3489{
3490	vfs_t *vfsp;
3491	uint_t count = 0;
3492	size_t rootpathlen = strlen(rootpath);
3493
3494	/*
3495	 * Holding zonehash_lock prevents race conditions with
3496	 * vfs_list_add()/vfs_list_remove() since we serialize with
3497	 * zone_find_by_path().
3498	 */
3499	ASSERT(MUTEX_HELD(&zonehash_lock));
3500	/*
3501	 * The rootpath must end with a '/'
3502	 */
3503	ASSERT(rootpath[rootpathlen - 1] == '/');
3504
3505	/*
3506	 * This intentionally does not count the rootpath itself if that
3507	 * happens to be a mount point.
3508	 */
3509	vfs_list_read_lock();
3510	vfsp = rootvfs;
3511	do {
3512		if (strncmp(rootpath, refstr_value(vfsp->vfs_mntpt),
3513		    rootpathlen) == 0)
3514			count++;
3515		vfsp = vfsp->vfs_next;
3516	} while (vfsp != rootvfs);
3517	vfs_list_unlock();
3518	return (count);
3519}
3520
3521/*
3522 * Helper function to make sure that a zone created on 'rootpath'
3523 * wouldn't end up containing other zones' rootpaths.
3524 */
3525static boolean_t
3526zone_is_nested(const char *rootpath)
3527{
3528	zone_t *zone;
3529	size_t rootpathlen = strlen(rootpath);
3530	size_t len;
3531
3532	ASSERT(MUTEX_HELD(&zonehash_lock));
3533
3534	for (zone = list_head(&zone_active); zone != NULL;
3535	    zone = list_next(&zone_active, zone)) {
3536		if (zone == global_zone)
3537			continue;
3538		len = strlen(zone->zone_rootpath);
3539		if (strncmp(rootpath, zone->zone_rootpath,
3540		    MIN(rootpathlen, len)) == 0)
3541			return (B_TRUE);
3542	}
3543	return (B_FALSE);
3544}
3545
3546static int
3547zone_set_privset(zone_t *zone, const priv_set_t *zone_privs,
3548    size_t zone_privssz)
3549{
3550	priv_set_t *privs = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
3551
3552	if (zone_privssz < sizeof (priv_set_t))
3553		return (set_errno(ENOMEM));
3554
3555	if (copyin(zone_privs, privs, sizeof (priv_set_t))) {
3556		kmem_free(privs, sizeof (priv_set_t));
3557		return (EFAULT);
3558	}
3559
3560	zone->zone_privset = privs;
3561	return (0);
3562}
3563
3564/*
3565 * We make creative use of nvlists to pass in rctls from userland.  The list is
3566 * a list of the following structures:
3567 *
3568 * (name = rctl_name, value = nvpair_list_array)
3569 *
3570 * Where each element of the nvpair_list_array is of the form:
3571 *
3572 * [(name = "privilege", value = RCPRIV_PRIVILEGED),
3573 * 	(name = "limit", value = uint64_t),
3574 * 	(name = "action", value = (RCTL_LOCAL_NOACTION || RCTL_LOCAL_DENY))]
3575 */
3576static int
3577parse_rctls(caddr_t ubuf, size_t buflen, nvlist_t **nvlp)
3578{
3579	nvpair_t *nvp = NULL;
3580	nvlist_t *nvl = NULL;
3581	char *kbuf;
3582	int error;
3583	rctl_val_t rv;
3584
3585	*nvlp = NULL;
3586
3587	if (buflen == 0)
3588		return (0);
3589
3590	if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL)
3591		return (ENOMEM);
3592	if (copyin(ubuf, kbuf, buflen)) {
3593		error = EFAULT;
3594		goto out;
3595	}
3596	if (nvlist_unpack(kbuf, buflen, &nvl, KM_SLEEP) != 0) {
3597		/*
3598		 * nvl may have been allocated/free'd, but the value set to
3599		 * non-NULL, so we reset it here.
3600		 */
3601		nvl = NULL;
3602		error = EINVAL;
3603		goto out;
3604	}
3605	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
3606		rctl_dict_entry_t *rde;
3607		rctl_hndl_t hndl;
3608		nvlist_t **nvlarray;
3609		uint_t i, nelem;
3610		char *name;
3611
3612		error = EINVAL;
3613		name = nvpair_name(nvp);
3614		if (strncmp(nvpair_name(nvp), "zone.", sizeof ("zone.") - 1)
3615		    != 0 || nvpair_type(nvp) != DATA_TYPE_NVLIST_ARRAY) {
3616			goto out;
3617		}
3618		if ((hndl = rctl_hndl_lookup(name)) == -1) {
3619			goto out;
3620		}
3621		rde = rctl_dict_lookup_hndl(hndl);
3622		error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
3623		ASSERT(error == 0);
3624		for (i = 0; i < nelem; i++) {
3625			if (error = nvlist2rctlval(nvlarray[i], &rv))
3626				goto out;
3627		}
3628		if (rctl_invalid_value(rde, &rv)) {
3629			error = EINVAL;
3630			goto out;
3631		}
3632	}
3633	error = 0;
3634	*nvlp = nvl;
3635out:
3636	kmem_free(kbuf, buflen);
3637	if (error && nvl != NULL)
3638		nvlist_free(nvl);
3639	return (error);
3640}
3641
3642int
3643zone_create_error(int er_error, int er_ext, int *er_out) {
3644	if (er_out != NULL) {
3645		if (copyout(&er_ext, er_out, sizeof (int))) {
3646			return (set_errno(EFAULT));
3647		}
3648	}
3649	return (set_errno(er_error));
3650}
3651
3652static int
3653zone_set_label(zone_t *zone, const bslabel_t *lab, uint32_t doi)
3654{
3655	ts_label_t *tsl;
3656	bslabel_t blab;
3657
3658	/* Get label from user */
3659	if (copyin(lab, &blab, sizeof (blab)) != 0)
3660		return (EFAULT);
3661	tsl = labelalloc(&blab, doi, KM_NOSLEEP);
3662	if (tsl == NULL)
3663		return (ENOMEM);
3664
3665	zone->zone_slabel = tsl;
3666	return (0);
3667}
3668
3669/*
3670 * Parses a comma-separated list of ZFS datasets into a per-zone dictionary.
3671 */
3672static int
3673parse_zfs(zone_t *zone, caddr_t ubuf, size_t buflen)
3674{
3675	char *kbuf;
3676	char *dataset, *next;
3677	zone_dataset_t *zd;
3678	size_t len;
3679
3680	if (ubuf == NULL || buflen == 0)
3681		return (0);
3682
3683	if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL)
3684		return (ENOMEM);
3685
3686	if (copyin(ubuf, kbuf, buflen) != 0) {
3687		kmem_free(kbuf, buflen);
3688		return (EFAULT);
3689	}
3690
3691	dataset = next = kbuf;
3692	for (;;) {
3693		zd = kmem_alloc(sizeof (zone_dataset_t), KM_SLEEP);
3694
3695		next = strchr(dataset, ',');
3696
3697		if (next == NULL)
3698			len = strlen(dataset);
3699		else
3700			len = next - dataset;
3701
3702		zd->zd_dataset = kmem_alloc(len + 1, KM_SLEEP);
3703		bcopy(dataset, zd->zd_dataset, len);
3704		zd->zd_dataset[len] = '\0';
3705
3706		list_insert_head(&zone->zone_datasets, zd);
3707
3708		if (next == NULL)
3709			break;
3710
3711		dataset = next + 1;
3712	}
3713
3714	kmem_free(kbuf, buflen);
3715	return (0);
3716}
3717
3718/*
3719 * System call to create/initialize a new zone named 'zone_name', rooted
3720 * at 'zone_root', with a zone-wide privilege limit set of 'zone_privs',
3721 * and initialized with the zone-wide rctls described in 'rctlbuf', and
3722 * with labeling set by 'match', 'doi', and 'label'.
3723 *
3724 * If extended error is non-null, we may use it to return more detailed
3725 * error information.
3726 */
3727static zoneid_t
3728zone_create(const char *zone_name, const char *zone_root,
3729    const priv_set_t *zone_privs, size_t zone_privssz,
3730    caddr_t rctlbuf, size_t rctlbufsz,
3731    caddr_t zfsbuf, size_t zfsbufsz, int *extended_error,
3732    int match, uint32_t doi, const bslabel_t *label,
3733    int flags)
3734{
3735	struct zsched_arg zarg;
3736	nvlist_t *rctls = NULL;
3737	proc_t *pp = curproc;
3738	zone_t *zone, *ztmp;
3739	zoneid_t zoneid;
3740	int error;
3741	int error2 = 0;
3742	char *str;
3743	cred_t *zkcr;
3744	boolean_t insert_label_hash;
3745
3746	if (secpolicy_zone_config(CRED()) != 0)
3747		return (set_errno(EPERM));
3748
3749	/* can't boot zone from within chroot environment */
3750	if (PTOU(pp)->u_rdir != NULL && PTOU(pp)->u_rdir != rootdir)
3751		return (zone_create_error(ENOTSUP, ZE_CHROOTED,
3752		    extended_error));
3753
3754	zone = kmem_zalloc(sizeof (zone_t), KM_SLEEP);
3755	zoneid = zone->zone_id = id_alloc(zoneid_space);
3756	zone->zone_status = ZONE_IS_UNINITIALIZED;
3757	zone->zone_pool = pool_default;
3758	zone->zone_pool_mod = gethrtime();
3759	zone->zone_psetid = ZONE_PS_INVAL;
3760	zone->zone_ncpus = 0;
3761	zone->zone_ncpus_online = 0;
3762	zone->zone_restart_init = B_TRUE;
3763	zone->zone_brand = &native_brand;
3764	zone->zone_initname = NULL;
3765	mutex_init(&zone->zone_lock, NULL, MUTEX_DEFAULT, NULL);
3766	mutex_init(&zone->zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
3767	mutex_init(&zone->zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
3768	cv_init(&zone->zone_cv, NULL, CV_DEFAULT, NULL);
3769	list_create(&zone->zone_zsd, sizeof (struct zsd_entry),
3770	    offsetof(struct zsd_entry, zsd_linkage));
3771	list_create(&zone->zone_datasets, sizeof (zone_dataset_t),
3772	    offsetof(zone_dataset_t, zd_linkage));
3773	rw_init(&zone->zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
3774
3775	if (flags & ZCF_NET_EXCL) {
3776		zone->zone_flags |= ZF_NET_EXCL;
3777	}
3778
3779	if ((error = zone_set_name(zone, zone_name)) != 0) {
3780		zone_free(zone);
3781		return (zone_create_error(error, 0, extended_error));
3782	}
3783
3784	if ((error = zone_set_root(zone, zone_root)) != 0) {
3785		zone_free(zone);
3786		return (zone_create_error(error, 0, extended_error));
3787	}
3788	if ((error = zone_set_privset(zone, zone_privs, zone_privssz)) != 0) {
3789		zone_free(zone);
3790		return (zone_create_error(error, 0, extended_error));
3791	}
3792
3793	/* initialize node name to be the same as zone name */
3794	zone->zone_nodename = kmem_alloc(_SYS_NMLN, KM_SLEEP);
3795	(void) strncpy(zone->zone_nodename, zone->zone_name, _SYS_NMLN);
3796	zone->zone_nodename[_SYS_NMLN - 1] = '\0';
3797
3798	zone->zone_domain = kmem_alloc(_SYS_NMLN, KM_SLEEP);
3799	zone->zone_domain[0] = '\0';
3800	zone->zone_shares = 1;
3801	zone->zone_shmmax = 0;
3802	zone->zone_ipc.ipcq_shmmni = 0;
3803	zone->zone_ipc.ipcq_semmni = 0;
3804	zone->zone_ipc.ipcq_msgmni = 0;
3805	zone->zone_bootargs = NULL;
3806	zone->zone_initname =
3807	    kmem_alloc(strlen(zone_default_initname) + 1, KM_SLEEP);
3808	(void) strcpy(zone->zone_initname, zone_default_initname);
3809	zone->zone_nlwps = 0;
3810	zone->zone_nlwps_ctl = INT_MAX;
3811	zone->zone_locked_mem = 0;
3812	zone->zone_locked_mem_ctl = UINT64_MAX;
3813	zone->zone_max_swap = 0;
3814	zone->zone_max_swap_ctl = UINT64_MAX;
3815	zone0.zone_lockedmem_kstat = NULL;
3816	zone0.zone_swapresv_kstat = NULL;
3817
3818	/*
3819	 * Zsched initializes the rctls.
3820	 */
3821	zone->zone_rctls = NULL;
3822
3823	if ((error = parse_rctls(rctlbuf, rctlbufsz, &rctls)) != 0) {
3824		zone_free(zone);
3825		return (zone_create_error(error, 0, extended_error));
3826	}
3827
3828	if ((error = parse_zfs(zone, zfsbuf, zfsbufsz)) != 0) {
3829		zone_free(zone);
3830		return (set_errno(error));
3831	}
3832
3833	/*
3834	 * Read in the trusted system parameters:
3835	 * match flag and sensitivity label.
3836	 */
3837	zone->zone_match = match;
3838	if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) {
3839		/* Fail if requested to set doi to anything but system's doi */
3840		if (doi != 0 && doi != default_doi) {
3841			zone_free(zone);
3842			return (set_errno(EINVAL));
3843		}
3844		/* Always apply system's doi to the zone */
3845		error = zone_set_label(zone, label, default_doi);
3846		if (error != 0) {
3847			zone_free(zone);
3848			return (set_errno(error));
3849		}
3850		insert_label_hash = B_TRUE;
3851	} else {
3852		/* all zones get an admin_low label if system is not labeled */
3853		zone->zone_slabel = l_admin_low;
3854		label_hold(l_admin_low);
3855		insert_label_hash = B_FALSE;
3856	}
3857
3858	/*
3859	 * Stop all lwps since that's what normally happens as part of fork().
3860	 * This needs to happen before we grab any locks to avoid deadlock
3861	 * (another lwp in the process could be waiting for the held lock).
3862	 */
3863	if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK)) {
3864		zone_free(zone);
3865		if (rctls)
3866			nvlist_free(rctls);
3867		return (zone_create_error(error, 0, extended_error));
3868	}
3869
3870	if (block_mounts() == 0) {
3871		mutex_enter(&pp->p_lock);
3872		if (curthread != pp->p_agenttp)
3873			continuelwps(pp);
3874		mutex_exit(&pp->p_lock);
3875		zone_free(zone);
3876		if (rctls)
3877			nvlist_free(rctls);
3878		return (zone_create_error(error, 0, extended_error));
3879	}
3880
3881	/*
3882	 * Set up credential for kernel access.  After this, any errors
3883	 * should go through the dance in errout rather than calling
3884	 * zone_free directly.
3885	 */
3886	zone->zone_kcred = crdup(kcred);
3887	crsetzone(zone->zone_kcred, zone);
3888	priv_intersect(zone->zone_privset, &CR_PPRIV(zone->zone_kcred));
3889	priv_intersect(zone->zone_privset, &CR_EPRIV(zone->zone_kcred));
3890	priv_intersect(zone->zone_privset, &CR_IPRIV(zone->zone_kcred));
3891	priv_intersect(zone->zone_privset, &CR_LPRIV(zone->zone_kcred));
3892
3893	mutex_enter(&zonehash_lock);
3894	/*
3895	 * Make sure zone doesn't already exist.
3896	 *
3897	 * If the system and zone are labeled,
3898	 * make sure no other zone exists that has the same label.
3899	 */
3900	if ((ztmp = zone_find_all_by_name(zone->zone_name)) != NULL ||
3901	    (insert_label_hash &&
3902	    (ztmp = zone_find_all_by_label(zone->zone_slabel)) != NULL)) {
3903		zone_status_t status;
3904
3905		status = zone_status_get(ztmp);
3906		if (status == ZONE_IS_READY || status == ZONE_IS_RUNNING)
3907			error = EEXIST;
3908		else
3909			error = EBUSY;
3910
3911		if (insert_label_hash)
3912			error2 = ZE_LABELINUSE;
3913
3914		goto errout;
3915	}
3916
3917	/*
3918	 * Don't allow zone creations which would cause one zone's rootpath to
3919	 * be accessible from that of another (non-global) zone.
3920	 */
3921	if (zone_is_nested(zone->zone_rootpath)) {
3922		error = EBUSY;
3923		goto errout;
3924	}
3925
3926	ASSERT(zonecount != 0);		/* check for leaks */
3927	if (zonecount + 1 > maxzones) {
3928		error = ENOMEM;
3929		goto errout;
3930	}
3931
3932	if (zone_mount_count(zone->zone_rootpath) != 0) {
3933		error = EBUSY;
3934		error2 = ZE_AREMOUNTS;
3935		goto errout;
3936	}
3937
3938	/*
3939	 * Zone is still incomplete, but we need to drop all locks while
3940	 * zsched() initializes this zone's kernel process.  We
3941	 * optimistically add the zone to the hashtable and associated
3942	 * lists so a parallel zone_create() doesn't try to create the
3943	 * same zone.
3944	 */
3945	zonecount++;
3946	(void) mod_hash_insert(zonehashbyid,
3947	    (mod_hash_key_t)(uintptr_t)zone->zone_id,
3948	    (mod_hash_val_t)(uintptr_t)zone);
3949	str = kmem_alloc(strlen(zone->zone_name) + 1, KM_SLEEP);
3950	(void) strcpy(str, zone->zone_name);
3951	(void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)str,
3952	    (mod_hash_val_t)(uintptr_t)zone);
3953	if (insert_label_hash) {
3954		(void) mod_hash_insert(zonehashbylabel,
3955		    (mod_hash_key_t)zone->zone_slabel, (mod_hash_val_t)zone);
3956		zone->zone_flags |= ZF_HASHED_LABEL;
3957	}
3958
3959	/*
3960	 * Insert into active list.  At this point there are no 'hold's
3961	 * on the zone, but everyone else knows not to use it, so we can
3962	 * continue to use it.  zsched() will do a zone_hold() if the
3963	 * newproc() is successful.
3964	 */
3965	list_insert_tail(&zone_active, zone);
3966	mutex_exit(&zonehash_lock);
3967
3968	zarg.zone = zone;
3969	zarg.nvlist = rctls;
3970	/*
3971	 * The process, task, and project rctls are probably wrong;
3972	 * we need an interface to get the default values of all rctls,
3973	 * and initialize zsched appropriately.  I'm not sure that that
3974	 * makes much of a difference, though.
3975	 */
3976	if (error = newproc(zsched, (void *)&zarg, syscid, minclsyspri, NULL)) {
3977		/*
3978		 * We need to undo all globally visible state.
3979		 */
3980		mutex_enter(&zonehash_lock);
3981		list_remove(&zone_active, zone);
3982		if (zone->zone_flags & ZF_HASHED_LABEL) {
3983			ASSERT(zone->zone_slabel != NULL);
3984			(void) mod_hash_destroy(zonehashbylabel,
3985			    (mod_hash_key_t)zone->zone_slabel);
3986		}
3987		(void) mod_hash_destroy(zonehashbyname,
3988		    (mod_hash_key_t)(uintptr_t)zone->zone_name);
3989		(void) mod_hash_destroy(zonehashbyid,
3990		    (mod_hash_key_t)(uintptr_t)zone->zone_id);
3991		ASSERT(zonecount > 1);
3992		zonecount--;
3993		goto errout;
3994	}
3995
3996	/*
3997	 * Zone creation can't fail from now on.
3998	 */
3999
4000	/*
4001	 * Create zone kstats
4002	 */
4003	zone_kstat_create(zone);
4004
4005	/*
4006	 * Let the other lwps continue.
4007	 */
4008	mutex_enter(&pp->p_lock);
4009	if (curthread != pp->p_agenttp)
4010		continuelwps(pp);
4011	mutex_exit(&pp->p_lock);
4012
4013	/*
4014	 * Wait for zsched to finish initializing the zone.
4015	 */
4016	zone_status_wait(zone, ZONE_IS_READY);
4017	/*
4018	 * The zone is fully visible, so we can let mounts progress.
4019	 */
4020	resume_mounts();
4021	if (rctls)
4022		nvlist_free(rctls);
4023
4024	return (zoneid);
4025
4026errout:
4027	mutex_exit(&zonehash_lock);
4028	/*
4029	 * Let the other lwps continue.
4030	 */
4031	mutex_enter(&pp->p_lock);
4032	if (curthread != pp->p_agenttp)
4033		continuelwps(pp);
4034	mutex_exit(&pp->p_lock);
4035
4036	resume_mounts();
4037	if (rctls)
4038		nvlist_free(rctls);
4039	/*
4040	 * There is currently one reference to the zone, a cred_ref from
4041	 * zone_kcred.  To free the zone, we call crfree, which will call
4042	 * zone_cred_rele, which will call zone_free.
4043	 */
4044	ASSERT(zone->zone_cred_ref == 1);	/* for zone_kcred */
4045	ASSERT(zone->zone_kcred->cr_ref == 1);
4046	ASSERT(zone->zone_ref == 0);
4047	zkcr = zone->zone_kcred;
4048	zone->zone_kcred = NULL;
4049	crfree(zkcr);				/* triggers call to zone_free */
4050	return (zone_create_error(error, error2, extended_error));
4051}
4052
4053/*
4054 * Cause the zone to boot.  This is pretty simple, since we let zoneadmd do
4055 * the heavy lifting.  initname is the path to the program to launch
4056 * at the "top" of the zone; if this is NULL, we use the system default,
4057 * which is stored at zone_default_initname.
4058 */
4059static int
4060zone_boot(zoneid_t zoneid)
4061{
4062	int err;
4063	zone_t *zone;
4064
4065	if (secpolicy_zone_config(CRED()) != 0)
4066		return (set_errno(EPERM));
4067	if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4068		return (set_errno(EINVAL));
4069
4070	mutex_enter(&zonehash_lock);
4071	/*
4072	 * Look for zone under hash lock to prevent races with calls to
4073	 * zone_shutdown, zone_destroy, etc.
4074	 */
4075	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4076		mutex_exit(&zonehash_lock);
4077		return (set_errno(EINVAL));
4078	}
4079
4080	mutex_enter(&zone_status_lock);
4081	if (zone_status_get(zone) != ZONE_IS_READY) {
4082		mutex_exit(&zone_status_lock);
4083		mutex_exit(&zonehash_lock);
4084		return (set_errno(EINVAL));
4085	}
4086	zone_status_set(zone, ZONE_IS_BOOTING);
4087	mutex_exit(&zone_status_lock);
4088
4089	zone_hold(zone);	/* so we can use the zone_t later */
4090	mutex_exit(&zonehash_lock);
4091
4092	if (zone_status_wait_sig(zone, ZONE_IS_RUNNING) == 0) {
4093		zone_rele(zone);
4094		return (set_errno(EINTR));
4095	}
4096
4097	/*
4098	 * Boot (starting init) might have failed, in which case the zone
4099	 * will go to the SHUTTING_DOWN state; an appropriate errno will
4100	 * be placed in zone->zone_boot_err, and so we return that.
4101	 */
4102	err = zone->zone_boot_err;
4103	zone_rele(zone);
4104	return (err ? set_errno(err) : 0);
4105}
4106
4107/*
4108 * Kills all user processes in the zone, waiting for them all to exit
4109 * before returning.
4110 */
4111static int
4112zone_empty(zone_t *zone)
4113{
4114	int waitstatus;
4115
4116	/*
4117	 * We need to drop zonehash_lock before killing all
4118	 * processes, otherwise we'll deadlock with zone_find_*
4119	 * which can be called from the exit path.
4120	 */
4121	ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
4122	while ((waitstatus = zone_status_timedwait_sig(zone, lbolt + hz,
4123	    ZONE_IS_EMPTY)) == -1) {
4124		killall(zone->zone_id);
4125	}
4126	/*
4127	 * return EINTR if we were signaled
4128	 */
4129	if (waitstatus == 0)
4130		return (EINTR);
4131	return (0);
4132}
4133
4134/*
4135 * This function implements the policy for zone visibility.
4136 *
4137 * In standard Solaris, a non-global zone can only see itself.
4138 *
4139 * In Trusted Extensions, a labeled zone can lookup any zone whose label
4140 * it dominates. For this test, the label of the global zone is treated as
4141 * admin_high so it is special-cased instead of being checked for dominance.
4142 *
4143 * Returns true if zone attributes are viewable, false otherwise.
4144 */
4145static boolean_t
4146zone_list_access(zone_t *zone)
4147{
4148
4149	if (curproc->p_zone == global_zone ||
4150	    curproc->p_zone == zone) {
4151		return (B_TRUE);
4152	} else if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) {
4153		bslabel_t *curproc_label;
4154		bslabel_t *zone_label;
4155
4156		curproc_label = label2bslabel(curproc->p_zone->zone_slabel);
4157		zone_label = label2bslabel(zone->zone_slabel);
4158
4159		if (zone->zone_id != GLOBAL_ZONEID &&
4160		    bldominates(curproc_label, zone_label)) {
4161			return (B_TRUE);
4162		} else {
4163			return (B_FALSE);
4164		}
4165	} else {
4166		return (B_FALSE);
4167	}
4168}
4169
4170/*
4171 * Systemcall to start the zone's halt sequence.  By the time this
4172 * function successfully returns, all user processes and kernel threads
4173 * executing in it will have exited, ZSD shutdown callbacks executed,
4174 * and the zone status set to ZONE_IS_DOWN.
4175 *
4176 * It is possible that the call will interrupt itself if the caller is the
4177 * parent of any process running in the zone, and doesn't have SIGCHLD blocked.
4178 */
4179static int
4180zone_shutdown(zoneid_t zoneid)
4181{
4182	int error;
4183	zone_t *zone;
4184	zone_status_t status;
4185
4186	if (secpolicy_zone_config(CRED()) != 0)
4187		return (set_errno(EPERM));
4188	if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4189		return (set_errno(EINVAL));
4190
4191	/*
4192	 * Block mounts so that VFS_MOUNT() can get an accurate view of
4193	 * the zone's status with regards to ZONE_IS_SHUTTING down.
4194	 *
4195	 * e.g. NFS can fail the mount if it determines that the zone
4196	 * has already begun the shutdown sequence.
4197	 */
4198	if (block_mounts() == 0)
4199		return (set_errno(EINTR));
4200	mutex_enter(&zonehash_lock);
4201	/*
4202	 * Look for zone under hash lock to prevent races with other
4203	 * calls to zone_shutdown and zone_destroy.
4204	 */
4205	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4206		mutex_exit(&zonehash_lock);
4207		resume_mounts();
4208		return (set_errno(EINVAL));
4209	}
4210	mutex_enter(&zone_status_lock);
4211	status = zone_status_get(zone);
4212	/*
4213	 * Fail if the zone isn't fully initialized yet.
4214	 */
4215	if (status < ZONE_IS_READY) {
4216		mutex_exit(&zone_status_lock);
4217		mutex_exit(&zonehash_lock);
4218		resume_mounts();
4219		return (set_errno(EINVAL));
4220	}
4221	/*
4222	 * If conditions required for zone_shutdown() to return have been met,
4223	 * return success.
4224	 */
4225	if (status >= ZONE_IS_DOWN) {
4226		mutex_exit(&zone_status_lock);
4227		mutex_exit(&zonehash_lock);
4228		resume_mounts();
4229		return (0);
4230	}
4231	/*
4232	 * If zone_shutdown() hasn't been called before, go through the motions.
4233	 * If it has, there's nothing to do but wait for the kernel threads to
4234	 * drain.
4235	 */
4236	if (status < ZONE_IS_EMPTY) {
4237		uint_t ntasks;
4238
4239		mutex_enter(&zone->zone_lock);
4240		if ((ntasks = zone->zone_ntasks) != 1) {
4241			/*
4242			 * There's still stuff running.
4243			 */
4244			zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
4245		}
4246		mutex_exit(&zone->zone_lock);
4247		if (ntasks == 1) {
4248			/*
4249			 * The only way to create another task is through
4250			 * zone_enter(), which will block until we drop
4251			 * zonehash_lock.  The zone is empty.
4252			 */
4253			if (zone->zone_kthreads == NULL) {
4254				/*
4255				 * Skip ahead to ZONE_IS_DOWN
4256				 */
4257				zone_status_set(zone, ZONE_IS_DOWN);
4258			} else {
4259				zone_status_set(zone, ZONE_IS_EMPTY);
4260			}
4261		}
4262	}
4263	zone_hold(zone);	/* so we can use the zone_t later */
4264	mutex_exit(&zone_status_lock);
4265	mutex_exit(&zonehash_lock);
4266	resume_mounts();
4267
4268	if (error = zone_empty(zone)) {
4269		zone_rele(zone);
4270		return (set_errno(error));
4271	}
4272	/*
4273	 * After the zone status goes to ZONE_IS_DOWN this zone will no
4274	 * longer be notified of changes to the pools configuration, so
4275	 * in order to not end up with a stale pool pointer, we point
4276	 * ourselves at the default pool and remove all resource
4277	 * visibility.  This is especially important as the zone_t may
4278	 * languish on the deathrow for a very long time waiting for
4279	 * cred's to drain out.
4280	 *
4281	 * This rebinding of the zone can happen multiple times
4282	 * (presumably due to interrupted or parallel systemcalls)
4283	 * without any adverse effects.
4284	 */
4285	if (pool_lock_intr() != 0) {
4286		zone_rele(zone);
4287		return (set_errno(EINTR));
4288	}
4289	if (pool_state == POOL_ENABLED) {
4290		mutex_enter(&cpu_lock);
4291		zone_pool_set(zone, pool_default);
4292		/*
4293		 * The zone no longer needs to be able to see any cpus.
4294		 */
4295		zone_pset_set(zone, ZONE_PS_INVAL);
4296		mutex_exit(&cpu_lock);
4297	}
4298	pool_unlock();
4299
4300	/*
4301	 * ZSD shutdown callbacks can be executed multiple times, hence
4302	 * it is safe to not be holding any locks across this call.
4303	 */
4304	zone_zsd_callbacks(zone, ZSD_SHUTDOWN);
4305
4306	mutex_enter(&zone_status_lock);
4307	if (zone->zone_kthreads == NULL && zone_status_get(zone) < ZONE_IS_DOWN)
4308		zone_status_set(zone, ZONE_IS_DOWN);
4309	mutex_exit(&zone_status_lock);
4310
4311	/*
4312	 * Wait for kernel threads to drain.
4313	 */
4314	if (!zone_status_wait_sig(zone, ZONE_IS_DOWN)) {
4315		zone_rele(zone);
4316		return (set_errno(EINTR));
4317	}
4318
4319	/*
4320	 * Zone can be become down/destroyable even if the above wait
4321	 * returns EINTR, so any code added here may never execute.
4322	 * (i.e. don't add code here)
4323	 */
4324
4325	zone_rele(zone);
4326	return (0);
4327}
4328
4329/*
4330 * Systemcall entry point to finalize the zone halt process.  The caller
4331 * must have already successfully called zone_shutdown().
4332 *
4333 * Upon successful completion, the zone will have been fully destroyed:
4334 * zsched will have exited, destructor callbacks executed, and the zone
4335 * removed from the list of active zones.
4336 */
4337static int
4338zone_destroy(zoneid_t zoneid)
4339{
4340	uint64_t uniqid;
4341	zone_t *zone;
4342	zone_status_t status;
4343
4344	if (secpolicy_zone_config(CRED()) != 0)
4345		return (set_errno(EPERM));
4346	if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4347		return (set_errno(EINVAL));
4348
4349	mutex_enter(&zonehash_lock);
4350	/*
4351	 * Look for zone under hash lock to prevent races with other
4352	 * calls to zone_destroy.
4353	 */
4354	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4355		mutex_exit(&zonehash_lock);
4356		return (set_errno(EINVAL));
4357	}
4358
4359	if (zone_mount_count(zone->zone_rootpath) != 0) {
4360		mutex_exit(&zonehash_lock);
4361		return (set_errno(EBUSY));
4362	}
4363	mutex_enter(&zone_status_lock);
4364	status = zone_status_get(zone);
4365	if (status < ZONE_IS_DOWN) {
4366		mutex_exit(&zone_status_lock);
4367		mutex_exit(&zonehash_lock);
4368		return (set_errno(EBUSY));
4369	} else if (status == ZONE_IS_DOWN) {
4370		zone_status_set(zone, ZONE_IS_DYING); /* Tell zsched to exit */
4371	}
4372	mutex_exit(&zone_status_lock);
4373	zone_hold(zone);
4374	mutex_exit(&zonehash_lock);
4375
4376	/*
4377	 * wait for zsched to exit
4378	 */
4379	zone_status_wait(zone, ZONE_IS_DEAD);
4380	zone_zsd_callbacks(zone, ZSD_DESTROY);
4381	zone->zone_netstack = NULL;
4382	uniqid = zone->zone_uniqid;
4383	zone_rele(zone);
4384	zone = NULL;	/* potentially free'd */
4385
4386	mutex_enter(&zonehash_lock);
4387	for (; /* ever */; ) {
4388		boolean_t unref;
4389
4390		if ((zone = zone_find_all_by_id(zoneid)) == NULL ||
4391		    zone->zone_uniqid != uniqid) {
4392			/*
4393			 * The zone has gone away.  Necessary conditions
4394			 * are met, so we return success.
4395			 */
4396			mutex_exit(&zonehash_lock);
4397			return (0);
4398		}
4399		mutex_enter(&zone->zone_lock);
4400		unref = ZONE_IS_UNREF(zone);
4401		mutex_exit(&zone->zone_lock);
4402		if (unref) {
4403			/*
4404			 * There is only one reference to the zone -- that
4405			 * added when the zone was added to the hashtables --
4406			 * and things will remain this way until we drop
4407			 * zonehash_lock... we can go ahead and cleanup the
4408			 * zone.
4409			 */
4410			break;
4411		}
4412
4413		if (cv_wait_sig(&zone_destroy_cv, &zonehash_lock) == 0) {
4414			/* Signaled */
4415			mutex_exit(&zonehash_lock);
4416			return (set_errno(EINTR));
4417		}
4418
4419	}
4420
4421	/*
4422	 * Remove CPU cap for this zone now since we're not going to
4423	 * fail below this point.
4424	 */
4425	cpucaps_zone_remove(zone);
4426
4427	/* Get rid of the zone's kstats */
4428	zone_kstat_delete(zone);
4429
4430	/* free brand specific data */
4431	if (ZONE_IS_BRANDED(zone))
4432		ZBROP(zone)->b_free_brand_data(zone);
4433
4434	/* Say goodbye to brand framework. */
4435	brand_unregister_zone(zone->zone_brand);
4436
4437	/*
4438	 * It is now safe to let the zone be recreated; remove it from the
4439	 * lists.  The memory will not be freed until the last cred
4440	 * reference goes away.
4441	 */
4442	ASSERT(zonecount > 1);	/* must be > 1; can't destroy global zone */
4443	zonecount--;
4444	/* remove from active list and hash tables */
4445	list_remove(&zone_active, zone);
4446	(void) mod_hash_destroy(zonehashbyname,
4447	    (mod_hash_key_t)zone->zone_name);
4448	(void) mod_hash_destroy(zonehashbyid,
4449	    (mod_hash_key_t)(uintptr_t)zone->zone_id);
4450	if (zone->zone_flags & ZF_HASHED_LABEL)
4451		(void) mod_hash_destroy(zonehashbylabel,
4452		    (mod_hash_key_t)zone->zone_slabel);
4453	mutex_exit(&zonehash_lock);
4454
4455	/*
4456	 * Release the root vnode; we're not using it anymore.  Nor should any
4457	 * other thread that might access it exist.
4458	 */
4459	if (zone->zone_rootvp != NULL) {
4460		VN_RELE(zone->zone_rootvp);
4461		zone->zone_rootvp = NULL;
4462	}
4463
4464	/* add to deathrow list */
4465	mutex_enter(&zone_deathrow_lock);
4466	list_insert_tail(&zone_deathrow, zone);
4467	mutex_exit(&zone_deathrow_lock);
4468
4469	/*
4470	 * Drop last reference (which was added by zsched()), this will
4471	 * free the zone unless there are outstanding cred references.
4472	 */
4473	zone_rele(zone);
4474	return (0);
4475}
4476
4477/*
4478 * Systemcall entry point for zone_getattr(2).
4479 */
4480static ssize_t
4481zone_getattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
4482{
4483	size_t size;
4484	int error = 0, err;
4485	zone_t *zone;
4486	char *zonepath;
4487	char *outstr;
4488	zone_status_t zone_status;
4489	pid_t initpid;
4490	boolean_t global = (curzone == global_zone);
4491	boolean_t inzone = (curzone->zone_id == zoneid);
4492	ushort_t flags;
4493
4494	mutex_enter(&zonehash_lock);
4495	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4496		mutex_exit(&zonehash_lock);
4497		return (set_errno(EINVAL));
4498	}
4499	zone_status = zone_status_get(zone);
4500	if (zone_status < ZONE_IS_INITIALIZED) {
4501		mutex_exit(&zonehash_lock);
4502		return (set_errno(EINVAL));
4503	}
4504	zone_hold(zone);
4505	mutex_exit(&zonehash_lock);
4506
4507	/*
4508	 * If not in the global zone, don't show information about other zones,
4509	 * unless the system is labeled and the local zone's label dominates
4510	 * the other zone.
4511	 */
4512	if (!zone_list_access(zone)) {
4513		zone_rele(zone);
4514		return (set_errno(EINVAL));
4515	}
4516
4517	switch (attr) {
4518	case ZONE_ATTR_ROOT:
4519		if (global) {
4520			/*
4521			 * Copy the path to trim the trailing "/" (except for
4522			 * the global zone).
4523			 */
4524			if (zone != global_zone)
4525				size = zone->zone_rootpathlen - 1;
4526			else
4527				size = zone->zone_rootpathlen;
4528			zonepath = kmem_alloc(size, KM_SLEEP);
4529			bcopy(zone->zone_rootpath, zonepath, size);
4530			zonepath[size - 1] = '\0';
4531		} else {
4532			if (inzone || !is_system_labeled()) {
4533				/*
4534				 * Caller is not in the global zone.
4535				 * if the query is on the current zone
4536				 * or the system is not labeled,
4537				 * just return faked-up path for current zone.
4538				 */
4539				zonepath = "/";
4540				size = 2;
4541			} else {
4542				/*
4543				 * Return related path for current zone.
4544				 */
4545				int prefix_len = strlen(zone_prefix);
4546				int zname_len = strlen(zone->zone_name);
4547
4548				size = prefix_len + zname_len + 1;
4549				zonepath = kmem_alloc(size, KM_SLEEP);
4550				bcopy(zone_prefix, zonepath, prefix_len);
4551				bcopy(zone->zone_name, zonepath +
4552				    prefix_len, zname_len);
4553				zonepath[size - 1] = '\0';
4554			}
4555		}
4556		if (bufsize > size)
4557			bufsize = size;
4558		if (buf != NULL) {
4559			err = copyoutstr(zonepath, buf, bufsize, NULL);
4560			if (err != 0 && err != ENAMETOOLONG)
4561				error = EFAULT;
4562		}
4563		if (global || (is_system_labeled() && !inzone))
4564			kmem_free(zonepath, size);
4565		break;
4566
4567	case ZONE_ATTR_NAME:
4568		size = strlen(zone->zone_name) + 1;
4569		if (bufsize > size)
4570			bufsize = size;
4571		if (buf != NULL) {
4572			err = copyoutstr(zone->zone_name, buf, bufsize, NULL);
4573			if (err != 0 && err != ENAMETOOLONG)
4574				error = EFAULT;
4575		}
4576		break;
4577
4578	case ZONE_ATTR_STATUS:
4579		/*
4580		 * Since we're not holding zonehash_lock, the zone status
4581		 * may be anything; leave it up to userland to sort it out.
4582		 */
4583		size = sizeof (zone_status);
4584		if (bufsize > size)
4585			bufsize = size;
4586		zone_status = zone_status_get(zone);
4587		if (buf != NULL &&
4588		    copyout(&zone_status, buf, bufsize) != 0)
4589			error = EFAULT;
4590		break;
4591	case ZONE_ATTR_FLAGS:
4592		size = sizeof (zone->zone_flags);
4593		if (bufsize > size)
4594			bufsize = size;
4595		flags = zone->zone_flags;
4596		if (buf != NULL &&
4597		    copyout(&flags, buf, bufsize) != 0)
4598			error = EFAULT;
4599		break;
4600	case ZONE_ATTR_PRIVSET:
4601		size = sizeof (priv_set_t);
4602		if (bufsize > size)
4603			bufsize = size;
4604		if (buf != NULL &&
4605		    copyout(zone->zone_privset, buf, bufsize) != 0)
4606			error = EFAULT;
4607		break;
4608	case ZONE_ATTR_UNIQID:
4609		size = sizeof (zone->zone_uniqid);
4610		if (bufsize > size)
4611			bufsize = size;
4612		if (buf != NULL &&
4613		    copyout(&zone->zone_uniqid, buf, bufsize) != 0)
4614			error = EFAULT;
4615		break;
4616	case ZONE_ATTR_POOLID:
4617		{
4618			pool_t *pool;
4619			poolid_t poolid;
4620
4621			if (pool_lock_intr() != 0) {
4622				error = EINTR;
4623				break;
4624			}
4625			pool = zone_pool_get(zone);
4626			poolid = pool->pool_id;
4627			pool_unlock();
4628			size = sizeof (poolid);
4629			if (bufsize > size)
4630				bufsize = size;
4631			if (buf != NULL && copyout(&poolid, buf, size) != 0)
4632				error = EFAULT;
4633		}
4634		break;
4635	case ZONE_ATTR_SLBL:
4636		size = sizeof (bslabel_t);
4637		if (bufsize > size)
4638			bufsize = size;
4639		if (zone->zone_slabel == NULL)
4640			error = EINVAL;
4641		else if (buf != NULL &&
4642		    copyout(label2bslabel(zone->zone_slabel), buf,
4643		    bufsize) != 0)
4644			error = EFAULT;
4645		break;
4646	case ZONE_ATTR_INITPID:
4647		size = sizeof (initpid);
4648		if (bufsize > size)
4649			bufsize = size;
4650		initpid = zone->zone_proc_initpid;
4651		if (initpid == -1) {
4652			error = ESRCH;
4653			break;
4654		}
4655		if (buf != NULL &&
4656		    copyout(&initpid, buf, bufsize) != 0)
4657			error = EFAULT;
4658		break;
4659	case ZONE_ATTR_BRAND:
4660		size = strlen(zone->zone_brand->b_name) + 1;
4661
4662		if (bufsize > size)
4663			bufsize = size;
4664		if (buf != NULL) {
4665			err = copyoutstr(zone->zone_brand->b_name, buf,
4666			    bufsize, NULL);
4667			if (err != 0 && err != ENAMETOOLONG)
4668				error = EFAULT;
4669		}
4670		break;
4671	case ZONE_ATTR_INITNAME:
4672		size = strlen(zone->zone_initname) + 1;
4673		if (bufsize > size)
4674			bufsize = size;
4675		if (buf != NULL) {
4676			err = copyoutstr(zone->zone_initname, buf, bufsize,
4677			    NULL);
4678			if (err != 0 && err != ENAMETOOLONG)
4679				error = EFAULT;
4680		}
4681		break;
4682	case ZONE_ATTR_BOOTARGS:
4683		if (zone->zone_bootargs == NULL)
4684			outstr = "";
4685		else
4686			outstr = zone->zone_bootargs;
4687		size = strlen(outstr) + 1;
4688		if (bufsize > size)
4689			bufsize = size;
4690		if (buf != NULL) {
4691			err = copyoutstr(outstr, buf, bufsize, NULL);
4692			if (err != 0 && err != ENAMETOOLONG)
4693				error = EFAULT;
4694		}
4695		break;
4696	case ZONE_ATTR_PHYS_MCAP:
4697		size = sizeof (zone->zone_phys_mcap);
4698		if (bufsize > size)
4699			bufsize = size;
4700		if (buf != NULL &&
4701		    copyout(&zone->zone_phys_mcap, buf, bufsize) != 0)
4702			error = EFAULT;
4703		break;
4704	case ZONE_ATTR_SCHED_CLASS:
4705		mutex_enter(&class_lock);
4706
4707		if (zone->zone_defaultcid >= loaded_classes)
4708			outstr = "";
4709		else
4710			outstr = sclass[zone->zone_defaultcid].cl_name;
4711		size = strlen(outstr) + 1;
4712		if (bufsize > size)
4713			bufsize = size;
4714		if (buf != NULL) {
4715			err = copyoutstr(outstr, buf, bufsize, NULL);
4716			if (err != 0 && err != ENAMETOOLONG)
4717				error = EFAULT;
4718		}
4719
4720		mutex_exit(&class_lock);
4721		break;
4722	default:
4723		if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone)) {
4724			size = bufsize;
4725			error = ZBROP(zone)->b_getattr(zone, attr, buf, &size);
4726		} else {
4727			error = EINVAL;
4728		}
4729	}
4730	zone_rele(zone);
4731
4732	if (error)
4733		return (set_errno(error));
4734	return ((ssize_t)size);
4735}
4736
4737/*
4738 * Systemcall entry point for zone_setattr(2).
4739 */
4740/*ARGSUSED*/
4741static int
4742zone_setattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
4743{
4744	zone_t *zone;
4745	zone_status_t zone_status;
4746	int err;
4747
4748	if (secpolicy_zone_config(CRED()) != 0)
4749		return (set_errno(EPERM));
4750
4751	/*
4752	 * Only the ZONE_ATTR_PHYS_MCAP attribute can be set on the
4753	 * global zone.
4754	 */
4755	if (zoneid == GLOBAL_ZONEID && attr != ZONE_ATTR_PHYS_MCAP) {
4756		return (set_errno(EINVAL));
4757	}
4758
4759	mutex_enter(&zonehash_lock);
4760	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4761		mutex_exit(&zonehash_lock);
4762		return (set_errno(EINVAL));
4763	}
4764	zone_hold(zone);
4765	mutex_exit(&zonehash_lock);
4766
4767	/*
4768	 * At present most attributes can only be set on non-running,
4769	 * non-global zones.
4770	 */
4771	zone_status = zone_status_get(zone);
4772	if (attr != ZONE_ATTR_PHYS_MCAP && zone_status > ZONE_IS_READY)
4773		goto done;
4774
4775	switch (attr) {
4776	case ZONE_ATTR_INITNAME:
4777		err = zone_set_initname(zone, (const char *)buf);
4778		break;
4779	case ZONE_ATTR_BOOTARGS:
4780		err = zone_set_bootargs(zone, (const char *)buf);
4781		break;
4782	case ZONE_ATTR_BRAND:
4783		err = zone_set_brand(zone, (const char *)buf);
4784		break;
4785	case ZONE_ATTR_PHYS_MCAP:
4786		err = zone_set_phys_mcap(zone, (const uint64_t *)buf);
4787		break;
4788	case ZONE_ATTR_SCHED_CLASS:
4789		err = zone_set_sched_class(zone, (const char *)buf);
4790		break;
4791	default:
4792		if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone))
4793			err = ZBROP(zone)->b_setattr(zone, attr, buf, bufsize);
4794		else
4795			err = EINVAL;
4796	}
4797
4798done:
4799	zone_rele(zone);
4800	return (err != 0 ? set_errno(err) : 0);
4801}
4802
4803/*
4804 * Return zero if the process has at least one vnode mapped in to its
4805 * address space which shouldn't be allowed to change zones.
4806 *
4807 * Also return zero if the process has any shared mappings which reserve
4808 * swap.  This is because the counting for zone.max-swap does not allow swap
4809 * reservation to be shared between zones.  zone swap reservation is counted
4810 * on zone->zone_max_swap.
4811 */
4812static int
4813as_can_change_zones(void)
4814{
4815	proc_t *pp = curproc;
4816	struct seg *seg;
4817	struct as *as = pp->p_as;
4818	vnode_t *vp;
4819	int allow = 1;
4820
4821	ASSERT(pp->p_as != &kas);
4822	AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
4823	for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) {
4824
4825		/*
4826		 * Cannot enter zone with shared anon memory which
4827		 * reserves swap.  See comment above.
4828		 */
4829		if (seg_can_change_zones(seg) == B_FALSE) {
4830			allow = 0;
4831			break;
4832		}
4833		/*
4834		 * if we can't get a backing vnode for this segment then skip
4835		 * it.
4836		 */
4837		vp = NULL;
4838		if (SEGOP_GETVP(seg, seg->s_base, &vp) != 0 || vp == NULL)
4839			continue;
4840		if (!vn_can_change_zones(vp)) { /* bail on first match */
4841			allow = 0;
4842			break;
4843		}
4844	}
4845	AS_LOCK_EXIT(as, &as->a_lock);
4846	return (allow);
4847}
4848
4849/*
4850 * Count swap reserved by curproc's address space
4851 */
4852static size_t
4853as_swresv(void)
4854{
4855	proc_t *pp = curproc;
4856	struct seg *seg;
4857	struct as *as = pp->p_as;
4858	size_t swap = 0;
4859
4860	ASSERT(pp->p_as != &kas);
4861	ASSERT(AS_WRITE_HELD(as, &as->a_lock));
4862	for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg))
4863		swap += seg_swresv(seg);
4864
4865	return (swap);
4866}
4867
4868/*
4869 * Systemcall entry point for zone_enter().
4870 *
4871 * The current process is injected into said zone.  In the process
4872 * it will change its project membership, privileges, rootdir/cwd,
4873 * zone-wide rctls, and pool association to match those of the zone.
4874 *
4875 * The first zone_enter() called while the zone is in the ZONE_IS_READY
4876 * state will transition it to ZONE_IS_RUNNING.  Processes may only
4877 * enter a zone that is "ready" or "running".
4878 */
4879static int
4880zone_enter(zoneid_t zoneid)
4881{
4882	zone_t *zone;
4883	vnode_t *vp;
4884	proc_t *pp = curproc;
4885	contract_t *ct;
4886	cont_process_t *ctp;
4887	task_t *tk, *oldtk;
4888	kproject_t *zone_proj0;
4889	cred_t *cr, *newcr;
4890	pool_t *oldpool, *newpool;
4891	sess_t *sp;
4892	uid_t uid;
4893	zone_status_t status;
4894	int err = 0;
4895	rctl_entity_p_t e;
4896	size_t swap;
4897	kthread_id_t t;
4898
4899	if (secpolicy_zone_config(CRED()) != 0)
4900		return (set_errno(EPERM));
4901	if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4902		return (set_errno(EINVAL));
4903
4904	/*
4905	 * Stop all lwps so we don't need to hold a lock to look at
4906	 * curproc->p_zone.  This needs to happen before we grab any
4907	 * locks to avoid deadlock (another lwp in the process could
4908	 * be waiting for the held lock).
4909	 */
4910	if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK))
4911		return (set_errno(EINTR));
4912
4913	/*
4914	 * Make sure we're not changing zones with files open or mapped in
4915	 * to our address space which shouldn't be changing zones.
4916	 */
4917	if (!files_can_change_zones()) {
4918		err = EBADF;
4919		goto out;
4920	}
4921	if (!as_can_change_zones()) {
4922		err = EFAULT;
4923		goto out;
4924	}
4925
4926	mutex_enter(&zonehash_lock);
4927	if (pp->p_zone != global_zone) {
4928		mutex_exit(&zonehash_lock);
4929		err = EINVAL;
4930		goto out;
4931	}
4932
4933	zone = zone_find_all_by_id(zoneid);
4934	if (zone == NULL) {
4935		mutex_exit(&zonehash_lock);
4936		err = EINVAL;
4937		goto out;
4938	}
4939
4940	/*
4941	 * To prevent processes in a zone from holding contracts on
4942	 * extrazonal resources, and to avoid process contract
4943	 * memberships which span zones, contract holders and processes
4944	 * which aren't the sole members of their encapsulating process
4945	 * contracts are not allowed to zone_enter.
4946	 */
4947	ctp = pp->p_ct_process;
4948	ct = &ctp->conp_contract;
4949	mutex_enter(&ct->ct_lock);
4950	mutex_enter(&pp->p_lock);
4951	if ((avl_numnodes(&pp->p_ct_held) != 0) || (ctp->conp_nmembers != 1)) {
4952		mutex_exit(&pp->p_lock);
4953		mutex_exit(&ct->ct_lock);
4954		mutex_exit(&zonehash_lock);
4955		err = EINVAL;
4956		goto out;
4957	}
4958
4959	/*
4960	 * Moreover, we don't allow processes whose encapsulating
4961	 * process contracts have inherited extrazonal contracts.
4962	 * While it would be easier to eliminate all process contracts
4963	 * with inherited contracts, we need to be able to give a
4964	 * restarted init (or other zone-penetrating process) its
4965	 * predecessor's contracts.
4966	 */
4967	if (ctp->conp_ninherited != 0) {
4968		contract_t *next;
4969		for (next = list_head(&ctp->conp_inherited); next;
4970		    next = list_next(&ctp->conp_inherited, next)) {
4971			if (contract_getzuniqid(next) != zone->zone_uniqid) {
4972				mutex_exit(&pp->p_lock);
4973				mutex_exit(&ct->ct_lock);
4974				mutex_exit(&zonehash_lock);
4975				err = EINVAL;
4976				goto out;
4977			}
4978		}
4979	}
4980
4981	mutex_exit(&pp->p_lock);
4982	mutex_exit(&ct->ct_lock);
4983
4984	status = zone_status_get(zone);
4985	if (status < ZONE_IS_READY || status >= ZONE_IS_SHUTTING_DOWN) {
4986		/*
4987		 * Can't join
4988		 */
4989		mutex_exit(&zonehash_lock);
4990		err = EINVAL;
4991		goto out;
4992	}
4993
4994	/*
4995	 * Make sure new priv set is within the permitted set for caller
4996	 */
4997	if (!priv_issubset(zone->zone_privset, &CR_OPPRIV(CRED()))) {
4998		mutex_exit(&zonehash_lock);
4999		err = EPERM;
5000		goto out;
5001	}
5002	/*
5003	 * We want to momentarily drop zonehash_lock while we optimistically
5004	 * bind curproc to the pool it should be running in.  This is safe
5005	 * since the zone can't disappear (we have a hold on it).
5006	 */
5007	zone_hold(zone);
5008	mutex_exit(&zonehash_lock);
5009
5010	/*
5011	 * Grab pool_lock to keep the pools configuration from changing
5012	 * and to stop ourselves from getting rebound to another pool
5013	 * until we join the zone.
5014	 */
5015	if (pool_lock_intr() != 0) {
5016		zone_rele(zone);
5017		err = EINTR;
5018		goto out;
5019	}
5020	ASSERT(secpolicy_pool(CRED()) == 0);
5021	/*
5022	 * Bind ourselves to the pool currently associated with the zone.
5023	 */
5024	oldpool = curproc->p_pool;
5025	newpool = zone_pool_get(zone);
5026	if (pool_state == POOL_ENABLED && newpool != oldpool &&
5027	    (err = pool_do_bind(newpool, P_PID, P_MYID,
5028	    POOL_BIND_ALL)) != 0) {
5029		pool_unlock();
5030		zone_rele(zone);
5031		goto out;
5032	}
5033
5034	/*
5035	 * Grab cpu_lock now; we'll need it later when we call
5036	 * task_join().
5037	 */
5038	mutex_enter(&cpu_lock);
5039	mutex_enter(&zonehash_lock);
5040	/*
5041	 * Make sure the zone hasn't moved on since we dropped zonehash_lock.
5042	 */
5043	if (zone_status_get(zone) >= ZONE_IS_SHUTTING_DOWN) {
5044		/*
5045		 * Can't join anymore.
5046		 */
5047		mutex_exit(&zonehash_lock);
5048		mutex_exit(&cpu_lock);
5049		if (pool_state == POOL_ENABLED &&
5050		    newpool != oldpool)
5051			(void) pool_do_bind(oldpool, P_PID, P_MYID,
5052			    POOL_BIND_ALL);
5053		pool_unlock();
5054		zone_rele(zone);
5055		err = EINVAL;
5056		goto out;
5057	}
5058
5059	/*
5060	 * a_lock must be held while transfering locked memory and swap
5061	 * reservation from the global zone to the non global zone because
5062	 * asynchronous faults on the processes' address space can lock
5063	 * memory and reserve swap via MCL_FUTURE and MAP_NORESERVE
5064	 * segments respectively.
5065	 */
5066	AS_LOCK_ENTER(pp->as, &pp->p_as->a_lock, RW_WRITER);
5067	swap = as_swresv();
5068	mutex_enter(&pp->p_lock);
5069	zone_proj0 = zone->zone_zsched->p_task->tk_proj;
5070	/* verify that we do not exceed and task or lwp limits */
5071	mutex_enter(&zone->zone_nlwps_lock);
5072	/* add new lwps to zone and zone's proj0 */
5073	zone_proj0->kpj_nlwps += pp->p_lwpcnt;
5074	zone->zone_nlwps += pp->p_lwpcnt;
5075	/* add 1 task to zone's proj0 */
5076	zone_proj0->kpj_ntasks += 1;
5077	mutex_exit(&zone->zone_nlwps_lock);
5078
5079	mutex_enter(&zone->zone_mem_lock);
5080	zone->zone_locked_mem += pp->p_locked_mem;
5081	zone_proj0->kpj_data.kpd_locked_mem += pp->p_locked_mem;
5082	zone->zone_max_swap += swap;
5083	mutex_exit(&zone->zone_mem_lock);
5084
5085	mutex_enter(&(zone_proj0->kpj_data.kpd_crypto_lock));
5086	zone_proj0->kpj_data.kpd_crypto_mem += pp->p_crypto_mem;
5087	mutex_exit(&(zone_proj0->kpj_data.kpd_crypto_lock));
5088
5089	/* remove lwps from proc's old zone and old project */
5090	mutex_enter(&pp->p_zone->zone_nlwps_lock);
5091	pp->p_zone->zone_nlwps -= pp->p_lwpcnt;
5092	pp->p_task->tk_proj->kpj_nlwps -= pp->p_lwpcnt;
5093	mutex_exit(&pp->p_zone->zone_nlwps_lock);
5094
5095	mutex_enter(&pp->p_zone->zone_mem_lock);
5096	pp->p_zone->zone_locked_mem -= pp->p_locked_mem;
5097	pp->p_task->tk_proj->kpj_data.kpd_locked_mem -= pp->p_locked_mem;
5098	pp->p_zone->zone_max_swap -= swap;
5099	mutex_exit(&pp->p_zone->zone_mem_lock);
5100
5101	mutex_enter(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock));
5102	pp->p_task->tk_proj->kpj_data.kpd_crypto_mem -= pp->p_crypto_mem;
5103	mutex_exit(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock));
5104
5105	mutex_exit(&pp->p_lock);
5106	AS_LOCK_EXIT(pp->p_as, &pp->p_as->a_lock);
5107
5108	/*
5109	 * Joining the zone cannot fail from now on.
5110	 *
5111	 * This means that a lot of the following code can be commonized and
5112	 * shared with zsched().
5113	 */
5114
5115	/*
5116	 * If the process contract fmri was inherited, we need to
5117	 * flag this so that any contract status will not leak
5118	 * extra zone information, svc_fmri in this case
5119	 */
5120	if (ctp->conp_svc_ctid != ct->ct_id) {
5121		mutex_enter(&ct->ct_lock);
5122		ctp->conp_svc_zone_enter = ct->ct_id;
5123		mutex_exit(&ct->ct_lock);
5124	}
5125
5126	/*
5127	 * Reset the encapsulating process contract's zone.
5128	 */
5129	ASSERT(ct->ct_mzuniqid == GLOBAL_ZONEUNIQID);
5130	contract_setzuniqid(ct, zone->zone_uniqid);
5131
5132	/*
5133	 * Create a new task and associate the process with the project keyed
5134	 * by (projid,zoneid).
5135	 *
5136	 * We might as well be in project 0; the global zone's projid doesn't
5137	 * make much sense in a zone anyhow.
5138	 *
5139	 * This also increments zone_ntasks, and returns with p_lock held.
5140	 */
5141	tk = task_create(0, zone);
5142	oldtk = task_join(tk, 0);
5143	mutex_exit(&cpu_lock);
5144
5145	pp->p_flag |= SZONETOP;
5146	pp->p_zone = zone;
5147
5148	/*
5149	 * call RCTLOP_SET functions on this proc
5150	 */
5151	e.rcep_p.zone = zone;
5152	e.rcep_t = RCENTITY_ZONE;
5153	(void) rctl_set_dup(NULL, NULL, pp, &e, zone->zone_rctls, NULL,
5154	    RCD_CALLBACK);
5155	mutex_exit(&pp->p_lock);
5156
5157	/*
5158	 * We don't need to hold any of zsched's locks here; not only do we know
5159	 * the process and zone aren't going away, we know its session isn't
5160	 * changing either.
5161	 *
5162	 * By joining zsched's session here, we mimic the behavior in the
5163	 * global zone of init's sid being the pid of sched.  We extend this
5164	 * to all zlogin-like zone_enter()'ing processes as well.
5165	 */
5166	mutex_enter(&pidlock);
5167	sp = zone->zone_zsched->p_sessp;
5168	sess_hold(zone->zone_zsched);
5169	mutex_enter(&pp->p_lock);
5170	pgexit(pp);
5171	sess_rele(pp->p_sessp, B_TRUE);
5172	pp->p_sessp = sp;
5173	pgjoin(pp, zone->zone_zsched->p_pidp);
5174
5175	/*
5176	 * If any threads are scheduled to be placed on zone wait queue they
5177	 * should abandon the idea since the wait queue is changing.
5178	 * We need to be holding pidlock & p_lock to do this.
5179	 */
5180	if ((t = pp->p_tlist) != NULL) {
5181		do {
5182			thread_lock(t);
5183			/*
5184			 * Kick this thread so that he doesn't sit
5185			 * on a wrong wait queue.
5186			 */
5187			if (ISWAITING(t))
5188				setrun_locked(t);
5189
5190			if (t->t_schedflag & TS_ANYWAITQ)
5191				t->t_schedflag &= ~ TS_ANYWAITQ;
5192
5193			thread_unlock(t);
5194		} while ((t = t->t_forw) != pp->p_tlist);
5195	}
5196
5197	/*
5198	 * If there is a default scheduling class for the zone and it is not
5199	 * the class we are currently in, change all of the threads in the
5200	 * process to the new class.  We need to be holding pidlock & p_lock
5201	 * when we call parmsset so this is a good place to do it.
5202	 */
5203	if (zone->zone_defaultcid > 0 &&
5204	    zone->zone_defaultcid != curthread->t_cid) {
5205		pcparms_t pcparms;
5206
5207		pcparms.pc_cid = zone->zone_defaultcid;
5208		pcparms.pc_clparms[0] = 0;
5209
5210		/*
5211		 * If setting the class fails, we still want to enter the zone.
5212		 */
5213		if ((t = pp->p_tlist) != NULL) {
5214			do {
5215				(void) parmsset(&pcparms, t);
5216			} while ((t = t->t_forw) != pp->p_tlist);
5217		}
5218	}
5219
5220	mutex_exit(&pp->p_lock);
5221	mutex_exit(&pidlock);
5222
5223	mutex_exit(&zonehash_lock);
5224	/*
5225	 * We're firmly in the zone; let pools progress.
5226	 */
5227	pool_unlock();
5228	task_rele(oldtk);
5229	/*
5230	 * We don't need to retain a hold on the zone since we already
5231	 * incremented zone_ntasks, so the zone isn't going anywhere.
5232	 */
5233	zone_rele(zone);
5234
5235	/*
5236	 * Chroot
5237	 */
5238	vp = zone->zone_rootvp;
5239	zone_chdir(vp, &PTOU(pp)->u_cdir, pp);
5240	zone_chdir(vp, &PTOU(pp)->u_rdir, pp);
5241
5242	/*
5243	 * Change process credentials
5244	 */
5245	newcr = cralloc();
5246	mutex_enter(&pp->p_crlock);
5247	cr = pp->p_cred;
5248	crcopy_to(cr, newcr);
5249	crsetzone(newcr, zone);
5250	pp->p_cred = newcr;
5251
5252	/*
5253	 * Restrict all process privilege sets to zone limit
5254	 */
5255	priv_intersect(zone->zone_privset, &CR_PPRIV(newcr));
5256	priv_intersect(zone->zone_privset, &CR_EPRIV(newcr));
5257	priv_intersect(zone->zone_privset, &CR_IPRIV(newcr));
5258	priv_intersect(zone->zone_privset, &CR_LPRIV(newcr));
5259	mutex_exit(&pp->p_crlock);
5260	crset(pp, newcr);
5261
5262	/*
5263	 * Adjust upcount to reflect zone entry.
5264	 */
5265	uid = crgetruid(newcr);
5266	mutex_enter(&pidlock);
5267	upcount_dec(uid, GLOBAL_ZONEID);
5268	upcount_inc(uid, zoneid);
5269	mutex_exit(&pidlock);
5270
5271	/*
5272	 * Set up core file path and content.
5273	 */
5274	set_core_defaults();
5275
5276out:
5277	/*
5278	 * Let the other lwps continue.
5279	 */
5280	mutex_enter(&pp->p_lock);
5281	if (curthread != pp->p_agenttp)
5282		continuelwps(pp);
5283	mutex_exit(&pp->p_lock);
5284
5285	return (err != 0 ? set_errno(err) : 0);
5286}
5287
5288/*
5289 * Systemcall entry point for zone_list(2).
5290 *
5291 * Processes running in a (non-global) zone only see themselves.
5292 * On labeled systems, they see all zones whose label they dominate.
5293 */
5294static int
5295zone_list(zoneid_t *zoneidlist, uint_t *numzones)
5296{
5297	zoneid_t *zoneids;
5298	zone_t *zone, *myzone;
5299	uint_t user_nzones, real_nzones;
5300	uint_t domi_nzones;
5301	int error;
5302
5303	if (copyin(numzones, &user_nzones, sizeof (uint_t)) != 0)
5304		return (set_errno(EFAULT));
5305
5306	myzone = curproc->p_zone;
5307	if (myzone != global_zone) {
5308		bslabel_t *mybslab;
5309
5310		if (!is_system_labeled()) {
5311			/* just return current zone */
5312			real_nzones = domi_nzones = 1;
5313			zoneids = kmem_alloc(sizeof (zoneid_t), KM_SLEEP);
5314			zoneids[0] = myzone->zone_id;
5315		} else {
5316			/* return all zones that are dominated */
5317			mutex_enter(&zonehash_lock);
5318			real_nzones = zonecount;
5319			domi_nzones = 0;
5320			if (real_nzones > 0) {
5321				zoneids = kmem_alloc(real_nzones *
5322				    sizeof (zoneid_t), KM_SLEEP);
5323				mybslab = label2bslabel(myzone->zone_slabel);
5324				for (zone = list_head(&zone_active);
5325				    zone != NULL;
5326				    zone = list_next(&zone_active, zone)) {
5327					if (zone->zone_id == GLOBAL_ZONEID)
5328						continue;
5329					if (zone != myzone &&
5330					    (zone->zone_flags & ZF_IS_SCRATCH))
5331						continue;
5332					/*
5333					 * Note that a label always dominates
5334					 * itself, so myzone is always included
5335					 * in the list.
5336					 */
5337					if (bldominates(mybslab,
5338					    label2bslabel(zone->zone_slabel))) {
5339						zoneids[domi_nzones++] =
5340						    zone->zone_id;
5341					}
5342				}
5343			}
5344			mutex_exit(&zonehash_lock);
5345		}
5346	} else {
5347		mutex_enter(&zonehash_lock);
5348		real_nzones = zonecount;
5349		domi_nzones = 0;
5350		if (real_nzones > 0) {
5351			zoneids = kmem_alloc(real_nzones * sizeof (zoneid_t),
5352			    KM_SLEEP);
5353			for (zone = list_head(&zone_active); zone != NULL;
5354			    zone = list_next(&zone_active, zone))
5355				zoneids[domi_nzones++] = zone->zone_id;
5356			ASSERT(domi_nzones == real_nzones);
5357		}
5358		mutex_exit(&zonehash_lock);
5359	}
5360
5361	/*
5362	 * If user has allocated space for fewer entries than we found, then
5363	 * return only up to his limit.  Either way, tell him exactly how many
5364	 * we found.
5365	 */
5366	if (domi_nzones < user_nzones)
5367		user_nzones = domi_nzones;
5368	error = 0;
5369	if (copyout(&domi_nzones, numzones, sizeof (uint_t)) != 0) {
5370		error = EFAULT;
5371	} else if (zoneidlist != NULL && user_nzones != 0) {
5372		if (copyout(zoneids, zoneidlist,
5373		    user_nzones * sizeof (zoneid_t)) != 0)
5374			error = EFAULT;
5375	}
5376
5377	if (real_nzones > 0)
5378		kmem_free(zoneids, real_nzones * sizeof (zoneid_t));
5379
5380	if (error != 0)
5381		return (set_errno(error));
5382	else
5383		return (0);
5384}
5385
5386/*
5387 * Systemcall entry point for zone_lookup(2).
5388 *
5389 * Non-global zones are only able to see themselves and (on labeled systems)
5390 * the zones they dominate.
5391 */
5392static zoneid_t
5393zone_lookup(const char *zone_name)
5394{
5395	char *kname;
5396	zone_t *zone;
5397	zoneid_t zoneid;
5398	int err;
5399
5400	if (zone_name == NULL) {
5401		/* return caller's zone id */
5402		return (getzoneid());
5403	}
5404
5405	kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
5406	if ((err = copyinstr(zone_name, kname, ZONENAME_MAX, NULL)) != 0) {
5407		kmem_free(kname, ZONENAME_MAX);
5408		return (set_errno(err));
5409	}
5410
5411	mutex_enter(&zonehash_lock);
5412	zone = zone_find_all_by_name(kname);
5413	kmem_free(kname, ZONENAME_MAX);
5414	/*
5415	 * In a non-global zone, can only lookup global and own name.
5416	 * In Trusted Extensions zone label dominance rules apply.
5417	 */
5418	if (zone == NULL ||
5419	    zone_status_get(zone) < ZONE_IS_READY ||
5420	    !zone_list_access(zone)) {
5421		mutex_exit(&zonehash_lock);
5422		return (set_errno(EINVAL));
5423	} else {
5424		zoneid = zone->zone_id;
5425		mutex_exit(&zonehash_lock);
5426		return (zoneid);
5427	}
5428}
5429
5430static int
5431zone_version(int *version_arg)
5432{
5433	int version = ZONE_SYSCALL_API_VERSION;
5434
5435	if (copyout(&version, version_arg, sizeof (int)) != 0)
5436		return (set_errno(EFAULT));
5437	return (0);
5438}
5439
5440/* ARGSUSED */
5441long
5442zone(int cmd, void *arg1, void *arg2, void *arg3, void *arg4)
5443{
5444	zone_def zs;
5445
5446	switch (cmd) {
5447	case ZONE_CREATE:
5448		if (get_udatamodel() == DATAMODEL_NATIVE) {
5449			if (copyin(arg1, &zs, sizeof (zone_def))) {
5450				return (set_errno(EFAULT));
5451			}
5452		} else {
5453#ifdef _SYSCALL32_IMPL
5454			zone_def32 zs32;
5455
5456			if (copyin(arg1, &zs32, sizeof (zone_def32))) {
5457				return (set_errno(EFAULT));
5458			}
5459			zs.zone_name =
5460			    (const char *)(unsigned long)zs32.zone_name;
5461			zs.zone_root =
5462			    (const char *)(unsigned long)zs32.zone_root;
5463			zs.zone_privs =
5464			    (const struct priv_set *)
5465			    (unsigned long)zs32.zone_privs;
5466			zs.zone_privssz = zs32.zone_privssz;
5467			zs.rctlbuf = (caddr_t)(unsigned long)zs32.rctlbuf;
5468			zs.rctlbufsz = zs32.rctlbufsz;
5469			zs.zfsbuf = (caddr_t)(unsigned long)zs32.zfsbuf;
5470			zs.zfsbufsz = zs32.zfsbufsz;
5471			zs.extended_error =
5472			    (int *)(unsigned long)zs32.extended_error;
5473			zs.match = zs32.match;
5474			zs.doi = zs32.doi;
5475			zs.label = (const bslabel_t *)(uintptr_t)zs32.label;
5476			zs.flags = zs32.flags;
5477#else
5478			panic("get_udatamodel() returned bogus result\n");
5479#endif
5480		}
5481
5482		return (zone_create(zs.zone_name, zs.zone_root,
5483		    zs.zone_privs, zs.zone_privssz,
5484		    (caddr_t)zs.rctlbuf, zs.rctlbufsz,
5485		    (caddr_t)zs.zfsbuf, zs.zfsbufsz,
5486		    zs.extended_error, zs.match, zs.doi,
5487		    zs.label, zs.flags));
5488	case ZONE_BOOT:
5489		return (zone_boot((zoneid_t)(uintptr_t)arg1));
5490	case ZONE_DESTROY:
5491		return (zone_destroy((zoneid_t)(uintptr_t)arg1));
5492	case ZONE_GETATTR:
5493		return (zone_getattr((zoneid_t)(uintptr_t)arg1,
5494		    (int)(uintptr_t)arg2, arg3, (size_t)arg4));
5495	case ZONE_SETATTR:
5496		return (zone_setattr((zoneid_t)(uintptr_t)arg1,
5497		    (int)(uintptr_t)arg2, arg3, (size_t)arg4));
5498	case ZONE_ENTER:
5499		return (zone_enter((zoneid_t)(uintptr_t)arg1));
5500	case ZONE_LIST:
5501		return (zone_list((zoneid_t *)arg1, (uint_t *)arg2));
5502	case ZONE_SHUTDOWN:
5503		return (zone_shutdown((zoneid_t)(uintptr_t)arg1));
5504	case ZONE_LOOKUP:
5505		return (zone_lookup((const char *)arg1));
5506	case ZONE_VERSION:
5507		return (zone_version((int *)arg1));
5508	case ZONE_ADD_DATALINK:
5509		return (zone_add_datalink((zoneid_t)(uintptr_t)arg1,
5510		    (char *)arg2));
5511	case ZONE_DEL_DATALINK:
5512		return (zone_remove_datalink((zoneid_t)(uintptr_t)arg1,
5513		    (char *)arg2));
5514	case ZONE_CHECK_DATALINK:
5515		return (zone_check_datalink((zoneid_t *)arg1, (char *)arg2));
5516	case ZONE_LIST_DATALINK:
5517		return (zone_list_datalink((zoneid_t)(uintptr_t)arg1,
5518		    (int *)arg2, (char *)arg3));
5519	default:
5520		return (set_errno(EINVAL));
5521	}
5522}
5523
5524struct zarg {
5525	zone_t *zone;
5526	zone_cmd_arg_t arg;
5527};
5528
5529static int
5530zone_lookup_door(const char *zone_name, door_handle_t *doorp)
5531{
5532	char *buf;
5533	size_t buflen;
5534	int error;
5535
5536	buflen = sizeof (ZONE_DOOR_PATH) + strlen(zone_name);
5537	buf = kmem_alloc(buflen, KM_SLEEP);
5538	(void) snprintf(buf, buflen, ZONE_DOOR_PATH, zone_name);
5539	error = door_ki_open(buf, doorp);
5540	kmem_free(buf, buflen);
5541	return (error);
5542}
5543
5544static void
5545zone_release_door(door_handle_t *doorp)
5546{
5547	door_ki_rele(*doorp);
5548	*doorp = NULL;
5549}
5550
5551static void
5552zone_ki_call_zoneadmd(struct zarg *zargp)
5553{
5554	door_handle_t door = NULL;
5555	door_arg_t darg, save_arg;
5556	char *zone_name;
5557	size_t zone_namelen;
5558	zoneid_t zoneid;
5559	zone_t *zone;
5560	zone_cmd_arg_t arg;
5561	uint64_t uniqid;
5562	size_t size;
5563	int error;
5564	int retry;
5565
5566	zone = zargp->zone;
5567	arg = zargp->arg;
5568	kmem_free(zargp, sizeof (*zargp));
5569
5570	zone_namelen = strlen(zone->zone_name) + 1;
5571	zone_name = kmem_alloc(zone_namelen, KM_SLEEP);
5572	bcopy(zone->zone_name, zone_name, zone_namelen);
5573	zoneid = zone->zone_id;
5574	uniqid = zone->zone_uniqid;
5575	/*
5576	 * zoneadmd may be down, but at least we can empty out the zone.
5577	 * We can ignore the return value of zone_empty() since we're called
5578	 * from a kernel thread and know we won't be delivered any signals.
5579	 */
5580	ASSERT(curproc == &p0);
5581	(void) zone_empty(zone);
5582	ASSERT(zone_status_get(zone) >= ZONE_IS_EMPTY);
5583	zone_rele(zone);
5584
5585	size = sizeof (arg);
5586	darg.rbuf = (char *)&arg;
5587	darg.data_ptr = (char *)&arg;
5588	darg.rsize = size;
5589	darg.data_size = size;
5590	darg.desc_ptr = NULL;
5591	darg.desc_num = 0;
5592
5593	save_arg = darg;
5594	/*
5595	 * Since we're not holding a reference to the zone, any number of
5596	 * things can go wrong, including the zone disappearing before we get a
5597	 * chance to talk to zoneadmd.
5598	 */
5599	for (retry = 0; /* forever */; retry++) {
5600		if (door == NULL &&
5601		    (error = zone_lookup_door(zone_name, &door)) != 0) {
5602			goto next;
5603		}
5604		ASSERT(door != NULL);
5605
5606		if ((error = door_ki_upcall(door, &darg)) == 0) {
5607			break;
5608		}
5609		switch (error) {
5610		case EINTR:
5611			/* FALLTHROUGH */
5612		case EAGAIN:	/* process may be forking */
5613			/*
5614			 * Back off for a bit
5615			 */
5616			break;
5617		case EBADF:
5618			zone_release_door(&door);
5619			if (zone_lookup_door(zone_name, &door) != 0) {
5620				/*
5621				 * zoneadmd may be dead, but it may come back to
5622				 * life later.
5623				 */
5624				break;
5625			}
5626			break;
5627		default:
5628			cmn_err(CE_WARN,
5629			    "zone_ki_call_zoneadmd: door_ki_upcall error %d\n",
5630			    error);
5631			goto out;
5632		}
5633next:
5634		/*
5635		 * If this isn't the same zone_t that we originally had in mind,
5636		 * then this is the same as if two kadmin requests come in at
5637		 * the same time: the first one wins.  This means we lose, so we
5638		 * bail.
5639		 */
5640		if ((zone = zone_find_by_id(zoneid)) == NULL) {
5641			/*
5642			 * Problem is solved.
5643			 */
5644			break;
5645		}
5646		if (zone->zone_uniqid != uniqid) {
5647			/*
5648			 * zoneid recycled
5649			 */
5650			zone_rele(zone);
5651			break;
5652		}
5653		/*
5654		 * We could zone_status_timedwait(), but there doesn't seem to
5655		 * be much point in doing that (plus, it would mean that
5656		 * zone_free() isn't called until this thread exits).
5657		 */
5658		zone_rele(zone);
5659		delay(hz);
5660		darg = save_arg;
5661	}
5662out:
5663	if (door != NULL) {
5664		zone_release_door(&door);
5665	}
5666	kmem_free(zone_name, zone_namelen);
5667	thread_exit();
5668}
5669
5670/*
5671 * Entry point for uadmin() to tell the zone to go away or reboot.  Analog to
5672 * kadmin().  The caller is a process in the zone.
5673 *
5674 * In order to shutdown the zone, we will hand off control to zoneadmd
5675 * (running in the global zone) via a door.  We do a half-hearted job at
5676 * killing all processes in the zone, create a kernel thread to contact
5677 * zoneadmd, and make note of the "uniqid" of the zone.  The uniqid is
5678 * a form of generation number used to let zoneadmd (as well as
5679 * zone_destroy()) know exactly which zone they're re talking about.
5680 */
5681int
5682zone_kadmin(int cmd, int fcn, const char *mdep, cred_t *credp)
5683{
5684	struct zarg *zargp;
5685	zone_cmd_t zcmd;
5686	zone_t *zone;
5687
5688	zone = curproc->p_zone;
5689	ASSERT(getzoneid() != GLOBAL_ZONEID);
5690
5691	switch (cmd) {
5692	case A_SHUTDOWN:
5693		switch (fcn) {
5694		case AD_HALT:
5695		case AD_POWEROFF:
5696			zcmd = Z_HALT;
5697			break;
5698		case AD_BOOT:
5699			zcmd = Z_REBOOT;
5700			break;
5701		case AD_IBOOT:
5702		case AD_SBOOT:
5703		case AD_SIBOOT:
5704		case AD_NOSYNC:
5705			return (ENOTSUP);
5706		default:
5707			return (EINVAL);
5708		}
5709		break;
5710	case A_REBOOT:
5711		zcmd = Z_REBOOT;
5712		break;
5713	case A_FTRACE:
5714	case A_REMOUNT:
5715	case A_FREEZE:
5716	case A_DUMP:
5717		return (ENOTSUP);
5718	default:
5719		ASSERT(cmd != A_SWAPCTL);	/* handled by uadmin() */
5720		return (EINVAL);
5721	}
5722
5723	if (secpolicy_zone_admin(credp, B_FALSE))
5724		return (EPERM);
5725	mutex_enter(&zone_status_lock);
5726
5727	/*
5728	 * zone_status can't be ZONE_IS_EMPTY or higher since curproc
5729	 * is in the zone.
5730	 */
5731	ASSERT(zone_status_get(zone) < ZONE_IS_EMPTY);
5732	if (zone_status_get(zone) > ZONE_IS_RUNNING) {
5733		/*
5734		 * This zone is already on its way down.
5735		 */
5736		mutex_exit(&zone_status_lock);
5737		return (0);
5738	}
5739	/*
5740	 * Prevent future zone_enter()s
5741	 */
5742	zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
5743	mutex_exit(&zone_status_lock);
5744
5745	/*
5746	 * Kill everyone now and call zoneadmd later.
5747	 * zone_ki_call_zoneadmd() will do a more thorough job of this
5748	 * later.
5749	 */
5750	killall(zone->zone_id);
5751	/*
5752	 * Now, create the thread to contact zoneadmd and do the rest of the
5753	 * work.  This thread can't be created in our zone otherwise
5754	 * zone_destroy() would deadlock.
5755	 */
5756	zargp = kmem_zalloc(sizeof (*zargp), KM_SLEEP);
5757	zargp->arg.cmd = zcmd;
5758	zargp->arg.uniqid = zone->zone_uniqid;
5759	zargp->zone = zone;
5760	(void) strcpy(zargp->arg.locale, "C");
5761	/* mdep was already copied in for us by uadmin */
5762	if (mdep != NULL)
5763		(void) strlcpy(zargp->arg.bootbuf, mdep,
5764		    sizeof (zargp->arg.bootbuf));
5765	zone_hold(zone);
5766
5767	(void) thread_create(NULL, 0, zone_ki_call_zoneadmd, zargp, 0, &p0,
5768	    TS_RUN, minclsyspri);
5769	exit(CLD_EXITED, 0);
5770
5771	return (EINVAL);
5772}
5773
5774/*
5775 * Entry point so kadmin(A_SHUTDOWN, ...) can set the global zone's
5776 * status to ZONE_IS_SHUTTING_DOWN.
5777 */
5778void
5779zone_shutdown_global(void)
5780{
5781	ASSERT(curproc->p_zone == global_zone);
5782
5783	mutex_enter(&zone_status_lock);
5784	ASSERT(zone_status_get(global_zone) == ZONE_IS_RUNNING);
5785	zone_status_set(global_zone, ZONE_IS_SHUTTING_DOWN);
5786	mutex_exit(&zone_status_lock);
5787}
5788
5789/*
5790 * Returns true if the named dataset is visible in the current zone.
5791 * The 'write' parameter is set to 1 if the dataset is also writable.
5792 */
5793int
5794zone_dataset_visible(const char *dataset, int *write)
5795{
5796	zone_dataset_t *zd;
5797	size_t len;
5798	zone_t *zone = curproc->p_zone;
5799
5800	if (dataset[0] == '\0')
5801		return (0);
5802
5803	/*
5804	 * Walk the list once, looking for datasets which match exactly, or
5805	 * specify a dataset underneath an exported dataset.  If found, return
5806	 * true and note that it is writable.
5807	 */
5808	for (zd = list_head(&zone->zone_datasets); zd != NULL;
5809	    zd = list_next(&zone->zone_datasets, zd)) {
5810
5811		len = strlen(zd->zd_dataset);
5812		if (strlen(dataset) >= len &&
5813		    bcmp(dataset, zd->zd_dataset, len) == 0 &&
5814		    (dataset[len] == '\0' || dataset[len] == '/' ||
5815		    dataset[len] == '@')) {
5816			if (write)
5817				*write = 1;
5818			return (1);
5819		}
5820	}
5821
5822	/*
5823	 * Walk the list a second time, searching for datasets which are parents
5824	 * of exported datasets.  These should be visible, but read-only.
5825	 *
5826	 * Note that we also have to support forms such as 'pool/dataset/', with
5827	 * a trailing slash.
5828	 */
5829	for (zd = list_head(&zone->zone_datasets); zd != NULL;
5830	    zd = list_next(&zone->zone_datasets, zd)) {
5831
5832		len = strlen(dataset);
5833		if (dataset[len - 1] == '/')
5834			len--;	/* Ignore trailing slash */
5835		if (len < strlen(zd->zd_dataset) &&
5836		    bcmp(dataset, zd->zd_dataset, len) == 0 &&
5837		    zd->zd_dataset[len] == '/') {
5838			if (write)
5839				*write = 0;
5840			return (1);
5841		}
5842	}
5843
5844	return (0);
5845}
5846
5847/*
5848 * zone_find_by_any_path() -
5849 *
5850 * kernel-private routine similar to zone_find_by_path(), but which
5851 * effectively compares against zone paths rather than zonerootpath
5852 * (i.e., the last component of zonerootpaths, which should be "root/",
5853 * are not compared.)  This is done in order to accurately identify all
5854 * paths, whether zone-visible or not, including those which are parallel
5855 * to /root/, such as /dev/, /home/, etc...
5856 *
5857 * If the specified path does not fall under any zone path then global
5858 * zone is returned.
5859 *
5860 * The treat_abs parameter indicates whether the path should be treated as
5861 * an absolute path although it does not begin with "/".  (This supports
5862 * nfs mount syntax such as host:any/path.)
5863 *
5864 * The caller is responsible for zone_rele of the returned zone.
5865 */
5866zone_t *
5867zone_find_by_any_path(const char *path, boolean_t treat_abs)
5868{
5869	zone_t *zone;
5870	int path_offset = 0;
5871
5872	if (path == NULL) {
5873		zone_hold(global_zone);
5874		return (global_zone);
5875	}
5876
5877	if (*path != '/') {
5878		ASSERT(treat_abs);
5879		path_offset = 1;
5880	}
5881
5882	mutex_enter(&zonehash_lock);
5883	for (zone = list_head(&zone_active); zone != NULL;
5884	    zone = list_next(&zone_active, zone)) {
5885		char	*c;
5886		size_t	pathlen;
5887		char *rootpath_start;
5888
5889		if (zone == global_zone)	/* skip global zone */
5890			continue;
5891
5892		/* scan backwards to find start of last component */
5893		c = zone->zone_rootpath + zone->zone_rootpathlen - 2;
5894		do {
5895			c--;
5896		} while (*c != '/');
5897
5898		pathlen = c - zone->zone_rootpath + 1 - path_offset;
5899		rootpath_start = (zone->zone_rootpath + path_offset);
5900		if (strncmp(path, rootpath_start, pathlen) == 0)
5901			break;
5902	}
5903	if (zone == NULL)
5904		zone = global_zone;
5905	zone_hold(zone);
5906	mutex_exit(&zonehash_lock);
5907	return (zone);
5908}
5909
5910/* List of data link names which are accessible from the zone */
5911struct dlnamelist {
5912	char			dlnl_name[LIFNAMSIZ];
5913	struct dlnamelist	*dlnl_next;
5914};
5915
5916
5917/*
5918 * Check whether the datalink name (dlname) itself is present.
5919 * Return true if found.
5920 */
5921static boolean_t
5922zone_dlname(zone_t *zone, char *dlname)
5923{
5924	struct dlnamelist *dlnl;
5925	boolean_t found = B_FALSE;
5926
5927	mutex_enter(&zone->zone_lock);
5928	for (dlnl = zone->zone_dl_list; dlnl != NULL; dlnl = dlnl->dlnl_next) {
5929		if (strncmp(dlnl->dlnl_name, dlname, LIFNAMSIZ) == 0) {
5930			found = B_TRUE;
5931			break;
5932		}
5933	}
5934	mutex_exit(&zone->zone_lock);
5935	return (found);
5936}
5937
5938/*
5939 * Add an data link name for the zone. Does not check for duplicates.
5940 */
5941static int
5942zone_add_datalink(zoneid_t zoneid, char *dlname)
5943{
5944	struct dlnamelist *dlnl;
5945	zone_t *zone;
5946	zone_t *thiszone;
5947	int err;
5948
5949	dlnl = kmem_zalloc(sizeof (struct dlnamelist), KM_SLEEP);
5950	if ((err = copyinstr(dlname, dlnl->dlnl_name, LIFNAMSIZ, NULL)) != 0) {
5951		kmem_free(dlnl, sizeof (struct dlnamelist));
5952		return (set_errno(err));
5953	}
5954
5955	thiszone = zone_find_by_id(zoneid);
5956	if (thiszone == NULL) {
5957		kmem_free(dlnl, sizeof (struct dlnamelist));
5958		return (set_errno(ENXIO));
5959	}
5960
5961	/*
5962	 * Verify that the datalink name isn't already used by a different
5963	 * zone while allowing duplicate entries for the same zone (e.g. due
5964	 * to both using IPv4 and IPv6 on an interface)
5965	 */
5966	mutex_enter(&zonehash_lock);
5967	for (zone = list_head(&zone_active); zone != NULL;
5968	    zone = list_next(&zone_active, zone)) {
5969		if (zone->zone_id == zoneid)
5970			continue;
5971
5972		if (zone_dlname(zone, dlnl->dlnl_name)) {
5973			mutex_exit(&zonehash_lock);
5974			zone_rele(thiszone);
5975			kmem_free(dlnl, sizeof (struct dlnamelist));
5976			return (set_errno(EPERM));
5977		}
5978	}
5979	mutex_enter(&thiszone->zone_lock);
5980	dlnl->dlnl_next = thiszone->zone_dl_list;
5981	thiszone->zone_dl_list = dlnl;
5982	mutex_exit(&thiszone->zone_lock);
5983	mutex_exit(&zonehash_lock);
5984	zone_rele(thiszone);
5985	return (0);
5986}
5987
5988static int
5989zone_remove_datalink(zoneid_t zoneid, char *dlname)
5990{
5991	struct dlnamelist *dlnl, *odlnl, **dlnlp;
5992	zone_t *zone;
5993	int err;
5994
5995	dlnl = kmem_zalloc(sizeof (struct dlnamelist), KM_SLEEP);
5996	if ((err = copyinstr(dlname, dlnl->dlnl_name, LIFNAMSIZ, NULL)) != 0) {
5997		kmem_free(dlnl, sizeof (struct dlnamelist));
5998		return (set_errno(err));
5999	}
6000	zone = zone_find_by_id(zoneid);
6001	if (zone == NULL) {
6002		kmem_free(dlnl, sizeof (struct dlnamelist));
6003		return (set_errno(EINVAL));
6004	}
6005
6006	mutex_enter(&zone->zone_lock);
6007	/* Look for match */
6008	dlnlp = &zone->zone_dl_list;
6009	while (*dlnlp != NULL) {
6010		if (strncmp(dlnl->dlnl_name, (*dlnlp)->dlnl_name,
6011		    LIFNAMSIZ) == 0)
6012			goto found;
6013		dlnlp = &((*dlnlp)->dlnl_next);
6014	}
6015	mutex_exit(&zone->zone_lock);
6016	zone_rele(zone);
6017	kmem_free(dlnl, sizeof (struct dlnamelist));
6018	return (set_errno(ENXIO));
6019
6020found:
6021	odlnl = *dlnlp;
6022	*dlnlp = (*dlnlp)->dlnl_next;
6023	kmem_free(odlnl, sizeof (struct dlnamelist));
6024
6025	mutex_exit(&zone->zone_lock);
6026	zone_rele(zone);
6027	kmem_free(dlnl, sizeof (struct dlnamelist));
6028	return (0);
6029}
6030
6031/*
6032 * Using the zoneidp as ALL_ZONES, we can lookup which zone is using datalink
6033 * name (dlname); otherwise we just check if the specified zoneidp has access
6034 * to the datalink name.
6035 */
6036static int
6037zone_check_datalink(zoneid_t *zoneidp, char *dlname)
6038{
6039	zoneid_t id;
6040	char *dln;
6041	zone_t *zone;
6042	int err = 0;
6043	boolean_t allzones = B_FALSE;
6044
6045	if (copyin(zoneidp, &id, sizeof (id)) != 0) {
6046		return (set_errno(EFAULT));
6047	}
6048	dln = kmem_zalloc(LIFNAMSIZ, KM_SLEEP);
6049	if ((err = copyinstr(dlname, dln, LIFNAMSIZ, NULL)) != 0) {
6050		kmem_free(dln, LIFNAMSIZ);
6051		return (set_errno(err));
6052	}
6053
6054	if (id == ALL_ZONES)
6055		allzones = B_TRUE;
6056
6057	/*
6058	 * Check whether datalink name is already used.
6059	 */
6060	mutex_enter(&zonehash_lock);
6061	for (zone = list_head(&zone_active); zone != NULL;
6062	    zone = list_next(&zone_active, zone)) {
6063		if (allzones || (id == zone->zone_id)) {
6064			if (!zone_dlname(zone, dln))
6065				continue;
6066			if (allzones)
6067				err = copyout(&zone->zone_id, zoneidp,
6068				    sizeof (*zoneidp));
6069
6070			mutex_exit(&zonehash_lock);
6071			kmem_free(dln, LIFNAMSIZ);
6072			return (err ? set_errno(EFAULT) : 0);
6073		}
6074	}
6075
6076	/* datalink name is not found in any active zone. */
6077	mutex_exit(&zonehash_lock);
6078	kmem_free(dln, LIFNAMSIZ);
6079	return (set_errno(ENXIO));
6080}
6081
6082/*
6083 * Get the names of the datalinks assigned to a zone.
6084 * Here *nump is the number of datalinks, and the assumption
6085 * is that the caller will guarantee that the the supplied buffer is
6086 * big enough to hold at least #*nump datalink names, that is,
6087 * LIFNAMSIZ X *nump
6088 * On return, *nump will be the "new" number of datalinks, if it
6089 * ever changed.
6090 */
6091static int
6092zone_list_datalink(zoneid_t zoneid, int *nump, char *buf)
6093{
6094	int num, dlcount;
6095	zone_t *zone;
6096	struct dlnamelist *dlnl;
6097	char *ptr;
6098
6099	if (copyin(nump, &dlcount, sizeof (dlcount)) != 0)
6100		return (set_errno(EFAULT));
6101
6102	zone = zone_find_by_id(zoneid);
6103	if (zone == NULL) {
6104		return (set_errno(ENXIO));
6105	}
6106
6107	num = 0;
6108	mutex_enter(&zone->zone_lock);
6109	ptr = buf;
6110	for (dlnl = zone->zone_dl_list; dlnl != NULL; dlnl = dlnl->dlnl_next) {
6111		/*
6112		 * If the list changed and the new number is bigger
6113		 * than what the caller supplied, just count, don't
6114		 * do copyout
6115		 */
6116		if (++num > dlcount)
6117			continue;
6118		if (copyout(dlnl->dlnl_name, ptr, LIFNAMSIZ) != 0) {
6119			mutex_exit(&zone->zone_lock);
6120			zone_rele(zone);
6121			return (set_errno(EFAULT));
6122		}
6123		ptr += LIFNAMSIZ;
6124	}
6125	mutex_exit(&zone->zone_lock);
6126	zone_rele(zone);
6127
6128	/* Increased or decreased, caller should be notified. */
6129	if (num != dlcount) {
6130		if (copyout(&num, nump, sizeof (num)) != 0) {
6131			return (set_errno(EFAULT));
6132		}
6133	}
6134	return (0);
6135}
6136
6137/*
6138 * Public interface for looking up a zone by zoneid. It's a customized version
6139 * for netstack_zone_create(). It can only be called from the zsd create
6140 * callbacks, since it doesn't have reference on the zone structure hence if
6141 * it is called elsewhere the zone could disappear after the zonehash_lock
6142 * is dropped.
6143 *
6144 * Furthermore it
6145 * 1. Doesn't check the status of the zone.
6146 * 2. It will be called even before zone_init is called, in that case the
6147 *    address of zone0 is returned directly, and netstack_zone_create()
6148 *    will only assign a value to zone0.zone_netstack, won't break anything.
6149 * 3. Returns without the zone being held.
6150 */
6151zone_t *
6152zone_find_by_id_nolock(zoneid_t zoneid)
6153{
6154	zone_t *zone;
6155
6156	mutex_enter(&zonehash_lock);
6157	if (zonehashbyid == NULL)
6158		zone = &zone0;
6159	else
6160		zone = zone_find_all_by_id(zoneid);
6161	mutex_exit(&zonehash_lock);
6162	return (zone);
6163}
6164
6165/*
6166 * Walk the datalinks for a given zone
6167 */
6168int
6169zone_datalink_walk(zoneid_t zoneid, int (*cb)(const char *, void *), void *data)
6170{
6171	zone_t *zone;
6172	struct dlnamelist *dlnl;
6173	int ret = 0;
6174
6175	if ((zone = zone_find_by_id(zoneid)) == NULL)
6176		return (ENOENT);
6177
6178	mutex_enter(&zone->zone_lock);
6179	for (dlnl = zone->zone_dl_list; dlnl != NULL; dlnl = dlnl->dlnl_next) {
6180		if ((ret = (*cb)(dlnl->dlnl_name, data)) != 0)
6181			break;
6182	}
6183	mutex_exit(&zone->zone_lock);
6184	zone_rele(zone);
6185	return (ret);
6186}
6187