xref: /illumos-gate/usr/src/uts/common/os/zone.c (revision 2918c4a3)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright 2015, Joyent Inc. All rights reserved.
25  * Copyright (c) 2016 by Delphix. All rights reserved.
26  * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
27  */
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  *   zone_rctl_lock: This is a per-zone lock used to protect other rctls,
166  *       currently just max_lofi
167  *   zsd_key_lock: This is a global lock protecting the key state for ZSD.
168  *   zone_deathrow_lock: This is a global lock protecting the "deathrow"
169  *       list (a list of zones in the ZONE_IS_DEAD state).
170  *
171  *   Ordering requirements:
172  *       pool_lock --> cpu_lock --> zonehash_lock --> zone_status_lock -->
173  *       zone_lock --> zsd_key_lock --> pidlock --> p_lock
174  *
175  *   When taking zone_mem_lock or zone_nlwps_lock, the lock ordering is:
176  *	zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_mem_lock
177  *	zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_nlwps_lock
178  *
179  *   Blocking memory allocations are permitted while holding any of the
180  *   zone locks.
181  *
182  *
183  *   System Call Interface:
184  *
185  *   The zone subsystem can be managed and queried from user level with
186  *   the following system calls (all subcodes of the primary "zone"
187  *   system call):
188  *   - zone_create: creates a zone with selected attributes (name,
189  *     root path, privileges, resource controls, ZFS datasets)
190  *   - zone_enter: allows the current process to enter a zone
191  *   - zone_getattr: reports attributes of a zone
192  *   - zone_setattr: set attributes of a zone
193  *   - zone_boot: set 'init' running for the zone
194  *   - zone_list: lists all zones active in the system
195  *   - zone_lookup: looks up zone id based on name
196  *   - zone_shutdown: initiates shutdown process (see states above)
197  *   - zone_destroy: completes shutdown process (see states above)
198  *
199  */
200 
201 #include <sys/priv_impl.h>
202 #include <sys/cred.h>
203 #include <c2/audit.h>
204 #include <sys/debug.h>
205 #include <sys/file.h>
206 #include <sys/kmem.h>
207 #include <sys/kstat.h>
208 #include <sys/mutex.h>
209 #include <sys/note.h>
210 #include <sys/pathname.h>
211 #include <sys/proc.h>
212 #include <sys/project.h>
213 #include <sys/sysevent.h>
214 #include <sys/task.h>
215 #include <sys/systm.h>
216 #include <sys/types.h>
217 #include <sys/utsname.h>
218 #include <sys/vnode.h>
219 #include <sys/vfs.h>
220 #include <sys/systeminfo.h>
221 #include <sys/policy.h>
222 #include <sys/cred_impl.h>
223 #include <sys/contract_impl.h>
224 #include <sys/contract/process_impl.h>
225 #include <sys/class.h>
226 #include <sys/pool.h>
227 #include <sys/pool_pset.h>
228 #include <sys/pset.h>
229 #include <sys/strlog.h>
230 #include <sys/sysmacros.h>
231 #include <sys/callb.h>
232 #include <sys/vmparam.h>
233 #include <sys/corectl.h>
234 #include <sys/ipc_impl.h>
235 #include <sys/klpd.h>
236 
237 #include <sys/door.h>
238 #include <sys/cpuvar.h>
239 #include <sys/sdt.h>
240 
241 #include <sys/uadmin.h>
242 #include <sys/session.h>
243 #include <sys/cmn_err.h>
244 #include <sys/modhash.h>
245 #include <sys/sunddi.h>
246 #include <sys/nvpair.h>
247 #include <sys/rctl.h>
248 #include <sys/fss.h>
249 #include <sys/brand.h>
250 #include <sys/zone.h>
251 #include <net/if.h>
252 #include <sys/cpucaps.h>
253 #include <vm/seg.h>
254 #include <sys/mac.h>
255 
256 /*
257  * This constant specifies the number of seconds that threads waiting for
258  * subsystems to release a zone's general-purpose references will wait before
259  * they log the zone's reference counts.  The constant's value shouldn't
260  * be so small that reference counts are unnecessarily reported for zones
261  * whose references are slowly released.  On the other hand, it shouldn't be so
262  * large that users reboot their systems out of frustration over hung zones
263  * before the system logs the zones' reference counts.
264  */
265 #define	ZONE_DESTROY_TIMEOUT_SECS	60
266 
267 /* List of data link IDs which are accessible from the zone */
268 typedef struct zone_dl {
269 	datalink_id_t	zdl_id;
270 	nvlist_t	*zdl_net;
271 	list_node_t	zdl_linkage;
272 } zone_dl_t;
273 
274 /*
275  * cv used to signal that all references to the zone have been released.  This
276  * needs to be global since there may be multiple waiters, and the first to
277  * wake up will free the zone_t, hence we cannot use zone->zone_cv.
278  */
279 static kcondvar_t zone_destroy_cv;
280 /*
281  * Lock used to serialize access to zone_cv.  This could have been per-zone,
282  * but then we'd need another lock for zone_destroy_cv, and why bother?
283  */
284 static kmutex_t zone_status_lock;
285 
286 /*
287  * ZSD-related global variables.
288  */
289 static kmutex_t zsd_key_lock;	/* protects the following two */
290 /*
291  * The next caller of zone_key_create() will be assigned a key of ++zsd_keyval.
292  */
293 static zone_key_t zsd_keyval = 0;
294 /*
295  * Global list of registered keys.  We use this when a new zone is created.
296  */
297 static list_t zsd_registered_keys;
298 
299 int zone_hash_size = 256;
300 static mod_hash_t *zonehashbyname, *zonehashbyid, *zonehashbylabel;
301 static kmutex_t zonehash_lock;
302 static uint_t zonecount;
303 static id_space_t *zoneid_space;
304 
305 /*
306  * The global zone (aka zone0) is the all-seeing, all-knowing zone in which the
307  * kernel proper runs, and which manages all other zones.
308  *
309  * Although not declared as static, the variable "zone0" should not be used
310  * except for by code that needs to reference the global zone early on in boot,
311  * before it is fully initialized.  All other consumers should use
312  * 'global_zone'.
313  */
314 zone_t zone0;
315 zone_t *global_zone = NULL;	/* Set when the global zone is initialized */
316 
317 /*
318  * List of active zones, protected by zonehash_lock.
319  */
320 static list_t zone_active;
321 
322 /*
323  * List of destroyed zones that still have outstanding cred references.
324  * Used for debugging.  Uses a separate lock to avoid lock ordering
325  * problems in zone_free.
326  */
327 static list_t zone_deathrow;
328 static kmutex_t zone_deathrow_lock;
329 
330 /* number of zones is limited by virtual interface limit in IP */
331 uint_t maxzones = 8192;
332 
333 /* Event channel to sent zone state change notifications */
334 evchan_t *zone_event_chan;
335 
336 /*
337  * This table holds the mapping from kernel zone states to
338  * states visible in the state notification API.
339  * The idea is that we only expose "obvious" states and
340  * do not expose states which are just implementation details.
341  */
342 const char  *zone_status_table[] = {
343 	ZONE_EVENT_UNINITIALIZED,	/* uninitialized */
344 	ZONE_EVENT_INITIALIZED,		/* initialized */
345 	ZONE_EVENT_READY,		/* ready */
346 	ZONE_EVENT_READY,		/* booting */
347 	ZONE_EVENT_RUNNING,		/* running */
348 	ZONE_EVENT_SHUTTING_DOWN,	/* shutting_down */
349 	ZONE_EVENT_SHUTTING_DOWN,	/* empty */
350 	ZONE_EVENT_SHUTTING_DOWN,	/* down */
351 	ZONE_EVENT_SHUTTING_DOWN,	/* dying */
352 	ZONE_EVENT_UNINITIALIZED,	/* dead */
353 };
354 
355 /*
356  * This array contains the names of the subsystems listed in zone_ref_subsys_t
357  * (see sys/zone.h).
358  */
359 static char *zone_ref_subsys_names[] = {
360 	"NFS",		/* ZONE_REF_NFS */
361 	"NFSv4",	/* ZONE_REF_NFSV4 */
362 	"SMBFS",	/* ZONE_REF_SMBFS */
363 	"MNTFS",	/* ZONE_REF_MNTFS */
364 	"LOFI",		/* ZONE_REF_LOFI */
365 	"VFS",		/* ZONE_REF_VFS */
366 	"IPC"		/* ZONE_REF_IPC */
367 };
368 
369 /*
370  * This isn't static so lint doesn't complain.
371  */
372 rctl_hndl_t rc_zone_cpu_shares;
373 rctl_hndl_t rc_zone_locked_mem;
374 rctl_hndl_t rc_zone_max_swap;
375 rctl_hndl_t rc_zone_max_lofi;
376 rctl_hndl_t rc_zone_cpu_cap;
377 rctl_hndl_t rc_zone_nlwps;
378 rctl_hndl_t rc_zone_nprocs;
379 rctl_hndl_t rc_zone_shmmax;
380 rctl_hndl_t rc_zone_shmmni;
381 rctl_hndl_t rc_zone_semmni;
382 rctl_hndl_t rc_zone_msgmni;
383 
384 const char * const zone_default_initname = "/sbin/init";
385 static char * const zone_prefix = "/zone/";
386 static int zone_shutdown(zoneid_t zoneid);
387 static int zone_add_datalink(zoneid_t, datalink_id_t);
388 static int zone_remove_datalink(zoneid_t, datalink_id_t);
389 static int zone_list_datalink(zoneid_t, int *, datalink_id_t *);
390 static int zone_set_network(zoneid_t, zone_net_data_t *);
391 static int zone_get_network(zoneid_t, zone_net_data_t *);
392 
393 typedef boolean_t zsd_applyfn_t(kmutex_t *, boolean_t, zone_t *, zone_key_t);
394 
395 static void zsd_apply_all_zones(zsd_applyfn_t *, zone_key_t);
396 static void zsd_apply_all_keys(zsd_applyfn_t *, zone_t *);
397 static boolean_t zsd_apply_create(kmutex_t *, boolean_t, zone_t *, zone_key_t);
398 static boolean_t zsd_apply_shutdown(kmutex_t *, boolean_t, zone_t *,
399     zone_key_t);
400 static boolean_t zsd_apply_destroy(kmutex_t *, boolean_t, zone_t *, zone_key_t);
401 static boolean_t zsd_wait_for_creator(zone_t *, struct zsd_entry *,
402     kmutex_t *);
403 static boolean_t zsd_wait_for_inprogress(zone_t *, struct zsd_entry *,
404     kmutex_t *);
405 
406 /*
407  * Bump this number when you alter the zone syscall interfaces; this is
408  * because we need to have support for previous API versions in libc
409  * to support patching; libc calls into the kernel to determine this number.
410  *
411  * Version 1 of the API is the version originally shipped with Solaris 10
412  * Version 2 alters the zone_create system call in order to support more
413  *     arguments by moving the args into a structure; and to do better
414  *     error reporting when zone_create() fails.
415  * Version 3 alters the zone_create system call in order to support the
416  *     import of ZFS datasets to zones.
417  * Version 4 alters the zone_create system call in order to support
418  *     Trusted Extensions.
419  * Version 5 alters the zone_boot system call, and converts its old
420  *     bootargs parameter to be set by the zone_setattr API instead.
421  * Version 6 adds the flag argument to zone_create.
422  */
423 static const int ZONE_SYSCALL_API_VERSION = 6;
424 
425 /*
426  * Certain filesystems (such as NFS and autofs) need to know which zone
427  * the mount is being placed in.  Because of this, we need to be able to
428  * ensure that a zone isn't in the process of being created/destroyed such
429  * that nfs_mount() thinks it is in the global/NGZ zone, while by the time
430  * it gets added the list of mounted zones, it ends up on the wrong zone's
431  * mount list. Since a zone can't reside on an NFS file system, we don't
432  * have to worry about the zonepath itself.
433  *
434  * The following functions: block_mounts()/resume_mounts() and
435  * mount_in_progress()/mount_completed() are used by zones and the VFS
436  * layer (respectively) to synchronize zone state transitions and new
437  * mounts within a zone. This syncronization is on a per-zone basis, so
438  * activity for one zone will not interfere with activity for another zone.
439  *
440  * The semantics are like a reader-reader lock such that there may
441  * either be multiple mounts (or zone state transitions, if that weren't
442  * serialized by zonehash_lock) in progress at the same time, but not
443  * both.
444  *
445  * We use cv's so the user can ctrl-C out of the operation if it's
446  * taking too long.
447  *
448  * The semantics are such that there is unfair bias towards the
449  * "current" operation.  This means that zone halt may starve if
450  * there is a rapid succession of new mounts coming in to the zone.
451  */
452 /*
453  * Prevent new mounts from progressing to the point of calling
454  * VFS_MOUNT().  If there are already mounts in this "region", wait for
455  * them to complete.
456  */
457 static int
458 block_mounts(zone_t *zp)
459 {
460 	int retval = 0;
461 
462 	/*
463 	 * Since it may block for a long time, block_mounts() shouldn't be
464 	 * called with zonehash_lock held.
465 	 */
466 	ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
467 	mutex_enter(&zp->zone_mount_lock);
468 	while (zp->zone_mounts_in_progress > 0) {
469 		if (cv_wait_sig(&zp->zone_mount_cv, &zp->zone_mount_lock) == 0)
470 			goto signaled;
471 	}
472 	/*
473 	 * A negative value of mounts_in_progress indicates that mounts
474 	 * have been blocked by (-mounts_in_progress) different callers
475 	 * (remotely possible if two threads enter zone_shutdown at the same
476 	 * time).
477 	 */
478 	zp->zone_mounts_in_progress--;
479 	retval = 1;
480 signaled:
481 	mutex_exit(&zp->zone_mount_lock);
482 	return (retval);
483 }
484 
485 /*
486  * The VFS layer may progress with new mounts as far as we're concerned.
487  * Allow them to progress if we were the last obstacle.
488  */
489 static void
490 resume_mounts(zone_t *zp)
491 {
492 	mutex_enter(&zp->zone_mount_lock);
493 	if (++zp->zone_mounts_in_progress == 0)
494 		cv_broadcast(&zp->zone_mount_cv);
495 	mutex_exit(&zp->zone_mount_lock);
496 }
497 
498 /*
499  * The VFS layer is busy with a mount; this zone should wait until all
500  * of its mounts are completed to progress.
501  */
502 void
503 mount_in_progress(zone_t *zp)
504 {
505 	mutex_enter(&zp->zone_mount_lock);
506 	while (zp->zone_mounts_in_progress < 0)
507 		cv_wait(&zp->zone_mount_cv, &zp->zone_mount_lock);
508 	zp->zone_mounts_in_progress++;
509 	mutex_exit(&zp->zone_mount_lock);
510 }
511 
512 /*
513  * VFS is done with one mount; wake up any waiting block_mounts()
514  * callers if this is the last mount.
515  */
516 void
517 mount_completed(zone_t *zp)
518 {
519 	mutex_enter(&zp->zone_mount_lock);
520 	if (--zp->zone_mounts_in_progress == 0)
521 		cv_broadcast(&zp->zone_mount_cv);
522 	mutex_exit(&zp->zone_mount_lock);
523 }
524 
525 /*
526  * ZSD routines.
527  *
528  * Zone Specific Data (ZSD) is modeled after Thread Specific Data as
529  * defined by the pthread_key_create() and related interfaces.
530  *
531  * Kernel subsystems may register one or more data items and/or
532  * callbacks to be executed when a zone is created, shutdown, or
533  * destroyed.
534  *
535  * Unlike the thread counterpart, destructor callbacks will be executed
536  * even if the data pointer is NULL and/or there are no constructor
537  * callbacks, so it is the responsibility of such callbacks to check for
538  * NULL data values if necessary.
539  *
540  * The locking strategy and overall picture is as follows:
541  *
542  * When someone calls zone_key_create(), a template ZSD entry is added to the
543  * global list "zsd_registered_keys", protected by zsd_key_lock.  While
544  * holding that lock all the existing zones are marked as
545  * ZSD_CREATE_NEEDED and a copy of the ZSD entry added to the per-zone
546  * zone_zsd list (protected by zone_lock). The global list is updated first
547  * (under zone_key_lock) to make sure that newly created zones use the
548  * most recent list of keys. Then under zonehash_lock we walk the zones
549  * and mark them.  Similar locking is used in zone_key_delete().
550  *
551  * The actual create, shutdown, and destroy callbacks are done without
552  * holding any lock. And zsd_flags are used to ensure that the operations
553  * completed so that when zone_key_create (and zone_create) is done, as well as
554  * zone_key_delete (and zone_destroy) is done, all the necessary callbacks
555  * are completed.
556  *
557  * When new zones are created constructor callbacks for all registered ZSD
558  * entries will be called. That also uses the above two phases of marking
559  * what needs to be done, and then running the callbacks without holding
560  * any locks.
561  *
562  * The framework does not provide any locking around zone_getspecific() and
563  * zone_setspecific() apart from that needed for internal consistency, so
564  * callers interested in atomic "test-and-set" semantics will need to provide
565  * their own locking.
566  */
567 
568 /*
569  * Helper function to find the zsd_entry associated with the key in the
570  * given list.
571  */
572 static struct zsd_entry *
573 zsd_find(list_t *l, zone_key_t key)
574 {
575 	struct zsd_entry *zsd;
576 
577 	for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
578 		if (zsd->zsd_key == key) {
579 			return (zsd);
580 		}
581 	}
582 	return (NULL);
583 }
584 
585 /*
586  * Helper function to find the zsd_entry associated with the key in the
587  * given list. Move it to the front of the list.
588  */
589 static struct zsd_entry *
590 zsd_find_mru(list_t *l, zone_key_t key)
591 {
592 	struct zsd_entry *zsd;
593 
594 	for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
595 		if (zsd->zsd_key == key) {
596 			/*
597 			 * Move to head of list to keep list in MRU order.
598 			 */
599 			if (zsd != list_head(l)) {
600 				list_remove(l, zsd);
601 				list_insert_head(l, zsd);
602 			}
603 			return (zsd);
604 		}
605 	}
606 	return (NULL);
607 }
608 
609 void
610 zone_key_create(zone_key_t *keyp, void *(*create)(zoneid_t),
611     void (*shutdown)(zoneid_t, void *), void (*destroy)(zoneid_t, void *))
612 {
613 	struct zsd_entry *zsdp;
614 	struct zsd_entry *t;
615 	struct zone *zone;
616 	zone_key_t  key;
617 
618 	zsdp = kmem_zalloc(sizeof (*zsdp), KM_SLEEP);
619 	zsdp->zsd_data = NULL;
620 	zsdp->zsd_create = create;
621 	zsdp->zsd_shutdown = shutdown;
622 	zsdp->zsd_destroy = destroy;
623 
624 	/*
625 	 * Insert in global list of callbacks. Makes future zone creations
626 	 * see it.
627 	 */
628 	mutex_enter(&zsd_key_lock);
629 	key = zsdp->zsd_key = ++zsd_keyval;
630 	ASSERT(zsd_keyval != 0);
631 	list_insert_tail(&zsd_registered_keys, zsdp);
632 	mutex_exit(&zsd_key_lock);
633 
634 	/*
635 	 * Insert for all existing zones and mark them as needing
636 	 * a create callback.
637 	 */
638 	mutex_enter(&zonehash_lock);	/* stop the world */
639 	for (zone = list_head(&zone_active); zone != NULL;
640 	    zone = list_next(&zone_active, zone)) {
641 		zone_status_t status;
642 
643 		mutex_enter(&zone->zone_lock);
644 
645 		/* Skip zones that are on the way down or not yet up */
646 		status = zone_status_get(zone);
647 		if (status >= ZONE_IS_DOWN ||
648 		    status == ZONE_IS_UNINITIALIZED) {
649 			mutex_exit(&zone->zone_lock);
650 			continue;
651 		}
652 
653 		t = zsd_find_mru(&zone->zone_zsd, key);
654 		if (t != NULL) {
655 			/*
656 			 * A zsd_configure already inserted it after
657 			 * we dropped zsd_key_lock above.
658 			 */
659 			mutex_exit(&zone->zone_lock);
660 			continue;
661 		}
662 		t = kmem_zalloc(sizeof (*t), KM_SLEEP);
663 		t->zsd_key = key;
664 		t->zsd_create = create;
665 		t->zsd_shutdown = shutdown;
666 		t->zsd_destroy = destroy;
667 		if (create != NULL) {
668 			t->zsd_flags = ZSD_CREATE_NEEDED;
669 			DTRACE_PROBE2(zsd__create__needed,
670 			    zone_t *, zone, zone_key_t, key);
671 		}
672 		list_insert_tail(&zone->zone_zsd, t);
673 		mutex_exit(&zone->zone_lock);
674 	}
675 	mutex_exit(&zonehash_lock);
676 
677 	if (create != NULL) {
678 		/* Now call the create callback for this key */
679 		zsd_apply_all_zones(zsd_apply_create, key);
680 	}
681 	/*
682 	 * It is safe for consumers to use the key now, make it
683 	 * globally visible. Specifically zone_getspecific() will
684 	 * always successfully return the zone specific data associated
685 	 * with the key.
686 	 */
687 	*keyp = key;
688 
689 }
690 
691 /*
692  * Function called when a module is being unloaded, or otherwise wishes
693  * to unregister its ZSD key and callbacks.
694  *
695  * Remove from the global list and determine the functions that need to
696  * be called under a global lock. Then call the functions without
697  * holding any locks. Finally free up the zone_zsd entries. (The apply
698  * functions need to access the zone_zsd entries to find zsd_data etc.)
699  */
700 int
701 zone_key_delete(zone_key_t key)
702 {
703 	struct zsd_entry *zsdp = NULL;
704 	zone_t *zone;
705 
706 	mutex_enter(&zsd_key_lock);
707 	zsdp = zsd_find_mru(&zsd_registered_keys, key);
708 	if (zsdp == NULL) {
709 		mutex_exit(&zsd_key_lock);
710 		return (-1);
711 	}
712 	list_remove(&zsd_registered_keys, zsdp);
713 	mutex_exit(&zsd_key_lock);
714 
715 	mutex_enter(&zonehash_lock);
716 	for (zone = list_head(&zone_active); zone != NULL;
717 	    zone = list_next(&zone_active, zone)) {
718 		struct zsd_entry *del;
719 
720 		mutex_enter(&zone->zone_lock);
721 		del = zsd_find_mru(&zone->zone_zsd, key);
722 		if (del == NULL) {
723 			/*
724 			 * Somebody else got here first e.g the zone going
725 			 * away.
726 			 */
727 			mutex_exit(&zone->zone_lock);
728 			continue;
729 		}
730 		ASSERT(del->zsd_shutdown == zsdp->zsd_shutdown);
731 		ASSERT(del->zsd_destroy == zsdp->zsd_destroy);
732 		if (del->zsd_shutdown != NULL &&
733 		    (del->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
734 			del->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
735 			DTRACE_PROBE2(zsd__shutdown__needed,
736 			    zone_t *, zone, zone_key_t, key);
737 		}
738 		if (del->zsd_destroy != NULL &&
739 		    (del->zsd_flags & ZSD_DESTROY_ALL) == 0) {
740 			del->zsd_flags |= ZSD_DESTROY_NEEDED;
741 			DTRACE_PROBE2(zsd__destroy__needed,
742 			    zone_t *, zone, zone_key_t, key);
743 		}
744 		mutex_exit(&zone->zone_lock);
745 	}
746 	mutex_exit(&zonehash_lock);
747 	kmem_free(zsdp, sizeof (*zsdp));
748 
749 	/* Now call the shutdown and destroy callback for this key */
750 	zsd_apply_all_zones(zsd_apply_shutdown, key);
751 	zsd_apply_all_zones(zsd_apply_destroy, key);
752 
753 	/* Now we can free up the zsdp structures in each zone */
754 	mutex_enter(&zonehash_lock);
755 	for (zone = list_head(&zone_active); zone != NULL;
756 	    zone = list_next(&zone_active, zone)) {
757 		struct zsd_entry *del;
758 
759 		mutex_enter(&zone->zone_lock);
760 		del = zsd_find(&zone->zone_zsd, key);
761 		if (del != NULL) {
762 			list_remove(&zone->zone_zsd, del);
763 			ASSERT(!(del->zsd_flags & ZSD_ALL_INPROGRESS));
764 			kmem_free(del, sizeof (*del));
765 		}
766 		mutex_exit(&zone->zone_lock);
767 	}
768 	mutex_exit(&zonehash_lock);
769 
770 	return (0);
771 }
772 
773 /*
774  * ZSD counterpart of pthread_setspecific().
775  *
776  * Since all zsd callbacks, including those with no create function,
777  * have an entry in zone_zsd, if the key is registered it is part of
778  * the zone_zsd list.
779  * Return an error if the key wasn't registerd.
780  */
781 int
782 zone_setspecific(zone_key_t key, zone_t *zone, const void *data)
783 {
784 	struct zsd_entry *t;
785 
786 	mutex_enter(&zone->zone_lock);
787 	t = zsd_find_mru(&zone->zone_zsd, key);
788 	if (t != NULL) {
789 		/*
790 		 * Replace old value with new
791 		 */
792 		t->zsd_data = (void *)data;
793 		mutex_exit(&zone->zone_lock);
794 		return (0);
795 	}
796 	mutex_exit(&zone->zone_lock);
797 	return (-1);
798 }
799 
800 /*
801  * ZSD counterpart of pthread_getspecific().
802  */
803 void *
804 zone_getspecific(zone_key_t key, zone_t *zone)
805 {
806 	struct zsd_entry *t;
807 	void *data;
808 
809 	mutex_enter(&zone->zone_lock);
810 	t = zsd_find_mru(&zone->zone_zsd, key);
811 	data = (t == NULL ? NULL : t->zsd_data);
812 	mutex_exit(&zone->zone_lock);
813 	return (data);
814 }
815 
816 /*
817  * Function used to initialize a zone's list of ZSD callbacks and data
818  * when the zone is being created.  The callbacks are initialized from
819  * the template list (zsd_registered_keys). The constructor callback is
820  * executed later (once the zone exists and with locks dropped).
821  */
822 static void
823 zone_zsd_configure(zone_t *zone)
824 {
825 	struct zsd_entry *zsdp;
826 	struct zsd_entry *t;
827 
828 	ASSERT(MUTEX_HELD(&zonehash_lock));
829 	ASSERT(list_head(&zone->zone_zsd) == NULL);
830 	mutex_enter(&zone->zone_lock);
831 	mutex_enter(&zsd_key_lock);
832 	for (zsdp = list_head(&zsd_registered_keys); zsdp != NULL;
833 	    zsdp = list_next(&zsd_registered_keys, zsdp)) {
834 		/*
835 		 * Since this zone is ZONE_IS_UNCONFIGURED, zone_key_create
836 		 * should not have added anything to it.
837 		 */
838 		ASSERT(zsd_find(&zone->zone_zsd, zsdp->zsd_key) == NULL);
839 
840 		t = kmem_zalloc(sizeof (*t), KM_SLEEP);
841 		t->zsd_key = zsdp->zsd_key;
842 		t->zsd_create = zsdp->zsd_create;
843 		t->zsd_shutdown = zsdp->zsd_shutdown;
844 		t->zsd_destroy = zsdp->zsd_destroy;
845 		if (zsdp->zsd_create != NULL) {
846 			t->zsd_flags = ZSD_CREATE_NEEDED;
847 			DTRACE_PROBE2(zsd__create__needed,
848 			    zone_t *, zone, zone_key_t, zsdp->zsd_key);
849 		}
850 		list_insert_tail(&zone->zone_zsd, t);
851 	}
852 	mutex_exit(&zsd_key_lock);
853 	mutex_exit(&zone->zone_lock);
854 }
855 
856 enum zsd_callback_type { ZSD_CREATE, ZSD_SHUTDOWN, ZSD_DESTROY };
857 
858 /*
859  * Helper function to execute shutdown or destructor callbacks.
860  */
861 static void
862 zone_zsd_callbacks(zone_t *zone, enum zsd_callback_type ct)
863 {
864 	struct zsd_entry *t;
865 
866 	ASSERT(ct == ZSD_SHUTDOWN || ct == ZSD_DESTROY);
867 	ASSERT(ct != ZSD_SHUTDOWN || zone_status_get(zone) >= ZONE_IS_EMPTY);
868 	ASSERT(ct != ZSD_DESTROY || zone_status_get(zone) >= ZONE_IS_DOWN);
869 
870 	/*
871 	 * Run the callback solely based on what is registered for the zone
872 	 * in zone_zsd. The global list can change independently of this
873 	 * as keys are registered and unregistered and we don't register new
874 	 * callbacks for a zone that is in the process of going away.
875 	 */
876 	mutex_enter(&zone->zone_lock);
877 	for (t = list_head(&zone->zone_zsd); t != NULL;
878 	    t = list_next(&zone->zone_zsd, t)) {
879 		zone_key_t key = t->zsd_key;
880 
881 		/* Skip if no callbacks registered */
882 
883 		if (ct == ZSD_SHUTDOWN) {
884 			if (t->zsd_shutdown != NULL &&
885 			    (t->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
886 				t->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
887 				DTRACE_PROBE2(zsd__shutdown__needed,
888 				    zone_t *, zone, zone_key_t, key);
889 			}
890 		} else {
891 			if (t->zsd_destroy != NULL &&
892 			    (t->zsd_flags & ZSD_DESTROY_ALL) == 0) {
893 				t->zsd_flags |= ZSD_DESTROY_NEEDED;
894 				DTRACE_PROBE2(zsd__destroy__needed,
895 				    zone_t *, zone, zone_key_t, key);
896 			}
897 		}
898 	}
899 	mutex_exit(&zone->zone_lock);
900 
901 	/* Now call the shutdown and destroy callback for this key */
902 	zsd_apply_all_keys(zsd_apply_shutdown, zone);
903 	zsd_apply_all_keys(zsd_apply_destroy, zone);
904 
905 }
906 
907 /*
908  * Called when the zone is going away; free ZSD-related memory, and
909  * destroy the zone_zsd list.
910  */
911 static void
912 zone_free_zsd(zone_t *zone)
913 {
914 	struct zsd_entry *t, *next;
915 
916 	/*
917 	 * Free all the zsd_entry's we had on this zone.
918 	 */
919 	mutex_enter(&zone->zone_lock);
920 	for (t = list_head(&zone->zone_zsd); t != NULL; t = next) {
921 		next = list_next(&zone->zone_zsd, t);
922 		list_remove(&zone->zone_zsd, t);
923 		ASSERT(!(t->zsd_flags & ZSD_ALL_INPROGRESS));
924 		kmem_free(t, sizeof (*t));
925 	}
926 	list_destroy(&zone->zone_zsd);
927 	mutex_exit(&zone->zone_lock);
928 
929 }
930 
931 /*
932  * Apply a function to all zones for particular key value.
933  *
934  * The applyfn has to drop zonehash_lock if it does some work, and
935  * then reacquire it before it returns.
936  * When the lock is dropped we don't follow list_next even
937  * if it is possible to do so without any hazards. This is
938  * because we want the design to allow for the list of zones
939  * to change in any arbitrary way during the time the
940  * lock was dropped.
941  *
942  * It is safe to restart the loop at list_head since the applyfn
943  * changes the zsd_flags as it does work, so a subsequent
944  * pass through will have no effect in applyfn, hence the loop will terminate
945  * in at worst O(N^2).
946  */
947 static void
948 zsd_apply_all_zones(zsd_applyfn_t *applyfn, zone_key_t key)
949 {
950 	zone_t *zone;
951 
952 	mutex_enter(&zonehash_lock);
953 	zone = list_head(&zone_active);
954 	while (zone != NULL) {
955 		if ((applyfn)(&zonehash_lock, B_FALSE, zone, key)) {
956 			/* Lock dropped - restart at head */
957 			zone = list_head(&zone_active);
958 		} else {
959 			zone = list_next(&zone_active, zone);
960 		}
961 	}
962 	mutex_exit(&zonehash_lock);
963 }
964 
965 /*
966  * Apply a function to all keys for a particular zone.
967  *
968  * The applyfn has to drop zonehash_lock if it does some work, and
969  * then reacquire it before it returns.
970  * When the lock is dropped we don't follow list_next even
971  * if it is possible to do so without any hazards. This is
972  * because we want the design to allow for the list of zsd callbacks
973  * to change in any arbitrary way during the time the
974  * lock was dropped.
975  *
976  * It is safe to restart the loop at list_head since the applyfn
977  * changes the zsd_flags as it does work, so a subsequent
978  * pass through will have no effect in applyfn, hence the loop will terminate
979  * in at worst O(N^2).
980  */
981 static void
982 zsd_apply_all_keys(zsd_applyfn_t *applyfn, zone_t *zone)
983 {
984 	struct zsd_entry *t;
985 
986 	mutex_enter(&zone->zone_lock);
987 	t = list_head(&zone->zone_zsd);
988 	while (t != NULL) {
989 		if ((applyfn)(NULL, B_TRUE, zone, t->zsd_key)) {
990 			/* Lock dropped - restart at head */
991 			t = list_head(&zone->zone_zsd);
992 		} else {
993 			t = list_next(&zone->zone_zsd, t);
994 		}
995 	}
996 	mutex_exit(&zone->zone_lock);
997 }
998 
999 /*
1000  * Call the create function for the zone and key if CREATE_NEEDED
1001  * is set.
1002  * If some other thread gets here first and sets CREATE_INPROGRESS, then
1003  * we wait for that thread to complete so that we can ensure that
1004  * all the callbacks are done when we've looped over all zones/keys.
1005  *
1006  * When we call the create function, we drop the global held by the
1007  * caller, and return true to tell the caller it needs to re-evalute the
1008  * state.
1009  * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1010  * remains held on exit.
1011  */
1012 static boolean_t
1013 zsd_apply_create(kmutex_t *lockp, boolean_t zone_lock_held,
1014     zone_t *zone, zone_key_t key)
1015 {
1016 	void *result;
1017 	struct zsd_entry *t;
1018 	boolean_t dropped;
1019 
1020 	if (lockp != NULL) {
1021 		ASSERT(MUTEX_HELD(lockp));
1022 	}
1023 	if (zone_lock_held) {
1024 		ASSERT(MUTEX_HELD(&zone->zone_lock));
1025 	} else {
1026 		mutex_enter(&zone->zone_lock);
1027 	}
1028 
1029 	t = zsd_find(&zone->zone_zsd, key);
1030 	if (t == NULL) {
1031 		/*
1032 		 * Somebody else got here first e.g the zone going
1033 		 * away.
1034 		 */
1035 		if (!zone_lock_held)
1036 			mutex_exit(&zone->zone_lock);
1037 		return (B_FALSE);
1038 	}
1039 	dropped = B_FALSE;
1040 	if (zsd_wait_for_inprogress(zone, t, lockp))
1041 		dropped = B_TRUE;
1042 
1043 	if (t->zsd_flags & ZSD_CREATE_NEEDED) {
1044 		t->zsd_flags &= ~ZSD_CREATE_NEEDED;
1045 		t->zsd_flags |= ZSD_CREATE_INPROGRESS;
1046 		DTRACE_PROBE2(zsd__create__inprogress,
1047 		    zone_t *, zone, zone_key_t, key);
1048 		mutex_exit(&zone->zone_lock);
1049 		if (lockp != NULL)
1050 			mutex_exit(lockp);
1051 
1052 		dropped = B_TRUE;
1053 		ASSERT(t->zsd_create != NULL);
1054 		DTRACE_PROBE2(zsd__create__start,
1055 		    zone_t *, zone, zone_key_t, key);
1056 
1057 		result = (*t->zsd_create)(zone->zone_id);
1058 
1059 		DTRACE_PROBE2(zsd__create__end,
1060 		    zone_t *, zone, voidn *, result);
1061 
1062 		ASSERT(result != NULL);
1063 		if (lockp != NULL)
1064 			mutex_enter(lockp);
1065 		mutex_enter(&zone->zone_lock);
1066 		t->zsd_data = result;
1067 		t->zsd_flags &= ~ZSD_CREATE_INPROGRESS;
1068 		t->zsd_flags |= ZSD_CREATE_COMPLETED;
1069 		cv_broadcast(&t->zsd_cv);
1070 		DTRACE_PROBE2(zsd__create__completed,
1071 		    zone_t *, zone, zone_key_t, key);
1072 	}
1073 	if (!zone_lock_held)
1074 		mutex_exit(&zone->zone_lock);
1075 	return (dropped);
1076 }
1077 
1078 /*
1079  * Call the shutdown function for the zone and key if SHUTDOWN_NEEDED
1080  * is set.
1081  * If some other thread gets here first and sets *_INPROGRESS, then
1082  * we wait for that thread to complete so that we can ensure that
1083  * all the callbacks are done when we've looped over all zones/keys.
1084  *
1085  * When we call the shutdown function, we drop the global held by the
1086  * caller, and return true to tell the caller it needs to re-evalute the
1087  * state.
1088  * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1089  * remains held on exit.
1090  */
1091 static boolean_t
1092 zsd_apply_shutdown(kmutex_t *lockp, boolean_t zone_lock_held,
1093     zone_t *zone, zone_key_t key)
1094 {
1095 	struct zsd_entry *t;
1096 	void *data;
1097 	boolean_t dropped;
1098 
1099 	if (lockp != NULL) {
1100 		ASSERT(MUTEX_HELD(lockp));
1101 	}
1102 	if (zone_lock_held) {
1103 		ASSERT(MUTEX_HELD(&zone->zone_lock));
1104 	} else {
1105 		mutex_enter(&zone->zone_lock);
1106 	}
1107 
1108 	t = zsd_find(&zone->zone_zsd, key);
1109 	if (t == NULL) {
1110 		/*
1111 		 * Somebody else got here first e.g the zone going
1112 		 * away.
1113 		 */
1114 		if (!zone_lock_held)
1115 			mutex_exit(&zone->zone_lock);
1116 		return (B_FALSE);
1117 	}
1118 	dropped = B_FALSE;
1119 	if (zsd_wait_for_creator(zone, t, lockp))
1120 		dropped = B_TRUE;
1121 
1122 	if (zsd_wait_for_inprogress(zone, t, lockp))
1123 		dropped = B_TRUE;
1124 
1125 	if (t->zsd_flags & ZSD_SHUTDOWN_NEEDED) {
1126 		t->zsd_flags &= ~ZSD_SHUTDOWN_NEEDED;
1127 		t->zsd_flags |= ZSD_SHUTDOWN_INPROGRESS;
1128 		DTRACE_PROBE2(zsd__shutdown__inprogress,
1129 		    zone_t *, zone, zone_key_t, key);
1130 		mutex_exit(&zone->zone_lock);
1131 		if (lockp != NULL)
1132 			mutex_exit(lockp);
1133 		dropped = B_TRUE;
1134 
1135 		ASSERT(t->zsd_shutdown != NULL);
1136 		data = t->zsd_data;
1137 
1138 		DTRACE_PROBE2(zsd__shutdown__start,
1139 		    zone_t *, zone, zone_key_t, key);
1140 
1141 		(t->zsd_shutdown)(zone->zone_id, data);
1142 		DTRACE_PROBE2(zsd__shutdown__end,
1143 		    zone_t *, zone, zone_key_t, key);
1144 
1145 		if (lockp != NULL)
1146 			mutex_enter(lockp);
1147 		mutex_enter(&zone->zone_lock);
1148 		t->zsd_flags &= ~ZSD_SHUTDOWN_INPROGRESS;
1149 		t->zsd_flags |= ZSD_SHUTDOWN_COMPLETED;
1150 		cv_broadcast(&t->zsd_cv);
1151 		DTRACE_PROBE2(zsd__shutdown__completed,
1152 		    zone_t *, zone, zone_key_t, key);
1153 	}
1154 	if (!zone_lock_held)
1155 		mutex_exit(&zone->zone_lock);
1156 	return (dropped);
1157 }
1158 
1159 /*
1160  * Call the destroy function for the zone and key if DESTROY_NEEDED
1161  * is set.
1162  * If some other thread gets here first and sets *_INPROGRESS, then
1163  * we wait for that thread to complete so that we can ensure that
1164  * all the callbacks are done when we've looped over all zones/keys.
1165  *
1166  * When we call the destroy function, we drop the global held by the
1167  * caller, and return true to tell the caller it needs to re-evalute the
1168  * state.
1169  * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1170  * remains held on exit.
1171  */
1172 static boolean_t
1173 zsd_apply_destroy(kmutex_t *lockp, boolean_t zone_lock_held,
1174     zone_t *zone, zone_key_t key)
1175 {
1176 	struct zsd_entry *t;
1177 	void *data;
1178 	boolean_t dropped;
1179 
1180 	if (lockp != NULL) {
1181 		ASSERT(MUTEX_HELD(lockp));
1182 	}
1183 	if (zone_lock_held) {
1184 		ASSERT(MUTEX_HELD(&zone->zone_lock));
1185 	} else {
1186 		mutex_enter(&zone->zone_lock);
1187 	}
1188 
1189 	t = zsd_find(&zone->zone_zsd, key);
1190 	if (t == NULL) {
1191 		/*
1192 		 * Somebody else got here first e.g the zone going
1193 		 * away.
1194 		 */
1195 		if (!zone_lock_held)
1196 			mutex_exit(&zone->zone_lock);
1197 		return (B_FALSE);
1198 	}
1199 	dropped = B_FALSE;
1200 	if (zsd_wait_for_creator(zone, t, lockp))
1201 		dropped = B_TRUE;
1202 
1203 	if (zsd_wait_for_inprogress(zone, t, lockp))
1204 		dropped = B_TRUE;
1205 
1206 	if (t->zsd_flags & ZSD_DESTROY_NEEDED) {
1207 		t->zsd_flags &= ~ZSD_DESTROY_NEEDED;
1208 		t->zsd_flags |= ZSD_DESTROY_INPROGRESS;
1209 		DTRACE_PROBE2(zsd__destroy__inprogress,
1210 		    zone_t *, zone, zone_key_t, key);
1211 		mutex_exit(&zone->zone_lock);
1212 		if (lockp != NULL)
1213 			mutex_exit(lockp);
1214 		dropped = B_TRUE;
1215 
1216 		ASSERT(t->zsd_destroy != NULL);
1217 		data = t->zsd_data;
1218 		DTRACE_PROBE2(zsd__destroy__start,
1219 		    zone_t *, zone, zone_key_t, key);
1220 
1221 		(t->zsd_destroy)(zone->zone_id, data);
1222 		DTRACE_PROBE2(zsd__destroy__end,
1223 		    zone_t *, zone, zone_key_t, key);
1224 
1225 		if (lockp != NULL)
1226 			mutex_enter(lockp);
1227 		mutex_enter(&zone->zone_lock);
1228 		t->zsd_data = NULL;
1229 		t->zsd_flags &= ~ZSD_DESTROY_INPROGRESS;
1230 		t->zsd_flags |= ZSD_DESTROY_COMPLETED;
1231 		cv_broadcast(&t->zsd_cv);
1232 		DTRACE_PROBE2(zsd__destroy__completed,
1233 		    zone_t *, zone, zone_key_t, key);
1234 	}
1235 	if (!zone_lock_held)
1236 		mutex_exit(&zone->zone_lock);
1237 	return (dropped);
1238 }
1239 
1240 /*
1241  * Wait for any CREATE_NEEDED flag to be cleared.
1242  * Returns true if lockp was temporarily dropped while waiting.
1243  */
1244 static boolean_t
1245 zsd_wait_for_creator(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
1246 {
1247 	boolean_t dropped = B_FALSE;
1248 
1249 	while (t->zsd_flags & ZSD_CREATE_NEEDED) {
1250 		DTRACE_PROBE2(zsd__wait__for__creator,
1251 		    zone_t *, zone, struct zsd_entry *, t);
1252 		if (lockp != NULL) {
1253 			dropped = B_TRUE;
1254 			mutex_exit(lockp);
1255 		}
1256 		cv_wait(&t->zsd_cv, &zone->zone_lock);
1257 		if (lockp != NULL) {
1258 			/* First drop zone_lock to preserve order */
1259 			mutex_exit(&zone->zone_lock);
1260 			mutex_enter(lockp);
1261 			mutex_enter(&zone->zone_lock);
1262 		}
1263 	}
1264 	return (dropped);
1265 }
1266 
1267 /*
1268  * Wait for any INPROGRESS flag to be cleared.
1269  * Returns true if lockp was temporarily dropped while waiting.
1270  */
1271 static boolean_t
1272 zsd_wait_for_inprogress(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
1273 {
1274 	boolean_t dropped = B_FALSE;
1275 
1276 	while (t->zsd_flags & ZSD_ALL_INPROGRESS) {
1277 		DTRACE_PROBE2(zsd__wait__for__inprogress,
1278 		    zone_t *, zone, struct zsd_entry *, t);
1279 		if (lockp != NULL) {
1280 			dropped = B_TRUE;
1281 			mutex_exit(lockp);
1282 		}
1283 		cv_wait(&t->zsd_cv, &zone->zone_lock);
1284 		if (lockp != NULL) {
1285 			/* First drop zone_lock to preserve order */
1286 			mutex_exit(&zone->zone_lock);
1287 			mutex_enter(lockp);
1288 			mutex_enter(&zone->zone_lock);
1289 		}
1290 	}
1291 	return (dropped);
1292 }
1293 
1294 /*
1295  * Frees memory associated with the zone dataset list.
1296  */
1297 static void
1298 zone_free_datasets(zone_t *zone)
1299 {
1300 	zone_dataset_t *t, *next;
1301 
1302 	for (t = list_head(&zone->zone_datasets); t != NULL; t = next) {
1303 		next = list_next(&zone->zone_datasets, t);
1304 		list_remove(&zone->zone_datasets, t);
1305 		kmem_free(t->zd_dataset, strlen(t->zd_dataset) + 1);
1306 		kmem_free(t, sizeof (*t));
1307 	}
1308 	list_destroy(&zone->zone_datasets);
1309 }
1310 
1311 /*
1312  * zone.cpu-shares resource control support.
1313  */
1314 /*ARGSUSED*/
1315 static rctl_qty_t
1316 zone_cpu_shares_usage(rctl_t *rctl, struct proc *p)
1317 {
1318 	ASSERT(MUTEX_HELD(&p->p_lock));
1319 	return (p->p_zone->zone_shares);
1320 }
1321 
1322 /*ARGSUSED*/
1323 static int
1324 zone_cpu_shares_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1325     rctl_qty_t nv)
1326 {
1327 	ASSERT(MUTEX_HELD(&p->p_lock));
1328 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1329 	if (e->rcep_p.zone == NULL)
1330 		return (0);
1331 
1332 	e->rcep_p.zone->zone_shares = nv;
1333 	return (0);
1334 }
1335 
1336 static rctl_ops_t zone_cpu_shares_ops = {
1337 	rcop_no_action,
1338 	zone_cpu_shares_usage,
1339 	zone_cpu_shares_set,
1340 	rcop_no_test
1341 };
1342 
1343 /*
1344  * zone.cpu-cap resource control support.
1345  */
1346 /*ARGSUSED*/
1347 static rctl_qty_t
1348 zone_cpu_cap_get(rctl_t *rctl, struct proc *p)
1349 {
1350 	ASSERT(MUTEX_HELD(&p->p_lock));
1351 	return (cpucaps_zone_get(p->p_zone));
1352 }
1353 
1354 /*ARGSUSED*/
1355 static int
1356 zone_cpu_cap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1357     rctl_qty_t nv)
1358 {
1359 	zone_t *zone = e->rcep_p.zone;
1360 
1361 	ASSERT(MUTEX_HELD(&p->p_lock));
1362 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1363 
1364 	if (zone == NULL)
1365 		return (0);
1366 
1367 	/*
1368 	 * set cap to the new value.
1369 	 */
1370 	return (cpucaps_zone_set(zone, nv));
1371 }
1372 
1373 static rctl_ops_t zone_cpu_cap_ops = {
1374 	rcop_no_action,
1375 	zone_cpu_cap_get,
1376 	zone_cpu_cap_set,
1377 	rcop_no_test
1378 };
1379 
1380 /*ARGSUSED*/
1381 static rctl_qty_t
1382 zone_lwps_usage(rctl_t *r, proc_t *p)
1383 {
1384 	rctl_qty_t nlwps;
1385 	zone_t *zone = p->p_zone;
1386 
1387 	ASSERT(MUTEX_HELD(&p->p_lock));
1388 
1389 	mutex_enter(&zone->zone_nlwps_lock);
1390 	nlwps = zone->zone_nlwps;
1391 	mutex_exit(&zone->zone_nlwps_lock);
1392 
1393 	return (nlwps);
1394 }
1395 
1396 /*ARGSUSED*/
1397 static int
1398 zone_lwps_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
1399     rctl_qty_t incr, uint_t flags)
1400 {
1401 	rctl_qty_t nlwps;
1402 
1403 	ASSERT(MUTEX_HELD(&p->p_lock));
1404 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1405 	if (e->rcep_p.zone == NULL)
1406 		return (0);
1407 	ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
1408 	nlwps = e->rcep_p.zone->zone_nlwps;
1409 
1410 	if (nlwps + incr > rcntl->rcv_value)
1411 		return (1);
1412 
1413 	return (0);
1414 }
1415 
1416 /*ARGSUSED*/
1417 static int
1418 zone_lwps_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv)
1419 {
1420 	ASSERT(MUTEX_HELD(&p->p_lock));
1421 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1422 	if (e->rcep_p.zone == NULL)
1423 		return (0);
1424 	e->rcep_p.zone->zone_nlwps_ctl = nv;
1425 	return (0);
1426 }
1427 
1428 static rctl_ops_t zone_lwps_ops = {
1429 	rcop_no_action,
1430 	zone_lwps_usage,
1431 	zone_lwps_set,
1432 	zone_lwps_test,
1433 };
1434 
1435 /*ARGSUSED*/
1436 static rctl_qty_t
1437 zone_procs_usage(rctl_t *r, proc_t *p)
1438 {
1439 	rctl_qty_t nprocs;
1440 	zone_t *zone = p->p_zone;
1441 
1442 	ASSERT(MUTEX_HELD(&p->p_lock));
1443 
1444 	mutex_enter(&zone->zone_nlwps_lock);
1445 	nprocs = zone->zone_nprocs;
1446 	mutex_exit(&zone->zone_nlwps_lock);
1447 
1448 	return (nprocs);
1449 }
1450 
1451 /*ARGSUSED*/
1452 static int
1453 zone_procs_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
1454     rctl_qty_t incr, uint_t flags)
1455 {
1456 	rctl_qty_t nprocs;
1457 
1458 	ASSERT(MUTEX_HELD(&p->p_lock));
1459 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1460 	if (e->rcep_p.zone == NULL)
1461 		return (0);
1462 	ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
1463 	nprocs = e->rcep_p.zone->zone_nprocs;
1464 
1465 	if (nprocs + incr > rcntl->rcv_value)
1466 		return (1);
1467 
1468 	return (0);
1469 }
1470 
1471 /*ARGSUSED*/
1472 static int
1473 zone_procs_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv)
1474 {
1475 	ASSERT(MUTEX_HELD(&p->p_lock));
1476 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1477 	if (e->rcep_p.zone == NULL)
1478 		return (0);
1479 	e->rcep_p.zone->zone_nprocs_ctl = nv;
1480 	return (0);
1481 }
1482 
1483 static rctl_ops_t zone_procs_ops = {
1484 	rcop_no_action,
1485 	zone_procs_usage,
1486 	zone_procs_set,
1487 	zone_procs_test,
1488 };
1489 
1490 /*ARGSUSED*/
1491 static rctl_qty_t
1492 zone_shmmax_usage(rctl_t *rctl, struct proc *p)
1493 {
1494 	ASSERT(MUTEX_HELD(&p->p_lock));
1495 	return (p->p_zone->zone_shmmax);
1496 }
1497 
1498 /*ARGSUSED*/
1499 static int
1500 zone_shmmax_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1501     rctl_qty_t incr, uint_t flags)
1502 {
1503 	rctl_qty_t v;
1504 	ASSERT(MUTEX_HELD(&p->p_lock));
1505 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1506 	v = e->rcep_p.zone->zone_shmmax + incr;
1507 	if (v > rval->rcv_value)
1508 		return (1);
1509 	return (0);
1510 }
1511 
1512 static rctl_ops_t zone_shmmax_ops = {
1513 	rcop_no_action,
1514 	zone_shmmax_usage,
1515 	rcop_no_set,
1516 	zone_shmmax_test
1517 };
1518 
1519 /*ARGSUSED*/
1520 static rctl_qty_t
1521 zone_shmmni_usage(rctl_t *rctl, struct proc *p)
1522 {
1523 	ASSERT(MUTEX_HELD(&p->p_lock));
1524 	return (p->p_zone->zone_ipc.ipcq_shmmni);
1525 }
1526 
1527 /*ARGSUSED*/
1528 static int
1529 zone_shmmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1530     rctl_qty_t incr, uint_t flags)
1531 {
1532 	rctl_qty_t v;
1533 	ASSERT(MUTEX_HELD(&p->p_lock));
1534 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1535 	v = e->rcep_p.zone->zone_ipc.ipcq_shmmni + incr;
1536 	if (v > rval->rcv_value)
1537 		return (1);
1538 	return (0);
1539 }
1540 
1541 static rctl_ops_t zone_shmmni_ops = {
1542 	rcop_no_action,
1543 	zone_shmmni_usage,
1544 	rcop_no_set,
1545 	zone_shmmni_test
1546 };
1547 
1548 /*ARGSUSED*/
1549 static rctl_qty_t
1550 zone_semmni_usage(rctl_t *rctl, struct proc *p)
1551 {
1552 	ASSERT(MUTEX_HELD(&p->p_lock));
1553 	return (p->p_zone->zone_ipc.ipcq_semmni);
1554 }
1555 
1556 /*ARGSUSED*/
1557 static int
1558 zone_semmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1559     rctl_qty_t incr, uint_t flags)
1560 {
1561 	rctl_qty_t v;
1562 	ASSERT(MUTEX_HELD(&p->p_lock));
1563 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1564 	v = e->rcep_p.zone->zone_ipc.ipcq_semmni + incr;
1565 	if (v > rval->rcv_value)
1566 		return (1);
1567 	return (0);
1568 }
1569 
1570 static rctl_ops_t zone_semmni_ops = {
1571 	rcop_no_action,
1572 	zone_semmni_usage,
1573 	rcop_no_set,
1574 	zone_semmni_test
1575 };
1576 
1577 /*ARGSUSED*/
1578 static rctl_qty_t
1579 zone_msgmni_usage(rctl_t *rctl, struct proc *p)
1580 {
1581 	ASSERT(MUTEX_HELD(&p->p_lock));
1582 	return (p->p_zone->zone_ipc.ipcq_msgmni);
1583 }
1584 
1585 /*ARGSUSED*/
1586 static int
1587 zone_msgmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1588     rctl_qty_t incr, uint_t flags)
1589 {
1590 	rctl_qty_t v;
1591 	ASSERT(MUTEX_HELD(&p->p_lock));
1592 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1593 	v = e->rcep_p.zone->zone_ipc.ipcq_msgmni + incr;
1594 	if (v > rval->rcv_value)
1595 		return (1);
1596 	return (0);
1597 }
1598 
1599 static rctl_ops_t zone_msgmni_ops = {
1600 	rcop_no_action,
1601 	zone_msgmni_usage,
1602 	rcop_no_set,
1603 	zone_msgmni_test
1604 };
1605 
1606 /*ARGSUSED*/
1607 static rctl_qty_t
1608 zone_locked_mem_usage(rctl_t *rctl, struct proc *p)
1609 {
1610 	rctl_qty_t q;
1611 	ASSERT(MUTEX_HELD(&p->p_lock));
1612 	mutex_enter(&p->p_zone->zone_mem_lock);
1613 	q = p->p_zone->zone_locked_mem;
1614 	mutex_exit(&p->p_zone->zone_mem_lock);
1615 	return (q);
1616 }
1617 
1618 /*ARGSUSED*/
1619 static int
1620 zone_locked_mem_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1621     rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1622 {
1623 	rctl_qty_t q;
1624 	zone_t *z;
1625 
1626 	z = e->rcep_p.zone;
1627 	ASSERT(MUTEX_HELD(&p->p_lock));
1628 	ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1629 	q = z->zone_locked_mem;
1630 	if (q + incr > rcntl->rcv_value)
1631 		return (1);
1632 	return (0);
1633 }
1634 
1635 /*ARGSUSED*/
1636 static int
1637 zone_locked_mem_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1638     rctl_qty_t nv)
1639 {
1640 	ASSERT(MUTEX_HELD(&p->p_lock));
1641 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1642 	if (e->rcep_p.zone == NULL)
1643 		return (0);
1644 	e->rcep_p.zone->zone_locked_mem_ctl = nv;
1645 	return (0);
1646 }
1647 
1648 static rctl_ops_t zone_locked_mem_ops = {
1649 	rcop_no_action,
1650 	zone_locked_mem_usage,
1651 	zone_locked_mem_set,
1652 	zone_locked_mem_test
1653 };
1654 
1655 /*ARGSUSED*/
1656 static rctl_qty_t
1657 zone_max_swap_usage(rctl_t *rctl, struct proc *p)
1658 {
1659 	rctl_qty_t q;
1660 	zone_t *z = p->p_zone;
1661 
1662 	ASSERT(MUTEX_HELD(&p->p_lock));
1663 	mutex_enter(&z->zone_mem_lock);
1664 	q = z->zone_max_swap;
1665 	mutex_exit(&z->zone_mem_lock);
1666 	return (q);
1667 }
1668 
1669 /*ARGSUSED*/
1670 static int
1671 zone_max_swap_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1672     rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1673 {
1674 	rctl_qty_t q;
1675 	zone_t *z;
1676 
1677 	z = e->rcep_p.zone;
1678 	ASSERT(MUTEX_HELD(&p->p_lock));
1679 	ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1680 	q = z->zone_max_swap;
1681 	if (q + incr > rcntl->rcv_value)
1682 		return (1);
1683 	return (0);
1684 }
1685 
1686 /*ARGSUSED*/
1687 static int
1688 zone_max_swap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1689     rctl_qty_t nv)
1690 {
1691 	ASSERT(MUTEX_HELD(&p->p_lock));
1692 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1693 	if (e->rcep_p.zone == NULL)
1694 		return (0);
1695 	e->rcep_p.zone->zone_max_swap_ctl = nv;
1696 	return (0);
1697 }
1698 
1699 static rctl_ops_t zone_max_swap_ops = {
1700 	rcop_no_action,
1701 	zone_max_swap_usage,
1702 	zone_max_swap_set,
1703 	zone_max_swap_test
1704 };
1705 
1706 /*ARGSUSED*/
1707 static rctl_qty_t
1708 zone_max_lofi_usage(rctl_t *rctl, struct proc *p)
1709 {
1710 	rctl_qty_t q;
1711 	zone_t *z = p->p_zone;
1712 
1713 	ASSERT(MUTEX_HELD(&p->p_lock));
1714 	mutex_enter(&z->zone_rctl_lock);
1715 	q = z->zone_max_lofi;
1716 	mutex_exit(&z->zone_rctl_lock);
1717 	return (q);
1718 }
1719 
1720 /*ARGSUSED*/
1721 static int
1722 zone_max_lofi_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1723     rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1724 {
1725 	rctl_qty_t q;
1726 	zone_t *z;
1727 
1728 	z = e->rcep_p.zone;
1729 	ASSERT(MUTEX_HELD(&p->p_lock));
1730 	ASSERT(MUTEX_HELD(&z->zone_rctl_lock));
1731 	q = z->zone_max_lofi;
1732 	if (q + incr > rcntl->rcv_value)
1733 		return (1);
1734 	return (0);
1735 }
1736 
1737 /*ARGSUSED*/
1738 static int
1739 zone_max_lofi_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1740     rctl_qty_t nv)
1741 {
1742 	ASSERT(MUTEX_HELD(&p->p_lock));
1743 	ASSERT(e->rcep_t == RCENTITY_ZONE);
1744 	if (e->rcep_p.zone == NULL)
1745 		return (0);
1746 	e->rcep_p.zone->zone_max_lofi_ctl = nv;
1747 	return (0);
1748 }
1749 
1750 static rctl_ops_t zone_max_lofi_ops = {
1751 	rcop_no_action,
1752 	zone_max_lofi_usage,
1753 	zone_max_lofi_set,
1754 	zone_max_lofi_test
1755 };
1756 
1757 /*
1758  * Helper function to brand the zone with a unique ID.
1759  */
1760 static void
1761 zone_uniqid(zone_t *zone)
1762 {
1763 	static uint64_t uniqid = 0;
1764 
1765 	ASSERT(MUTEX_HELD(&zonehash_lock));
1766 	zone->zone_uniqid = uniqid++;
1767 }
1768 
1769 /*
1770  * Returns a held pointer to the "kcred" for the specified zone.
1771  */
1772 struct cred *
1773 zone_get_kcred(zoneid_t zoneid)
1774 {
1775 	zone_t *zone;
1776 	cred_t *cr;
1777 
1778 	if ((zone = zone_find_by_id(zoneid)) == NULL)
1779 		return (NULL);
1780 	cr = zone->zone_kcred;
1781 	crhold(cr);
1782 	zone_rele(zone);
1783 	return (cr);
1784 }
1785 
1786 static int
1787 zone_lockedmem_kstat_update(kstat_t *ksp, int rw)
1788 {
1789 	zone_t *zone = ksp->ks_private;
1790 	zone_kstat_t *zk = ksp->ks_data;
1791 
1792 	if (rw == KSTAT_WRITE)
1793 		return (EACCES);
1794 
1795 	zk->zk_usage.value.ui64 = zone->zone_locked_mem;
1796 	zk->zk_value.value.ui64 = zone->zone_locked_mem_ctl;
1797 	return (0);
1798 }
1799 
1800 static int
1801 zone_nprocs_kstat_update(kstat_t *ksp, int rw)
1802 {
1803 	zone_t *zone = ksp->ks_private;
1804 	zone_kstat_t *zk = ksp->ks_data;
1805 
1806 	if (rw == KSTAT_WRITE)
1807 		return (EACCES);
1808 
1809 	zk->zk_usage.value.ui64 = zone->zone_nprocs;
1810 	zk->zk_value.value.ui64 = zone->zone_nprocs_ctl;
1811 	return (0);
1812 }
1813 
1814 static int
1815 zone_swapresv_kstat_update(kstat_t *ksp, int rw)
1816 {
1817 	zone_t *zone = ksp->ks_private;
1818 	zone_kstat_t *zk = ksp->ks_data;
1819 
1820 	if (rw == KSTAT_WRITE)
1821 		return (EACCES);
1822 
1823 	zk->zk_usage.value.ui64 = zone->zone_max_swap;
1824 	zk->zk_value.value.ui64 = zone->zone_max_swap_ctl;
1825 	return (0);
1826 }
1827 
1828 static kstat_t *
1829 zone_kstat_create_common(zone_t *zone, char *name,
1830     int (*updatefunc) (kstat_t *, int))
1831 {
1832 	kstat_t *ksp;
1833 	zone_kstat_t *zk;
1834 
1835 	ksp = rctl_kstat_create_zone(zone, name, KSTAT_TYPE_NAMED,
1836 	    sizeof (zone_kstat_t) / sizeof (kstat_named_t),
1837 	    KSTAT_FLAG_VIRTUAL);
1838 
1839 	if (ksp == NULL)
1840 		return (NULL);
1841 
1842 	zk = ksp->ks_data = kmem_alloc(sizeof (zone_kstat_t), KM_SLEEP);
1843 	ksp->ks_data_size += strlen(zone->zone_name) + 1;
1844 	kstat_named_init(&zk->zk_zonename, "zonename", KSTAT_DATA_STRING);
1845 	kstat_named_setstr(&zk->zk_zonename, zone->zone_name);
1846 	kstat_named_init(&zk->zk_usage, "usage", KSTAT_DATA_UINT64);
1847 	kstat_named_init(&zk->zk_value, "value", KSTAT_DATA_UINT64);
1848 	ksp->ks_update = updatefunc;
1849 	ksp->ks_private = zone;
1850 	kstat_install(ksp);
1851 	return (ksp);
1852 }
1853 
1854 
1855 static int
1856 zone_mcap_kstat_update(kstat_t *ksp, int rw)
1857 {
1858 	zone_t *zone = ksp->ks_private;
1859 	zone_mcap_kstat_t *zmp = ksp->ks_data;
1860 
1861 	if (rw == KSTAT_WRITE)
1862 		return (EACCES);
1863 
1864 	zmp->zm_pgpgin.value.ui64 = zone->zone_pgpgin;
1865 	zmp->zm_anonpgin.value.ui64 = zone->zone_anonpgin;
1866 	zmp->zm_execpgin.value.ui64 = zone->zone_execpgin;
1867 	zmp->zm_fspgin.value.ui64 = zone->zone_fspgin;
1868 	zmp->zm_anon_alloc_fail.value.ui64 = zone->zone_anon_alloc_fail;
1869 
1870 	return (0);
1871 }
1872 
1873 static kstat_t *
1874 zone_mcap_kstat_create(zone_t *zone)
1875 {
1876 	kstat_t *ksp;
1877 	zone_mcap_kstat_t *zmp;
1878 
1879 	if ((ksp = kstat_create_zone("memory_cap", zone->zone_id,
1880 	    zone->zone_name, "zone_memory_cap", KSTAT_TYPE_NAMED,
1881 	    sizeof (zone_mcap_kstat_t) / sizeof (kstat_named_t),
1882 	    KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
1883 		return (NULL);
1884 
1885 	if (zone->zone_id != GLOBAL_ZONEID)
1886 		kstat_zone_add(ksp, GLOBAL_ZONEID);
1887 
1888 	zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_mcap_kstat_t), KM_SLEEP);
1889 	ksp->ks_data_size += strlen(zone->zone_name) + 1;
1890 	ksp->ks_lock = &zone->zone_mcap_lock;
1891 	zone->zone_mcap_stats = zmp;
1892 
1893 	/* The kstat "name" field is not large enough for a full zonename */
1894 	kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
1895 	kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
1896 	kstat_named_init(&zmp->zm_pgpgin, "pgpgin", KSTAT_DATA_UINT64);
1897 	kstat_named_init(&zmp->zm_anonpgin, "anonpgin", KSTAT_DATA_UINT64);
1898 	kstat_named_init(&zmp->zm_execpgin, "execpgin", KSTAT_DATA_UINT64);
1899 	kstat_named_init(&zmp->zm_fspgin, "fspgin", KSTAT_DATA_UINT64);
1900 	kstat_named_init(&zmp->zm_anon_alloc_fail, "anon_alloc_fail",
1901 	    KSTAT_DATA_UINT64);
1902 
1903 	ksp->ks_update = zone_mcap_kstat_update;
1904 	ksp->ks_private = zone;
1905 
1906 	kstat_install(ksp);
1907 	return (ksp);
1908 }
1909 
1910 static int
1911 zone_misc_kstat_update(kstat_t *ksp, int rw)
1912 {
1913 	zone_t *zone = ksp->ks_private;
1914 	zone_misc_kstat_t *zmp = ksp->ks_data;
1915 	hrtime_t hrtime;
1916 	uint64_t tmp;
1917 
1918 	if (rw == KSTAT_WRITE)
1919 		return (EACCES);
1920 
1921 	tmp = cpu_uarray_sum(zone->zone_ustate, ZONE_USTATE_STIME);
1922 	hrtime = UINT64_OVERFLOW_TO_INT64(tmp);
1923 	scalehrtime(&hrtime);
1924 	zmp->zm_stime.value.ui64 = hrtime;
1925 
1926 	tmp = cpu_uarray_sum(zone->zone_ustate, ZONE_USTATE_UTIME);
1927 	hrtime = UINT64_OVERFLOW_TO_INT64(tmp);
1928 	scalehrtime(&hrtime);
1929 	zmp->zm_utime.value.ui64 = hrtime;
1930 
1931 	tmp = cpu_uarray_sum(zone->zone_ustate, ZONE_USTATE_WTIME);
1932 	hrtime = UINT64_OVERFLOW_TO_INT64(tmp);
1933 	scalehrtime(&hrtime);
1934 	zmp->zm_wtime.value.ui64 = hrtime;
1935 
1936 	zmp->zm_avenrun1.value.ui32 = zone->zone_avenrun[0];
1937 	zmp->zm_avenrun5.value.ui32 = zone->zone_avenrun[1];
1938 	zmp->zm_avenrun15.value.ui32 = zone->zone_avenrun[2];
1939 
1940 	zmp->zm_ffcap.value.ui32 = zone->zone_ffcap;
1941 	zmp->zm_ffnoproc.value.ui32 = zone->zone_ffnoproc;
1942 	zmp->zm_ffnomem.value.ui32 = zone->zone_ffnomem;
1943 	zmp->zm_ffmisc.value.ui32 = zone->zone_ffmisc;
1944 
1945 	zmp->zm_nested_intp.value.ui32 = zone->zone_nested_intp;
1946 
1947 	zmp->zm_init_pid.value.ui32 = zone->zone_proc_initpid;
1948 	zmp->zm_boot_time.value.ui64 = (uint64_t)zone->zone_boot_time;
1949 
1950 	return (0);
1951 }
1952 
1953 static kstat_t *
1954 zone_misc_kstat_create(zone_t *zone)
1955 {
1956 	kstat_t *ksp;
1957 	zone_misc_kstat_t *zmp;
1958 
1959 	if ((ksp = kstat_create_zone("zones", zone->zone_id,
1960 	    zone->zone_name, "zone_misc", KSTAT_TYPE_NAMED,
1961 	    sizeof (zone_misc_kstat_t) / sizeof (kstat_named_t),
1962 	    KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
1963 		return (NULL);
1964 
1965 	if (zone->zone_id != GLOBAL_ZONEID)
1966 		kstat_zone_add(ksp, GLOBAL_ZONEID);
1967 
1968 	zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_misc_kstat_t), KM_SLEEP);
1969 	ksp->ks_data_size += strlen(zone->zone_name) + 1;
1970 	ksp->ks_lock = &zone->zone_misc_lock;
1971 	zone->zone_misc_stats = zmp;
1972 
1973 	/* The kstat "name" field is not large enough for a full zonename */
1974 	kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
1975 	kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
1976 	kstat_named_init(&zmp->zm_utime, "nsec_user", KSTAT_DATA_UINT64);
1977 	kstat_named_init(&zmp->zm_stime, "nsec_sys", KSTAT_DATA_UINT64);
1978 	kstat_named_init(&zmp->zm_wtime, "nsec_waitrq", KSTAT_DATA_UINT64);
1979 	kstat_named_init(&zmp->zm_avenrun1, "avenrun_1min", KSTAT_DATA_UINT32);
1980 	kstat_named_init(&zmp->zm_avenrun5, "avenrun_5min", KSTAT_DATA_UINT32);
1981 	kstat_named_init(&zmp->zm_avenrun15, "avenrun_15min",
1982 	    KSTAT_DATA_UINT32);
1983 	kstat_named_init(&zmp->zm_ffcap, "forkfail_cap", KSTAT_DATA_UINT32);
1984 	kstat_named_init(&zmp->zm_ffnoproc, "forkfail_noproc",
1985 	    KSTAT_DATA_UINT32);
1986 	kstat_named_init(&zmp->zm_ffnomem, "forkfail_nomem", KSTAT_DATA_UINT32);
1987 	kstat_named_init(&zmp->zm_ffmisc, "forkfail_misc", KSTAT_DATA_UINT32);
1988 	kstat_named_init(&zmp->zm_nested_intp, "nested_interp",
1989 	    KSTAT_DATA_UINT32);
1990 	kstat_named_init(&zmp->zm_init_pid, "init_pid", KSTAT_DATA_UINT32);
1991 	kstat_named_init(&zmp->zm_boot_time, "boot_time", KSTAT_DATA_UINT64);
1992 
1993 	ksp->ks_update = zone_misc_kstat_update;
1994 	ksp->ks_private = zone;
1995 
1996 	kstat_install(ksp);
1997 	return (ksp);
1998 }
1999 
2000 static void
2001 zone_kstat_create(zone_t *zone)
2002 {
2003 	zone->zone_lockedmem_kstat = zone_kstat_create_common(zone,
2004 	    "lockedmem", zone_lockedmem_kstat_update);
2005 	zone->zone_swapresv_kstat = zone_kstat_create_common(zone,
2006 	    "swapresv", zone_swapresv_kstat_update);
2007 	zone->zone_nprocs_kstat = zone_kstat_create_common(zone,
2008 	    "nprocs", zone_nprocs_kstat_update);
2009 
2010 	if ((zone->zone_mcap_ksp = zone_mcap_kstat_create(zone)) == NULL) {
2011 		zone->zone_mcap_stats = kmem_zalloc(
2012 		    sizeof (zone_mcap_kstat_t), KM_SLEEP);
2013 	}
2014 
2015 	if ((zone->zone_misc_ksp = zone_misc_kstat_create(zone)) == NULL) {
2016 		zone->zone_misc_stats = kmem_zalloc(
2017 		    sizeof (zone_misc_kstat_t), KM_SLEEP);
2018 	}
2019 }
2020 
2021 static void
2022 zone_kstat_delete_common(kstat_t **pkstat, size_t datasz)
2023 {
2024 	void *data;
2025 
2026 	if (*pkstat != NULL) {
2027 		data = (*pkstat)->ks_data;
2028 		kstat_delete(*pkstat);
2029 		kmem_free(data, datasz);
2030 		*pkstat = NULL;
2031 	}
2032 }
2033 
2034 static void
2035 zone_kstat_delete(zone_t *zone)
2036 {
2037 	zone_kstat_delete_common(&zone->zone_lockedmem_kstat,
2038 	    sizeof (zone_kstat_t));
2039 	zone_kstat_delete_common(&zone->zone_swapresv_kstat,
2040 	    sizeof (zone_kstat_t));
2041 	zone_kstat_delete_common(&zone->zone_nprocs_kstat,
2042 	    sizeof (zone_kstat_t));
2043 	zone_kstat_delete_common(&zone->zone_mcap_ksp,
2044 	    sizeof (zone_mcap_kstat_t));
2045 	zone_kstat_delete_common(&zone->zone_misc_ksp,
2046 	    sizeof (zone_misc_kstat_t));
2047 }
2048 
2049 /*
2050  * Called very early on in boot to initialize the ZSD list so that
2051  * zone_key_create() can be called before zone_init().  It also initializes
2052  * portions of zone0 which may be used before zone_init() is called.  The
2053  * variable "global_zone" will be set when zone0 is fully initialized by
2054  * zone_init().
2055  */
2056 void
2057 zone_zsd_init(void)
2058 {
2059 	mutex_init(&zonehash_lock, NULL, MUTEX_DEFAULT, NULL);
2060 	mutex_init(&zsd_key_lock, NULL, MUTEX_DEFAULT, NULL);
2061 	list_create(&zsd_registered_keys, sizeof (struct zsd_entry),
2062 	    offsetof(struct zsd_entry, zsd_linkage));
2063 	list_create(&zone_active, sizeof (zone_t),
2064 	    offsetof(zone_t, zone_linkage));
2065 	list_create(&zone_deathrow, sizeof (zone_t),
2066 	    offsetof(zone_t, zone_linkage));
2067 
2068 	mutex_init(&zone0.zone_lock, NULL, MUTEX_DEFAULT, NULL);
2069 	mutex_init(&zone0.zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
2070 	mutex_init(&zone0.zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
2071 	zone0.zone_shares = 1;
2072 	zone0.zone_nlwps = 0;
2073 	zone0.zone_nlwps_ctl = INT_MAX;
2074 	zone0.zone_nprocs = 0;
2075 	zone0.zone_nprocs_ctl = INT_MAX;
2076 	zone0.zone_locked_mem = 0;
2077 	zone0.zone_locked_mem_ctl = UINT64_MAX;
2078 	ASSERT(zone0.zone_max_swap == 0);
2079 	zone0.zone_max_swap_ctl = UINT64_MAX;
2080 	zone0.zone_max_lofi = 0;
2081 	zone0.zone_max_lofi_ctl = UINT64_MAX;
2082 	zone0.zone_shmmax = 0;
2083 	zone0.zone_ipc.ipcq_shmmni = 0;
2084 	zone0.zone_ipc.ipcq_semmni = 0;
2085 	zone0.zone_ipc.ipcq_msgmni = 0;
2086 	zone0.zone_name = GLOBAL_ZONENAME;
2087 	zone0.zone_nodename = utsname.nodename;
2088 	zone0.zone_domain = srpc_domain;
2089 	zone0.zone_hostid = HW_INVALID_HOSTID;
2090 	zone0.zone_fs_allowed = NULL;
2091 	psecflags_default(&zone0.zone_secflags);
2092 	zone0.zone_ref = 1;
2093 	zone0.zone_id = GLOBAL_ZONEID;
2094 	zone0.zone_status = ZONE_IS_RUNNING;
2095 	zone0.zone_rootpath = "/";
2096 	zone0.zone_rootpathlen = 2;
2097 	zone0.zone_psetid = ZONE_PS_INVAL;
2098 	zone0.zone_ncpus = 0;
2099 	zone0.zone_ncpus_online = 0;
2100 	zone0.zone_proc_initpid = 1;
2101 	zone0.zone_initname = initname;
2102 	zone0.zone_lockedmem_kstat = NULL;
2103 	zone0.zone_swapresv_kstat = NULL;
2104 	zone0.zone_nprocs_kstat = NULL;
2105 
2106 	list_create(&zone0.zone_ref_list, sizeof (zone_ref_t),
2107 	    offsetof(zone_ref_t, zref_linkage));
2108 	list_create(&zone0.zone_zsd, sizeof (struct zsd_entry),
2109 	    offsetof(struct zsd_entry, zsd_linkage));
2110 	list_insert_head(&zone_active, &zone0);
2111 
2112 	/*
2113 	 * The root filesystem is not mounted yet, so zone_rootvp cannot be set
2114 	 * to anything meaningful.  It is assigned to be 'rootdir' in
2115 	 * vfs_mountroot().
2116 	 */
2117 	zone0.zone_rootvp = NULL;
2118 	zone0.zone_vfslist = NULL;
2119 	zone0.zone_bootargs = initargs;
2120 	zone0.zone_privset = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
2121 	/*
2122 	 * The global zone has all privileges
2123 	 */
2124 	priv_fillset(zone0.zone_privset);
2125 	/*
2126 	 * Add p0 to the global zone
2127 	 */
2128 	zone0.zone_zsched = &p0;
2129 	p0.p_zone = &zone0;
2130 }
2131 
2132 /*
2133  * Compute a hash value based on the contents of the label and the DOI.  The
2134  * hash algorithm is somewhat arbitrary, but is based on the observation that
2135  * humans will likely pick labels that differ by amounts that work out to be
2136  * multiples of the number of hash chains, and thus stirring in some primes
2137  * should help.
2138  */
2139 static uint_t
2140 hash_bylabel(void *hdata, mod_hash_key_t key)
2141 {
2142 	const ts_label_t *lab = (ts_label_t *)key;
2143 	const uint32_t *up, *ue;
2144 	uint_t hash;
2145 	int i;
2146 
2147 	_NOTE(ARGUNUSED(hdata));
2148 
2149 	hash = lab->tsl_doi + (lab->tsl_doi << 1);
2150 	/* we depend on alignment of label, but not representation */
2151 	up = (const uint32_t *)&lab->tsl_label;
2152 	ue = up + sizeof (lab->tsl_label) / sizeof (*up);
2153 	i = 1;
2154 	while (up < ue) {
2155 		/* using 2^n + 1, 1 <= n <= 16 as source of many primes */
2156 		hash += *up + (*up << ((i % 16) + 1));
2157 		up++;
2158 		i++;
2159 	}
2160 	return (hash);
2161 }
2162 
2163 /*
2164  * All that mod_hash cares about here is zero (equal) versus non-zero (not
2165  * equal).  This may need to be changed if less than / greater than is ever
2166  * needed.
2167  */
2168 static int
2169 hash_labelkey_cmp(mod_hash_key_t key1, mod_hash_key_t key2)
2170 {
2171 	ts_label_t *lab1 = (ts_label_t *)key1;
2172 	ts_label_t *lab2 = (ts_label_t *)key2;
2173 
2174 	return (label_equal(lab1, lab2) ? 0 : 1);
2175 }
2176 
2177 /*
2178  * Called by main() to initialize the zones framework.
2179  */
2180 void
2181 zone_init(void)
2182 {
2183 	rctl_dict_entry_t *rde;
2184 	rctl_val_t *dval;
2185 	rctl_set_t *set;
2186 	rctl_alloc_gp_t *gp;
2187 	rctl_entity_p_t e;
2188 	int res;
2189 
2190 	ASSERT(curproc == &p0);
2191 
2192 	/*
2193 	 * Create ID space for zone IDs.  ID 0 is reserved for the
2194 	 * global zone.
2195 	 */
2196 	zoneid_space = id_space_create("zoneid_space", 1, MAX_ZONEID);
2197 
2198 	/*
2199 	 * Initialize generic zone resource controls, if any.
2200 	 */
2201 	rc_zone_cpu_shares = rctl_register("zone.cpu-shares",
2202 	    RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_NEVER |
2203 	    RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT | RCTL_GLOBAL_SYSLOG_NEVER,
2204 	    FSS_MAXSHARES, FSS_MAXSHARES, &zone_cpu_shares_ops);
2205 
2206 	rc_zone_cpu_cap = rctl_register("zone.cpu-cap",
2207 	    RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_ALWAYS |
2208 	    RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |RCTL_GLOBAL_SYSLOG_NEVER |
2209 	    RCTL_GLOBAL_INFINITE,
2210 	    MAXCAP, MAXCAP, &zone_cpu_cap_ops);
2211 
2212 	rc_zone_nlwps = rctl_register("zone.max-lwps", RCENTITY_ZONE,
2213 	    RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
2214 	    INT_MAX, INT_MAX, &zone_lwps_ops);
2215 
2216 	rc_zone_nprocs = rctl_register("zone.max-processes", RCENTITY_ZONE,
2217 	    RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
2218 	    INT_MAX, INT_MAX, &zone_procs_ops);
2219 
2220 	/*
2221 	 * System V IPC resource controls
2222 	 */
2223 	rc_zone_msgmni = rctl_register("zone.max-msg-ids",
2224 	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2225 	    RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_msgmni_ops);
2226 
2227 	rc_zone_semmni = rctl_register("zone.max-sem-ids",
2228 	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2229 	    RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_semmni_ops);
2230 
2231 	rc_zone_shmmni = rctl_register("zone.max-shm-ids",
2232 	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2233 	    RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_shmmni_ops);
2234 
2235 	rc_zone_shmmax = rctl_register("zone.max-shm-memory",
2236 	    RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2237 	    RCTL_GLOBAL_BYTES, UINT64_MAX, UINT64_MAX, &zone_shmmax_ops);
2238 
2239 	/*
2240 	 * Create a rctl_val with PRIVILEGED, NOACTION, value = 1.  Then attach
2241 	 * this at the head of the rctl_dict_entry for ``zone.cpu-shares''.
2242 	 */
2243 	dval = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
2244 	bzero(dval, sizeof (rctl_val_t));
2245 	dval->rcv_value = 1;
2246 	dval->rcv_privilege = RCPRIV_PRIVILEGED;
2247 	dval->rcv_flagaction = RCTL_LOCAL_NOACTION;
2248 	dval->rcv_action_recip_pid = -1;
2249 
2250 	rde = rctl_dict_lookup("zone.cpu-shares");
2251 	(void) rctl_val_list_insert(&rde->rcd_default_value, dval);
2252 
2253 	rc_zone_locked_mem = rctl_register("zone.max-locked-memory",
2254 	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
2255 	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2256 	    &zone_locked_mem_ops);
2257 
2258 	rc_zone_max_swap = rctl_register("zone.max-swap",
2259 	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
2260 	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2261 	    &zone_max_swap_ops);
2262 
2263 	rc_zone_max_lofi = rctl_register("zone.max-lofi",
2264 	    RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |
2265 	    RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2266 	    &zone_max_lofi_ops);
2267 
2268 	/*
2269 	 * Initialize the ``global zone''.
2270 	 */
2271 	set = rctl_set_create();
2272 	gp = rctl_set_init_prealloc(RCENTITY_ZONE);
2273 	mutex_enter(&p0.p_lock);
2274 	e.rcep_p.zone = &zone0;
2275 	e.rcep_t = RCENTITY_ZONE;
2276 	zone0.zone_rctls = rctl_set_init(RCENTITY_ZONE, &p0, &e, set,
2277 	    gp);
2278 
2279 	zone0.zone_nlwps = p0.p_lwpcnt;
2280 	zone0.zone_nprocs = 1;
2281 	zone0.zone_ntasks = 1;
2282 	mutex_exit(&p0.p_lock);
2283 	zone0.zone_restart_init = B_TRUE;
2284 	zone0.zone_brand = &native_brand;
2285 	rctl_prealloc_destroy(gp);
2286 	/*
2287 	 * pool_default hasn't been initialized yet, so we let pool_init()
2288 	 * take care of making sure the global zone is in the default pool.
2289 	 */
2290 
2291 	/*
2292 	 * Initialize global zone kstats
2293 	 */
2294 	zone_kstat_create(&zone0);
2295 
2296 	/*
2297 	 * Initialize zone label.
2298 	 * mlp are initialized when tnzonecfg is loaded.
2299 	 */
2300 	zone0.zone_slabel = l_admin_low;
2301 	rw_init(&zone0.zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
2302 	label_hold(l_admin_low);
2303 
2304 	/*
2305 	 * Initialise the lock for the database structure used by mntfs.
2306 	 */
2307 	rw_init(&zone0.zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL);
2308 
2309 	zone0.zone_ustate = cpu_uarray_zalloc(ZONE_USTATE_MAX, KM_SLEEP);
2310 
2311 	mutex_enter(&zonehash_lock);
2312 	zone_uniqid(&zone0);
2313 	ASSERT(zone0.zone_uniqid == GLOBAL_ZONEUNIQID);
2314 
2315 	zonehashbyid = mod_hash_create_idhash("zone_by_id", zone_hash_size,
2316 	    mod_hash_null_valdtor);
2317 	zonehashbyname = mod_hash_create_strhash("zone_by_name",
2318 	    zone_hash_size, mod_hash_null_valdtor);
2319 	/*
2320 	 * maintain zonehashbylabel only for labeled systems
2321 	 */
2322 	if (is_system_labeled())
2323 		zonehashbylabel = mod_hash_create_extended("zone_by_label",
2324 		    zone_hash_size, mod_hash_null_keydtor,
2325 		    mod_hash_null_valdtor, hash_bylabel, NULL,
2326 		    hash_labelkey_cmp, KM_SLEEP);
2327 	zonecount = 1;
2328 
2329 	(void) mod_hash_insert(zonehashbyid, (mod_hash_key_t)GLOBAL_ZONEID,
2330 	    (mod_hash_val_t)&zone0);
2331 	(void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)zone0.zone_name,
2332 	    (mod_hash_val_t)&zone0);
2333 	if (is_system_labeled()) {
2334 		zone0.zone_flags |= ZF_HASHED_LABEL;
2335 		(void) mod_hash_insert(zonehashbylabel,
2336 		    (mod_hash_key_t)zone0.zone_slabel, (mod_hash_val_t)&zone0);
2337 	}
2338 	mutex_exit(&zonehash_lock);
2339 
2340 	/*
2341 	 * We avoid setting zone_kcred until now, since kcred is initialized
2342 	 * sometime after zone_zsd_init() and before zone_init().
2343 	 */
2344 	zone0.zone_kcred = kcred;
2345 	/*
2346 	 * The global zone is fully initialized (except for zone_rootvp which
2347 	 * will be set when the root filesystem is mounted).
2348 	 */
2349 	global_zone = &zone0;
2350 
2351 	/*
2352 	 * Setup an event channel to send zone status change notifications on
2353 	 */
2354 	res = sysevent_evc_bind(ZONE_EVENT_CHANNEL, &zone_event_chan,
2355 	    EVCH_CREAT);
2356 
2357 	if (res)
2358 		panic("Sysevent_evc_bind failed during zone setup.\n");
2359 
2360 }
2361 
2362 static void
2363 zone_free(zone_t *zone)
2364 {
2365 	ASSERT(zone != global_zone);
2366 	ASSERT(zone->zone_ntasks == 0);
2367 	ASSERT(zone->zone_nlwps == 0);
2368 	ASSERT(zone->zone_nprocs == 0);
2369 	ASSERT(zone->zone_cred_ref == 0);
2370 	ASSERT(zone->zone_kcred == NULL);
2371 	ASSERT(zone_status_get(zone) == ZONE_IS_DEAD ||
2372 	    zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
2373 	ASSERT(list_is_empty(&zone->zone_ref_list));
2374 
2375 	/*
2376 	 * Remove any zone caps.
2377 	 */
2378 	cpucaps_zone_remove(zone);
2379 
2380 	ASSERT(zone->zone_cpucap == NULL);
2381 
2382 	/* remove from deathrow list */
2383 	if (zone_status_get(zone) == ZONE_IS_DEAD) {
2384 		ASSERT(zone->zone_ref == 0);
2385 		mutex_enter(&zone_deathrow_lock);
2386 		list_remove(&zone_deathrow, zone);
2387 		mutex_exit(&zone_deathrow_lock);
2388 	}
2389 
2390 	list_destroy(&zone->zone_ref_list);
2391 	zone_free_zsd(zone);
2392 	zone_free_datasets(zone);
2393 	list_destroy(&zone->zone_dl_list);
2394 
2395 	cpu_uarray_free(zone->zone_ustate);
2396 
2397 	if (zone->zone_rootvp != NULL)
2398 		VN_RELE(zone->zone_rootvp);
2399 	if (zone->zone_rootpath)
2400 		kmem_free(zone->zone_rootpath, zone->zone_rootpathlen);
2401 	if (zone->zone_name != NULL)
2402 		kmem_free(zone->zone_name, ZONENAME_MAX);
2403 	if (zone->zone_slabel != NULL)
2404 		label_rele(zone->zone_slabel);
2405 	if (zone->zone_nodename != NULL)
2406 		kmem_free(zone->zone_nodename, _SYS_NMLN);
2407 	if (zone->zone_domain != NULL)
2408 		kmem_free(zone->zone_domain, _SYS_NMLN);
2409 	if (zone->zone_privset != NULL)
2410 		kmem_free(zone->zone_privset, sizeof (priv_set_t));
2411 	if (zone->zone_rctls != NULL)
2412 		rctl_set_free(zone->zone_rctls);
2413 	if (zone->zone_bootargs != NULL)
2414 		strfree(zone->zone_bootargs);
2415 	if (zone->zone_initname != NULL)
2416 		strfree(zone->zone_initname);
2417 	if (zone->zone_fs_allowed != NULL)
2418 		strfree(zone->zone_fs_allowed);
2419 	if (zone->zone_pfexecd != NULL)
2420 		klpd_freelist(&zone->zone_pfexecd);
2421 	id_free(zoneid_space, zone->zone_id);
2422 	mutex_destroy(&zone->zone_lock);
2423 	cv_destroy(&zone->zone_cv);
2424 	rw_destroy(&zone->zone_mlps.mlpl_rwlock);
2425 	rw_destroy(&zone->zone_mntfs_db_lock);
2426 	kmem_free(zone, sizeof (zone_t));
2427 }
2428 
2429 /*
2430  * See block comment at the top of this file for information about zone
2431  * status values.
2432  */
2433 /*
2434  * Convenience function for setting zone status.
2435  */
2436 static void
2437 zone_status_set(zone_t *zone, zone_status_t status)
2438 {
2439 
2440 	nvlist_t *nvl = NULL;
2441 	ASSERT(MUTEX_HELD(&zone_status_lock));
2442 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE &&
2443 	    status >= zone_status_get(zone));
2444 
2445 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) ||
2446 	    nvlist_add_string(nvl, ZONE_CB_NAME, zone->zone_name) ||
2447 	    nvlist_add_string(nvl, ZONE_CB_NEWSTATE,
2448 	    zone_status_table[status]) ||
2449 	    nvlist_add_string(nvl, ZONE_CB_OLDSTATE,
2450 	    zone_status_table[zone->zone_status]) ||
2451 	    nvlist_add_int32(nvl, ZONE_CB_ZONEID, zone->zone_id) ||
2452 	    nvlist_add_uint64(nvl, ZONE_CB_TIMESTAMP, (uint64_t)gethrtime()) ||
2453 	    sysevent_evc_publish(zone_event_chan, ZONE_EVENT_STATUS_CLASS,
2454 	    ZONE_EVENT_STATUS_SUBCLASS, "sun.com", "kernel", nvl, EVCH_SLEEP)) {
2455 #ifdef DEBUG
2456 		(void) printf(
2457 		    "Failed to allocate and send zone state change event.\n");
2458 #endif
2459 	}
2460 	nvlist_free(nvl);
2461 
2462 	zone->zone_status = status;
2463 
2464 	cv_broadcast(&zone->zone_cv);
2465 }
2466 
2467 /*
2468  * Public function to retrieve the zone status.  The zone status may
2469  * change after it is retrieved.
2470  */
2471 zone_status_t
2472 zone_status_get(zone_t *zone)
2473 {
2474 	return (zone->zone_status);
2475 }
2476 
2477 static int
2478 zone_set_bootargs(zone_t *zone, const char *zone_bootargs)
2479 {
2480 	char *buf = kmem_zalloc(BOOTARGS_MAX, KM_SLEEP);
2481 	int err = 0;
2482 
2483 	ASSERT(zone != global_zone);
2484 	if ((err = copyinstr(zone_bootargs, buf, BOOTARGS_MAX, NULL)) != 0)
2485 		goto done;	/* EFAULT or ENAMETOOLONG */
2486 
2487 	if (zone->zone_bootargs != NULL)
2488 		strfree(zone->zone_bootargs);
2489 
2490 	zone->zone_bootargs = strdup(buf);
2491 
2492 done:
2493 	kmem_free(buf, BOOTARGS_MAX);
2494 	return (err);
2495 }
2496 
2497 static int
2498 zone_set_brand(zone_t *zone, const char *brand)
2499 {
2500 	struct brand_attr *attrp;
2501 	brand_t *bp;
2502 
2503 	attrp = kmem_alloc(sizeof (struct brand_attr), KM_SLEEP);
2504 	if (copyin(brand, attrp, sizeof (struct brand_attr)) != 0) {
2505 		kmem_free(attrp, sizeof (struct brand_attr));
2506 		return (EFAULT);
2507 	}
2508 
2509 	bp = brand_register_zone(attrp);
2510 	kmem_free(attrp, sizeof (struct brand_attr));
2511 	if (bp == NULL)
2512 		return (EINVAL);
2513 
2514 	/*
2515 	 * This is the only place where a zone can change it's brand.
2516 	 * We already need to hold zone_status_lock to check the zone
2517 	 * status, so we'll just use that lock to serialize zone
2518 	 * branding requests as well.
2519 	 */
2520 	mutex_enter(&zone_status_lock);
2521 
2522 	/* Re-Branding is not allowed and the zone can't be booted yet */
2523 	if ((ZONE_IS_BRANDED(zone)) ||
2524 	    (zone_status_get(zone) >= ZONE_IS_BOOTING)) {
2525 		mutex_exit(&zone_status_lock);
2526 		brand_unregister_zone(bp);
2527 		return (EINVAL);
2528 	}
2529 
2530 	/* set up the brand specific data */
2531 	zone->zone_brand = bp;
2532 	ZBROP(zone)->b_init_brand_data(zone);
2533 
2534 	mutex_exit(&zone_status_lock);
2535 	return (0);
2536 }
2537 
2538 static int
2539 zone_set_secflags(zone_t *zone, const psecflags_t *zone_secflags)
2540 {
2541 	int err = 0;
2542 	psecflags_t psf;
2543 
2544 	ASSERT(zone != global_zone);
2545 
2546 	if ((err = copyin(zone_secflags, &psf, sizeof (psf))) != 0)
2547 		return (err);
2548 
2549 	if (zone_status_get(zone) > ZONE_IS_READY)
2550 		return (EINVAL);
2551 
2552 	if (!psecflags_validate(&psf))
2553 		return (EINVAL);
2554 
2555 	(void) memcpy(&zone->zone_secflags, &psf, sizeof (psf));
2556 
2557 	/* Set security flags on the zone's zsched */
2558 	(void) memcpy(&zone->zone_zsched->p_secflags, &zone->zone_secflags,
2559 	    sizeof (zone->zone_zsched->p_secflags));
2560 
2561 	return (0);
2562 }
2563 
2564 static int
2565 zone_set_fs_allowed(zone_t *zone, const char *zone_fs_allowed)
2566 {
2567 	char *buf = kmem_zalloc(ZONE_FS_ALLOWED_MAX, KM_SLEEP);
2568 	int err = 0;
2569 
2570 	ASSERT(zone != global_zone);
2571 	if ((err = copyinstr(zone_fs_allowed, buf,
2572 	    ZONE_FS_ALLOWED_MAX, NULL)) != 0)
2573 		goto done;
2574 
2575 	if (zone->zone_fs_allowed != NULL)
2576 		strfree(zone->zone_fs_allowed);
2577 
2578 	zone->zone_fs_allowed = strdup(buf);
2579 
2580 done:
2581 	kmem_free(buf, ZONE_FS_ALLOWED_MAX);
2582 	return (err);
2583 }
2584 
2585 static int
2586 zone_set_initname(zone_t *zone, const char *zone_initname)
2587 {
2588 	char initname[INITNAME_SZ];
2589 	size_t len;
2590 	int err = 0;
2591 
2592 	ASSERT(zone != global_zone);
2593 	if ((err = copyinstr(zone_initname, initname, INITNAME_SZ, &len)) != 0)
2594 		return (err);	/* EFAULT or ENAMETOOLONG */
2595 
2596 	if (zone->zone_initname != NULL)
2597 		strfree(zone->zone_initname);
2598 
2599 	zone->zone_initname = kmem_alloc(strlen(initname) + 1, KM_SLEEP);
2600 	(void) strcpy(zone->zone_initname, initname);
2601 	return (0);
2602 }
2603 
2604 static int
2605 zone_set_phys_mcap(zone_t *zone, const uint64_t *zone_mcap)
2606 {
2607 	uint64_t mcap;
2608 	int err = 0;
2609 
2610 	if ((err = copyin(zone_mcap, &mcap, sizeof (uint64_t))) == 0)
2611 		zone->zone_phys_mcap = mcap;
2612 
2613 	return (err);
2614 }
2615 
2616 static int
2617 zone_set_sched_class(zone_t *zone, const char *new_class)
2618 {
2619 	char sched_class[PC_CLNMSZ];
2620 	id_t classid;
2621 	int err;
2622 
2623 	ASSERT(zone != global_zone);
2624 	if ((err = copyinstr(new_class, sched_class, PC_CLNMSZ, NULL)) != 0)
2625 		return (err);	/* EFAULT or ENAMETOOLONG */
2626 
2627 	if (getcid(sched_class, &classid) != 0 || CLASS_KERNEL(classid))
2628 		return (set_errno(EINVAL));
2629 	zone->zone_defaultcid = classid;
2630 	ASSERT(zone->zone_defaultcid > 0 &&
2631 	    zone->zone_defaultcid < loaded_classes);
2632 
2633 	return (0);
2634 }
2635 
2636 /*
2637  * Block indefinitely waiting for (zone_status >= status)
2638  */
2639 void
2640 zone_status_wait(zone_t *zone, zone_status_t status)
2641 {
2642 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2643 
2644 	mutex_enter(&zone_status_lock);
2645 	while (zone->zone_status < status) {
2646 		cv_wait(&zone->zone_cv, &zone_status_lock);
2647 	}
2648 	mutex_exit(&zone_status_lock);
2649 }
2650 
2651 /*
2652  * Private CPR-safe version of zone_status_wait().
2653  */
2654 static void
2655 zone_status_wait_cpr(zone_t *zone, zone_status_t status, char *str)
2656 {
2657 	callb_cpr_t cprinfo;
2658 
2659 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2660 
2661 	CALLB_CPR_INIT(&cprinfo, &zone_status_lock, callb_generic_cpr,
2662 	    str);
2663 	mutex_enter(&zone_status_lock);
2664 	while (zone->zone_status < status) {
2665 		CALLB_CPR_SAFE_BEGIN(&cprinfo);
2666 		cv_wait(&zone->zone_cv, &zone_status_lock);
2667 		CALLB_CPR_SAFE_END(&cprinfo, &zone_status_lock);
2668 	}
2669 	/*
2670 	 * zone_status_lock is implicitly released by the following.
2671 	 */
2672 	CALLB_CPR_EXIT(&cprinfo);
2673 }
2674 
2675 /*
2676  * Block until zone enters requested state or signal is received.  Return (0)
2677  * if signaled, non-zero otherwise.
2678  */
2679 int
2680 zone_status_wait_sig(zone_t *zone, zone_status_t status)
2681 {
2682 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2683 
2684 	mutex_enter(&zone_status_lock);
2685 	while (zone->zone_status < status) {
2686 		if (!cv_wait_sig(&zone->zone_cv, &zone_status_lock)) {
2687 			mutex_exit(&zone_status_lock);
2688 			return (0);
2689 		}
2690 	}
2691 	mutex_exit(&zone_status_lock);
2692 	return (1);
2693 }
2694 
2695 /*
2696  * Block until the zone enters the requested state or the timeout expires,
2697  * whichever happens first.  Return (-1) if operation timed out, time remaining
2698  * otherwise.
2699  */
2700 clock_t
2701 zone_status_timedwait(zone_t *zone, clock_t tim, zone_status_t status)
2702 {
2703 	clock_t timeleft = 0;
2704 
2705 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2706 
2707 	mutex_enter(&zone_status_lock);
2708 	while (zone->zone_status < status && timeleft != -1) {
2709 		timeleft = cv_timedwait(&zone->zone_cv, &zone_status_lock, tim);
2710 	}
2711 	mutex_exit(&zone_status_lock);
2712 	return (timeleft);
2713 }
2714 
2715 /*
2716  * Block until the zone enters the requested state, the current process is
2717  * signaled,  or the timeout expires, whichever happens first.  Return (-1) if
2718  * operation timed out, 0 if signaled, time remaining otherwise.
2719  */
2720 clock_t
2721 zone_status_timedwait_sig(zone_t *zone, clock_t tim, zone_status_t status)
2722 {
2723 	clock_t timeleft = tim - ddi_get_lbolt();
2724 
2725 	ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2726 
2727 	mutex_enter(&zone_status_lock);
2728 	while (zone->zone_status < status) {
2729 		timeleft = cv_timedwait_sig(&zone->zone_cv, &zone_status_lock,
2730 		    tim);
2731 		if (timeleft <= 0)
2732 			break;
2733 	}
2734 	mutex_exit(&zone_status_lock);
2735 	return (timeleft);
2736 }
2737 
2738 /*
2739  * Zones have two reference counts: one for references from credential
2740  * structures (zone_cred_ref), and one (zone_ref) for everything else.
2741  * This is so we can allow a zone to be rebooted while there are still
2742  * outstanding cred references, since certain drivers cache dblks (which
2743  * implicitly results in cached creds).  We wait for zone_ref to drop to
2744  * 0 (actually 1), but not zone_cred_ref.  The zone structure itself is
2745  * later freed when the zone_cred_ref drops to 0, though nothing other
2746  * than the zone id and privilege set should be accessed once the zone
2747  * is "dead".
2748  *
2749  * A debugging flag, zone_wait_for_cred, can be set to a non-zero value
2750  * to force halt/reboot to block waiting for the zone_cred_ref to drop
2751  * to 0.  This can be useful to flush out other sources of cached creds
2752  * that may be less innocuous than the driver case.
2753  *
2754  * Zones also provide a tracked reference counting mechanism in which zone
2755  * references are represented by "crumbs" (zone_ref structures).  Crumbs help
2756  * debuggers determine the sources of leaked zone references.  See
2757  * zone_hold_ref() and zone_rele_ref() below for more information.
2758  */
2759 
2760 int zone_wait_for_cred = 0;
2761 
2762 static void
2763 zone_hold_locked(zone_t *z)
2764 {
2765 	ASSERT(MUTEX_HELD(&z->zone_lock));
2766 	z->zone_ref++;
2767 	ASSERT(z->zone_ref != 0);
2768 }
2769 
2770 /*
2771  * Increment the specified zone's reference count.  The zone's zone_t structure
2772  * will not be freed as long as the zone's reference count is nonzero.
2773  * Decrement the zone's reference count via zone_rele().
2774  *
2775  * NOTE: This function should only be used to hold zones for short periods of
2776  * time.  Use zone_hold_ref() if the zone must be held for a long time.
2777  */
2778 void
2779 zone_hold(zone_t *z)
2780 {
2781 	mutex_enter(&z->zone_lock);
2782 	zone_hold_locked(z);
2783 	mutex_exit(&z->zone_lock);
2784 }
2785 
2786 /*
2787  * If the non-cred ref count drops to 1 and either the cred ref count
2788  * is 0 or we aren't waiting for cred references, the zone is ready to
2789  * be destroyed.
2790  */
2791 #define	ZONE_IS_UNREF(zone)	((zone)->zone_ref == 1 && \
2792 	    (!zone_wait_for_cred || (zone)->zone_cred_ref == 0))
2793 
2794 /*
2795  * Common zone reference release function invoked by zone_rele() and
2796  * zone_rele_ref().  If subsys is ZONE_REF_NUM_SUBSYS, then the specified
2797  * zone's subsystem-specific reference counters are not affected by the
2798  * release.  If ref is not NULL, then the zone_ref_t to which it refers is
2799  * removed from the specified zone's reference list.  ref must be non-NULL iff
2800  * subsys is not ZONE_REF_NUM_SUBSYS.
2801  */
2802 static void
2803 zone_rele_common(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
2804 {
2805 	boolean_t wakeup;
2806 
2807 	mutex_enter(&z->zone_lock);
2808 	ASSERT(z->zone_ref != 0);
2809 	z->zone_ref--;
2810 	if (subsys != ZONE_REF_NUM_SUBSYS) {
2811 		ASSERT(z->zone_subsys_ref[subsys] != 0);
2812 		z->zone_subsys_ref[subsys]--;
2813 		list_remove(&z->zone_ref_list, ref);
2814 	}
2815 	if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
2816 		/* no more refs, free the structure */
2817 		mutex_exit(&z->zone_lock);
2818 		zone_free(z);
2819 		return;
2820 	}
2821 	/* signal zone_destroy so the zone can finish halting */
2822 	wakeup = (ZONE_IS_UNREF(z) && zone_status_get(z) >= ZONE_IS_DEAD);
2823 	mutex_exit(&z->zone_lock);
2824 
2825 	if (wakeup) {
2826 		/*
2827 		 * Grabbing zonehash_lock here effectively synchronizes with
2828 		 * zone_destroy() to avoid missed signals.
2829 		 */
2830 		mutex_enter(&zonehash_lock);
2831 		cv_broadcast(&zone_destroy_cv);
2832 		mutex_exit(&zonehash_lock);
2833 	}
2834 }
2835 
2836 /*
2837  * Decrement the specified zone's reference count.  The specified zone will
2838  * cease to exist after this function returns if the reference count drops to
2839  * zero.  This function should be paired with zone_hold().
2840  */
2841 void
2842 zone_rele(zone_t *z)
2843 {
2844 	zone_rele_common(z, NULL, ZONE_REF_NUM_SUBSYS);
2845 }
2846 
2847 /*
2848  * Initialize a zone reference structure.  This function must be invoked for
2849  * a reference structure before the structure is passed to zone_hold_ref().
2850  */
2851 void
2852 zone_init_ref(zone_ref_t *ref)
2853 {
2854 	ref->zref_zone = NULL;
2855 	list_link_init(&ref->zref_linkage);
2856 }
2857 
2858 /*
2859  * Acquire a reference to zone z.  The caller must specify the
2860  * zone_ref_subsys_t constant associated with its subsystem.  The specified
2861  * zone_ref_t structure will represent a reference to the specified zone.  Use
2862  * zone_rele_ref() to release the reference.
2863  *
2864  * The referenced zone_t structure will not be freed as long as the zone_t's
2865  * zone_status field is not ZONE_IS_DEAD and the zone has outstanding
2866  * references.
2867  *
2868  * NOTE: The zone_ref_t structure must be initialized before it is used.
2869  * See zone_init_ref() above.
2870  */
2871 void
2872 zone_hold_ref(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
2873 {
2874 	ASSERT(subsys >= 0 && subsys < ZONE_REF_NUM_SUBSYS);
2875 
2876 	/*
2877 	 * Prevent consumers from reusing a reference structure before
2878 	 * releasing it.
2879 	 */
2880 	VERIFY(ref->zref_zone == NULL);
2881 
2882 	ref->zref_zone = z;
2883 	mutex_enter(&z->zone_lock);
2884 	zone_hold_locked(z);
2885 	z->zone_subsys_ref[subsys]++;
2886 	ASSERT(z->zone_subsys_ref[subsys] != 0);
2887 	list_insert_head(&z->zone_ref_list, ref);
2888 	mutex_exit(&z->zone_lock);
2889 }
2890 
2891 /*
2892  * Release the zone reference represented by the specified zone_ref_t.
2893  * The reference is invalid after it's released; however, the zone_ref_t
2894  * structure can be reused without having to invoke zone_init_ref().
2895  * subsys should be the same value that was passed to zone_hold_ref()
2896  * when the reference was acquired.
2897  */
2898 void
2899 zone_rele_ref(zone_ref_t *ref, zone_ref_subsys_t subsys)
2900 {
2901 	zone_rele_common(ref->zref_zone, ref, subsys);
2902 
2903 	/*
2904 	 * Set the zone_ref_t's zref_zone field to NULL to generate panics
2905 	 * when consumers dereference the reference.  This helps us catch
2906 	 * consumers who use released references.  Furthermore, this lets
2907 	 * consumers reuse the zone_ref_t structure without having to
2908 	 * invoke zone_init_ref().
2909 	 */
2910 	ref->zref_zone = NULL;
2911 }
2912 
2913 void
2914 zone_cred_hold(zone_t *z)
2915 {
2916 	mutex_enter(&z->zone_lock);
2917 	z->zone_cred_ref++;
2918 	ASSERT(z->zone_cred_ref != 0);
2919 	mutex_exit(&z->zone_lock);
2920 }
2921 
2922 void
2923 zone_cred_rele(zone_t *z)
2924 {
2925 	boolean_t wakeup;
2926 
2927 	mutex_enter(&z->zone_lock);
2928 	ASSERT(z->zone_cred_ref != 0);
2929 	z->zone_cred_ref--;
2930 	if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
2931 		/* no more refs, free the structure */
2932 		mutex_exit(&z->zone_lock);
2933 		zone_free(z);
2934 		return;
2935 	}
2936 	/*
2937 	 * If zone_destroy is waiting for the cred references to drain
2938 	 * out, and they have, signal it.
2939 	 */
2940 	wakeup = (zone_wait_for_cred && ZONE_IS_UNREF(z) &&
2941 	    zone_status_get(z) >= ZONE_IS_DEAD);
2942 	mutex_exit(&z->zone_lock);
2943 
2944 	if (wakeup) {
2945 		/*
2946 		 * Grabbing zonehash_lock here effectively synchronizes with
2947 		 * zone_destroy() to avoid missed signals.
2948 		 */
2949 		mutex_enter(&zonehash_lock);
2950 		cv_broadcast(&zone_destroy_cv);
2951 		mutex_exit(&zonehash_lock);
2952 	}
2953 }
2954 
2955 void
2956 zone_task_hold(zone_t *z)
2957 {
2958 	mutex_enter(&z->zone_lock);
2959 	z->zone_ntasks++;
2960 	ASSERT(z->zone_ntasks != 0);
2961 	mutex_exit(&z->zone_lock);
2962 }
2963 
2964 void
2965 zone_task_rele(zone_t *zone)
2966 {
2967 	uint_t refcnt;
2968 
2969 	mutex_enter(&zone->zone_lock);
2970 	ASSERT(zone->zone_ntasks != 0);
2971 	refcnt = --zone->zone_ntasks;
2972 	if (refcnt > 1)	{	/* Common case */
2973 		mutex_exit(&zone->zone_lock);
2974 		return;
2975 	}
2976 	zone_hold_locked(zone);	/* so we can use the zone_t later */
2977 	mutex_exit(&zone->zone_lock);
2978 	if (refcnt == 1) {
2979 		/*
2980 		 * See if the zone is shutting down.
2981 		 */
2982 		mutex_enter(&zone_status_lock);
2983 		if (zone_status_get(zone) != ZONE_IS_SHUTTING_DOWN) {
2984 			goto out;
2985 		}
2986 
2987 		/*
2988 		 * Make sure the ntasks didn't change since we
2989 		 * dropped zone_lock.
2990 		 */
2991 		mutex_enter(&zone->zone_lock);
2992 		if (refcnt != zone->zone_ntasks) {
2993 			mutex_exit(&zone->zone_lock);
2994 			goto out;
2995 		}
2996 		mutex_exit(&zone->zone_lock);
2997 
2998 		/*
2999 		 * No more user processes in the zone.  The zone is empty.
3000 		 */
3001 		zone_status_set(zone, ZONE_IS_EMPTY);
3002 		goto out;
3003 	}
3004 
3005 	ASSERT(refcnt == 0);
3006 	/*
3007 	 * zsched has exited; the zone is dead.
3008 	 */
3009 	zone->zone_zsched = NULL;		/* paranoia */
3010 	mutex_enter(&zone_status_lock);
3011 	zone_status_set(zone, ZONE_IS_DEAD);
3012 out:
3013 	mutex_exit(&zone_status_lock);
3014 	zone_rele(zone);
3015 }
3016 
3017 zoneid_t
3018 getzoneid(void)
3019 {
3020 	return (curproc->p_zone->zone_id);
3021 }
3022 
3023 /*
3024  * Internal versions of zone_find_by_*().  These don't zone_hold() or
3025  * check the validity of a zone's state.
3026  */
3027 static zone_t *
3028 zone_find_all_by_id(zoneid_t zoneid)
3029 {
3030 	mod_hash_val_t hv;
3031 	zone_t *zone = NULL;
3032 
3033 	ASSERT(MUTEX_HELD(&zonehash_lock));
3034 
3035 	if (mod_hash_find(zonehashbyid,
3036 	    (mod_hash_key_t)(uintptr_t)zoneid, &hv) == 0)
3037 		zone = (zone_t *)hv;
3038 	return (zone);
3039 }
3040 
3041 static zone_t *
3042 zone_find_all_by_label(const ts_label_t *label)
3043 {
3044 	mod_hash_val_t hv;
3045 	zone_t *zone = NULL;
3046 
3047 	ASSERT(MUTEX_HELD(&zonehash_lock));
3048 
3049 	/*
3050 	 * zonehashbylabel is not maintained for unlabeled systems
3051 	 */
3052 	if (!is_system_labeled())
3053 		return (NULL);
3054 	if (mod_hash_find(zonehashbylabel, (mod_hash_key_t)label, &hv) == 0)
3055 		zone = (zone_t *)hv;
3056 	return (zone);
3057 }
3058 
3059 static zone_t *
3060 zone_find_all_by_name(char *name)
3061 {
3062 	mod_hash_val_t hv;
3063 	zone_t *zone = NULL;
3064 
3065 	ASSERT(MUTEX_HELD(&zonehash_lock));
3066 
3067 	if (mod_hash_find(zonehashbyname, (mod_hash_key_t)name, &hv) == 0)
3068 		zone = (zone_t *)hv;
3069 	return (zone);
3070 }
3071 
3072 /*
3073  * Public interface for looking up a zone by zoneid.  Only returns the zone if
3074  * it is fully initialized, and has not yet begun the zone_destroy() sequence.
3075  * Caller must call zone_rele() once it is done with the zone.
3076  *
3077  * The zone may begin the zone_destroy() sequence immediately after this
3078  * function returns, but may be safely used until zone_rele() is called.
3079  */
3080 zone_t *
3081 zone_find_by_id(zoneid_t zoneid)
3082 {
3083 	zone_t *zone;
3084 	zone_status_t status;
3085 
3086 	mutex_enter(&zonehash_lock);
3087 	if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
3088 		mutex_exit(&zonehash_lock);
3089 		return (NULL);
3090 	}
3091 	status = zone_status_get(zone);
3092 	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3093 		/*
3094 		 * For all practical purposes the zone doesn't exist.
3095 		 */
3096 		mutex_exit(&zonehash_lock);
3097 		return (NULL);
3098 	}
3099 	zone_hold(zone);
3100 	mutex_exit(&zonehash_lock);
3101 	return (zone);
3102 }
3103 
3104 /*
3105  * Similar to zone_find_by_id, but using zone label as the key.
3106  */
3107 zone_t *
3108 zone_find_by_label(const ts_label_t *label)
3109 {
3110 	zone_t *zone;
3111 	zone_status_t status;
3112 
3113 	mutex_enter(&zonehash_lock);
3114 	if ((zone = zone_find_all_by_label(label)) == NULL) {
3115 		mutex_exit(&zonehash_lock);
3116 		return (NULL);
3117 	}
3118 
3119 	status = zone_status_get(zone);
3120 	if (status > ZONE_IS_DOWN) {
3121 		/*
3122 		 * For all practical purposes the zone doesn't exist.
3123 		 */
3124 		mutex_exit(&zonehash_lock);
3125 		return (NULL);
3126 	}
3127 	zone_hold(zone);
3128 	mutex_exit(&zonehash_lock);
3129 	return (zone);
3130 }
3131 
3132 /*
3133  * Similar to zone_find_by_id, but using zone name as the key.
3134  */
3135 zone_t *
3136 zone_find_by_name(char *name)
3137 {
3138 	zone_t *zone;
3139 	zone_status_t status;
3140 
3141 	mutex_enter(&zonehash_lock);
3142 	if ((zone = zone_find_all_by_name(name)) == NULL) {
3143 		mutex_exit(&zonehash_lock);
3144 		return (NULL);
3145 	}
3146 	status = zone_status_get(zone);
3147 	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3148 		/*
3149 		 * For all practical purposes the zone doesn't exist.
3150 		 */
3151 		mutex_exit(&zonehash_lock);
3152 		return (NULL);
3153 	}
3154 	zone_hold(zone);
3155 	mutex_exit(&zonehash_lock);
3156 	return (zone);
3157 }
3158 
3159 /*
3160  * Similar to zone_find_by_id(), using the path as a key.  For instance,
3161  * if there is a zone "foo" rooted at /foo/root, and the path argument
3162  * is "/foo/root/proc", it will return the held zone_t corresponding to
3163  * zone "foo".
3164  *
3165  * zone_find_by_path() always returns a non-NULL value, since at the
3166  * very least every path will be contained in the global zone.
3167  *
3168  * As with the other zone_find_by_*() functions, the caller is
3169  * responsible for zone_rele()ing the return value of this function.
3170  */
3171 zone_t *
3172 zone_find_by_path(const char *path)
3173 {
3174 	zone_t *zone;
3175 	zone_t *zret = NULL;
3176 	zone_status_t status;
3177 
3178 	if (path == NULL) {
3179 		/*
3180 		 * Call from rootconf().
3181 		 */
3182 		zone_hold(global_zone);
3183 		return (global_zone);
3184 	}
3185 	ASSERT(*path == '/');
3186 	mutex_enter(&zonehash_lock);
3187 	for (zone = list_head(&zone_active); zone != NULL;
3188 	    zone = list_next(&zone_active, zone)) {
3189 		if (ZONE_PATH_VISIBLE(path, zone))
3190 			zret = zone;
3191 	}
3192 	ASSERT(zret != NULL);
3193 	status = zone_status_get(zret);
3194 	if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3195 		/*
3196 		 * Zone practically doesn't exist.
3197 		 */
3198 		zret = global_zone;
3199 	}
3200 	zone_hold(zret);
3201 	mutex_exit(&zonehash_lock);
3202 	return (zret);
3203 }
3204 
3205 /*
3206  * Public interface for updating per-zone load averages.  Called once per
3207  * second.
3208  *
3209  * Based on loadavg_update(), genloadavg() and calcloadavg() from clock.c.
3210  */
3211 void
3212 zone_loadavg_update(void)
3213 {
3214 	zone_t *zp;
3215 	zone_status_t status;
3216 	struct loadavg_s *lavg;
3217 	hrtime_t zone_total;
3218 	uint64_t tmp;
3219 	int i;
3220 	hrtime_t hr_avg;
3221 	int nrun;
3222 	static int64_t f[3] = { 135, 27, 9 };
3223 	int64_t q, r;
3224 
3225 	mutex_enter(&zonehash_lock);
3226 	for (zp = list_head(&zone_active); zp != NULL;
3227 	    zp = list_next(&zone_active, zp)) {
3228 		mutex_enter(&zp->zone_lock);
3229 
3230 		/* Skip zones that are on the way down or not yet up */
3231 		status = zone_status_get(zp);
3232 		if (status < ZONE_IS_READY || status >= ZONE_IS_DOWN) {
3233 			/* For all practical purposes the zone doesn't exist. */
3234 			mutex_exit(&zp->zone_lock);
3235 			continue;
3236 		}
3237 
3238 		/*
3239 		 * Update the 10 second moving average data in zone_loadavg.
3240 		 */
3241 		lavg = &zp->zone_loadavg;
3242 
3243 		tmp = cpu_uarray_sum_all(zp->zone_ustate);
3244 		zone_total = UINT64_OVERFLOW_TO_INT64(tmp);
3245 
3246 		scalehrtime(&zone_total);
3247 
3248 		/* The zone_total should always be increasing. */
3249 		lavg->lg_loads[lavg->lg_cur] = (zone_total > lavg->lg_total) ?
3250 		    zone_total - lavg->lg_total : 0;
3251 		lavg->lg_cur = (lavg->lg_cur + 1) % S_LOADAVG_SZ;
3252 		/* lg_total holds the prev. 1 sec. total */
3253 		lavg->lg_total = zone_total;
3254 
3255 		/*
3256 		 * To simplify the calculation, we don't calculate the load avg.
3257 		 * until the zone has been up for at least 10 seconds and our
3258 		 * moving average is thus full.
3259 		 */
3260 		if ((lavg->lg_len + 1) < S_LOADAVG_SZ) {
3261 			lavg->lg_len++;
3262 			mutex_exit(&zp->zone_lock);
3263 			continue;
3264 		}
3265 
3266 		/* Now calculate the 1min, 5min, 15 min load avg. */
3267 		hr_avg = 0;
3268 		for (i = 0; i < S_LOADAVG_SZ; i++)
3269 			hr_avg += lavg->lg_loads[i];
3270 		hr_avg = hr_avg / S_LOADAVG_SZ;
3271 		nrun = hr_avg / (NANOSEC / LGRP_LOADAVG_IN_THREAD_MAX);
3272 
3273 		/* Compute load avg. See comment in calcloadavg() */
3274 		for (i = 0; i < 3; i++) {
3275 			q = (zp->zone_hp_avenrun[i] >> 16) << 7;
3276 			r = (zp->zone_hp_avenrun[i] & 0xffff) << 7;
3277 			zp->zone_hp_avenrun[i] +=
3278 			    ((nrun - q) * f[i] - ((r * f[i]) >> 16)) >> 4;
3279 
3280 			/* avenrun[] can only hold 31 bits of load avg. */
3281 			if (zp->zone_hp_avenrun[i] <
3282 			    ((uint64_t)1<<(31+16-FSHIFT)))
3283 				zp->zone_avenrun[i] = (int32_t)
3284 				    (zp->zone_hp_avenrun[i] >> (16 - FSHIFT));
3285 			else
3286 				zp->zone_avenrun[i] = 0x7fffffff;
3287 		}
3288 
3289 		mutex_exit(&zp->zone_lock);
3290 	}
3291 	mutex_exit(&zonehash_lock);
3292 }
3293 
3294 /*
3295  * Get the number of cpus visible to this zone.  The system-wide global
3296  * 'ncpus' is returned if pools are disabled, the caller is in the
3297  * global zone, or a NULL zone argument is passed in.
3298  */
3299 int
3300 zone_ncpus_get(zone_t *zone)
3301 {
3302 	int myncpus = zone == NULL ? 0 : zone->zone_ncpus;
3303 
3304 	return (myncpus != 0 ? myncpus : ncpus);
3305 }
3306 
3307 /*
3308  * Get the number of online cpus visible to this zone.  The system-wide
3309  * global 'ncpus_online' is returned if pools are disabled, the caller
3310  * is in the global zone, or a NULL zone argument is passed in.
3311  */
3312 int
3313 zone_ncpus_online_get(zone_t *zone)
3314 {
3315 	int myncpus_online = zone == NULL ? 0 : zone->zone_ncpus_online;
3316 
3317 	return (myncpus_online != 0 ? myncpus_online : ncpus_online);
3318 }
3319 
3320 /*
3321  * Return the pool to which the zone is currently bound.
3322  */
3323 pool_t *
3324 zone_pool_get(zone_t *zone)
3325 {
3326 	ASSERT(pool_lock_held());
3327 
3328 	return (zone->zone_pool);
3329 }
3330 
3331 /*
3332  * Set the zone's pool pointer and update the zone's visibility to match
3333  * the resources in the new pool.
3334  */
3335 void
3336 zone_pool_set(zone_t *zone, pool_t *pool)
3337 {
3338 	ASSERT(pool_lock_held());
3339 	ASSERT(MUTEX_HELD(&cpu_lock));
3340 
3341 	zone->zone_pool = pool;
3342 	zone_pset_set(zone, pool->pool_pset->pset_id);
3343 }
3344 
3345 /*
3346  * Return the cached value of the id of the processor set to which the
3347  * zone is currently bound.  The value will be ZONE_PS_INVAL if the pools
3348  * facility is disabled.
3349  */
3350 psetid_t
3351 zone_pset_get(zone_t *zone)
3352 {
3353 	ASSERT(MUTEX_HELD(&cpu_lock));
3354 
3355 	return (zone->zone_psetid);
3356 }
3357 
3358 /*
3359  * Set the cached value of the id of the processor set to which the zone
3360  * is currently bound.  Also update the zone's visibility to match the
3361  * resources in the new processor set.
3362  */
3363 void
3364 zone_pset_set(zone_t *zone, psetid_t newpsetid)
3365 {
3366 	psetid_t oldpsetid;
3367 
3368 	ASSERT(MUTEX_HELD(&cpu_lock));
3369 	oldpsetid = zone_pset_get(zone);
3370 
3371 	if (oldpsetid == newpsetid)
3372 		return;
3373 	/*
3374 	 * Global zone sees all.
3375 	 */
3376 	if (zone != global_zone) {
3377 		zone->zone_psetid = newpsetid;
3378 		if (newpsetid != ZONE_PS_INVAL)
3379 			pool_pset_visibility_add(newpsetid, zone);
3380 		if (oldpsetid != ZONE_PS_INVAL)
3381 			pool_pset_visibility_remove(oldpsetid, zone);
3382 	}
3383 	/*
3384 	 * Disabling pools, so we should start using the global values
3385 	 * for ncpus and ncpus_online.
3386 	 */
3387 	if (newpsetid == ZONE_PS_INVAL) {
3388 		zone->zone_ncpus = 0;
3389 		zone->zone_ncpus_online = 0;
3390 	}
3391 }
3392 
3393 /*
3394  * Walk the list of active zones and issue the provided callback for
3395  * each of them.
3396  *
3397  * Caller must not be holding any locks that may be acquired under
3398  * zonehash_lock.  See comment at the beginning of the file for a list of
3399  * common locks and their interactions with zones.
3400  */
3401 int
3402 zone_walk(int (*cb)(zone_t *, void *), void *data)
3403 {
3404 	zone_t *zone;
3405 	int ret = 0;
3406 	zone_status_t status;
3407 
3408 	mutex_enter(&zonehash_lock);
3409 	for (zone = list_head(&zone_active); zone != NULL;
3410 	    zone = list_next(&zone_active, zone)) {
3411 		/*
3412 		 * Skip zones that shouldn't be externally visible.
3413 		 */
3414 		status = zone_status_get(zone);
3415 		if (status < ZONE_IS_READY || status > ZONE_IS_DOWN)
3416 			continue;
3417 		/*
3418 		 * Bail immediately if any callback invocation returns a
3419 		 * non-zero value.
3420 		 */
3421 		ret = (*cb)(zone, data);
3422 		if (ret != 0)
3423 			break;
3424 	}
3425 	mutex_exit(&zonehash_lock);
3426 	return (ret);
3427 }
3428 
3429 static int
3430 zone_set_root(zone_t *zone, const char *upath)
3431 {
3432 	vnode_t *vp;
3433 	int trycount;
3434 	int error = 0;
3435 	char *path;
3436 	struct pathname upn, pn;
3437 	size_t pathlen;
3438 
3439 	if ((error = pn_get((char *)upath, UIO_USERSPACE, &upn)) != 0)
3440 		return (error);
3441 
3442