1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2015, Joyent Inc. All rights reserved.
25 */
26
27 /*
28 * Zones
29 *
30 * A zone is a named collection of processes, namespace constraints,
31 * and other system resources which comprise a secure and manageable
32 * application containment facility.
33 *
34 * Zones (represented by the reference counted zone_t) are tracked in
35 * the kernel in the zonehash. Elsewhere in the kernel, Zone IDs
36 * (zoneid_t) are used to track zone association. Zone IDs are
37 * dynamically generated when the zone is created; if a persistent
38 * identifier is needed (core files, accounting logs, audit trail,
39 * etc.), the zone name should be used.
40 *
41 *
42 * Global Zone:
43 *
44 * The global zone (zoneid 0) is automatically associated with all
45 * system resources that have not been bound to a user-created zone.
46 * This means that even systems where zones are not in active use
47 * have a global zone, and all processes, mounts, etc. are
48 * associated with that zone. The global zone is generally
49 * unconstrained in terms of privileges and access, though the usual
50 * credential and privilege based restrictions apply.
51 *
52 *
53 * Zone States:
54 *
55 * The states in which a zone may be in and the transitions are as
56 * follows:
57 *
58 * ZONE_IS_UNINITIALIZED: primordial state for a zone. The partially
59 * initialized zone is added to the list of active zones on the system but
60 * isn't accessible.
61 *
62 * ZONE_IS_INITIALIZED: Initialization complete except the ZSD callbacks are
63 * not yet completed. Not possible to enter the zone, but attributes can
64 * be retrieved.
65 *
66 * ZONE_IS_READY: zsched (the kernel dummy process for a zone) is
67 * ready. The zone is made visible after the ZSD constructor callbacks are
68 * executed. A zone remains in this state until it transitions into
69 * the ZONE_IS_BOOTING state as a result of a call to zone_boot().
70 *
71 * ZONE_IS_BOOTING: in this shortlived-state, zsched attempts to start
72 * init. Should that fail, the zone proceeds to the ZONE_IS_SHUTTING_DOWN
73 * state.
74 *
75 * ZONE_IS_RUNNING: The zone is open for business: zsched has
76 * successfully started init. A zone remains in this state until
77 * zone_shutdown() is called.
78 *
79 * ZONE_IS_SHUTTING_DOWN: zone_shutdown() has been called, the system is
80 * killing all processes running in the zone. The zone remains
81 * in this state until there are no more user processes running in the zone.
82 * zone_create(), zone_enter(), and zone_destroy() on this zone will fail.
83 * Since zone_shutdown() is restartable, it may be called successfully
84 * multiple times for the same zone_t. Setting of the zone's state to
85 * ZONE_IS_SHUTTING_DOWN is synchronized with mounts, so VOP_MOUNT() may check
86 * the zone's status without worrying about it being a moving target.
87 *
88 * ZONE_IS_EMPTY: zone_shutdown() has been called, and there
89 * are no more user processes in the zone. The zone remains in this
90 * state until there are no more kernel threads associated with the
91 * zone. zone_create(), zone_enter(), and zone_destroy() on this zone will
92 * fail.
93 *
94 * ZONE_IS_DOWN: All kernel threads doing work on behalf of the zone
95 * have exited. zone_shutdown() returns. Henceforth it is not possible to
96 * join the zone or create kernel threads therein.
97 *
98 * ZONE_IS_DYING: zone_destroy() has been called on the zone; zone
99 * remains in this state until zsched exits. Calls to zone_find_by_*()
100 * return NULL from now on.
101 *
102 * ZONE_IS_DEAD: zsched has exited (zone_ntasks == 0). There are no
103 * processes or threads doing work on behalf of the zone. The zone is
104 * removed from the list of active zones. zone_destroy() returns, and
105 * the zone can be recreated.
106 *
107 * ZONE_IS_FREE (internal state): zone_ref goes to 0, ZSD destructor
108 * callbacks are executed, and all memory associated with the zone is
109 * freed.
110 *
111 * Threads can wait for the zone to enter a requested state by using
112 * zone_status_wait() or zone_status_timedwait() with the desired
113 * state passed in as an argument. Zone state transitions are
114 * uni-directional; it is not possible to move back to an earlier state.
115 *
116 *
117 * Zone-Specific Data:
118 *
119 * Subsystems needing to maintain zone-specific data can store that
120 * data using the ZSD mechanism. This provides a zone-specific data
121 * store, similar to thread-specific data (see pthread_getspecific(3C)
122 * or the TSD code in uts/common/disp/thread.c. Also, ZSD can be used
123 * to register callbacks to be invoked when a zone is created, shut
124 * down, or destroyed. This can be used to initialize zone-specific
125 * data for new zones and to clean up when zones go away.
126 *
127 *
128 * Data Structures:
129 *
130 * The per-zone structure (zone_t) is reference counted, and freed
131 * when all references are released. zone_hold and zone_rele can be
132 * used to adjust the reference count. In addition, reference counts
133 * associated with the cred_t structure are tracked separately using
134 * zone_cred_hold and zone_cred_rele.
135 *
136 * Pointers to active zone_t's are stored in two hash tables; one
137 * for searching by id, the other for searching by name. Lookups
138 * can be performed on either basis, using zone_find_by_id and
139 * zone_find_by_name. Both return zone_t pointers with the zone
140 * held, so zone_rele should be called when the pointer is no longer
141 * needed. Zones can also be searched by path; zone_find_by_path
142 * returns the zone with which a path name is associated (global
143 * zone if the path is not within some other zone's file system
144 * hierarchy). This currently requires iterating through each zone,
145 * so it is slower than an id or name search via a hash table.
146 *
147 *
148 * Locking:
149 *
150 * zonehash_lock: This is a top-level global lock used to protect the
151 * zone hash tables and lists. Zones cannot be created or destroyed
152 * while this lock is held.
153 * zone_status_lock: This is a global lock protecting zone state.
154 * Zones cannot change state while this lock is held. It also
155 * protects the list of kernel threads associated with a zone.
156 * zone_lock: This is a per-zone lock used to protect several fields of
157 * the zone_t (see <sys/zone.h> for details). In addition, holding
158 * this lock means that the zone cannot go away.
159 * zone_nlwps_lock: This is a per-zone lock used to protect the fields
160 * related to the zone.max-lwps rctl.
161 * zone_mem_lock: This is a per-zone lock used to protect the fields
162 * related to the zone.max-locked-memory and zone.max-swap rctls.
163 * zone_rctl_lock: This is a per-zone lock used to protect other rctls,
164 * currently just max_lofi
165 * zsd_key_lock: This is a global lock protecting the key state for ZSD.
166 * zone_deathrow_lock: This is a global lock protecting the "deathrow"
167 * list (a list of zones in the ZONE_IS_DEAD state).
168 *
169 * Ordering requirements:
170 * pool_lock --> cpu_lock --> zonehash_lock --> zone_status_lock -->
171 * zone_lock --> zsd_key_lock --> pidlock --> p_lock
172 *
173 * When taking zone_mem_lock or zone_nlwps_lock, the lock ordering is:
174 * zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_mem_lock
175 * zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_nlwps_lock
176 *
177 * Blocking memory allocations are permitted while holding any of the
178 * zone locks.
179 *
180 *
181 * System Call Interface:
182 *
183 * The zone subsystem can be managed and queried from user level with
184 * the following system calls (all subcodes of the primary "zone"
185 * system call):
186 * - zone_create: creates a zone with selected attributes (name,
187 * root path, privileges, resource controls, ZFS datasets)
188 * - zone_enter: allows the current process to enter a zone
189 * - zone_getattr: reports attributes of a zone
190 * - zone_setattr: set attributes of a zone
191 * - zone_boot: set 'init' running for the zone
192 * - zone_list: lists all zones active in the system
193 * - zone_lookup: looks up zone id based on name
194 * - zone_shutdown: initiates shutdown process (see states above)
195 * - zone_destroy: completes shutdown process (see states above)
196 *
197 */
198
199 #include <sys/priv_impl.h>
200 #include <sys/cred.h>
201 #include <c2/audit.h>
202 #include <sys/debug.h>
203 #include <sys/file.h>
204 #include <sys/kmem.h>
205 #include <sys/kstat.h>
206 #include <sys/mutex.h>
207 #include <sys/note.h>
208 #include <sys/pathname.h>
209 #include <sys/proc.h>
210 #include <sys/project.h>
211 #include <sys/sysevent.h>
212 #include <sys/task.h>
213 #include <sys/systm.h>
214 #include <sys/types.h>
215 #include <sys/utsname.h>
216 #include <sys/vnode.h>
217 #include <sys/vfs.h>
218 #include <sys/systeminfo.h>
219 #include <sys/policy.h>
220 #include <sys/cred_impl.h>
221 #include <sys/contract_impl.h>
222 #include <sys/contract/process_impl.h>
223 #include <sys/class.h>
224 #include <sys/pool.h>
225 #include <sys/pool_pset.h>
226 #include <sys/pset.h>
227 #include <sys/strlog.h>
228 #include <sys/sysmacros.h>
229 #include <sys/callb.h>
230 #include <sys/vmparam.h>
231 #include <sys/corectl.h>
232 #include <sys/ipc_impl.h>
233 #include <sys/klpd.h>
234
235 #include <sys/door.h>
236 #include <sys/cpuvar.h>
237 #include <sys/sdt.h>
238
239 #include <sys/uadmin.h>
240 #include <sys/session.h>
241 #include <sys/cmn_err.h>
242 #include <sys/modhash.h>
243 #include <sys/sunddi.h>
244 #include <sys/nvpair.h>
245 #include <sys/rctl.h>
246 #include <sys/fss.h>
247 #include <sys/brand.h>
248 #include <sys/zone.h>
249 #include <net/if.h>
250 #include <sys/cpucaps.h>
251 #include <vm/seg.h>
252 #include <sys/mac.h>
253
254 /*
255 * This constant specifies the number of seconds that threads waiting for
256 * subsystems to release a zone's general-purpose references will wait before
257 * they log the zone's reference counts. The constant's value shouldn't
258 * be so small that reference counts are unnecessarily reported for zones
259 * whose references are slowly released. On the other hand, it shouldn't be so
260 * large that users reboot their systems out of frustration over hung zones
261 * before the system logs the zones' reference counts.
262 */
263 #define ZONE_DESTROY_TIMEOUT_SECS 60
264
265 /* List of data link IDs which are accessible from the zone */
266 typedef struct zone_dl {
267 datalink_id_t zdl_id;
268 nvlist_t *zdl_net;
269 list_node_t zdl_linkage;
270 } zone_dl_t;
271
272 /*
273 * cv used to signal that all references to the zone have been released. This
274 * needs to be global since there may be multiple waiters, and the first to
275 * wake up will free the zone_t, hence we cannot use zone->zone_cv.
276 */
277 static kcondvar_t zone_destroy_cv;
278 /*
279 * Lock used to serialize access to zone_cv. This could have been per-zone,
280 * but then we'd need another lock for zone_destroy_cv, and why bother?
281 */
282 static kmutex_t zone_status_lock;
283
284 /*
285 * ZSD-related global variables.
286 */
287 static kmutex_t zsd_key_lock; /* protects the following two */
288 /*
289 * The next caller of zone_key_create() will be assigned a key of ++zsd_keyval.
290 */
291 static zone_key_t zsd_keyval = 0;
292 /*
293 * Global list of registered keys. We use this when a new zone is created.
294 */
295 static list_t zsd_registered_keys;
296
297 int zone_hash_size = 256;
298 static mod_hash_t *zonehashbyname, *zonehashbyid, *zonehashbylabel;
299 static kmutex_t zonehash_lock;
300 static uint_t zonecount;
301 static id_space_t *zoneid_space;
302
303 /*
304 * The global zone (aka zone0) is the all-seeing, all-knowing zone in which the
305 * kernel proper runs, and which manages all other zones.
306 *
307 * Although not declared as static, the variable "zone0" should not be used
308 * except for by code that needs to reference the global zone early on in boot,
309 * before it is fully initialized. All other consumers should use
310 * 'global_zone'.
311 */
312 zone_t zone0;
313 zone_t *global_zone = NULL; /* Set when the global zone is initialized */
314
315 /*
316 * List of active zones, protected by zonehash_lock.
317 */
318 static list_t zone_active;
319
320 /*
321 * List of destroyed zones that still have outstanding cred references.
322 * Used for debugging. Uses a separate lock to avoid lock ordering
323 * problems in zone_free.
324 */
325 static list_t zone_deathrow;
326 static kmutex_t zone_deathrow_lock;
327
328 /* number of zones is limited by virtual interface limit in IP */
329 uint_t maxzones = 8192;
330
331 /* Event channel to sent zone state change notifications */
332 evchan_t *zone_event_chan;
333
334 /*
335 * This table holds the mapping from kernel zone states to
336 * states visible in the state notification API.
337 * The idea is that we only expose "obvious" states and
338 * do not expose states which are just implementation details.
339 */
340 const char *zone_status_table[] = {
341 ZONE_EVENT_UNINITIALIZED, /* uninitialized */
342 ZONE_EVENT_INITIALIZED, /* initialized */
343 ZONE_EVENT_READY, /* ready */
344 ZONE_EVENT_READY, /* booting */
345 ZONE_EVENT_RUNNING, /* running */
346 ZONE_EVENT_SHUTTING_DOWN, /* shutting_down */
347 ZONE_EVENT_SHUTTING_DOWN, /* empty */
348 ZONE_EVENT_SHUTTING_DOWN, /* down */
349 ZONE_EVENT_SHUTTING_DOWN, /* dying */
350 ZONE_EVENT_UNINITIALIZED, /* dead */
351 };
352
353 /*
354 * This array contains the names of the subsystems listed in zone_ref_subsys_t
355 * (see sys/zone.h).
356 */
357 static char *zone_ref_subsys_names[] = {
358 "NFS", /* ZONE_REF_NFS */
359 "NFSv4", /* ZONE_REF_NFSV4 */
360 "SMBFS", /* ZONE_REF_SMBFS */
361 "MNTFS", /* ZONE_REF_MNTFS */
362 "LOFI", /* ZONE_REF_LOFI */
363 "VFS", /* ZONE_REF_VFS */
364 "IPC" /* ZONE_REF_IPC */
365 };
366
367 /*
368 * This isn't static so lint doesn't complain.
369 */
370 rctl_hndl_t rc_zone_cpu_shares;
371 rctl_hndl_t rc_zone_locked_mem;
372 rctl_hndl_t rc_zone_max_swap;
373 rctl_hndl_t rc_zone_max_lofi;
374 rctl_hndl_t rc_zone_cpu_cap;
375 rctl_hndl_t rc_zone_nlwps;
376 rctl_hndl_t rc_zone_nprocs;
377 rctl_hndl_t rc_zone_shmmax;
378 rctl_hndl_t rc_zone_shmmni;
379 rctl_hndl_t rc_zone_semmni;
380 rctl_hndl_t rc_zone_msgmni;
381
382 const char * const zone_default_initname = "/sbin/init";
383 static char * const zone_prefix = "/zone/";
384 static int zone_shutdown(zoneid_t zoneid);
385 static int zone_add_datalink(zoneid_t, datalink_id_t);
386 static int zone_remove_datalink(zoneid_t, datalink_id_t);
387 static int zone_list_datalink(zoneid_t, int *, datalink_id_t *);
388 static int zone_set_network(zoneid_t, zone_net_data_t *);
389 static int zone_get_network(zoneid_t, zone_net_data_t *);
390
391 typedef boolean_t zsd_applyfn_t(kmutex_t *, boolean_t, zone_t *, zone_key_t);
392
393 static void zsd_apply_all_zones(zsd_applyfn_t *, zone_key_t);
394 static void zsd_apply_all_keys(zsd_applyfn_t *, zone_t *);
395 static boolean_t zsd_apply_create(kmutex_t *, boolean_t, zone_t *, zone_key_t);
396 static boolean_t zsd_apply_shutdown(kmutex_t *, boolean_t, zone_t *,
397 zone_key_t);
398 static boolean_t zsd_apply_destroy(kmutex_t *, boolean_t, zone_t *, zone_key_t);
399 static boolean_t zsd_wait_for_creator(zone_t *, struct zsd_entry *,
400 kmutex_t *);
401 static boolean_t zsd_wait_for_inprogress(zone_t *, struct zsd_entry *,
402 kmutex_t *);
403
404 /*
405 * Bump this number when you alter the zone syscall interfaces; this is
406 * because we need to have support for previous API versions in libc
407 * to support patching; libc calls into the kernel to determine this number.
408 *
409 * Version 1 of the API is the version originally shipped with Solaris 10
410 * Version 2 alters the zone_create system call in order to support more
411 * arguments by moving the args into a structure; and to do better
412 * error reporting when zone_create() fails.
413 * Version 3 alters the zone_create system call in order to support the
414 * import of ZFS datasets to zones.
415 * Version 4 alters the zone_create system call in order to support
416 * Trusted Extensions.
417 * Version 5 alters the zone_boot system call, and converts its old
418 * bootargs parameter to be set by the zone_setattr API instead.
419 * Version 6 adds the flag argument to zone_create.
420 */
421 static const int ZONE_SYSCALL_API_VERSION = 6;
422
423 /*
424 * Certain filesystems (such as NFS and autofs) need to know which zone
425 * the mount is being placed in. Because of this, we need to be able to
426 * ensure that a zone isn't in the process of being created/destroyed such
427 * that nfs_mount() thinks it is in the global/NGZ zone, while by the time
428 * it gets added the list of mounted zones, it ends up on the wrong zone's
429 * mount list. Since a zone can't reside on an NFS file system, we don't
430 * have to worry about the zonepath itself.
431 *
432 * The following functions: block_mounts()/resume_mounts() and
433 * mount_in_progress()/mount_completed() are used by zones and the VFS
434 * layer (respectively) to synchronize zone state transitions and new
435 * mounts within a zone. This syncronization is on a per-zone basis, so
436 * activity for one zone will not interfere with activity for another zone.
437 *
438 * The semantics are like a reader-reader lock such that there may
439 * either be multiple mounts (or zone state transitions, if that weren't
440 * serialized by zonehash_lock) in progress at the same time, but not
441 * both.
442 *
443 * We use cv's so the user can ctrl-C out of the operation if it's
444 * taking too long.
445 *
446 * The semantics are such that there is unfair bias towards the
447 * "current" operation. This means that zone halt may starve if
448 * there is a rapid succession of new mounts coming in to the zone.
449 */
450 /*
451 * Prevent new mounts from progressing to the point of calling
452 * VFS_MOUNT(). If there are already mounts in this "region", wait for
453 * them to complete.
454 */
455 static int
456 block_mounts(zone_t *zp)
457 {
458 int retval = 0;
459
460 /*
461 * Since it may block for a long time, block_mounts() shouldn't be
462 * called with zonehash_lock held.
463 */
464 ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
465 mutex_enter(&zp->zone_mount_lock);
466 while (zp->zone_mounts_in_progress > 0) {
467 if (cv_wait_sig(&zp->zone_mount_cv, &zp->zone_mount_lock) == 0)
468 goto signaled;
469 }
470 /*
471 * A negative value of mounts_in_progress indicates that mounts
472 * have been blocked by (-mounts_in_progress) different callers
473 * (remotely possible if two threads enter zone_shutdown at the same
474 * time).
475 */
476 zp->zone_mounts_in_progress--;
477 retval = 1;
478 signaled:
479 mutex_exit(&zp->zone_mount_lock);
480 return (retval);
481 }
482
483 /*
484 * The VFS layer may progress with new mounts as far as we're concerned.
485 * Allow them to progress if we were the last obstacle.
486 */
487 static void
488 resume_mounts(zone_t *zp)
489 {
490 mutex_enter(&zp->zone_mount_lock);
491 if (++zp->zone_mounts_in_progress == 0)
492 cv_broadcast(&zp->zone_mount_cv);
493 mutex_exit(&zp->zone_mount_lock);
494 }
495
496 /*
497 * The VFS layer is busy with a mount; this zone should wait until all
498 * of its mounts are completed to progress.
499 */
500 void
501 mount_in_progress(zone_t *zp)
502 {
503 mutex_enter(&zp->zone_mount_lock);
504 while (zp->zone_mounts_in_progress < 0)
505 cv_wait(&zp->zone_mount_cv, &zp->zone_mount_lock);
506 zp->zone_mounts_in_progress++;
507 mutex_exit(&zp->zone_mount_lock);
508 }
509
510 /*
511 * VFS is done with one mount; wake up any waiting block_mounts()
512 * callers if this is the last mount.
513 */
514 void
515 mount_completed(zone_t *zp)
516 {
517 mutex_enter(&zp->zone_mount_lock);
518 if (--zp->zone_mounts_in_progress == 0)
519 cv_broadcast(&zp->zone_mount_cv);
520 mutex_exit(&zp->zone_mount_lock);
521 }
522
523 /*
524 * ZSD routines.
525 *
526 * Zone Specific Data (ZSD) is modeled after Thread Specific Data as
527 * defined by the pthread_key_create() and related interfaces.
528 *
529 * Kernel subsystems may register one or more data items and/or
530 * callbacks to be executed when a zone is created, shutdown, or
531 * destroyed.
532 *
533 * Unlike the thread counterpart, destructor callbacks will be executed
534 * even if the data pointer is NULL and/or there are no constructor
535 * callbacks, so it is the responsibility of such callbacks to check for
536 * NULL data values if necessary.
537 *
538 * The locking strategy and overall picture is as follows:
539 *
540 * When someone calls zone_key_create(), a template ZSD entry is added to the
541 * global list "zsd_registered_keys", protected by zsd_key_lock. While
542 * holding that lock all the existing zones are marked as
543 * ZSD_CREATE_NEEDED and a copy of the ZSD entry added to the per-zone
544 * zone_zsd list (protected by zone_lock). The global list is updated first
545 * (under zone_key_lock) to make sure that newly created zones use the
546 * most recent list of keys. Then under zonehash_lock we walk the zones
547 * and mark them. Similar locking is used in zone_key_delete().
548 *
549 * The actual create, shutdown, and destroy callbacks are done without
550 * holding any lock. And zsd_flags are used to ensure that the operations
551 * completed so that when zone_key_create (and zone_create) is done, as well as
552 * zone_key_delete (and zone_destroy) is done, all the necessary callbacks
553 * are completed.
554 *
555 * When new zones are created constructor callbacks for all registered ZSD
556 * entries will be called. That also uses the above two phases of marking
557 * what needs to be done, and then running the callbacks without holding
558 * any locks.
559 *
560 * The framework does not provide any locking around zone_getspecific() and
561 * zone_setspecific() apart from that needed for internal consistency, so
562 * callers interested in atomic "test-and-set" semantics will need to provide
563 * their own locking.
564 */
565
566 /*
567 * Helper function to find the zsd_entry associated with the key in the
568 * given list.
569 */
570 static struct zsd_entry *
571 zsd_find(list_t *l, zone_key_t key)
572 {
573 struct zsd_entry *zsd;
574
575 for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
576 if (zsd->zsd_key == key) {
577 return (zsd);
578 }
579 }
580 return (NULL);
581 }
582
583 /*
584 * Helper function to find the zsd_entry associated with the key in the
585 * given list. Move it to the front of the list.
586 */
587 static struct zsd_entry *
588 zsd_find_mru(list_t *l, zone_key_t key)
589 {
590 struct zsd_entry *zsd;
591
592 for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
593 if (zsd->zsd_key == key) {
594 /*
595 * Move to head of list to keep list in MRU order.
596 */
597 if (zsd != list_head(l)) {
598 list_remove(l, zsd);
599 list_insert_head(l, zsd);
600 }
601 return (zsd);
602 }
603 }
604 return (NULL);
605 }
606
607 void
608 zone_key_create(zone_key_t *keyp, void *(*create)(zoneid_t),
609 void (*shutdown)(zoneid_t, void *), void (*destroy)(zoneid_t, void *))
610 {
611 struct zsd_entry *zsdp;
612 struct zsd_entry *t;
613 struct zone *zone;
614 zone_key_t key;
615
616 zsdp = kmem_zalloc(sizeof (*zsdp), KM_SLEEP);
617 zsdp->zsd_data = NULL;
618 zsdp->zsd_create = create;
619 zsdp->zsd_shutdown = shutdown;
620 zsdp->zsd_destroy = destroy;
621
622 /*
623 * Insert in global list of callbacks. Makes future zone creations
624 * see it.
625 */
626 mutex_enter(&zsd_key_lock);
627 key = zsdp->zsd_key = ++zsd_keyval;
628 ASSERT(zsd_keyval != 0);
629 list_insert_tail(&zsd_registered_keys, zsdp);
630 mutex_exit(&zsd_key_lock);
631
632 /*
633 * Insert for all existing zones and mark them as needing
634 * a create callback.
635 */
636 mutex_enter(&zonehash_lock); /* stop the world */
637 for (zone = list_head(&zone_active); zone != NULL;
638 zone = list_next(&zone_active, zone)) {
639 zone_status_t status;
640
641 mutex_enter(&zone->zone_lock);
642
643 /* Skip zones that are on the way down or not yet up */
644 status = zone_status_get(zone);
645 if (status >= ZONE_IS_DOWN ||
646 status == ZONE_IS_UNINITIALIZED) {
647 mutex_exit(&zone->zone_lock);
648 continue;
649 }
650
651 t = zsd_find_mru(&zone->zone_zsd, key);
652 if (t != NULL) {
653 /*
654 * A zsd_configure already inserted it after
655 * we dropped zsd_key_lock above.
656 */
657 mutex_exit(&zone->zone_lock);
658 continue;
659 }
660 t = kmem_zalloc(sizeof (*t), KM_SLEEP);
661 t->zsd_key = key;
662 t->zsd_create = create;
663 t->zsd_shutdown = shutdown;
664 t->zsd_destroy = destroy;
665 if (create != NULL) {
666 t->zsd_flags = ZSD_CREATE_NEEDED;
667 DTRACE_PROBE2(zsd__create__needed,
668 zone_t *, zone, zone_key_t, key);
669 }
670 list_insert_tail(&zone->zone_zsd, t);
671 mutex_exit(&zone->zone_lock);
672 }
673 mutex_exit(&zonehash_lock);
674
675 if (create != NULL) {
676 /* Now call the create callback for this key */
677 zsd_apply_all_zones(zsd_apply_create, key);
678 }
679 /*
680 * It is safe for consumers to use the key now, make it
681 * globally visible. Specifically zone_getspecific() will
682 * always successfully return the zone specific data associated
683 * with the key.
684 */
685 *keyp = key;
686
687 }
688
689 /*
690 * Function called when a module is being unloaded, or otherwise wishes
691 * to unregister its ZSD key and callbacks.
692 *
693 * Remove from the global list and determine the functions that need to
694 * be called under a global lock. Then call the functions without
695 * holding any locks. Finally free up the zone_zsd entries. (The apply
696 * functions need to access the zone_zsd entries to find zsd_data etc.)
697 */
698 int
699 zone_key_delete(zone_key_t key)
700 {
701 struct zsd_entry *zsdp = NULL;
702 zone_t *zone;
703
704 mutex_enter(&zsd_key_lock);
705 zsdp = zsd_find_mru(&zsd_registered_keys, key);
706 if (zsdp == NULL) {
707 mutex_exit(&zsd_key_lock);
708 return (-1);
709 }
710 list_remove(&zsd_registered_keys, zsdp);
711 mutex_exit(&zsd_key_lock);
712
713 mutex_enter(&zonehash_lock);
714 for (zone = list_head(&zone_active); zone != NULL;
715 zone = list_next(&zone_active, zone)) {
716 struct zsd_entry *del;
717
718 mutex_enter(&zone->zone_lock);
719 del = zsd_find_mru(&zone->zone_zsd, key);
720 if (del == NULL) {
721 /*
722 * Somebody else got here first e.g the zone going
723 * away.
724 */
725 mutex_exit(&zone->zone_lock);
726 continue;
727 }
728 ASSERT(del->zsd_shutdown == zsdp->zsd_shutdown);
729 ASSERT(del->zsd_destroy == zsdp->zsd_destroy);
730 if (del->zsd_shutdown != NULL &&
731 (del->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
732 del->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
733 DTRACE_PROBE2(zsd__shutdown__needed,
734 zone_t *, zone, zone_key_t, key);
735 }
736 if (del->zsd_destroy != NULL &&
737 (del->zsd_flags & ZSD_DESTROY_ALL) == 0) {
738 del->zsd_flags |= ZSD_DESTROY_NEEDED;
739 DTRACE_PROBE2(zsd__destroy__needed,
740 zone_t *, zone, zone_key_t, key);
741 }
742 mutex_exit(&zone->zone_lock);
743 }
744 mutex_exit(&zonehash_lock);
745 kmem_free(zsdp, sizeof (*zsdp));
746
747 /* Now call the shutdown and destroy callback for this key */
748 zsd_apply_all_zones(zsd_apply_shutdown, key);
749 zsd_apply_all_zones(zsd_apply_destroy, key);
750
751 /* Now we can free up the zsdp structures in each zone */
752 mutex_enter(&zonehash_lock);
753 for (zone = list_head(&zone_active); zone != NULL;
754 zone = list_next(&zone_active, zone)) {
755 struct zsd_entry *del;
756
757 mutex_enter(&zone->zone_lock);
758 del = zsd_find(&zone->zone_zsd, key);
759 if (del != NULL) {
760 list_remove(&zone->zone_zsd, del);
761 ASSERT(!(del->zsd_flags & ZSD_ALL_INPROGRESS));
762 kmem_free(del, sizeof (*del));
763 }
764 mutex_exit(&zone->zone_lock);
765 }
766 mutex_exit(&zonehash_lock);
767
768 return (0);
769 }
770
771 /*
772 * ZSD counterpart of pthread_setspecific().
773 *
774 * Since all zsd callbacks, including those with no create function,
775 * have an entry in zone_zsd, if the key is registered it is part of
776 * the zone_zsd list.
777 * Return an error if the key wasn't registerd.
778 */
779 int
780 zone_setspecific(zone_key_t key, zone_t *zone, const void *data)
781 {
782 struct zsd_entry *t;
783
784 mutex_enter(&zone->zone_lock);
785 t = zsd_find_mru(&zone->zone_zsd, key);
786 if (t != NULL) {
787 /*
788 * Replace old value with new
789 */
790 t->zsd_data = (void *)data;
791 mutex_exit(&zone->zone_lock);
792 return (0);
793 }
794 mutex_exit(&zone->zone_lock);
795 return (-1);
796 }
797
798 /*
799 * ZSD counterpart of pthread_getspecific().
800 */
801 void *
802 zone_getspecific(zone_key_t key, zone_t *zone)
803 {
804 struct zsd_entry *t;
805 void *data;
806
807 mutex_enter(&zone->zone_lock);
808 t = zsd_find_mru(&zone->zone_zsd, key);
809 data = (t == NULL ? NULL : t->zsd_data);
810 mutex_exit(&zone->zone_lock);
811 return (data);
812 }
813
814 /*
815 * Function used to initialize a zone's list of ZSD callbacks and data
816 * when the zone is being created. The callbacks are initialized from
817 * the template list (zsd_registered_keys). The constructor callback is
818 * executed later (once the zone exists and with locks dropped).
819 */
820 static void
821 zone_zsd_configure(zone_t *zone)
822 {
823 struct zsd_entry *zsdp;
824 struct zsd_entry *t;
825
826 ASSERT(MUTEX_HELD(&zonehash_lock));
827 ASSERT(list_head(&zone->zone_zsd) == NULL);
828 mutex_enter(&zone->zone_lock);
829 mutex_enter(&zsd_key_lock);
830 for (zsdp = list_head(&zsd_registered_keys); zsdp != NULL;
831 zsdp = list_next(&zsd_registered_keys, zsdp)) {
832 /*
833 * Since this zone is ZONE_IS_UNCONFIGURED, zone_key_create
834 * should not have added anything to it.
835 */
836 ASSERT(zsd_find(&zone->zone_zsd, zsdp->zsd_key) == NULL);
837
838 t = kmem_zalloc(sizeof (*t), KM_SLEEP);
839 t->zsd_key = zsdp->zsd_key;
840 t->zsd_create = zsdp->zsd_create;
841 t->zsd_shutdown = zsdp->zsd_shutdown;
842 t->zsd_destroy = zsdp->zsd_destroy;
843 if (zsdp->zsd_create != NULL) {
844 t->zsd_flags = ZSD_CREATE_NEEDED;
845 DTRACE_PROBE2(zsd__create__needed,
846 zone_t *, zone, zone_key_t, zsdp->zsd_key);
847 }
848 list_insert_tail(&zone->zone_zsd, t);
849 }
850 mutex_exit(&zsd_key_lock);
851 mutex_exit(&zone->zone_lock);
852 }
853
854 enum zsd_callback_type { ZSD_CREATE, ZSD_SHUTDOWN, ZSD_DESTROY };
855
856 /*
857 * Helper function to execute shutdown or destructor callbacks.
858 */
859 static void
860 zone_zsd_callbacks(zone_t *zone, enum zsd_callback_type ct)
861 {
862 struct zsd_entry *t;
863
864 ASSERT(ct == ZSD_SHUTDOWN || ct == ZSD_DESTROY);
865 ASSERT(ct != ZSD_SHUTDOWN || zone_status_get(zone) >= ZONE_IS_EMPTY);
866 ASSERT(ct != ZSD_DESTROY || zone_status_get(zone) >= ZONE_IS_DOWN);
867
868 /*
869 * Run the callback solely based on what is registered for the zone
870 * in zone_zsd. The global list can change independently of this
871 * as keys are registered and unregistered and we don't register new
872 * callbacks for a zone that is in the process of going away.
873 */
874 mutex_enter(&zone->zone_lock);
875 for (t = list_head(&zone->zone_zsd); t != NULL;
876 t = list_next(&zone->zone_zsd, t)) {
877 zone_key_t key = t->zsd_key;
878
879 /* Skip if no callbacks registered */
880
881 if (ct == ZSD_SHUTDOWN) {
882 if (t->zsd_shutdown != NULL &&
883 (t->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
884 t->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
885 DTRACE_PROBE2(zsd__shutdown__needed,
886 zone_t *, zone, zone_key_t, key);
887 }
888 } else {
889 if (t->zsd_destroy != NULL &&
890 (t->zsd_flags & ZSD_DESTROY_ALL) == 0) {
891 t->zsd_flags |= ZSD_DESTROY_NEEDED;
892 DTRACE_PROBE2(zsd__destroy__needed,
893 zone_t *, zone, zone_key_t, key);
894 }
895 }
896 }
897 mutex_exit(&zone->zone_lock);
898
899 /* Now call the shutdown and destroy callback for this key */
900 zsd_apply_all_keys(zsd_apply_shutdown, zone);
901 zsd_apply_all_keys(zsd_apply_destroy, zone);
902
903 }
904
905 /*
906 * Called when the zone is going away; free ZSD-related memory, and
907 * destroy the zone_zsd list.
908 */
909 static void
910 zone_free_zsd(zone_t *zone)
911 {
912 struct zsd_entry *t, *next;
913
914 /*
915 * Free all the zsd_entry's we had on this zone.
916 */
917 mutex_enter(&zone->zone_lock);
918 for (t = list_head(&zone->zone_zsd); t != NULL; t = next) {
919 next = list_next(&zone->zone_zsd, t);
920 list_remove(&zone->zone_zsd, t);
921 ASSERT(!(t->zsd_flags & ZSD_ALL_INPROGRESS));
922 kmem_free(t, sizeof (*t));
923 }
924 list_destroy(&zone->zone_zsd);
925 mutex_exit(&zone->zone_lock);
926
927 }
928
929 /*
930 * Apply a function to all zones for particular key value.
931 *
932 * The applyfn has to drop zonehash_lock if it does some work, and
933 * then reacquire it before it returns.
934 * When the lock is dropped we don't follow list_next even
935 * if it is possible to do so without any hazards. This is
936 * because we want the design to allow for the list of zones
937 * to change in any arbitrary way during the time the
938 * lock was dropped.
939 *
940 * It is safe to restart the loop at list_head since the applyfn
941 * changes the zsd_flags as it does work, so a subsequent
942 * pass through will have no effect in applyfn, hence the loop will terminate
943 * in at worst O(N^2).
944 */
945 static void
946 zsd_apply_all_zones(zsd_applyfn_t *applyfn, zone_key_t key)
947 {
948 zone_t *zone;
949
950 mutex_enter(&zonehash_lock);
951 zone = list_head(&zone_active);
952 while (zone != NULL) {
953 if ((applyfn)(&zonehash_lock, B_FALSE, zone, key)) {
954 /* Lock dropped - restart at head */
955 zone = list_head(&zone_active);
956 } else {
957 zone = list_next(&zone_active, zone);
958 }
959 }
960 mutex_exit(&zonehash_lock);
961 }
962
963 /*
964 * Apply a function to all keys for a particular zone.
965 *
966 * The applyfn has to drop zonehash_lock if it does some work, and
967 * then reacquire it before it returns.
968 * When the lock is dropped we don't follow list_next even
969 * if it is possible to do so without any hazards. This is
970 * because we want the design to allow for the list of zsd callbacks
971 * to change in any arbitrary way during the time the
972 * lock was dropped.
973 *
974 * It is safe to restart the loop at list_head since the applyfn
975 * changes the zsd_flags as it does work, so a subsequent
976 * pass through will have no effect in applyfn, hence the loop will terminate
977 * in at worst O(N^2).
978 */
979 static void
980 zsd_apply_all_keys(zsd_applyfn_t *applyfn, zone_t *zone)
981 {
982 struct zsd_entry *t;
983
984 mutex_enter(&zone->zone_lock);
985 t = list_head(&zone->zone_zsd);
986 while (t != NULL) {
987 if ((applyfn)(NULL, B_TRUE, zone, t->zsd_key)) {
988 /* Lock dropped - restart at head */
989 t = list_head(&zone->zone_zsd);
990 } else {
991 t = list_next(&zone->zone_zsd, t);
992 }
993 }
994 mutex_exit(&zone->zone_lock);
995 }
996
997 /*
998 * Call the create function for the zone and key if CREATE_NEEDED
999 * is set.
1000 * If some other thread gets here first and sets CREATE_INPROGRESS, then
1001 * we wait for that thread to complete so that we can ensure that
1002 * all the callbacks are done when we've looped over all zones/keys.
1003 *
1004 * When we call the create function, we drop the global held by the
1005 * caller, and return true to tell the caller it needs to re-evalute the
1006 * state.
1007 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1008 * remains held on exit.
1009 */
1010 static boolean_t
1011 zsd_apply_create(kmutex_t *lockp, boolean_t zone_lock_held,
1012 zone_t *zone, zone_key_t key)
1013 {
1014 void *result;
1015 struct zsd_entry *t;
1016 boolean_t dropped;
1017
1018 if (lockp != NULL) {
1019 ASSERT(MUTEX_HELD(lockp));
1020 }
1021 if (zone_lock_held) {
1022 ASSERT(MUTEX_HELD(&zone->zone_lock));
1023 } else {
1024 mutex_enter(&zone->zone_lock);
1025 }
1026
1027 t = zsd_find(&zone->zone_zsd, key);
1028 if (t == NULL) {
1029 /*
1030 * Somebody else got here first e.g the zone going
1031 * away.
1032 */
1033 if (!zone_lock_held)
1034 mutex_exit(&zone->zone_lock);
1035 return (B_FALSE);
1036 }
1037 dropped = B_FALSE;
1038 if (zsd_wait_for_inprogress(zone, t, lockp))
1039 dropped = B_TRUE;
1040
1041 if (t->zsd_flags & ZSD_CREATE_NEEDED) {
1042 t->zsd_flags &= ~ZSD_CREATE_NEEDED;
1043 t->zsd_flags |= ZSD_CREATE_INPROGRESS;
1044 DTRACE_PROBE2(zsd__create__inprogress,
1045 zone_t *, zone, zone_key_t, key);
1046 mutex_exit(&zone->zone_lock);
1047 if (lockp != NULL)
1048 mutex_exit(lockp);
1049
1050 dropped = B_TRUE;
1051 ASSERT(t->zsd_create != NULL);
1052 DTRACE_PROBE2(zsd__create__start,
1053 zone_t *, zone, zone_key_t, key);
1054
1055 result = (*t->zsd_create)(zone->zone_id);
1056
1057 DTRACE_PROBE2(zsd__create__end,
1058 zone_t *, zone, voidn *, result);
1059
1060 ASSERT(result != NULL);
1061 if (lockp != NULL)
1062 mutex_enter(lockp);
1063 mutex_enter(&zone->zone_lock);
1064 t->zsd_data = result;
1065 t->zsd_flags &= ~ZSD_CREATE_INPROGRESS;
1066 t->zsd_flags |= ZSD_CREATE_COMPLETED;
1067 cv_broadcast(&t->zsd_cv);
1068 DTRACE_PROBE2(zsd__create__completed,
1069 zone_t *, zone, zone_key_t, key);
1070 }
1071 if (!zone_lock_held)
1072 mutex_exit(&zone->zone_lock);
1073 return (dropped);
1074 }
1075
1076 /*
1077 * Call the shutdown function for the zone and key if SHUTDOWN_NEEDED
1078 * is set.
1079 * If some other thread gets here first and sets *_INPROGRESS, then
1080 * we wait for that thread to complete so that we can ensure that
1081 * all the callbacks are done when we've looped over all zones/keys.
1082 *
1083 * When we call the shutdown function, we drop the global held by the
1084 * caller, and return true to tell the caller it needs to re-evalute the
1085 * state.
1086 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1087 * remains held on exit.
1088 */
1089 static boolean_t
1090 zsd_apply_shutdown(kmutex_t *lockp, boolean_t zone_lock_held,
1091 zone_t *zone, zone_key_t key)
1092 {
1093 struct zsd_entry *t;
1094 void *data;
1095 boolean_t dropped;
1096
1097 if (lockp != NULL) {
1098 ASSERT(MUTEX_HELD(lockp));
1099 }
1100 if (zone_lock_held) {
1101 ASSERT(MUTEX_HELD(&zone->zone_lock));
1102 } else {
1103 mutex_enter(&zone->zone_lock);
1104 }
1105
1106 t = zsd_find(&zone->zone_zsd, key);
1107 if (t == NULL) {
1108 /*
1109 * Somebody else got here first e.g the zone going
1110 * away.
1111 */
1112 if (!zone_lock_held)
1113 mutex_exit(&zone->zone_lock);
1114 return (B_FALSE);
1115 }
1116 dropped = B_FALSE;
1117 if (zsd_wait_for_creator(zone, t, lockp))
1118 dropped = B_TRUE;
1119
1120 if (zsd_wait_for_inprogress(zone, t, lockp))
1121 dropped = B_TRUE;
1122
1123 if (t->zsd_flags & ZSD_SHUTDOWN_NEEDED) {
1124 t->zsd_flags &= ~ZSD_SHUTDOWN_NEEDED;
1125 t->zsd_flags |= ZSD_SHUTDOWN_INPROGRESS;
1126 DTRACE_PROBE2(zsd__shutdown__inprogress,
1127 zone_t *, zone, zone_key_t, key);
1128 mutex_exit(&zone->zone_lock);
1129 if (lockp != NULL)
1130 mutex_exit(lockp);
1131 dropped = B_TRUE;
1132
1133 ASSERT(t->zsd_shutdown != NULL);
1134 data = t->zsd_data;
1135
1136 DTRACE_PROBE2(zsd__shutdown__start,
1137 zone_t *, zone, zone_key_t, key);
1138
1139 (t->zsd_shutdown)(zone->zone_id, data);
1140 DTRACE_PROBE2(zsd__shutdown__end,
1141 zone_t *, zone, zone_key_t, key);
1142
1143 if (lockp != NULL)
1144 mutex_enter(lockp);
1145 mutex_enter(&zone->zone_lock);
1146 t->zsd_flags &= ~ZSD_SHUTDOWN_INPROGRESS;
1147 t->zsd_flags |= ZSD_SHUTDOWN_COMPLETED;
1148 cv_broadcast(&t->zsd_cv);
1149 DTRACE_PROBE2(zsd__shutdown__completed,
1150 zone_t *, zone, zone_key_t, key);
1151 }
1152 if (!zone_lock_held)
1153 mutex_exit(&zone->zone_lock);
1154 return (dropped);
1155 }
1156
1157 /*
1158 * Call the destroy function for the zone and key if DESTROY_NEEDED
1159 * is set.
1160 * If some other thread gets here first and sets *_INPROGRESS, then
1161 * we wait for that thread to complete so that we can ensure that
1162 * all the callbacks are done when we've looped over all zones/keys.
1163 *
1164 * When we call the destroy function, we drop the global held by the
1165 * caller, and return true to tell the caller it needs to re-evalute the
1166 * state.
1167 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1168 * remains held on exit.
1169 */
1170 static boolean_t
1171 zsd_apply_destroy(kmutex_t *lockp, boolean_t zone_lock_held,
1172 zone_t *zone, zone_key_t key)
1173 {
1174 struct zsd_entry *t;
1175 void *data;
1176 boolean_t dropped;
1177
1178 if (lockp != NULL) {
1179 ASSERT(MUTEX_HELD(lockp));
1180 }
1181 if (zone_lock_held) {
1182 ASSERT(MUTEX_HELD(&zone->zone_lock));
1183 } else {
1184 mutex_enter(&zone->zone_lock);
1185 }
1186
1187 t = zsd_find(&zone->zone_zsd, key);
1188 if (t == NULL) {
1189 /*
1190 * Somebody else got here first e.g the zone going
1191 * away.
1192 */
1193 if (!zone_lock_held)
1194 mutex_exit(&zone->zone_lock);
1195 return (B_FALSE);
1196 }
1197 dropped = B_FALSE;
1198 if (zsd_wait_for_creator(zone, t, lockp))
1199 dropped = B_TRUE;
1200
1201 if (zsd_wait_for_inprogress(zone, t, lockp))
1202 dropped = B_TRUE;
1203
1204 if (t->zsd_flags & ZSD_DESTROY_NEEDED) {
1205 t->zsd_flags &= ~ZSD_DESTROY_NEEDED;
1206 t->zsd_flags |= ZSD_DESTROY_INPROGRESS;
1207 DTRACE_PROBE2(zsd__destroy__inprogress,
1208 zone_t *, zone, zone_key_t, key);
1209 mutex_exit(&zone->zone_lock);
1210 if (lockp != NULL)
1211 mutex_exit(lockp);
1212 dropped = B_TRUE;
1213
1214 ASSERT(t->zsd_destroy != NULL);
1215 data = t->zsd_data;
1216 DTRACE_PROBE2(zsd__destroy__start,
1217 zone_t *, zone, zone_key_t, key);
1218
1219 (t->zsd_destroy)(zone->zone_id, data);
1220 DTRACE_PROBE2(zsd__destroy__end,
1221 zone_t *, zone, zone_key_t, key);
1222
1223 if (lockp != NULL)
1224 mutex_enter(lockp);
1225 mutex_enter(&zone->zone_lock);
1226 t->zsd_data = NULL;
1227 t->zsd_flags &= ~ZSD_DESTROY_INPROGRESS;
1228 t->zsd_flags |= ZSD_DESTROY_COMPLETED;
1229 cv_broadcast(&t->zsd_cv);
1230 DTRACE_PROBE2(zsd__destroy__completed,
1231 zone_t *, zone, zone_key_t, key);
1232 }
1233 if (!zone_lock_held)
1234 mutex_exit(&zone->zone_lock);
1235 return (dropped);
1236 }
1237
1238 /*
1239 * Wait for any CREATE_NEEDED flag to be cleared.
1240 * Returns true if lockp was temporarily dropped while waiting.
1241 */
1242 static boolean_t
1243 zsd_wait_for_creator(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
1244 {
1245 boolean_t dropped = B_FALSE;
1246
1247 while (t->zsd_flags & ZSD_CREATE_NEEDED) {
1248 DTRACE_PROBE2(zsd__wait__for__creator,
1249 zone_t *, zone, struct zsd_entry *, t);
1250 if (lockp != NULL) {
1251 dropped = B_TRUE;
1252 mutex_exit(lockp);
1253 }
1254 cv_wait(&t->zsd_cv, &zone->zone_lock);
1255 if (lockp != NULL) {
1256 /* First drop zone_lock to preserve order */
1257 mutex_exit(&zone->zone_lock);
1258 mutex_enter(lockp);
1259 mutex_enter(&zone->zone_lock);
1260 }
1261 }
1262 return (dropped);
1263 }
1264
1265 /*
1266 * Wait for any INPROGRESS flag to be cleared.
1267 * Returns true if lockp was temporarily dropped while waiting.
1268 */
1269 static boolean_t
1270 zsd_wait_for_inprogress(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
1271 {
1272 boolean_t dropped = B_FALSE;
1273
1274 while (t->zsd_flags & ZSD_ALL_INPROGRESS) {
1275 DTRACE_PROBE2(zsd__wait__for__inprogress,
1276 zone_t *, zone, struct zsd_entry *, t);
1277 if (lockp != NULL) {
1278 dropped = B_TRUE;
1279 mutex_exit(lockp);
1280 }
1281 cv_wait(&t->zsd_cv, &zone->zone_lock);
1282 if (lockp != NULL) {
1283 /* First drop zone_lock to preserve order */
1284 mutex_exit(&zone->zone_lock);
1285 mutex_enter(lockp);
1286 mutex_enter(&zone->zone_lock);
1287 }
1288 }
1289 return (dropped);
1290 }
1291
1292 /*
1293 * Frees memory associated with the zone dataset list.
1294 */
1295 static void
1296 zone_free_datasets(zone_t *zone)
1297 {
1298 zone_dataset_t *t, *next;
1299
1300 for (t = list_head(&zone->zone_datasets); t != NULL; t = next) {
1301 next = list_next(&zone->zone_datasets, t);
1302 list_remove(&zone->zone_datasets, t);
1303 kmem_free(t->zd_dataset, strlen(t->zd_dataset) + 1);
1304 kmem_free(t, sizeof (*t));
1305 }
1306 list_destroy(&zone->zone_datasets);
1307 }
1308
1309 /*
1310 * zone.cpu-shares resource control support.
1311 */
1312 /*ARGSUSED*/
1313 static rctl_qty_t
1314 zone_cpu_shares_usage(rctl_t *rctl, struct proc *p)
1315 {
1316 ASSERT(MUTEX_HELD(&p->p_lock));
1317 return (p->p_zone->zone_shares);
1318 }
1319
1320 /*ARGSUSED*/
1321 static int
1322 zone_cpu_shares_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1323 rctl_qty_t nv)
1324 {
1325 ASSERT(MUTEX_HELD(&p->p_lock));
1326 ASSERT(e->rcep_t == RCENTITY_ZONE);
1327 if (e->rcep_p.zone == NULL)
1328 return (0);
1329
1330 e->rcep_p.zone->zone_shares = nv;
1331 return (0);
1332 }
1333
1334 static rctl_ops_t zone_cpu_shares_ops = {
1335 rcop_no_action,
1336 zone_cpu_shares_usage,
1337 zone_cpu_shares_set,
1338 rcop_no_test
1339 };
1340
1341 /*
1342 * zone.cpu-cap resource control support.
1343 */
1344 /*ARGSUSED*/
1345 static rctl_qty_t
1346 zone_cpu_cap_get(rctl_t *rctl, struct proc *p)
1347 {
1348 ASSERT(MUTEX_HELD(&p->p_lock));
1349 return (cpucaps_zone_get(p->p_zone));
1350 }
1351
1352 /*ARGSUSED*/
1353 static int
1354 zone_cpu_cap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1355 rctl_qty_t nv)
1356 {
1357 zone_t *zone = e->rcep_p.zone;
1358
1359 ASSERT(MUTEX_HELD(&p->p_lock));
1360 ASSERT(e->rcep_t == RCENTITY_ZONE);
1361
1362 if (zone == NULL)
1363 return (0);
1364
1365 /*
1366 * set cap to the new value.
1367 */
1368 return (cpucaps_zone_set(zone, nv));
1369 }
1370
1371 static rctl_ops_t zone_cpu_cap_ops = {
1372 rcop_no_action,
1373 zone_cpu_cap_get,
1374 zone_cpu_cap_set,
1375 rcop_no_test
1376 };
1377
1378 /*ARGSUSED*/
1379 static rctl_qty_t
1380 zone_lwps_usage(rctl_t *r, proc_t *p)
1381 {
1382 rctl_qty_t nlwps;
1383 zone_t *zone = p->p_zone;
1384
1385 ASSERT(MUTEX_HELD(&p->p_lock));
1386
1387 mutex_enter(&zone->zone_nlwps_lock);
1388 nlwps = zone->zone_nlwps;
1389 mutex_exit(&zone->zone_nlwps_lock);
1390
1391 return (nlwps);
1392 }
1393
1394 /*ARGSUSED*/
1395 static int
1396 zone_lwps_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
1397 rctl_qty_t incr, uint_t flags)
1398 {
1399 rctl_qty_t nlwps;
1400
1401 ASSERT(MUTEX_HELD(&p->p_lock));
1402 ASSERT(e->rcep_t == RCENTITY_ZONE);
1403 if (e->rcep_p.zone == NULL)
1404 return (0);
1405 ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
1406 nlwps = e->rcep_p.zone->zone_nlwps;
1407
1408 if (nlwps + incr > rcntl->rcv_value)
1409 return (1);
1410
1411 return (0);
1412 }
1413
1414 /*ARGSUSED*/
1415 static int
1416 zone_lwps_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv)
1417 {
1418 ASSERT(MUTEX_HELD(&p->p_lock));
1419 ASSERT(e->rcep_t == RCENTITY_ZONE);
1420 if (e->rcep_p.zone == NULL)
1421 return (0);
1422 e->rcep_p.zone->zone_nlwps_ctl = nv;
1423 return (0);
1424 }
1425
1426 static rctl_ops_t zone_lwps_ops = {
1427 rcop_no_action,
1428 zone_lwps_usage,
1429 zone_lwps_set,
1430 zone_lwps_test,
1431 };
1432
1433 /*ARGSUSED*/
1434 static rctl_qty_t
1435 zone_procs_usage(rctl_t *r, proc_t *p)
1436 {
1437 rctl_qty_t nprocs;
1438 zone_t *zone = p->p_zone;
1439
1440 ASSERT(MUTEX_HELD(&p->p_lock));
1441
1442 mutex_enter(&zone->zone_nlwps_lock);
1443 nprocs = zone->zone_nprocs;
1444 mutex_exit(&zone->zone_nlwps_lock);
1445
1446 return (nprocs);
1447 }
1448
1449 /*ARGSUSED*/
1450 static int
1451 zone_procs_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
1452 rctl_qty_t incr, uint_t flags)
1453 {
1454 rctl_qty_t nprocs;
1455
1456 ASSERT(MUTEX_HELD(&p->p_lock));
1457 ASSERT(e->rcep_t == RCENTITY_ZONE);
1458 if (e->rcep_p.zone == NULL)
1459 return (0);
1460 ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
1461 nprocs = e->rcep_p.zone->zone_nprocs;
1462
1463 if (nprocs + incr > rcntl->rcv_value)
1464 return (1);
1465
1466 return (0);
1467 }
1468
1469 /*ARGSUSED*/
1470 static int
1471 zone_procs_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv)
1472 {
1473 ASSERT(MUTEX_HELD(&p->p_lock));
1474 ASSERT(e->rcep_t == RCENTITY_ZONE);
1475 if (e->rcep_p.zone == NULL)
1476 return (0);
1477 e->rcep_p.zone->zone_nprocs_ctl = nv;
1478 return (0);
1479 }
1480
1481 static rctl_ops_t zone_procs_ops = {
1482 rcop_no_action,
1483 zone_procs_usage,
1484 zone_procs_set,
1485 zone_procs_test,
1486 };
1487
1488 /*ARGSUSED*/
1489 static int
1490 zone_shmmax_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1491 rctl_qty_t incr, uint_t flags)
1492 {
1493 rctl_qty_t v;
1494 ASSERT(MUTEX_HELD(&p->p_lock));
1495 ASSERT(e->rcep_t == RCENTITY_ZONE);
1496 v = e->rcep_p.zone->zone_shmmax + incr;
1497 if (v > rval->rcv_value)
1498 return (1);
1499 return (0);
1500 }
1501
1502 static rctl_ops_t zone_shmmax_ops = {
1503 rcop_no_action,
1504 rcop_no_usage,
1505 rcop_no_set,
1506 zone_shmmax_test
1507 };
1508
1509 /*ARGSUSED*/
1510 static int
1511 zone_shmmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1512 rctl_qty_t incr, uint_t flags)
1513 {
1514 rctl_qty_t v;
1515 ASSERT(MUTEX_HELD(&p->p_lock));
1516 ASSERT(e->rcep_t == RCENTITY_ZONE);
1517 v = e->rcep_p.zone->zone_ipc.ipcq_shmmni + incr;
1518 if (v > rval->rcv_value)
1519 return (1);
1520 return (0);
1521 }
1522
1523 static rctl_ops_t zone_shmmni_ops = {
1524 rcop_no_action,
1525 rcop_no_usage,
1526 rcop_no_set,
1527 zone_shmmni_test
1528 };
1529
1530 /*ARGSUSED*/
1531 static int
1532 zone_semmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1533 rctl_qty_t incr, uint_t flags)
1534 {
1535 rctl_qty_t v;
1536 ASSERT(MUTEX_HELD(&p->p_lock));
1537 ASSERT(e->rcep_t == RCENTITY_ZONE);
1538 v = e->rcep_p.zone->zone_ipc.ipcq_semmni + incr;
1539 if (v > rval->rcv_value)
1540 return (1);
1541 return (0);
1542 }
1543
1544 static rctl_ops_t zone_semmni_ops = {
1545 rcop_no_action,
1546 rcop_no_usage,
1547 rcop_no_set,
1548 zone_semmni_test
1549 };
1550
1551 /*ARGSUSED*/
1552 static int
1553 zone_msgmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1554 rctl_qty_t incr, uint_t flags)
1555 {
1556 rctl_qty_t v;
1557 ASSERT(MUTEX_HELD(&p->p_lock));
1558 ASSERT(e->rcep_t == RCENTITY_ZONE);
1559 v = e->rcep_p.zone->zone_ipc.ipcq_msgmni + incr;
1560 if (v > rval->rcv_value)
1561 return (1);
1562 return (0);
1563 }
1564
1565 static rctl_ops_t zone_msgmni_ops = {
1566 rcop_no_action,
1567 rcop_no_usage,
1568 rcop_no_set,
1569 zone_msgmni_test
1570 };
1571
1572 /*ARGSUSED*/
1573 static rctl_qty_t
1574 zone_locked_mem_usage(rctl_t *rctl, struct proc *p)
1575 {
1576 rctl_qty_t q;
1577 ASSERT(MUTEX_HELD(&p->p_lock));
1578 mutex_enter(&p->p_zone->zone_mem_lock);
1579 q = p->p_zone->zone_locked_mem;
1580 mutex_exit(&p->p_zone->zone_mem_lock);
1581 return (q);
1582 }
1583
1584 /*ARGSUSED*/
1585 static int
1586 zone_locked_mem_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1587 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1588 {
1589 rctl_qty_t q;
1590 zone_t *z;
1591
1592 z = e->rcep_p.zone;
1593 ASSERT(MUTEX_HELD(&p->p_lock));
1594 ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1595 q = z->zone_locked_mem;
1596 if (q + incr > rcntl->rcv_value)
1597 return (1);
1598 return (0);
1599 }
1600
1601 /*ARGSUSED*/
1602 static int
1603 zone_locked_mem_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1604 rctl_qty_t nv)
1605 {
1606 ASSERT(MUTEX_HELD(&p->p_lock));
1607 ASSERT(e->rcep_t == RCENTITY_ZONE);
1608 if (e->rcep_p.zone == NULL)
1609 return (0);
1610 e->rcep_p.zone->zone_locked_mem_ctl = nv;
1611 return (0);
1612 }
1613
1614 static rctl_ops_t zone_locked_mem_ops = {
1615 rcop_no_action,
1616 zone_locked_mem_usage,
1617 zone_locked_mem_set,
1618 zone_locked_mem_test
1619 };
1620
1621 /*ARGSUSED*/
1622 static rctl_qty_t
1623 zone_max_swap_usage(rctl_t *rctl, struct proc *p)
1624 {
1625 rctl_qty_t q;
1626 zone_t *z = p->p_zone;
1627
1628 ASSERT(MUTEX_HELD(&p->p_lock));
1629 mutex_enter(&z->zone_mem_lock);
1630 q = z->zone_max_swap;
1631 mutex_exit(&z->zone_mem_lock);
1632 return (q);
1633 }
1634
1635 /*ARGSUSED*/
1636 static int
1637 zone_max_swap_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1638 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1639 {
1640 rctl_qty_t q;
1641 zone_t *z;
1642
1643 z = e->rcep_p.zone;
1644 ASSERT(MUTEX_HELD(&p->p_lock));
1645 ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1646 q = z->zone_max_swap;
1647 if (q + incr > rcntl->rcv_value)
1648 return (1);
1649 return (0);
1650 }
1651
1652 /*ARGSUSED*/
1653 static int
1654 zone_max_swap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1655 rctl_qty_t nv)
1656 {
1657 ASSERT(MUTEX_HELD(&p->p_lock));
1658 ASSERT(e->rcep_t == RCENTITY_ZONE);
1659 if (e->rcep_p.zone == NULL)
1660 return (0);
1661 e->rcep_p.zone->zone_max_swap_ctl = nv;
1662 return (0);
1663 }
1664
1665 static rctl_ops_t zone_max_swap_ops = {
1666 rcop_no_action,
1667 zone_max_swap_usage,
1668 zone_max_swap_set,
1669 zone_max_swap_test
1670 };
1671
1672 /*ARGSUSED*/
1673 static rctl_qty_t
1674 zone_max_lofi_usage(rctl_t *rctl, struct proc *p)
1675 {
1676 rctl_qty_t q;
1677 zone_t *z = p->p_zone;
1678
1679 ASSERT(MUTEX_HELD(&p->p_lock));
1680 mutex_enter(&z->zone_rctl_lock);
1681 q = z->zone_max_lofi;
1682 mutex_exit(&z->zone_rctl_lock);
1683 return (q);
1684 }
1685
1686 /*ARGSUSED*/
1687 static int
1688 zone_max_lofi_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1689 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1690 {
1691 rctl_qty_t q;
1692 zone_t *z;
1693
1694 z = e->rcep_p.zone;
1695 ASSERT(MUTEX_HELD(&p->p_lock));
1696 ASSERT(MUTEX_HELD(&z->zone_rctl_lock));
1697 q = z->zone_max_lofi;
1698 if (q + incr > rcntl->rcv_value)
1699 return (1);
1700 return (0);
1701 }
1702
1703 /*ARGSUSED*/
1704 static int
1705 zone_max_lofi_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1706 rctl_qty_t nv)
1707 {
1708 ASSERT(MUTEX_HELD(&p->p_lock));
1709 ASSERT(e->rcep_t == RCENTITY_ZONE);
1710 if (e->rcep_p.zone == NULL)
1711 return (0);
1712 e->rcep_p.zone->zone_max_lofi_ctl = nv;
1713 return (0);
1714 }
1715
1716 static rctl_ops_t zone_max_lofi_ops = {
1717 rcop_no_action,
1718 zone_max_lofi_usage,
1719 zone_max_lofi_set,
1720 zone_max_lofi_test
1721 };
1722
1723 /*
1724 * Helper function to brand the zone with a unique ID.
1725 */
1726 static void
1727 zone_uniqid(zone_t *zone)
1728 {
1729 static uint64_t uniqid = 0;
1730
1731 ASSERT(MUTEX_HELD(&zonehash_lock));
1732 zone->zone_uniqid = uniqid++;
1733 }
1734
1735 /*
1736 * Returns a held pointer to the "kcred" for the specified zone.
1737 */
1738 struct cred *
1739 zone_get_kcred(zoneid_t zoneid)
1740 {
1741 zone_t *zone;
1742 cred_t *cr;
1743
1744 if ((zone = zone_find_by_id(zoneid)) == NULL)
1745 return (NULL);
1746 cr = zone->zone_kcred;
1747 crhold(cr);
1748 zone_rele(zone);
1749 return (cr);
1750 }
1751
1752 static int
1753 zone_lockedmem_kstat_update(kstat_t *ksp, int rw)
1754 {
1755 zone_t *zone = ksp->ks_private;
1756 zone_kstat_t *zk = ksp->ks_data;
1757
1758 if (rw == KSTAT_WRITE)
1759 return (EACCES);
1760
1761 zk->zk_usage.value.ui64 = zone->zone_locked_mem;
1762 zk->zk_value.value.ui64 = zone->zone_locked_mem_ctl;
1763 return (0);
1764 }
1765
1766 static int
1767 zone_nprocs_kstat_update(kstat_t *ksp, int rw)
1768 {
1769 zone_t *zone = ksp->ks_private;
1770 zone_kstat_t *zk = ksp->ks_data;
1771
1772 if (rw == KSTAT_WRITE)
1773 return (EACCES);
1774
1775 zk->zk_usage.value.ui64 = zone->zone_nprocs;
1776 zk->zk_value.value.ui64 = zone->zone_nprocs_ctl;
1777 return (0);
1778 }
1779
1780 static int
1781 zone_swapresv_kstat_update(kstat_t *ksp, int rw)
1782 {
1783 zone_t *zone = ksp->ks_private;
1784 zone_kstat_t *zk = ksp->ks_data;
1785
1786 if (rw == KSTAT_WRITE)
1787 return (EACCES);
1788
1789 zk->zk_usage.value.ui64 = zone->zone_max_swap;
1790 zk->zk_value.value.ui64 = zone->zone_max_swap_ctl;
1791 return (0);
1792 }
1793
1794 static kstat_t *
1795 zone_kstat_create_common(zone_t *zone, char *name,
1796 int (*updatefunc) (kstat_t *, int))
1797 {
1798 kstat_t *ksp;
1799 zone_kstat_t *zk;
1800
1801 ksp = rctl_kstat_create_zone(zone, name, KSTAT_TYPE_NAMED,
1802 sizeof (zone_kstat_t) / sizeof (kstat_named_t),
1803 KSTAT_FLAG_VIRTUAL);
1804
1805 if (ksp == NULL)
1806 return (NULL);
1807
1808 zk = ksp->ks_data = kmem_alloc(sizeof (zone_kstat_t), KM_SLEEP);
1809 ksp->ks_data_size += strlen(zone->zone_name) + 1;
1810 kstat_named_init(&zk->zk_zonename, "zonename", KSTAT_DATA_STRING);
1811 kstat_named_setstr(&zk->zk_zonename, zone->zone_name);
1812 kstat_named_init(&zk->zk_usage, "usage", KSTAT_DATA_UINT64);
1813 kstat_named_init(&zk->zk_value, "value", KSTAT_DATA_UINT64);
1814 ksp->ks_update = updatefunc;
1815 ksp->ks_private = zone;
1816 kstat_install(ksp);
1817 return (ksp);
1818 }
1819
1820
1821 static int
1822 zone_mcap_kstat_update(kstat_t *ksp, int rw)
1823 {
1824 zone_t *zone = ksp->ks_private;
1825 zone_mcap_kstat_t *zmp = ksp->ks_data;
1826
1827 if (rw == KSTAT_WRITE)
1828 return (EACCES);
1829
1830 zmp->zm_pgpgin.value.ui64 = zone->zone_pgpgin;
1831 zmp->zm_anonpgin.value.ui64 = zone->zone_anonpgin;
1832 zmp->zm_execpgin.value.ui64 = zone->zone_execpgin;
1833 zmp->zm_fspgin.value.ui64 = zone->zone_fspgin;
1834 zmp->zm_anon_alloc_fail.value.ui64 = zone->zone_anon_alloc_fail;
1835
1836 return (0);
1837 }
1838
1839 static kstat_t *
1840 zone_mcap_kstat_create(zone_t *zone)
1841 {
1842 kstat_t *ksp;
1843 zone_mcap_kstat_t *zmp;
1844
1845 if ((ksp = kstat_create_zone("memory_cap", zone->zone_id,
1846 zone->zone_name, "zone_memory_cap", KSTAT_TYPE_NAMED,
1847 sizeof (zone_mcap_kstat_t) / sizeof (kstat_named_t),
1848 KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
1849 return (NULL);
1850
1851 if (zone->zone_id != GLOBAL_ZONEID)
1852 kstat_zone_add(ksp, GLOBAL_ZONEID);
1853
1854 zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_mcap_kstat_t), KM_SLEEP);
1855 ksp->ks_data_size += strlen(zone->zone_name) + 1;
1856 ksp->ks_lock = &zone->zone_mcap_lock;
1857 zone->zone_mcap_stats = zmp;
1858
1859 /* The kstat "name" field is not large enough for a full zonename */
1860 kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
1861 kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
1862 kstat_named_init(&zmp->zm_pgpgin, "pgpgin", KSTAT_DATA_UINT64);
1863 kstat_named_init(&zmp->zm_anonpgin, "anonpgin", KSTAT_DATA_UINT64);
1864 kstat_named_init(&zmp->zm_execpgin, "execpgin", KSTAT_DATA_UINT64);
1865 kstat_named_init(&zmp->zm_fspgin, "fspgin", KSTAT_DATA_UINT64);
1866 kstat_named_init(&zmp->zm_anon_alloc_fail, "anon_alloc_fail",
1867 KSTAT_DATA_UINT64);
1868
1869 ksp->ks_update = zone_mcap_kstat_update;
1870 ksp->ks_private = zone;
1871
1872 kstat_install(ksp);
1873 return (ksp);
1874 }
1875
1876 static int
1877 zone_misc_kstat_update(kstat_t *ksp, int rw)
1878 {
1879 zone_t *zone = ksp->ks_private;
1880 zone_misc_kstat_t *zmp = ksp->ks_data;
1881 hrtime_t tmp;
1882
1883 if (rw == KSTAT_WRITE)
1884 return (EACCES);
1885
1886 tmp = zone->zone_utime;
1887 scalehrtime(&tmp);
1888 zmp->zm_utime.value.ui64 = tmp;
1889 tmp = zone->zone_stime;
1890 scalehrtime(&tmp);
1891 zmp->zm_stime.value.ui64 = tmp;
1892 tmp = zone->zone_wtime;
1893 scalehrtime(&tmp);
1894 zmp->zm_wtime.value.ui64 = tmp;
1895
1896 zmp->zm_avenrun1.value.ui32 = zone->zone_avenrun[0];
1897 zmp->zm_avenrun5.value.ui32 = zone->zone_avenrun[1];
1898 zmp->zm_avenrun15.value.ui32 = zone->zone_avenrun[2];
1899
1900 zmp->zm_ffcap.value.ui32 = zone->zone_ffcap;
1901 zmp->zm_ffnoproc.value.ui32 = zone->zone_ffnoproc;
1902 zmp->zm_ffnomem.value.ui32 = zone->zone_ffnomem;
1903 zmp->zm_ffmisc.value.ui32 = zone->zone_ffmisc;
1904
1905 zmp->zm_nested_intp.value.ui32 = zone->zone_nested_intp;
1906
1907 zmp->zm_init_pid.value.ui32 = zone->zone_proc_initpid;
1908 zmp->zm_boot_time.value.ui64 = (uint64_t)zone->zone_boot_time;
1909
1910 return (0);
1911 }
1912
1913 static kstat_t *
1914 zone_misc_kstat_create(zone_t *zone)
1915 {
1916 kstat_t *ksp;
1917 zone_misc_kstat_t *zmp;
1918
1919 if ((ksp = kstat_create_zone("zones", zone->zone_id,
1920 zone->zone_name, "zone_misc", KSTAT_TYPE_NAMED,
1921 sizeof (zone_misc_kstat_t) / sizeof (kstat_named_t),
1922 KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
1923 return (NULL);
1924
1925 if (zone->zone_id != GLOBAL_ZONEID)
1926 kstat_zone_add(ksp, GLOBAL_ZONEID);
1927
1928 zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_misc_kstat_t), KM_SLEEP);
1929 ksp->ks_data_size += strlen(zone->zone_name) + 1;
1930 ksp->ks_lock = &zone->zone_misc_lock;
1931 zone->zone_misc_stats = zmp;
1932
1933 /* The kstat "name" field is not large enough for a full zonename */
1934 kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
1935 kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
1936 kstat_named_init(&zmp->zm_utime, "nsec_user", KSTAT_DATA_UINT64);
1937 kstat_named_init(&zmp->zm_stime, "nsec_sys", KSTAT_DATA_UINT64);
1938 kstat_named_init(&zmp->zm_wtime, "nsec_waitrq", KSTAT_DATA_UINT64);
1939 kstat_named_init(&zmp->zm_avenrun1, "avenrun_1min", KSTAT_DATA_UINT32);
1940 kstat_named_init(&zmp->zm_avenrun5, "avenrun_5min", KSTAT_DATA_UINT32);
1941 kstat_named_init(&zmp->zm_avenrun15, "avenrun_15min",
1942 KSTAT_DATA_UINT32);
1943 kstat_named_init(&zmp->zm_ffcap, "forkfail_cap", KSTAT_DATA_UINT32);
1944 kstat_named_init(&zmp->zm_ffnoproc, "forkfail_noproc",
1945 KSTAT_DATA_UINT32);
1946 kstat_named_init(&zmp->zm_ffnomem, "forkfail_nomem", KSTAT_DATA_UINT32);
1947 kstat_named_init(&zmp->zm_ffmisc, "forkfail_misc", KSTAT_DATA_UINT32);
1948 kstat_named_init(&zmp->zm_nested_intp, "nested_interp",
1949 KSTAT_DATA_UINT32);
1950 kstat_named_init(&zmp->zm_init_pid, "init_pid", KSTAT_DATA_UINT32);
1951 kstat_named_init(&zmp->zm_boot_time, "boot_time", KSTAT_DATA_UINT64);
1952
1953 ksp->ks_update = zone_misc_kstat_update;
1954 ksp->ks_private = zone;
1955
1956 kstat_install(ksp);
1957 return (ksp);
1958 }
1959
1960 static void
1961 zone_kstat_create(zone_t *zone)
1962 {
1963 zone->zone_lockedmem_kstat = zone_kstat_create_common(zone,
1964 "lockedmem", zone_lockedmem_kstat_update);
1965 zone->zone_swapresv_kstat = zone_kstat_create_common(zone,
1966 "swapresv", zone_swapresv_kstat_update);
1967 zone->zone_nprocs_kstat = zone_kstat_create_common(zone,
1968 "nprocs", zone_nprocs_kstat_update);
1969
1970 if ((zone->zone_mcap_ksp = zone_mcap_kstat_create(zone)) == NULL) {
1971 zone->zone_mcap_stats = kmem_zalloc(
1972 sizeof (zone_mcap_kstat_t), KM_SLEEP);
1973 }
1974
1975 if ((zone->zone_misc_ksp = zone_misc_kstat_create(zone)) == NULL) {
1976 zone->zone_misc_stats = kmem_zalloc(
1977 sizeof (zone_misc_kstat_t), KM_SLEEP);
1978 }
1979 }
1980
1981 static void
1982 zone_kstat_delete_common(kstat_t **pkstat, size_t datasz)
1983 {
1984 void *data;
1985
1986 if (*pkstat != NULL) {
1987 data = (*pkstat)->ks_data;
1988 kstat_delete(*pkstat);
1989 kmem_free(data, datasz);
1990 *pkstat = NULL;
1991 }
1992 }
1993
1994 static void
1995 zone_kstat_delete(zone_t *zone)
1996 {
1997 zone_kstat_delete_common(&zone->zone_lockedmem_kstat,
1998 sizeof (zone_kstat_t));
1999 zone_kstat_delete_common(&zone->zone_swapresv_kstat,
2000 sizeof (zone_kstat_t));
2001 zone_kstat_delete_common(&zone->zone_nprocs_kstat,
2002 sizeof (zone_kstat_t));
2003 zone_kstat_delete_common(&zone->zone_mcap_ksp,
2004 sizeof (zone_mcap_kstat_t));
2005 zone_kstat_delete_common(&zone->zone_misc_ksp,
2006 sizeof (zone_misc_kstat_t));
2007 }
2008
2009 /*
2010 * Called very early on in boot to initialize the ZSD list so that
2011 * zone_key_create() can be called before zone_init(). It also initializes
2012 * portions of zone0 which may be used before zone_init() is called. The
2013 * variable "global_zone" will be set when zone0 is fully initialized by
2014 * zone_init().
2015 */
2016 void
2017 zone_zsd_init(void)
2018 {
2019 mutex_init(&zonehash_lock, NULL, MUTEX_DEFAULT, NULL);
2020 mutex_init(&zsd_key_lock, NULL, MUTEX_DEFAULT, NULL);
2021 list_create(&zsd_registered_keys, sizeof (struct zsd_entry),
2022 offsetof(struct zsd_entry, zsd_linkage));
2023 list_create(&zone_active, sizeof (zone_t),
2024 offsetof(zone_t, zone_linkage));
2025 list_create(&zone_deathrow, sizeof (zone_t),
2026 offsetof(zone_t, zone_linkage));
2027
2028 mutex_init(&zone0.zone_lock, NULL, MUTEX_DEFAULT, NULL);
2029 mutex_init(&zone0.zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
2030 mutex_init(&zone0.zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
2031 zone0.zone_shares = 1;
2032 zone0.zone_nlwps = 0;
2033 zone0.zone_nlwps_ctl = INT_MAX;
2034 zone0.zone_nprocs = 0;
2035 zone0.zone_nprocs_ctl = INT_MAX;
2036 zone0.zone_locked_mem = 0;
2037 zone0.zone_locked_mem_ctl = UINT64_MAX;
2038 ASSERT(zone0.zone_max_swap == 0);
2039 zone0.zone_max_swap_ctl = UINT64_MAX;
2040 zone0.zone_max_lofi = 0;
2041 zone0.zone_max_lofi_ctl = UINT64_MAX;
2042 zone0.zone_shmmax = 0;
2043 zone0.zone_ipc.ipcq_shmmni = 0;
2044 zone0.zone_ipc.ipcq_semmni = 0;
2045 zone0.zone_ipc.ipcq_msgmni = 0;
2046 zone0.zone_name = GLOBAL_ZONENAME;
2047 zone0.zone_nodename = utsname.nodename;
2048 zone0.zone_domain = srpc_domain;
2049 zone0.zone_hostid = HW_INVALID_HOSTID;
2050 zone0.zone_fs_allowed = NULL;
2051 psecflags_default(&zone0.zone_secflags);
2052 zone0.zone_ref = 1;
2053 zone0.zone_id = GLOBAL_ZONEID;
2054 zone0.zone_status = ZONE_IS_RUNNING;
2055 zone0.zone_rootpath = "/";
2056 zone0.zone_rootpathlen = 2;
2057 zone0.zone_psetid = ZONE_PS_INVAL;
2058 zone0.zone_ncpus = 0;
2059 zone0.zone_ncpus_online = 0;
2060 zone0.zone_proc_initpid = 1;
2061 zone0.zone_initname = initname;
2062 zone0.zone_lockedmem_kstat = NULL;
2063 zone0.zone_swapresv_kstat = NULL;
2064 zone0.zone_nprocs_kstat = NULL;
2065
2066 zone0.zone_stime = 0;
2067 zone0.zone_utime = 0;
2068 zone0.zone_wtime = 0;
2069
2070 list_create(&zone0.zone_ref_list, sizeof (zone_ref_t),
2071 offsetof(zone_ref_t, zref_linkage));
2072 list_create(&zone0.zone_zsd, sizeof (struct zsd_entry),
2073 offsetof(struct zsd_entry, zsd_linkage));
2074 list_insert_head(&zone_active, &zone0);
2075
2076 /*
2077 * The root filesystem is not mounted yet, so zone_rootvp cannot be set
2078 * to anything meaningful. It is assigned to be 'rootdir' in
2079 * vfs_mountroot().
2080 */
2081 zone0.zone_rootvp = NULL;
2082 zone0.zone_vfslist = NULL;
2083 zone0.zone_bootargs = initargs;
2084 zone0.zone_privset = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
2085 /*
2086 * The global zone has all privileges
2087 */
2088 priv_fillset(zone0.zone_privset);
2089 /*
2090 * Add p0 to the global zone
2091 */
2092 zone0.zone_zsched = &p0;
2093 p0.p_zone = &zone0;
2094 }
2095
2096 /*
2097 * Compute a hash value based on the contents of the label and the DOI. The
2098 * hash algorithm is somewhat arbitrary, but is based on the observation that
2099 * humans will likely pick labels that differ by amounts that work out to be
2100 * multiples of the number of hash chains, and thus stirring in some primes
2101 * should help.
2102 */
2103 static uint_t
2104 hash_bylabel(void *hdata, mod_hash_key_t key)
2105 {
2106 const ts_label_t *lab = (ts_label_t *)key;
2107 const uint32_t *up, *ue;
2108 uint_t hash;
2109 int i;
2110
2111 _NOTE(ARGUNUSED(hdata));
2112
2113 hash = lab->tsl_doi + (lab->tsl_doi << 1);
2114 /* we depend on alignment of label, but not representation */
2115 up = (const uint32_t *)&lab->tsl_label;
2116 ue = up + sizeof (lab->tsl_label) / sizeof (*up);
2117 i = 1;
2118 while (up < ue) {
2119 /* using 2^n + 1, 1 <= n <= 16 as source of many primes */
2120 hash += *up + (*up << ((i % 16) + 1));
2121 up++;
2122 i++;
2123 }
2124 return (hash);
2125 }
2126
2127 /*
2128 * All that mod_hash cares about here is zero (equal) versus non-zero (not
2129 * equal). This may need to be changed if less than / greater than is ever
2130 * needed.
2131 */
2132 static int
2133 hash_labelkey_cmp(mod_hash_key_t key1, mod_hash_key_t key2)
2134 {
2135 ts_label_t *lab1 = (ts_label_t *)key1;
2136 ts_label_t *lab2 = (ts_label_t *)key2;
2137
2138 return (label_equal(lab1, lab2) ? 0 : 1);
2139 }
2140
2141 /*
2142 * Called by main() to initialize the zones framework.
2143 */
2144 void
2145 zone_init(void)
2146 {
2147 rctl_dict_entry_t *rde;
2148 rctl_val_t *dval;
2149 rctl_set_t *set;
2150 rctl_alloc_gp_t *gp;
2151 rctl_entity_p_t e;
2152 int res;
2153
2154 ASSERT(curproc == &p0);
2155
2156 /*
2157 * Create ID space for zone IDs. ID 0 is reserved for the
2158 * global zone.
2159 */
2160 zoneid_space = id_space_create("zoneid_space", 1, MAX_ZONEID);
2161
2162 /*
2163 * Initialize generic zone resource controls, if any.
2164 */
2165 rc_zone_cpu_shares = rctl_register("zone.cpu-shares",
2166 RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_NEVER |
2167 RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT | RCTL_GLOBAL_SYSLOG_NEVER,
2168 FSS_MAXSHARES, FSS_MAXSHARES, &zone_cpu_shares_ops);
2169
2170 rc_zone_cpu_cap = rctl_register("zone.cpu-cap",
2171 RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_ALWAYS |
2172 RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |RCTL_GLOBAL_SYSLOG_NEVER |
2173 RCTL_GLOBAL_INFINITE,
2174 MAXCAP, MAXCAP, &zone_cpu_cap_ops);
2175
2176 rc_zone_nlwps = rctl_register("zone.max-lwps", RCENTITY_ZONE,
2177 RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
2178 INT_MAX, INT_MAX, &zone_lwps_ops);
2179
2180 rc_zone_nprocs = rctl_register("zone.max-processes", RCENTITY_ZONE,
2181 RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
2182 INT_MAX, INT_MAX, &zone_procs_ops);
2183
2184 /*
2185 * System V IPC resource controls
2186 */
2187 rc_zone_msgmni = rctl_register("zone.max-msg-ids",
2188 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2189 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_msgmni_ops);
2190
2191 rc_zone_semmni = rctl_register("zone.max-sem-ids",
2192 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2193 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_semmni_ops);
2194
2195 rc_zone_shmmni = rctl_register("zone.max-shm-ids",
2196 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2197 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_shmmni_ops);
2198
2199 rc_zone_shmmax = rctl_register("zone.max-shm-memory",
2200 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2201 RCTL_GLOBAL_BYTES, UINT64_MAX, UINT64_MAX, &zone_shmmax_ops);
2202
2203 /*
2204 * Create a rctl_val with PRIVILEGED, NOACTION, value = 1. Then attach
2205 * this at the head of the rctl_dict_entry for ``zone.cpu-shares''.
2206 */
2207 dval = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
2208 bzero(dval, sizeof (rctl_val_t));
2209 dval->rcv_value = 1;
2210 dval->rcv_privilege = RCPRIV_PRIVILEGED;
2211 dval->rcv_flagaction = RCTL_LOCAL_NOACTION;
2212 dval->rcv_action_recip_pid = -1;
2213
2214 rde = rctl_dict_lookup("zone.cpu-shares");
2215 (void) rctl_val_list_insert(&rde->rcd_default_value, dval);
2216
2217 rc_zone_locked_mem = rctl_register("zone.max-locked-memory",
2218 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
2219 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2220 &zone_locked_mem_ops);
2221
2222 rc_zone_max_swap = rctl_register("zone.max-swap",
2223 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
2224 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2225 &zone_max_swap_ops);
2226
2227 rc_zone_max_lofi = rctl_register("zone.max-lofi",
2228 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |
2229 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2230 &zone_max_lofi_ops);
2231
2232 /*
2233 * Initialize the ``global zone''.
2234 */
2235 set = rctl_set_create();
2236 gp = rctl_set_init_prealloc(RCENTITY_ZONE);
2237 mutex_enter(&p0.p_lock);
2238 e.rcep_p.zone = &zone0;
2239 e.rcep_t = RCENTITY_ZONE;
2240 zone0.zone_rctls = rctl_set_init(RCENTITY_ZONE, &p0, &e, set,
2241 gp);
2242
2243 zone0.zone_nlwps = p0.p_lwpcnt;
2244 zone0.zone_nprocs = 1;
2245 zone0.zone_ntasks = 1;
2246 mutex_exit(&p0.p_lock);
2247 zone0.zone_restart_init = B_TRUE;
2248 zone0.zone_brand = &native_brand;
2249 rctl_prealloc_destroy(gp);
2250 /*
2251 * pool_default hasn't been initialized yet, so we let pool_init()
2252 * take care of making sure the global zone is in the default pool.
2253 */
2254
2255 /*
2256 * Initialize global zone kstats
2257 */
2258 zone_kstat_create(&zone0);
2259
2260 /*
2261 * Initialize zone label.
2262 * mlp are initialized when tnzonecfg is loaded.
2263 */
2264 zone0.zone_slabel = l_admin_low;
2265 rw_init(&zone0.zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
2266 label_hold(l_admin_low);
2267
2268 /*
2269 * Initialise the lock for the database structure used by mntfs.
2270 */
2271 rw_init(&zone0.zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL);
2272
2273 mutex_enter(&zonehash_lock);
2274 zone_uniqid(&zone0);
2275 ASSERT(zone0.zone_uniqid == GLOBAL_ZONEUNIQID);
2276
2277 zonehashbyid = mod_hash_create_idhash("zone_by_id", zone_hash_size,
2278 mod_hash_null_valdtor);
2279 zonehashbyname = mod_hash_create_strhash("zone_by_name",
2280 zone_hash_size, mod_hash_null_valdtor);
2281 /*
2282 * maintain zonehashbylabel only for labeled systems
2283 */
2284 if (is_system_labeled())
2285 zonehashbylabel = mod_hash_create_extended("zone_by_label",
2286 zone_hash_size, mod_hash_null_keydtor,
2287 mod_hash_null_valdtor, hash_bylabel, NULL,
2288 hash_labelkey_cmp, KM_SLEEP);
2289 zonecount = 1;
2290
2291 (void) mod_hash_insert(zonehashbyid, (mod_hash_key_t)GLOBAL_ZONEID,
2292 (mod_hash_val_t)&zone0);
2293 (void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)zone0.zone_name,
2294 (mod_hash_val_t)&zone0);
2295 if (is_system_labeled()) {
2296 zone0.zone_flags |= ZF_HASHED_LABEL;
2297 (void) mod_hash_insert(zonehashbylabel,
2298 (mod_hash_key_t)zone0.zone_slabel, (mod_hash_val_t)&zone0);
2299 }
2300 mutex_exit(&zonehash_lock);
2301
2302 /*
2303 * We avoid setting zone_kcred until now, since kcred is initialized
2304 * sometime after zone_zsd_init() and before zone_init().
2305 */
2306 zone0.zone_kcred = kcred;
2307 /*
2308 * The global zone is fully initialized (except for zone_rootvp which
2309 * will be set when the root filesystem is mounted).
2310 */
2311 global_zone = &zone0;
2312
2313 /*
2314 * Setup an event channel to send zone status change notifications on
2315 */
2316 res = sysevent_evc_bind(ZONE_EVENT_CHANNEL, &zone_event_chan,
2317 EVCH_CREAT);
2318
2319 if (res)
2320 panic("Sysevent_evc_bind failed during zone setup.\n");
2321
2322 }
2323
2324 static void
2325 zone_free(zone_t *zone)
2326 {
2327 ASSERT(zone != global_zone);
2328 ASSERT(zone->zone_ntasks == 0);
2329 ASSERT(zone->zone_nlwps == 0);
2330 ASSERT(zone->zone_nprocs == 0);
2331 ASSERT(zone->zone_cred_ref == 0);
2332 ASSERT(zone->zone_kcred == NULL);
2333 ASSERT(zone_status_get(zone) == ZONE_IS_DEAD ||
2334 zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
2335 ASSERT(list_is_empty(&zone->zone_ref_list));
2336
2337 /*
2338 * Remove any zone caps.
2339 */
2340 cpucaps_zone_remove(zone);
2341
2342 ASSERT(zone->zone_cpucap == NULL);
2343
2344 /* remove from deathrow list */
2345 if (zone_status_get(zone) == ZONE_IS_DEAD) {
2346 ASSERT(zone->zone_ref == 0);
2347 mutex_enter(&zone_deathrow_lock);
2348 list_remove(&zone_deathrow, zone);
2349 mutex_exit(&zone_deathrow_lock);
2350 }
2351
2352 list_destroy(&zone->zone_ref_list);
2353 zone_free_zsd(zone);
2354 zone_free_datasets(zone);
2355 list_destroy(&zone->zone_dl_list);
2356
2357 if (zone->zone_rootvp != NULL)
2358 VN_RELE(zone->zone_rootvp);
2359 if (zone->zone_rootpath)
2360 kmem_free(zone->zone_rootpath, zone->zone_rootpathlen);
2361 if (zone->zone_name != NULL)
2362 kmem_free(zone->zone_name, ZONENAME_MAX);
2363 if (zone->zone_slabel != NULL)
2364 label_rele(zone->zone_slabel);
2365 if (zone->zone_nodename != NULL)
2366 kmem_free(zone->zone_nodename, _SYS_NMLN);
2367 if (zone->zone_domain != NULL)
2368 kmem_free(zone->zone_domain, _SYS_NMLN);
2369 if (zone->zone_privset != NULL)
2370 kmem_free(zone->zone_privset, sizeof (priv_set_t));
2371 if (zone->zone_rctls != NULL)
2372 rctl_set_free(zone->zone_rctls);
2373 if (zone->zone_bootargs != NULL)
2374 strfree(zone->zone_bootargs);
2375 if (zone->zone_initname != NULL)
2376 strfree(zone->zone_initname);
2377 if (zone->zone_fs_allowed != NULL)
2378 strfree(zone->zone_fs_allowed);
2379 if (zone->zone_pfexecd != NULL)
2380 klpd_freelist(&zone->zone_pfexecd);
2381 id_free(zoneid_space, zone->zone_id);
2382 mutex_destroy(&zone->zone_lock);
2383 cv_destroy(&zone->zone_cv);
2384 rw_destroy(&zone->zone_mlps.mlpl_rwlock);
2385 rw_destroy(&zone->zone_mntfs_db_lock);
2386 kmem_free(zone, sizeof (zone_t));
2387 }
2388
2389 /*
2390 * See block comment at the top of this file for information about zone
2391 * status values.
2392 */
2393 /*
2394 * Convenience function for setting zone status.
2395 */
2396 static void
2397 zone_status_set(zone_t *zone, zone_status_t status)
2398 {
2399
2400 nvlist_t *nvl = NULL;
2401 ASSERT(MUTEX_HELD(&zone_status_lock));
2402 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE &&
2403 status >= zone_status_get(zone));
2404
2405 if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) ||
2406 nvlist_add_string(nvl, ZONE_CB_NAME, zone->zone_name) ||
2407 nvlist_add_string(nvl, ZONE_CB_NEWSTATE,
2408 zone_status_table[status]) ||
2409 nvlist_add_string(nvl, ZONE_CB_OLDSTATE,
2410 zone_status_table[zone->zone_status]) ||
2411 nvlist_add_int32(nvl, ZONE_CB_ZONEID, zone->zone_id) ||
2412 nvlist_add_uint64(nvl, ZONE_CB_TIMESTAMP, (uint64_t)gethrtime()) ||
2413 sysevent_evc_publish(zone_event_chan, ZONE_EVENT_STATUS_CLASS,
2414 ZONE_EVENT_STATUS_SUBCLASS, "sun.com", "kernel", nvl, EVCH_SLEEP)) {
2415 #ifdef DEBUG
2416 (void) printf(
2417 "Failed to allocate and send zone state change event.\n");
2418 #endif
2419 }
2420 nvlist_free(nvl);
2421
2422 zone->zone_status = status;
2423
2424 cv_broadcast(&zone->zone_cv);
2425 }
2426
2427 /*
2428 * Public function to retrieve the zone status. The zone status may
2429 * change after it is retrieved.
2430 */
2431 zone_status_t
2432 zone_status_get(zone_t *zone)
2433 {
2434 return (zone->zone_status);
2435 }
2436
2437 static int
2438 zone_set_bootargs(zone_t *zone, const char *zone_bootargs)
2439 {
2440 char *buf = kmem_zalloc(BOOTARGS_MAX, KM_SLEEP);
2441 int err = 0;
2442
2443 ASSERT(zone != global_zone);
2444 if ((err = copyinstr(zone_bootargs, buf, BOOTARGS_MAX, NULL)) != 0)
2445 goto done; /* EFAULT or ENAMETOOLONG */
2446
2447 if (zone->zone_bootargs != NULL)
2448 strfree(zone->zone_bootargs);
2449
2450 zone->zone_bootargs = strdup(buf);
2451
2452 done:
2453 kmem_free(buf, BOOTARGS_MAX);
2454 return (err);
2455 }
2456
2457 static int
2458 zone_set_brand(zone_t *zone, const char *brand)
2459 {
2460 struct brand_attr *attrp;
2461 brand_t *bp;
2462
2463 attrp = kmem_alloc(sizeof (struct brand_attr), KM_SLEEP);
2464 if (copyin(brand, attrp, sizeof (struct brand_attr)) != 0) {
2465 kmem_free(attrp, sizeof (struct brand_attr));
2466 return (EFAULT);
2467 }
2468
2469 bp = brand_register_zone(attrp);
2470 kmem_free(attrp, sizeof (struct brand_attr));
2471 if (bp == NULL)
2472 return (EINVAL);
2473
2474 /*
2475 * This is the only place where a zone can change it's brand.
2476 * We already need to hold zone_status_lock to check the zone
2477 * status, so we'll just use that lock to serialize zone
2478 * branding requests as well.
2479 */
2480 mutex_enter(&zone_status_lock);
2481
2482 /* Re-Branding is not allowed and the zone can't be booted yet */
2483 if ((ZONE_IS_BRANDED(zone)) ||
2484 (zone_status_get(zone) >= ZONE_IS_BOOTING)) {
2485 mutex_exit(&zone_status_lock);
2486 brand_unregister_zone(bp);
2487 return (EINVAL);
2488 }
2489
2490 /* set up the brand specific data */
2491 zone->zone_brand = bp;
2492 ZBROP(zone)->b_init_brand_data(zone);
2493
2494 mutex_exit(&zone_status_lock);
2495 return (0);
2496 }
2497
2498 static int
2499 zone_set_secflags(zone_t *zone, const psecflags_t *zone_secflags)
2500 {
2501 int err = 0;
2502 psecflags_t psf;
2503
2504 ASSERT(zone != global_zone);
2505
2506 if ((err = copyin(zone_secflags, &psf, sizeof (psf))) != 0)
2507 return (err);
2508
2509 if (zone_status_get(zone) > ZONE_IS_READY)
2510 return (EINVAL);
2511
2512 if (!psecflags_validate(&psf))
2513 return (EINVAL);
2514
2515 (void) memcpy(&zone->zone_secflags, &psf, sizeof (psf));
2516
2517 /* Set security flags on the zone's zsched */
2518 (void) memcpy(&zone->zone_zsched->p_secflags, &zone->zone_secflags,
2519 sizeof (zone->zone_zsched->p_secflags));
2520
2521 return (0);
2522 }
2523
2524 static int
2525 zone_set_fs_allowed(zone_t *zone, const char *zone_fs_allowed)
2526 {
2527 char *buf = kmem_zalloc(ZONE_FS_ALLOWED_MAX, KM_SLEEP);
2528 int err = 0;
2529
2530 ASSERT(zone != global_zone);
2531 if ((err = copyinstr(zone_fs_allowed, buf,
2532 ZONE_FS_ALLOWED_MAX, NULL)) != 0)
2533 goto done;
2534
2535 if (zone->zone_fs_allowed != NULL)
2536 strfree(zone->zone_fs_allowed);
2537
2538 zone->zone_fs_allowed = strdup(buf);
2539
2540 done:
2541 kmem_free(buf, ZONE_FS_ALLOWED_MAX);
2542 return (err);
2543 }
2544
2545 static int
2546 zone_set_initname(zone_t *zone, const char *zone_initname)
2547 {
2548 char initname[INITNAME_SZ];
2549 size_t len;
2550 int err = 0;
2551
2552 ASSERT(zone != global_zone);
2553 if ((err = copyinstr(zone_initname, initname, INITNAME_SZ, &len)) != 0)
2554 return (err); /* EFAULT or ENAMETOOLONG */
2555
2556 if (zone->zone_initname != NULL)
2557 strfree(zone->zone_initname);
2558
2559 zone->zone_initname = kmem_alloc(strlen(initname) + 1, KM_SLEEP);
2560 (void) strcpy(zone->zone_initname, initname);
2561 return (0);
2562 }
2563
2564 static int
2565 zone_set_phys_mcap(zone_t *zone, const uint64_t *zone_mcap)
2566 {
2567 uint64_t mcap;
2568 int err = 0;
2569
2570 if ((err = copyin(zone_mcap, &mcap, sizeof (uint64_t))) == 0)
2571 zone->zone_phys_mcap = mcap;
2572
2573 return (err);
2574 }
2575
2576 static int
2577 zone_set_sched_class(zone_t *zone, const char *new_class)
2578 {
2579 char sched_class[PC_CLNMSZ];
2580 id_t classid;
2581 int err;
2582
2583 ASSERT(zone != global_zone);
2584 if ((err = copyinstr(new_class, sched_class, PC_CLNMSZ, NULL)) != 0)
2585 return (err); /* EFAULT or ENAMETOOLONG */
2586
2587 if (getcid(sched_class, &classid) != 0 || CLASS_KERNEL(classid))
2588 return (set_errno(EINVAL));
2589 zone->zone_defaultcid = classid;
2590 ASSERT(zone->zone_defaultcid > 0 &&
2591 zone->zone_defaultcid < loaded_classes);
2592
2593 return (0);
2594 }
2595
2596 /*
2597 * Block indefinitely waiting for (zone_status >= status)
2598 */
2599 void
2600 zone_status_wait(zone_t *zone, zone_status_t status)
2601 {
2602 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2603
2604 mutex_enter(&zone_status_lock);
2605 while (zone->zone_status < status) {
2606 cv_wait(&zone->zone_cv, &zone_status_lock);
2607 }
2608 mutex_exit(&zone_status_lock);
2609 }
2610
2611 /*
2612 * Private CPR-safe version of zone_status_wait().
2613 */
2614 static void
2615 zone_status_wait_cpr(zone_t *zone, zone_status_t status, char *str)
2616 {
2617 callb_cpr_t cprinfo;
2618
2619 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2620
2621 CALLB_CPR_INIT(&cprinfo, &zone_status_lock, callb_generic_cpr,
2622 str);
2623 mutex_enter(&zone_status_lock);
2624 while (zone->zone_status < status) {
2625 CALLB_CPR_SAFE_BEGIN(&cprinfo);
2626 cv_wait(&zone->zone_cv, &zone_status_lock);
2627 CALLB_CPR_SAFE_END(&cprinfo, &zone_status_lock);
2628 }
2629 /*
2630 * zone_status_lock is implicitly released by the following.
2631 */
2632 CALLB_CPR_EXIT(&cprinfo);
2633 }
2634
2635 /*
2636 * Block until zone enters requested state or signal is received. Return (0)
2637 * if signaled, non-zero otherwise.
2638 */
2639 int
2640 zone_status_wait_sig(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 if (!cv_wait_sig(&zone->zone_cv, &zone_status_lock)) {
2647 mutex_exit(&zone_status_lock);
2648 return (0);
2649 }
2650 }
2651 mutex_exit(&zone_status_lock);
2652 return (1);
2653 }
2654
2655 /*
2656 * Block until the zone enters the requested state or the timeout expires,
2657 * whichever happens first. Return (-1) if operation timed out, time remaining
2658 * otherwise.
2659 */
2660 clock_t
2661 zone_status_timedwait(zone_t *zone, clock_t tim, zone_status_t status)
2662 {
2663 clock_t timeleft = 0;
2664
2665 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2666
2667 mutex_enter(&zone_status_lock);
2668 while (zone->zone_status < status && timeleft != -1) {
2669 timeleft = cv_timedwait(&zone->zone_cv, &zone_status_lock, tim);
2670 }
2671 mutex_exit(&zone_status_lock);
2672 return (timeleft);
2673 }
2674
2675 /*
2676 * Block until the zone enters the requested state, the current process is
2677 * signaled, or the timeout expires, whichever happens first. Return (-1) if
2678 * operation timed out, 0 if signaled, time remaining otherwise.
2679 */
2680 clock_t
2681 zone_status_timedwait_sig(zone_t *zone, clock_t tim, zone_status_t status)
2682 {
2683 clock_t timeleft = tim - ddi_get_lbolt();
2684
2685 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2686
2687 mutex_enter(&zone_status_lock);
2688 while (zone->zone_status < status) {
2689 timeleft = cv_timedwait_sig(&zone->zone_cv, &zone_status_lock,
2690 tim);
2691 if (timeleft <= 0)
2692 break;
2693 }
2694 mutex_exit(&zone_status_lock);
2695 return (timeleft);
2696 }
2697
2698 /*
2699 * Zones have two reference counts: one for references from credential
2700 * structures (zone_cred_ref), and one (zone_ref) for everything else.
2701 * This is so we can allow a zone to be rebooted while there are still
2702 * outstanding cred references, since certain drivers cache dblks (which
2703 * implicitly results in cached creds). We wait for zone_ref to drop to
2704 * 0 (actually 1), but not zone_cred_ref. The zone structure itself is
2705 * later freed when the zone_cred_ref drops to 0, though nothing other
2706 * than the zone id and privilege set should be accessed once the zone
2707 * is "dead".
2708 *
2709 * A debugging flag, zone_wait_for_cred, can be set to a non-zero value
2710 * to force halt/reboot to block waiting for the zone_cred_ref to drop
2711 * to 0. This can be useful to flush out other sources of cached creds
2712 * that may be less innocuous than the driver case.
2713 *
2714 * Zones also provide a tracked reference counting mechanism in which zone
2715 * references are represented by "crumbs" (zone_ref structures). Crumbs help
2716 * debuggers determine the sources of leaked zone references. See
2717 * zone_hold_ref() and zone_rele_ref() below for more information.
2718 */
2719
2720 int zone_wait_for_cred = 0;
2721
2722 static void
2723 zone_hold_locked(zone_t *z)
2724 {
2725 ASSERT(MUTEX_HELD(&z->zone_lock));
2726 z->zone_ref++;
2727 ASSERT(z->zone_ref != 0);
2728 }
2729
2730 /*
2731 * Increment the specified zone's reference count. The zone's zone_t structure
2732 * will not be freed as long as the zone's reference count is nonzero.
2733 * Decrement the zone's reference count via zone_rele().
2734 *
2735 * NOTE: This function should only be used to hold zones for short periods of
2736 * time. Use zone_hold_ref() if the zone must be held for a long time.
2737 */
2738 void
2739 zone_hold(zone_t *z)
2740 {
2741 mutex_enter(&z->zone_lock);
2742 zone_hold_locked(z);
2743 mutex_exit(&z->zone_lock);
2744 }
2745
2746 /*
2747 * If the non-cred ref count drops to 1 and either the cred ref count
2748 * is 0 or we aren't waiting for cred references, the zone is ready to
2749 * be destroyed.
2750 */
2751 #define ZONE_IS_UNREF(zone) ((zone)->zone_ref == 1 && \
2752 (!zone_wait_for_cred || (zone)->zone_cred_ref == 0))
2753
2754 /*
2755 * Common zone reference release function invoked by zone_rele() and
2756 * zone_rele_ref(). If subsys is ZONE_REF_NUM_SUBSYS, then the specified
2757 * zone's subsystem-specific reference counters are not affected by the
2758 * release. If ref is not NULL, then the zone_ref_t to which it refers is
2759 * removed from the specified zone's reference list. ref must be non-NULL iff
2760 * subsys is not ZONE_REF_NUM_SUBSYS.
2761 */
2762 static void
2763 zone_rele_common(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
2764 {
2765 boolean_t wakeup;
2766
2767 mutex_enter(&z->zone_lock);
2768 ASSERT(z->zone_ref != 0);
2769 z->zone_ref--;
2770 if (subsys != ZONE_REF_NUM_SUBSYS) {
2771 ASSERT(z->zone_subsys_ref[subsys] != 0);
2772 z->zone_subsys_ref[subsys]--;
2773 list_remove(&z->zone_ref_list, ref);
2774 }
2775 if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
2776 /* no more refs, free the structure */
2777 mutex_exit(&z->zone_lock);
2778 zone_free(z);
2779 return;
2780 }
2781 /* signal zone_destroy so the zone can finish halting */
2782 wakeup = (ZONE_IS_UNREF(z) && zone_status_get(z) >= ZONE_IS_DEAD);
2783 mutex_exit(&z->zone_lock);
2784
2785 if (wakeup) {
2786 /*
2787 * Grabbing zonehash_lock here effectively synchronizes with
2788 * zone_destroy() to avoid missed signals.
2789 */
2790 mutex_enter(&zonehash_lock);
2791 cv_broadcast(&zone_destroy_cv);
2792 mutex_exit(&zonehash_lock);
2793 }
2794 }
2795
2796 /*
2797 * Decrement the specified zone's reference count. The specified zone will
2798 * cease to exist after this function returns if the reference count drops to
2799 * zero. This function should be paired with zone_hold().
2800 */
2801 void
2802 zone_rele(zone_t *z)
2803 {
2804 zone_rele_common(z, NULL, ZONE_REF_NUM_SUBSYS);
2805 }
2806
2807 /*
2808 * Initialize a zone reference structure. This function must be invoked for
2809 * a reference structure before the structure is passed to zone_hold_ref().
2810 */
2811 void
2812 zone_init_ref(zone_ref_t *ref)
2813 {
2814 ref->zref_zone = NULL;
2815 list_link_init(&ref->zref_linkage);
2816 }
2817
2818 /*
2819 * Acquire a reference to zone z. The caller must specify the
2820 * zone_ref_subsys_t constant associated with its subsystem. The specified
2821 * zone_ref_t structure will represent a reference to the specified zone. Use
2822 * zone_rele_ref() to release the reference.
2823 *
2824 * The referenced zone_t structure will not be freed as long as the zone_t's
2825 * zone_status field is not ZONE_IS_DEAD and the zone has outstanding
2826 * references.
2827 *
2828 * NOTE: The zone_ref_t structure must be initialized before it is used.
2829 * See zone_init_ref() above.
2830 */
2831 void
2832 zone_hold_ref(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
2833 {
2834 ASSERT(subsys >= 0 && subsys < ZONE_REF_NUM_SUBSYS);
2835
2836 /*
2837 * Prevent consumers from reusing a reference structure before
2838 * releasing it.
2839 */
2840 VERIFY(ref->zref_zone == NULL);
2841
2842 ref->zref_zone = z;
2843 mutex_enter(&z->zone_lock);
2844 zone_hold_locked(z);
2845 z->zone_subsys_ref[subsys]++;
2846 ASSERT(z->zone_subsys_ref[subsys] != 0);
2847 list_insert_head(&z->zone_ref_list, ref);
2848 mutex_exit(&z->zone_lock);
2849 }
2850
2851 /*
2852 * Release the zone reference represented by the specified zone_ref_t.
2853 * The reference is invalid after it's released; however, the zone_ref_t
2854 * structure can be reused without having to invoke zone_init_ref().
2855 * subsys should be the same value that was passed to zone_hold_ref()
2856 * when the reference was acquired.
2857 */
2858 void
2859 zone_rele_ref(zone_ref_t *ref, zone_ref_subsys_t subsys)
2860 {
2861 zone_rele_common(ref->zref_zone, ref, subsys);
2862
2863 /*
2864 * Set the zone_ref_t's zref_zone field to NULL to generate panics
2865 * when consumers dereference the reference. This helps us catch
2866 * consumers who use released references. Furthermore, this lets
2867 * consumers reuse the zone_ref_t structure without having to
2868 * invoke zone_init_ref().
2869 */
2870 ref->zref_zone = NULL;
2871 }
2872
2873 void
2874 zone_cred_hold(zone_t *z)
2875 {
2876 mutex_enter(&z->zone_lock);
2877 z->zone_cred_ref++;
2878 ASSERT(z->zone_cred_ref != 0);
2879 mutex_exit(&z->zone_lock);
2880 }
2881
2882 void
2883 zone_cred_rele(zone_t *z)
2884 {
2885 boolean_t wakeup;
2886
2887 mutex_enter(&z->zone_lock);
2888 ASSERT(z->zone_cred_ref != 0);
2889 z->zone_cred_ref--;
2890 if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
2891 /* no more refs, free the structure */
2892 mutex_exit(&z->zone_lock);
2893 zone_free(z);
2894 return;
2895 }
2896 /*
2897 * If zone_destroy is waiting for the cred references to drain
2898 * out, and they have, signal it.
2899 */
2900 wakeup = (zone_wait_for_cred && ZONE_IS_UNREF(z) &&
2901 zone_status_get(z) >= ZONE_IS_DEAD);
2902 mutex_exit(&z->zone_lock);
2903
2904 if (wakeup) {
2905 /*
2906 * Grabbing zonehash_lock here effectively synchronizes with
2907 * zone_destroy() to avoid missed signals.
2908 */
2909 mutex_enter(&zonehash_lock);
2910 cv_broadcast(&zone_destroy_cv);
2911 mutex_exit(&zonehash_lock);
2912 }
2913 }
2914
2915 void
2916 zone_task_hold(zone_t *z)
2917 {
2918 mutex_enter(&z->zone_lock);
2919 z->zone_ntasks++;
2920 ASSERT(z->zone_ntasks != 0);
2921 mutex_exit(&z->zone_lock);
2922 }
2923
2924 void
2925 zone_task_rele(zone_t *zone)
2926 {
2927 uint_t refcnt;
2928
2929 mutex_enter(&zone->zone_lock);
2930 ASSERT(zone->zone_ntasks != 0);
2931 refcnt = --zone->zone_ntasks;
2932 if (refcnt > 1) { /* Common case */
2933 mutex_exit(&zone->zone_lock);
2934 return;
2935 }
2936 zone_hold_locked(zone); /* so we can use the zone_t later */
2937 mutex_exit(&zone->zone_lock);
2938 if (refcnt == 1) {
2939 /*
2940 * See if the zone is shutting down.
2941 */
2942 mutex_enter(&zone_status_lock);
2943 if (zone_status_get(zone) != ZONE_IS_SHUTTING_DOWN) {
2944 goto out;
2945 }
2946
2947 /*
2948 * Make sure the ntasks didn't change since we
2949 * dropped zone_lock.
2950 */
2951 mutex_enter(&zone->zone_lock);
2952 if (refcnt != zone->zone_ntasks) {
2953 mutex_exit(&zone->zone_lock);
2954 goto out;
2955 }
2956 mutex_exit(&zone->zone_lock);
2957
2958 /*
2959 * No more user processes in the zone. The zone is empty.
2960 */
2961 zone_status_set(zone, ZONE_IS_EMPTY);
2962 goto out;
2963 }
2964
2965 ASSERT(refcnt == 0);
2966 /*
2967 * zsched has exited; the zone is dead.
2968 */
2969 zone->zone_zsched = NULL; /* paranoia */
2970 mutex_enter(&zone_status_lock);
2971 zone_status_set(zone, ZONE_IS_DEAD);
2972 out:
2973 mutex_exit(&zone_status_lock);
2974 zone_rele(zone);
2975 }
2976
2977 zoneid_t
2978 getzoneid(void)
2979 {
2980 return (curproc->p_zone->zone_id);
2981 }
2982
2983 /*
2984 * Internal versions of zone_find_by_*(). These don't zone_hold() or
2985 * check the validity of a zone's state.
2986 */
2987 static zone_t *
2988 zone_find_all_by_id(zoneid_t zoneid)
2989 {
2990 mod_hash_val_t hv;
2991 zone_t *zone = NULL;
2992
2993 ASSERT(MUTEX_HELD(&zonehash_lock));
2994
2995 if (mod_hash_find(zonehashbyid,
2996 (mod_hash_key_t)(uintptr_t)zoneid, &hv) == 0)
2997 zone = (zone_t *)hv;
2998 return (zone);
2999 }
3000
3001 static zone_t *
3002 zone_find_all_by_label(const ts_label_t *label)
3003 {
3004 mod_hash_val_t hv;
3005 zone_t *zone = NULL;
3006
3007 ASSERT(MUTEX_HELD(&zonehash_lock));
3008
3009 /*
3010 * zonehashbylabel is not maintained for unlabeled systems
3011 */
3012 if (!is_system_labeled())
3013 return (NULL);
3014 if (mod_hash_find(zonehashbylabel, (mod_hash_key_t)label, &hv) == 0)
3015 zone = (zone_t *)hv;
3016 return (zone);
3017 }
3018
3019 static zone_t *
3020 zone_find_all_by_name(char *name)
3021 {
3022 mod_hash_val_t hv;
3023 zone_t *zone = NULL;
3024
3025 ASSERT(MUTEX_HELD(&zonehash_lock));
3026
3027 if (mod_hash_find(zonehashbyname, (mod_hash_key_t)name, &hv) == 0)
3028 zone = (zone_t *)hv;
3029 return (zone);
3030 }
3031
3032 /*
3033 * Public interface for looking up a zone by zoneid. Only returns the zone if
3034 * it is fully initialized, and has not yet begun the zone_destroy() sequence.
3035 * Caller must call zone_rele() once it is done with the zone.
3036 *
3037 * The zone may begin the zone_destroy() sequence immediately after this
3038 * function returns, but may be safely used until zone_rele() is called.
3039 */
3040 zone_t *
3041 zone_find_by_id(zoneid_t zoneid)
3042 {
3043 zone_t *zone;
3044 zone_status_t status;
3045
3046 mutex_enter(&zonehash_lock);
3047 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
3048 mutex_exit(&zonehash_lock);
3049 return (NULL);
3050 }
3051 status = zone_status_get(zone);
3052 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3053 /*
3054 * For all practical purposes the zone doesn't exist.
3055 */
3056 mutex_exit(&zonehash_lock);
3057 return (NULL);
3058 }
3059 zone_hold(zone);
3060 mutex_exit(&zonehash_lock);
3061 return (zone);
3062 }
3063
3064 /*
3065 * Similar to zone_find_by_id, but using zone label as the key.
3066 */
3067 zone_t *
3068 zone_find_by_label(const ts_label_t *label)
3069 {
3070 zone_t *zone;
3071 zone_status_t status;
3072
3073 mutex_enter(&zonehash_lock);
3074 if ((zone = zone_find_all_by_label(label)) == NULL) {
3075 mutex_exit(&zonehash_lock);
3076 return (NULL);
3077 }
3078
3079 status = zone_status_get(zone);
3080 if (status > ZONE_IS_DOWN) {
3081 /*
3082 * For all practical purposes the zone doesn't exist.
3083 */
3084 mutex_exit(&zonehash_lock);
3085 return (NULL);
3086 }
3087 zone_hold(zone);
3088 mutex_exit(&zonehash_lock);
3089 return (zone);
3090 }
3091
3092 /*
3093 * Similar to zone_find_by_id, but using zone name as the key.
3094 */
3095 zone_t *
3096 zone_find_by_name(char *name)
3097 {
3098 zone_t *zone;
3099 zone_status_t status;
3100
3101 mutex_enter(&zonehash_lock);
3102 if ((zone = zone_find_all_by_name(name)) == NULL) {
3103 mutex_exit(&zonehash_lock);
3104 return (NULL);
3105 }
3106 status = zone_status_get(zone);
3107 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3108 /*
3109 * For all practical purposes the zone doesn't exist.
3110 */
3111 mutex_exit(&zonehash_lock);
3112 return (NULL);
3113 }
3114 zone_hold(zone);
3115 mutex_exit(&zonehash_lock);
3116 return (zone);
3117 }
3118
3119 /*
3120 * Similar to zone_find_by_id(), using the path as a key. For instance,
3121 * if there is a zone "foo" rooted at /foo/root, and the path argument
3122 * is "/foo/root/proc", it will return the held zone_t corresponding to
3123 * zone "foo".
3124 *
3125 * zone_find_by_path() always returns a non-NULL value, since at the
3126 * very least every path will be contained in the global zone.
3127 *
3128 * As with the other zone_find_by_*() functions, the caller is
3129 * responsible for zone_rele()ing the return value of this function.
3130 */
3131 zone_t *
3132 zone_find_by_path(const char *path)
3133 {
3134 zone_t *zone;
3135 zone_t *zret = NULL;
3136 zone_status_t status;
3137
3138 if (path == NULL) {
3139 /*
3140 * Call from rootconf().
3141 */
3142 zone_hold(global_zone);
3143 return (global_zone);
3144 }
3145 ASSERT(*path == '/');
3146 mutex_enter(&zonehash_lock);
3147 for (zone = list_head(&zone_active); zone != NULL;
3148 zone = list_next(&zone_active, zone)) {
3149 if (ZONE_PATH_VISIBLE(path, zone))
3150 zret = zone;
3151 }
3152 ASSERT(zret != NULL);
3153 status = zone_status_get(zret);
3154 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3155 /*
3156 * Zone practically doesn't exist.
3157 */
3158 zret = global_zone;
3159 }
3160 zone_hold(zret);
3161 mutex_exit(&zonehash_lock);
3162 return (zret);
3163 }
3164
3165 /*
3166 * Public interface for updating per-zone load averages. Called once per
3167 * second.
3168 *
3169 * Based on loadavg_update(), genloadavg() and calcloadavg() from clock.c.
3170 */
3171 void
3172 zone_loadavg_update()
3173 {
3174 zone_t *zp;
3175 zone_status_t status;
3176 struct loadavg_s *lavg;
3177 hrtime_t zone_total;
3178 int i;
3179 hrtime_t hr_avg;
3180 int nrun;
3181 static int64_t f[3] = { 135, 27, 9 };
3182 int64_t q, r;
3183
3184 mutex_enter(&zonehash_lock);
3185 for (zp = list_head(&zone_active); zp != NULL;
3186 zp = list_next(&zone_active, zp)) {
3187 mutex_enter(&zp->zone_lock);
3188
3189 /* Skip zones that are on the way down or not yet up */
3190 status = zone_status_get(zp);
3191 if (status < ZONE_IS_READY || status >= ZONE_IS_DOWN) {
3192 /* For all practical purposes the zone doesn't exist. */
3193 mutex_exit(&zp->zone_lock);
3194 continue;
3195 }
3196
3197 /*
3198 * Update the 10 second moving average data in zone_loadavg.
3199 */
3200 lavg = &zp->zone_loadavg;
3201
3202 zone_total = zp->zone_utime + zp->zone_stime + zp->zone_wtime;
3203 scalehrtime(&zone_total);
3204
3205 /* The zone_total should always be increasing. */
3206 lavg->lg_loads[lavg->lg_cur] = (zone_total > lavg->lg_total) ?
3207 zone_total - lavg->lg_total : 0;
3208 lavg->lg_cur = (lavg->lg_cur + 1) % S_LOADAVG_SZ;
3209 /* lg_total holds the prev. 1 sec. total */
3210 lavg->lg_total = zone_total;
3211
3212 /*
3213 * To simplify the calculation, we don't calculate the load avg.
3214 * until the zone has been up for at least 10 seconds and our
3215 * moving average is thus full.
3216 */
3217 if ((lavg->lg_len + 1) < S_LOADAVG_SZ) {
3218 lavg->lg_len++;
3219 mutex_exit(&zp->zone_lock);
3220 continue;
3221 }
3222
3223 /* Now calculate the 1min, 5min, 15 min load avg. */
3224 hr_avg = 0;
3225 for (i = 0; i < S_LOADAVG_SZ; i++)
3226 hr_avg += lavg->lg_loads[i];
3227 hr_avg = hr_avg / S_LOADAVG_SZ;
3228 nrun = hr_avg / (NANOSEC / LGRP_LOADAVG_IN_THREAD_MAX);
3229
3230 /* Compute load avg. See comment in calcloadavg() */
3231 for (i = 0; i < 3; i++) {
3232 q = (zp->zone_hp_avenrun[i] >> 16) << 7;
3233 r = (zp->zone_hp_avenrun[i] & 0xffff) << 7;
3234 zp->zone_hp_avenrun[i] +=
3235 ((nrun - q) * f[i] - ((r * f[i]) >> 16)) >> 4;
3236
3237 /* avenrun[] can only hold 31 bits of load avg. */
3238 if (zp->zone_hp_avenrun[i] <
3239 ((uint64_t)1<<(31+16-FSHIFT)))
3240 zp->zone_avenrun[i] = (int32_t)
3241 (zp->zone_hp_avenrun[i] >> (16 - FSHIFT));
3242 else
3243 zp->zone_avenrun[i] = 0x7fffffff;
3244 }
3245
3246 mutex_exit(&zp->zone_lock);
3247 }
3248 mutex_exit(&zonehash_lock);
3249 }
3250
3251 /*
3252 * Get the number of cpus visible to this zone. The system-wide global
3253 * 'ncpus' is returned if pools are disabled, the caller is in the
3254 * global zone, or a NULL zone argument is passed in.
3255 */
3256 int
3257 zone_ncpus_get(zone_t *zone)
3258 {
3259 int myncpus = zone == NULL ? 0 : zone->zone_ncpus;
3260
3261 return (myncpus != 0 ? myncpus : ncpus);
3262 }
3263
3264 /*
3265 * Get the number of online cpus visible to this zone. The system-wide
3266 * global 'ncpus_online' is returned if pools are disabled, the caller
3267 * is in the global zone, or a NULL zone argument is passed in.
3268 */
3269 int
3270 zone_ncpus_online_get(zone_t *zone)
3271 {
3272 int myncpus_online = zone == NULL ? 0 : zone->zone_ncpus_online;
3273
3274 return (myncpus_online != 0 ? myncpus_online : ncpus_online);
3275 }
3276
3277 /*
3278 * Return the pool to which the zone is currently bound.
3279 */
3280 pool_t *
3281 zone_pool_get(zone_t *zone)
3282 {
3283 ASSERT(pool_lock_held());
3284
3285 return (zone->zone_pool);
3286 }
3287
3288 /*
3289 * Set the zone's pool pointer and update the zone's visibility to match
3290 * the resources in the new pool.
3291 */
3292 void
3293 zone_pool_set(zone_t *zone, pool_t *pool)
3294 {
3295 ASSERT(pool_lock_held());
3296 ASSERT(MUTEX_HELD(&cpu_lock));
3297
3298 zone->zone_pool = pool;
3299 zone_pset_set(zone, pool->pool_pset->pset_id);
3300 }
3301
3302 /*
3303 * Return the cached value of the id of the processor set to which the
3304 * zone is currently bound. The value will be ZONE_PS_INVAL if the pools
3305 * facility is disabled.
3306 */
3307 psetid_t
3308 zone_pset_get(zone_t *zone)
3309 {
3310 ASSERT(MUTEX_HELD(&cpu_lock));
3311
3312 return (zone->zone_psetid);
3313 }
3314
3315 /*
3316 * Set the cached value of the id of the processor set to which the zone
3317 * is currently bound. Also update the zone's visibility to match the
3318 * resources in the new processor set.
3319 */
3320 void
3321 zone_pset_set(zone_t *zone, psetid_t newpsetid)
3322 {
3323 psetid_t oldpsetid;
3324
3325 ASSERT(MUTEX_HELD(&cpu_lock));
3326 oldpsetid = zone_pset_get(zone);
3327
3328 if (oldpsetid == newpsetid)
3329 return;
3330 /*
3331 * Global zone sees all.
3332 */
3333 if (zone != global_zone) {
3334 zone->zone_psetid = newpsetid;
3335 if (newpsetid != ZONE_PS_INVAL)
3336 pool_pset_visibility_add(newpsetid, zone);
3337 if (oldpsetid != ZONE_PS_INVAL)
3338 pool_pset_visibility_remove(oldpsetid, zone);
3339 }
3340 /*
3341 * Disabling pools, so we should start using the global values
3342 * for ncpus and ncpus_online.
3343 */
3344 if (newpsetid == ZONE_PS_INVAL) {
3345 zone->zone_ncpus = 0;
3346 zone->zone_ncpus_online = 0;
3347 }
3348 }
3349
3350 /*
3351 * Walk the list of active zones and issue the provided callback for
3352 * each of them.
3353 *
3354 * Caller must not be holding any locks that may be acquired under
3355 * zonehash_lock. See comment at the beginning of the file for a list of
3356 * common locks and their interactions with zones.
3357 */
3358 int
3359 zone_walk(int (*cb)(zone_t *, void *), void *data)
3360 {
3361 zone_t *zone;
3362 int ret = 0;
3363 zone_status_t status;
3364
3365 mutex_enter(&zonehash_lock);
3366 for (zone = list_head(&zone_active); zone != NULL;
3367 zone = list_next(&zone_active, zone)) {
3368 /*
3369 * Skip zones that shouldn't be externally visible.
3370 */
3371 status = zone_status_get(zone);
3372 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN)
3373 continue;
3374 /*
3375 * Bail immediately if any callback invocation returns a
3376 * non-zero value.
3377 */
3378 ret = (*cb)(zone, data);
3379 if (ret != 0)
3380 break;
3381 }
3382 mutex_exit(&zonehash_lock);
3383 return (ret);
3384 }
3385
3386 static int
3387 zone_set_root(zone_t *zone, const char *upath)
3388 {
3389 vnode_t *vp;
3390 int trycount;
3391 int error = 0;
3392 char *path;
3393 struct pathname upn, pn;
3394 size_t pathlen;
3395
3396 if ((error = pn_get((char *)upath, UIO_USERSPACE, &upn)) != 0)
3397 return (error);
3398
3399 pn_alloc(&pn);
3400
3401 /* prevent infinite loop */
3402 trycount = 10;
3403 for (;;) {
3404 if (--trycount <= 0) {
3405 error = ESTALE;
3406 goto out;
3407 }
3408
3409 if ((error = lookuppn(&upn, &pn, FOLLOW, NULLVPP, &vp)) == 0) {
3410 /*
3411 * VOP_ACCESS() may cover 'vp' with a new
3412 * filesystem, if 'vp' is an autoFS vnode.
3413 * Get the new 'vp' if so.
3414 */
3415 if ((error =
3416 VOP_ACCESS(vp, VEXEC, 0, CRED(), NULL)) == 0 &&
3417 (!vn_ismntpt(vp) ||
3418 (error = traverse(&vp)) == 0)) {
3419 pathlen = pn.pn_pathlen + 2;
3420 path = kmem_alloc(pathlen, KM_SLEEP);
3421 (void) strncpy(path, pn.pn_path,
3422 pn.pn_pathlen + 1);
3423 path[pathlen - 2] = '/';
3424 path[pathlen - 1] = '\0';
3425 pn_free(&pn);
3426 pn_free(&upn);
3427
3428 /* Success! */
3429 break;
3430 }
3431 VN_RELE(vp);
3432 }
3433 if (error != ESTALE)
3434 goto out;
3435 }
3436
3437 ASSERT(error == 0);
3438 zone->zone_rootvp = vp; /* we hold a reference to vp */
3439 zone->zone_rootpath = path;
3440 zone->zone_rootpathlen = pathlen;
3441 if (pathlen > 5 && strcmp(path + pathlen - 5, "/lu/") == 0)
3442 zone->zone_flags |= ZF_IS_SCRATCH;
3443 return (0);
3444
3445 out:
3446 pn_free(&pn);
3447 pn_free(&upn);
3448 return (error);
3449 }
3450
3451 #define isalnum(c) (((c) >= '0' && (c) <= '9') || \
3452 ((c) >= 'a' && (c) <= 'z') || \
3453 ((c) >= 'A' && (c) <= 'Z'))
3454
3455 static int
3456 zone_set_name(zone_t *zone, const char *uname)
3457 {
3458 char *kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
3459 size_t len;
3460 int i, err;
3461
3462 if ((err = copyinstr(uname, kname, ZONENAME_MAX, &len)) != 0) {
3463 kmem_free(kname, ZONENAME_MAX);
3464 return (err); /* EFAULT or ENAMETOOLONG */
3465 }
3466
3467 /* must be less than ZONENAME_MAX */
3468 if (len == ZONENAME_MAX && kname[ZONENAME_MAX - 1] != '\0') {
3469 kmem_free(kname, ZONENAME_MAX);
3470 return (EINVAL);
3471 }
3472
3473 /*
3474 * Name must start with an alphanumeric and must contain only
3475 * alphanumerics, '-', '_' and '.'.
3476 */
3477 if (!isalnum(kname[0])) {
3478 kmem_free(kname, ZONENAME_MAX);
3479 return (EINVAL);
3480 }
3481 for (i = 1; i < len - 1; i++) {
3482 if (!isalnum(kname[i]) && kname[i] != '-' && kname[i] != '_' &&
3483 kname[i] != '.') {
3484 kmem_free(kname, ZONENAME_MAX);
3485 return (EINVAL);
3486 }
3487 }
3488
3489 zone->zone_name = kname;
3490 return (0);
3491 }
3492
3493 /*
3494 * Gets the 32-bit hostid of the specified zone as an unsigned int. If 'zonep'
3495 * is NULL or it points to a zone with no hostid emulation, then the machine's
3496 * hostid (i.e., the global zone's hostid) is returned. This function returns
3497 * zero if neither the zone nor the host machine (global zone) have hostids. It
3498 * returns HW_INVALID_HOSTID if the function attempts to return the machine's
3499 * hostid and the machine's hostid is invalid.
3500 */
3501 uint32_t
3502 zone_get_hostid(zone_t *zonep)
3503 {
3504 unsigned long machine_hostid;
3505
3506 if (zonep == NULL || zonep->zone_hostid == HW_INVALID_HOSTID) {
3507 if (ddi_strtoul(hw_serial, NULL, 10, &machine_hostid) != 0)
3508 return (HW_INVALID_HOSTID);
3509 return ((uint32_t)machine_hostid);
3510 }
3511 return (zonep->zone_hostid);
3512 }
3513
3514 /*
3515 * Similar to thread_create(), but makes sure the thread is in the appropriate
3516 * zone's zsched process (curproc->p_zone->zone_zsched) before returning.
3517 */
3518 /*ARGSUSED*/
3519 kthread_t *
3520 zthread_create(
3521 caddr_t stk,
3522 size_t stksize,
3523 void (*proc)(),
3524 void *arg,
3525 size_t len,
3526 pri_t pri)
3527 {
3528 kthread_t *t;
3529 zone_t *zone = curproc->p_zone;
3530 proc_t *pp = zone->zone_zsched;
3531
3532 zone_hold(zone); /* Reference to be dropped when thread exits */
3533
3534 /*
3535 * No-one should be trying to create threads if the zone is shutting
3536 * down and there aren't any kernel threads around. See comment
3537 * in zthread_exit().
3538 */
3539 ASSERT(!(zone->zone_kthreads == NULL &&
3540 zone_status_get(zone) >= ZONE_IS_EMPTY));
3541 /*
3542 * Create a thread, but don't let it run until we've finished setting
3543 * things up.
3544 */
3545 t = thread_create(stk, stksize, proc, arg, len, pp, TS_STOPPED, pri);
3546 ASSERT(t->t_forw == NULL);
3547 mutex_enter(&zone_status_lock);
3548 if (zone->zone_kthreads == NULL) {
3549 t->t_forw = t->t_back = t;
3550 } else {
3551 kthread_t *tx = zone->zone_kthreads;
3552
3553 t->t_forw = tx;
3554 t->t_back = tx->t_back;
3555 tx->t_back->t_forw = t;
3556 tx->t_back = t;
3557 }
3558 zone->zone_kthreads = t;
3559 mutex_exit(&zone_status_lock);
3560
3561 mutex_enter(&pp->p_lock);
3562 t->t_proc_flag |= TP_ZTHREAD;
3563 project_rele(t->t_proj);
3564 t->t_proj = project_hold(pp->p_task->tk_proj);
3565
3566 /*
3567 * Setup complete, let it run.
3568 */
3569 thread_lock(t);
3570 t->t_schedflag |= TS_ALLSTART;
3571 setrun_locked(t);
3572 thread_unlock(t);
3573
3574 mutex_exit(&pp->p_lock);
3575
3576 return (t);
3577 }
3578
3579 /*
3580 * Similar to thread_exit(). Must be called by threads created via
3581 * zthread_exit().
3582 */
3583 void
3584 zthread_exit(void)
3585 {
3586 kthread_t *t = curthread;
3587 proc_t *pp = curproc;
3588 zone_t *zone = pp->p_zone;
3589
3590 mutex_enter(&zone_status_lock);
3591
3592 /*
3593 * Reparent to p0
3594 */
3595 kpreempt_disable();
3596 mutex_enter(&pp->p_lock);
3597 t->t_proc_flag &= ~TP_ZTHREAD;
3598 t->t_procp = &p0;
3599 hat_thread_exit(t);
3600 mutex_exit(&pp->p_lock);
3601 kpreempt_enable();
3602
3603 if (t->t_back == t) {
3604 ASSERT(t->t_forw == t);
3605 /*
3606 * If the zone is empty, once the thread count
3607 * goes to zero no further kernel threads can be
3608 * created. This is because if the creator is a process
3609 * in the zone, then it must have exited before the zone
3610 * state could be set to ZONE_IS_EMPTY.
3611 * Otherwise, if the creator is a kernel thread in the
3612 * zone, the thread count is non-zero.
3613 *
3614 * This really means that non-zone kernel threads should
3615 * not create zone kernel threads.
3616 */
3617 zone->zone_kthreads = NULL;
3618 if (zone_status_get(zone) == ZONE_IS_EMPTY) {
3619 zone_status_set(zone, ZONE_IS_DOWN);
3620 /*
3621 * Remove any CPU caps on this zone.
3622 */
3623 cpucaps_zone_remove(zone);
3624 }
3625 } else {
3626 t->t_forw->t_back = t->t_back;
3627 t->t_back->t_forw = t->t_forw;
3628 if (zone->zone_kthreads == t)
3629 zone->zone_kthreads = t->t_forw;
3630 }
3631 mutex_exit(&zone_status_lock);
3632 zone_rele(zone);
3633 thread_exit();
3634 /* NOTREACHED */
3635 }
3636
3637 static void
3638 zone_chdir(vnode_t *vp, vnode_t **vpp, proc_t *pp)
3639 {
3640 vnode_t *oldvp;
3641
3642 /* we're going to hold a reference here to the directory */
3643 VN_HOLD(vp);
3644
3645 /* update abs cwd/root path see c2/audit.c */
3646 if (AU_AUDITING())
3647 audit_chdirec(vp, vpp);
3648
3649 mutex_enter(&pp->p_lock);
3650 oldvp = *vpp;
3651 *vpp = vp;
3652 mutex_exit(&pp->p_lock);
3653 if (oldvp != NULL)
3654 VN_RELE(oldvp);
3655 }
3656
3657 /*
3658 * Convert an rctl value represented by an nvlist_t into an rctl_val_t.
3659 */
3660 static int
3661 nvlist2rctlval(nvlist_t *nvl, rctl_val_t *rv)
3662 {
3663 nvpair_t *nvp = NULL;
3664 boolean_t priv_set = B_FALSE;
3665 boolean_t limit_set = B_FALSE;
3666 boolean_t action_set = B_FALSE;
3667
3668 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
3669 const char *name;
3670 uint64_t ui64;
3671
3672 name = nvpair_name(nvp);
3673 if (nvpair_type(nvp) != DATA_TYPE_UINT64)
3674 return (EINVAL);
3675 (void) nvpair_value_uint64(nvp, &ui64);
3676 if (strcmp(name, "privilege") == 0) {
3677 /*
3678 * Currently only privileged values are allowed, but
3679 * this may change in the future.
3680 */
3681 if (ui64 != RCPRIV_PRIVILEGED)
3682 return (EINVAL);
3683 rv->rcv_privilege = ui64;
3684 priv_set = B_TRUE;
3685 } else if (strcmp(name, "limit") == 0) {
3686 rv->rcv_value = ui64;
3687 limit_set = B_TRUE;
3688 } else if (strcmp(name, "action") == 0) {
3689 if (ui64 != RCTL_LOCAL_NOACTION &&
3690 ui64 != RCTL_LOCAL_DENY)
3691 return (EINVAL);
3692 rv->rcv_flagaction = ui64;
3693 action_set = B_TRUE;
3694 } else {
3695 return (EINVAL);
3696 }
3697 }
3698
3699 if (!(priv_set && limit_set && action_set))
3700 return (EINVAL);
3701 rv->rcv_action_signal = 0;
3702 rv->rcv_action_recipient = NULL;
3703 rv->rcv_action_recip_pid = -1;
3704 rv->rcv_firing_time = 0;
3705
3706 return (0);
3707 }
3708
3709 /*
3710 * Non-global zone version of start_init.
3711 */
3712 void
3713 zone_start_init(void)
3714 {
3715 proc_t *p = ttoproc(curthread);
3716 zone_t *z = p->p_zone;
3717
3718 ASSERT(!INGLOBALZONE(curproc));
3719
3720 /*
3721 * For all purposes (ZONE_ATTR_INITPID and restart_init),
3722 * storing just the pid of init is sufficient.
3723 */
3724 z->zone_proc_initpid = p->p_pid;
3725
3726 /*
3727 * We maintain zone_boot_err so that we can return the cause of the
3728 * failure back to the caller of the zone_boot syscall.
3729 */
3730 p->p_zone->zone_boot_err = start_init_common();
3731
3732 /*
3733 * We will prevent booting zones from becoming running zones if the
3734 * global zone is shutting down.
3735 */
3736 mutex_enter(&zone_status_lock);
3737 if (z->zone_boot_err != 0 || zone_status_get(global_zone) >=
3738 ZONE_IS_SHUTTING_DOWN) {
3739 /*
3740 * Make sure we are still in the booting state-- we could have
3741 * raced and already be shutting down, or even further along.
3742 */
3743 if (zone_status_get(z) == ZONE_IS_BOOTING) {
3744 zone_status_set(z, ZONE_IS_SHUTTING_DOWN);
3745 }
3746 mutex_exit(&zone_status_lock);
3747 /* It's gone bad, dispose of the process */
3748 if (proc_exit(CLD_EXITED, z->zone_boot_err) != 0) {
3749 mutex_enter(&p->p_lock);
3750 ASSERT(p->p_flag & SEXITLWPS);
3751 lwp_exit();
3752 }
3753 } else {
3754 if (zone_status_get(z) == ZONE_IS_BOOTING)
3755 zone_status_set(z, ZONE_IS_RUNNING);
3756 mutex_exit(&zone_status_lock);
3757 /* cause the process to return to userland. */
3758 lwp_rtt();
3759 }
3760 }
3761
3762 struct zsched_arg {
3763 zone_t *zone;
3764 nvlist_t *nvlist;
3765 };
3766
3767 /*
3768 * Per-zone "sched" workalike. The similarity to "sched" doesn't have
3769 * anything to do with scheduling, but rather with the fact that
3770 * per-zone kernel threads are parented to zsched, just like regular
3771 * kernel threads are parented to sched (p0).
3772 *
3773 * zsched is also responsible for launching init for the zone.
3774 */
3775 static void
3776 zsched(void *arg)
3777 {
3778 struct zsched_arg *za = arg;
3779 proc_t *pp = curproc;
3780 proc_t *initp = proc_init;
3781 zone_t *zone = za->zone;
3782 cred_t *cr, *oldcred;
3783 rctl_set_t *set;
3784 rctl_alloc_gp_t *gp;
3785 contract_t *ct = NULL;
3786 task_t *tk, *oldtk;
3787 rctl_entity_p_t e;
3788 kproject_t *pj;
3789
3790 nvlist_t *nvl = za->nvlist;
3791 nvpair_t *nvp = NULL;
3792
3793 bcopy("zsched", PTOU(pp)->u_psargs, sizeof ("zsched"));
3794 bcopy("zsched", PTOU(pp)->u_comm, sizeof ("zsched"));
3795 PTOU(pp)->u_argc = 0;
3796 PTOU(pp)->u_argv = NULL;
3797 PTOU(pp)->u_envp = NULL;
3798 closeall(P_FINFO(pp));
3799
3800 /*
3801 * We are this zone's "zsched" process. As the zone isn't generally
3802 * visible yet we don't need to grab any locks before initializing its
3803 * zone_proc pointer.
3804 */
3805 zone_hold(zone); /* this hold is released by zone_destroy() */
3806 zone->zone_zsched = pp;
3807 mutex_enter(&pp->p_lock);
3808 pp->p_zone = zone;
3809 mutex_exit(&pp->p_lock);
3810
3811 /*
3812 * Disassociate process from its 'parent'; parent ourselves to init
3813 * (pid 1) and change other values as needed.
3814 */
3815 sess_create();
3816
3817 mutex_enter(&pidlock);
3818 proc_detach(pp);
3819 pp->p_ppid = 1;
3820 pp->p_flag |= SZONETOP;
3821 pp->p_ancpid = 1;
3822 pp->p_parent = initp;
3823 pp->p_psibling = NULL;
3824 if (initp->p_child)
3825 initp->p_child->p_psibling = pp;
3826 pp->p_sibling = initp->p_child;
3827 initp->p_child = pp;
3828
3829 /* Decrement what newproc() incremented. */
3830 upcount_dec(crgetruid(CRED()), GLOBAL_ZONEID);
3831 /*
3832 * Our credentials are about to become kcred-like, so we don't care
3833 * about the caller's ruid.
3834 */
3835 upcount_inc(crgetruid(kcred), zone->zone_id);
3836 mutex_exit(&pidlock);
3837
3838 /*
3839 * getting out of global zone, so decrement lwp and process counts
3840 */
3841 pj = pp->p_task->tk_proj;
3842 mutex_enter(&global_zone->zone_nlwps_lock);
3843 pj->kpj_nlwps -= pp->p_lwpcnt;
3844 global_zone->zone_nlwps -= pp->p_lwpcnt;
3845 pj->kpj_nprocs--;
3846 global_zone->zone_nprocs--;
3847 mutex_exit(&global_zone->zone_nlwps_lock);
3848
3849 /*
3850 * Decrement locked memory counts on old zone and project.
3851 */
3852 mutex_enter(&global_zone->zone_mem_lock);
3853 global_zone->zone_locked_mem -= pp->p_locked_mem;
3854 pj->kpj_data.kpd_locked_mem -= pp->p_locked_mem;
3855 mutex_exit(&global_zone->zone_mem_lock);
3856
3857 /*
3858 * Create and join a new task in project '0' of this zone.
3859 *
3860 * We don't need to call holdlwps() since we know we're the only lwp in
3861 * this process.
3862 *
3863 * task_join() returns with p_lock held.
3864 */
3865 tk = task_create(0, zone);
3866 mutex_enter(&cpu_lock);
3867 oldtk = task_join(tk, 0);
3868
3869 pj = pp->p_task->tk_proj;
3870
3871 mutex_enter(&zone->zone_mem_lock);
3872 zone->zone_locked_mem += pp->p_locked_mem;
3873 pj->kpj_data.kpd_locked_mem += pp->p_locked_mem;
3874 mutex_exit(&zone->zone_mem_lock);
3875
3876 /*
3877 * add lwp and process counts to zsched's zone, and increment
3878 * project's task and process count due to the task created in
3879 * the above task_create.
3880 */
3881 mutex_enter(&zone->zone_nlwps_lock);
3882 pj->kpj_nlwps += pp->p_lwpcnt;
3883 pj->kpj_ntasks += 1;
3884 zone->zone_nlwps += pp->p_lwpcnt;
3885 pj->kpj_nprocs++;
3886 zone->zone_nprocs++;
3887 mutex_exit(&zone->zone_nlwps_lock);
3888
3889 mutex_exit(&curproc->p_lock);
3890 mutex_exit(&cpu_lock);
3891 task_rele(oldtk);
3892
3893 /*
3894 * The process was created by a process in the global zone, hence the
3895 * credentials are wrong. We might as well have kcred-ish credentials.
3896 */
3897 cr = zone->zone_kcred;
3898 crhold(cr);
3899 mutex_enter(&pp->p_crlock);
3900 oldcred = pp->p_cred;
3901 pp->p_cred = cr;
3902 mutex_exit(&pp->p_crlock);
3903 crfree(oldcred);
3904
3905 /*
3906 * Hold credentials again (for thread)
3907 */
3908 crhold(cr);
3909
3910 /*
3911 * p_lwpcnt can't change since this is a kernel process.
3912 */
3913 crset(pp, cr);
3914
3915 /*
3916 * Chroot
3917 */
3918 zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_cdir, pp);
3919 zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_rdir, pp);
3920
3921 /*
3922 * Initialize zone's rctl set.
3923 */
3924 set = rctl_set_create();
3925 gp = rctl_set_init_prealloc(RCENTITY_ZONE);
3926 mutex_enter(&pp->p_lock);
3927 e.rcep_p.zone = zone;
3928 e.rcep_t = RCENTITY_ZONE;
3929 zone->zone_rctls = rctl_set_init(RCENTITY_ZONE, pp, &e, set, gp);
3930 mutex_exit(&pp->p_lock);
3931 rctl_prealloc_destroy(gp);
3932
3933 /*
3934 * Apply the rctls passed in to zone_create(). This is basically a list
3935 * assignment: all of the old values are removed and the new ones
3936 * inserted. That is, if an empty list is passed in, all values are
3937 * removed.
3938 */
3939 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
3940 rctl_dict_entry_t *rde;
3941 rctl_hndl_t hndl;
3942 char *name;
3943 nvlist_t **nvlarray;
3944 uint_t i, nelem;
3945 int error; /* For ASSERT()s */
3946
3947 name = nvpair_name(nvp);
3948 hndl = rctl_hndl_lookup(name);
3949 ASSERT(hndl != -1);
3950 rde = rctl_dict_lookup_hndl(hndl);
3951 ASSERT(rde != NULL);
3952
3953 for (; /* ever */; ) {
3954 rctl_val_t oval;
3955
3956 mutex_enter(&pp->p_lock);
3957 error = rctl_local_get(hndl, NULL, &oval, pp);
3958 mutex_exit(&pp->p_lock);
3959 ASSERT(error == 0); /* Can't fail for RCTL_FIRST */
3960 ASSERT(oval.rcv_privilege != RCPRIV_BASIC);
3961 if (oval.rcv_privilege == RCPRIV_SYSTEM)
3962 break;
3963 mutex_enter(&pp->p_lock);
3964 error = rctl_local_delete(hndl, &oval, pp);
3965 mutex_exit(&pp->p_lock);
3966 ASSERT(error == 0);
3967 }
3968 error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
3969 ASSERT(error == 0);
3970 for (i = 0; i < nelem; i++) {
3971 rctl_val_t *nvalp;
3972
3973 nvalp = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
3974 error = nvlist2rctlval(nvlarray[i], nvalp);
3975 ASSERT(error == 0);
3976 /*
3977 * rctl_local_insert can fail if the value being
3978 * inserted is a duplicate; this is OK.
3979 */
3980 mutex_enter(&pp->p_lock);
3981 if (rctl_local_insert(hndl, nvalp, pp) != 0)
3982 kmem_cache_free(rctl_val_cache, nvalp);
3983 mutex_exit(&pp->p_lock);
3984 }
3985 }
3986
3987 /*
3988 * Tell the world that we're done setting up.
3989 *
3990 * At this point we want to set the zone status to ZONE_IS_INITIALIZED
3991 * and atomically set the zone's processor set visibility. Once
3992 * we drop pool_lock() this zone will automatically get updated
3993 * to reflect any future changes to the pools configuration.
3994 *
3995 * Note that after we drop the locks below (zonehash_lock in
3996 * particular) other operations such as a zone_getattr call can
3997 * now proceed and observe the zone. That is the reason for doing a
3998 * state transition to the INITIALIZED state.
3999 */
4000 pool_lock();
4001 mutex_enter(&cpu_lock);
4002 mutex_enter(&zonehash_lock);
4003 zone_uniqid(zone);
4004 zone_zsd_configure(zone);
4005 if (pool_state == POOL_ENABLED)
4006 zone_pset_set(zone, pool_default->pool_pset->pset_id);
4007 mutex_enter(&zone_status_lock);
4008 ASSERT(zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
4009 zone_status_set(zone, ZONE_IS_INITIALIZED);
4010 mutex_exit(&zone_status_lock);
4011 mutex_exit(&zonehash_lock);
4012 mutex_exit(&cpu_lock);
4013 pool_unlock();
4014
4015 /* Now call the create callback for this key */
4016 zsd_apply_all_keys(zsd_apply_create, zone);
4017
4018 /* The callbacks are complete. Mark ZONE_IS_READY */
4019 mutex_enter(&zone_status_lock);
4020 ASSERT(zone_status_get(zone) == ZONE_IS_INITIALIZED);
4021 zone_status_set(zone, ZONE_IS_READY);
4022 mutex_exit(&zone_status_lock);
4023
4024 /*
4025 * Once we see the zone transition to the ZONE_IS_BOOTING state,
4026 * we launch init, and set the state to running.
4027 */
4028 zone_status_wait_cpr(zone, ZONE_IS_BOOTING, "zsched");
4029
4030 if (zone_status_get(zone) == ZONE_IS_BOOTING) {
4031 id_t cid;
4032
4033 /*
4034 * Ok, this is a little complicated. We need to grab the
4035 * zone's pool's scheduling class ID; note that by now, we
4036 * are already bound to a pool if we need to be (zoneadmd
4037 * will have done that to us while we're in the READY
4038 * state). *But* the scheduling class for the zone's 'init'
4039 * must be explicitly passed to newproc, which doesn't
4040 * respect pool bindings.
4041 *
4042 * We hold the pool_lock across the call to newproc() to
4043 * close the obvious race: the pool's scheduling class
4044 * could change before we manage to create the LWP with
4045 * classid 'cid'.
4046 */
4047 pool_lock();
4048 if (zone->zone_defaultcid > 0)
4049 cid = zone->zone_defaultcid;
4050 else
4051 cid = pool_get_class(zone->zone_pool);
4052 if (cid == -1)
4053 cid = defaultcid;
4054
4055 /*
4056 * If this fails, zone_boot will ultimately fail. The
4057 * state of the zone will be set to SHUTTING_DOWN-- userland
4058 * will have to tear down the zone, and fail, or try again.
4059 */
4060 if ((zone->zone_boot_err = newproc(zone_start_init, NULL, cid,
4061 minclsyspri - 1, &ct, 0)) != 0) {
4062 mutex_enter(&zone_status_lock);
4063 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
4064 mutex_exit(&zone_status_lock);
4065 } else {
4066 zone->zone_boot_time = gethrestime_sec();
4067 }
4068
4069 pool_unlock();
4070 }
4071
4072 /*
4073 * Wait for zone_destroy() to be called. This is what we spend
4074 * most of our life doing.
4075 */
4076 zone_status_wait_cpr(zone, ZONE_IS_DYING, "zsched");
4077
4078 if (ct)
4079 /*
4080 * At this point the process contract should be empty.
4081 * (Though if it isn't, it's not the end of the world.)
4082 */
4083 VERIFY(contract_abandon(ct, curproc, B_TRUE) == 0);
4084
4085 /*
4086 * Allow kcred to be freed when all referring processes
4087 * (including this one) go away. We can't just do this in
4088 * zone_free because we need to wait for the zone_cred_ref to
4089 * drop to 0 before calling zone_free, and the existence of
4090 * zone_kcred will prevent that. Thus, we call crfree here to
4091 * balance the crdup in zone_create. The crhold calls earlier
4092 * in zsched will be dropped when the thread and process exit.
4093 */
4094 crfree(zone->zone_kcred);
4095 zone->zone_kcred = NULL;
4096
4097 exit(CLD_EXITED, 0);
4098 }
4099
4100 /*
4101 * Helper function to determine if there are any submounts of the
4102 * provided path. Used to make sure the zone doesn't "inherit" any
4103 * mounts from before it is created.
4104 */
4105 static uint_t
4106 zone_mount_count(const char *rootpath)
4107 {
4108 vfs_t *vfsp;
4109 uint_t count = 0;
4110 size_t rootpathlen = strlen(rootpath);
4111
4112 /*
4113 * Holding zonehash_lock prevents race conditions with
4114 * vfs_list_add()/vfs_list_remove() since we serialize with
4115 * zone_find_by_path().
4116 */
4117 ASSERT(MUTEX_HELD(&zonehash_lock));
4118 /*
4119 * The rootpath must end with a '/'
4120 */
4121 ASSERT(rootpath[rootpathlen - 1] == '/');
4122
4123 /*
4124 * This intentionally does not count the rootpath itself if that
4125 * happens to be a mount point.
4126 */
4127 vfs_list_read_lock();
4128 vfsp = rootvfs;
4129 do {
4130 if (strncmp(rootpath, refstr_value(vfsp->vfs_mntpt),
4131 rootpathlen) == 0)
4132 count++;
4133 vfsp = vfsp->vfs_next;
4134 } while (vfsp != rootvfs);
4135 vfs_list_unlock();
4136 return (count);
4137 }
4138
4139 /*
4140 * Helper function to make sure that a zone created on 'rootpath'
4141 * wouldn't end up containing other zones' rootpaths.
4142 */
4143 static boolean_t
4144 zone_is_nested(const char *rootpath)
4145 {
4146 zone_t *zone;
4147 size_t rootpathlen = strlen(rootpath);
4148 size_t len;
4149
4150 ASSERT(MUTEX_HELD(&zonehash_lock));
4151
4152 /*
4153 * zone_set_root() appended '/' and '\0' at the end of rootpath
4154 */
4155 if ((rootpathlen <= 3) && (rootpath[0] == '/') &&
4156 (rootpath[1] == '/') && (rootpath[2] == '\0'))
4157 return (B_TRUE);
4158
4159 for (zone = list_head(&zone_active); zone != NULL;
4160 zone = list_next(&zone_active, zone)) {
4161 if (zone == global_zone)
4162 continue;
4163 len = strlen(zone->zone_rootpath);
4164 if (strncmp(rootpath, zone->zone_rootpath,
4165 MIN(rootpathlen, len)) == 0)
4166 return (B_TRUE);
4167 }
4168 return (B_FALSE);
4169 }
4170
4171 static int
4172 zone_set_privset(zone_t *zone, const priv_set_t *zone_privs,
4173 size_t zone_privssz)
4174 {
4175 priv_set_t *privs;
4176
4177 if (zone_privssz < sizeof (priv_set_t))
4178 return (ENOMEM);
4179
4180 privs = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
4181
4182 if (copyin(zone_privs, privs, sizeof (priv_set_t))) {
4183 kmem_free(privs, sizeof (priv_set_t));
4184 return (EFAULT);
4185 }
4186
4187 zone->zone_privset = privs;
4188 return (0);
4189 }
4190
4191 /*
4192 * We make creative use of nvlists to pass in rctls from userland. The list is
4193 * a list of the following structures:
4194 *
4195 * (name = rctl_name, value = nvpair_list_array)
4196 *
4197 * Where each element of the nvpair_list_array is of the form:
4198 *
4199 * [(name = "privilege", value = RCPRIV_PRIVILEGED),
4200 * (name = "limit", value = uint64_t),
4201 * (name = "action", value = (RCTL_LOCAL_NOACTION || RCTL_LOCAL_DENY))]
4202 */
4203 static int
4204 parse_rctls(caddr_t ubuf, size_t buflen, nvlist_t **nvlp)
4205 {
4206 nvpair_t *nvp = NULL;
4207 nvlist_t *nvl = NULL;
4208 char *kbuf;
4209 int error;
4210 rctl_val_t rv;
4211
4212 *nvlp = NULL;
4213
4214 if (buflen == 0)
4215 return (0);
4216
4217 if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL)
4218 return (ENOMEM);
4219 if (copyin(ubuf, kbuf, buflen)) {
4220 error = EFAULT;
4221 goto out;
4222 }
4223 if (nvlist_unpack(kbuf, buflen, &nvl, KM_SLEEP) != 0) {
4224 /*
4225 * nvl may have been allocated/free'd, but the value set to
4226 * non-NULL, so we reset it here.
4227 */
4228 nvl = NULL;
4229 error = EINVAL;
4230 goto out;
4231 }
4232 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
4233 rctl_dict_entry_t *rde;
4234 rctl_hndl_t hndl;
4235 nvlist_t **nvlarray;
4236 uint_t i, nelem;
4237 char *name;
4238
4239 error = EINVAL;
4240 name = nvpair_name(nvp);
4241 if (strncmp(nvpair_name(nvp), "zone.", sizeof ("zone.") - 1)
4242 != 0 || nvpair_type(nvp) != DATA_TYPE_NVLIST_ARRAY) {
4243 goto out;
4244 }
4245 if ((hndl = rctl_hndl_lookup(name)) == -1) {
4246 goto out;
4247 }
4248 rde = rctl_dict_lookup_hndl(hndl);
4249 error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
4250 ASSERT(error == 0);
4251 for (i = 0; i < nelem; i++) {
4252 if (error = nvlist2rctlval(nvlarray[i], &rv))
4253 goto out;
4254 }
4255 if (rctl_invalid_value(rde, &rv)) {
4256 error = EINVAL;
4257 goto out;
4258 }
4259 }
4260 error = 0;
4261 *nvlp = nvl;
4262 out:
4263 kmem_free(kbuf, buflen);
4264 if (error && nvl != NULL)
4265 nvlist_free(nvl);
4266 return (error);
4267 }
4268
4269 int
4270 zone_create_error(int er_error, int er_ext, int *er_out)
4271 {
4272 if (er_out != NULL) {
4273 if (copyout(&er_ext, er_out, sizeof (int))) {
4274 return (set_errno(EFAULT));
4275 }
4276 }
4277 return (set_errno(er_error));
4278 }
4279
4280 static int
4281 zone_set_label(zone_t *zone, const bslabel_t *lab, uint32_t doi)
4282 {
4283 ts_label_t *tsl;
4284 bslabel_t blab;
4285
4286 /* Get label from user */
4287 if (copyin(lab, &blab, sizeof (blab)) != 0)
4288 return (EFAULT);
4289 tsl = labelalloc(&blab, doi, KM_NOSLEEP);
4290 if (tsl == NULL)
4291 return (ENOMEM);
4292
4293 zone->zone_slabel = tsl;
4294 return (0);
4295 }
4296
4297 /*
4298 * Parses a comma-separated list of ZFS datasets into a per-zone dictionary.
4299 */
4300 static int
4301 parse_zfs(zone_t *zone, caddr_t ubuf, size_t buflen)
4302 {
4303 char *kbuf;
4304 char *dataset, *next;
4305 zone_dataset_t *zd;
4306 size_t len;
4307
4308 if (ubuf == NULL || buflen == 0)
4309 return (0);
4310
4311 if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL)
4312 return (ENOMEM);
4313
4314 if (copyin(ubuf, kbuf, buflen) != 0) {
4315 kmem_free(kbuf, buflen);
4316 return (EFAULT);
4317 }
4318
4319 dataset = next = kbuf;
4320 for (;;) {
4321 zd = kmem_alloc(sizeof (zone_dataset_t), KM_SLEEP);
4322
4323 next = strchr(dataset, ',');
4324
4325 if (next == NULL)
4326 len = strlen(dataset);
4327 else
4328 len = next - dataset;
4329
4330 zd->zd_dataset = kmem_alloc(len + 1, KM_SLEEP);
4331 bcopy(dataset, zd->zd_dataset, len);
4332 zd->zd_dataset[len] = '\0';
4333
4334 list_insert_head(&zone->zone_datasets, zd);
4335
4336 if (next == NULL)
4337 break;
4338
4339 dataset = next + 1;
4340 }
4341
4342 kmem_free(kbuf, buflen);
4343 return (0);
4344 }
4345
4346 /*
4347 * System call to create/initialize a new zone named 'zone_name', rooted
4348 * at 'zone_root', with a zone-wide privilege limit set of 'zone_privs',
4349 * and initialized with the zone-wide rctls described in 'rctlbuf', and
4350 * with labeling set by 'match', 'doi', and 'label'.
4351 *
4352 * If extended error is non-null, we may use it to return more detailed
4353 * error information.
4354 */
4355 static zoneid_t
4356 zone_create(const char *zone_name, const char *zone_root,
4357 const priv_set_t *zone_privs, size_t zone_privssz,
4358 caddr_t rctlbuf, size_t rctlbufsz,
4359 caddr_t zfsbuf, size_t zfsbufsz, int *extended_error,
4360 int match, uint32_t doi, const bslabel_t *label,
4361 int flags)
4362 {
4363 struct zsched_arg zarg;
4364 nvlist_t *rctls = NULL;
4365 proc_t *pp = curproc;
4366 zone_t *zone, *ztmp;
4367 zoneid_t zoneid;
4368 int error;
4369 int error2 = 0;
4370 char *str;
4371 cred_t *zkcr;
4372 boolean_t insert_label_hash;
4373
4374 if (secpolicy_zone_config(CRED()) != 0)
4375 return (set_errno(EPERM));
4376
4377 /* can't boot zone from within chroot environment */
4378 if (PTOU(pp)->u_rdir != NULL && PTOU(pp)->u_rdir != rootdir)
4379 return (zone_create_error(ENOTSUP, ZE_CHROOTED,
4380 extended_error));
4381
4382 zone = kmem_zalloc(sizeof (zone_t), KM_SLEEP);
4383 zoneid = zone->zone_id = id_alloc(zoneid_space);
4384 zone->zone_status = ZONE_IS_UNINITIALIZED;
4385 zone->zone_pool = pool_default;
4386 zone->zone_pool_mod = gethrtime();
4387 zone->zone_psetid = ZONE_PS_INVAL;
4388 zone->zone_ncpus = 0;
4389 zone->zone_ncpus_online = 0;
4390 zone->zone_restart_init = B_TRUE;
4391 zone->zone_brand = &native_brand;
4392 zone->zone_initname = NULL;
4393 mutex_init(&zone->zone_lock, NULL, MUTEX_DEFAULT, NULL);
4394 mutex_init(&zone->zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
4395 mutex_init(&zone->zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
4396 cv_init(&zone->zone_cv, NULL, CV_DEFAULT, NULL);
4397 list_create(&zone->zone_ref_list, sizeof (zone_ref_t),
4398 offsetof(zone_ref_t, zref_linkage));
4399 list_create(&zone->zone_zsd, sizeof (struct zsd_entry),
4400 offsetof(struct zsd_entry, zsd_linkage));
4401 list_create(&zone->zone_datasets, sizeof (zone_dataset_t),
4402 offsetof(zone_dataset_t, zd_linkage));
4403 list_create(&zone->zone_dl_list, sizeof (zone_dl_t),
4404 offsetof(zone_dl_t, zdl_linkage));
4405 rw_init(&zone->zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
4406 rw_init(&zone->zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL);
4407
4408 if (flags & ZCF_NET_EXCL) {
4409 zone->zone_flags |= ZF_NET_EXCL;
4410 }
4411
4412 if ((error = zone_set_name(zone, zone_name)) != 0) {
4413 zone_free(zone);
4414 return (zone_create_error(error, 0, extended_error));
4415 }
4416
4417 if ((error = zone_set_root(zone, zone_root)) != 0) {
4418 zone_free(zone);
4419 return (zone_create_error(error, 0, extended_error));
4420 }
4421 if ((error = zone_set_privset(zone, zone_privs, zone_privssz)) != 0) {
4422 zone_free(zone);
4423 return (zone_create_error(error, 0, extended_error));
4424 }
4425
4426 /* initialize node name to be the same as zone name */
4427 zone->zone_nodename = kmem_alloc(_SYS_NMLN, KM_SLEEP);
4428 (void) strncpy(zone->zone_nodename, zone->zone_name, _SYS_NMLN);
4429 zone->zone_nodename[_SYS_NMLN - 1] = '\0';
4430
4431 zone->zone_domain = kmem_alloc(_SYS_NMLN, KM_SLEEP);
4432 zone->zone_domain[0] = '\0';
4433 zone->zone_hostid = HW_INVALID_HOSTID;
4434 zone->zone_shares = 1;
4435 zone->zone_shmmax = 0;
4436 zone->zone_ipc.ipcq_shmmni = 0;
4437 zone->zone_ipc.ipcq_semmni = 0;
4438 zone->zone_ipc.ipcq_msgmni = 0;
4439 zone->zone_bootargs = NULL;
4440 zone->zone_fs_allowed = NULL;
4441
4442 secflags_zero(&zone0.zone_secflags.psf_lower);
4443 secflags_zero(&zone0.zone_secflags.psf_effective);
4444 secflags_zero(&zone0.zone_secflags.psf_inherit);
4445 secflags_fullset(&zone0.zone_secflags.psf_upper);
4446
4447 zone->zone_initname =
4448 kmem_alloc(strlen(zone_default_initname) + 1, KM_SLEEP);
4449 (void) strcpy(zone->zone_initname, zone_default_initname);
4450 zone->zone_nlwps = 0;
4451 zone->zone_nlwps_ctl = INT_MAX;
4452 zone->zone_nprocs = 0;
4453 zone->zone_nprocs_ctl = INT_MAX;
4454 zone->zone_locked_mem = 0;
4455 zone->zone_locked_mem_ctl = UINT64_MAX;
4456 zone->zone_max_swap = 0;
4457 zone->zone_max_swap_ctl = UINT64_MAX;
4458 zone->zone_max_lofi = 0;
4459 zone->zone_max_lofi_ctl = UINT64_MAX;
4460 zone0.zone_lockedmem_kstat = NULL;
4461 zone0.zone_swapresv_kstat = NULL;
4462
4463 /*
4464 * Zsched initializes the rctls.
4465 */
4466 zone->zone_rctls = NULL;
4467
4468 if ((error = parse_rctls(rctlbuf, rctlbufsz, &rctls)) != 0) {
4469 zone_free(zone);
4470 return (zone_create_error(error, 0, extended_error));
4471 }
4472
4473 if ((error = parse_zfs(zone, zfsbuf, zfsbufsz)) != 0) {
4474 zone_free(zone);
4475 return (set_errno(error));
4476 }
4477
4478 /*
4479 * Read in the trusted system parameters:
4480 * match flag and sensitivity label.
4481 */
4482 zone->zone_match = match;
4483 if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) {
4484 /* Fail if requested to set doi to anything but system's doi */
4485 if (doi != 0 && doi != default_doi) {
4486 zone_free(zone);
4487 return (set_errno(EINVAL));
4488 }
4489 /* Always apply system's doi to the zone */
4490 error = zone_set_label(zone, label, default_doi);
4491 if (error != 0) {
4492 zone_free(zone);
4493 return (set_errno(error));
4494 }
4495 insert_label_hash = B_TRUE;
4496 } else {
4497 /* all zones get an admin_low label if system is not labeled */
4498 zone->zone_slabel = l_admin_low;
4499 label_hold(l_admin_low);
4500 insert_label_hash = B_FALSE;
4501 }
4502
4503 /*
4504 * Stop all lwps since that's what normally happens as part of fork().
4505 * This needs to happen before we grab any locks to avoid deadlock
4506 * (another lwp in the process could be waiting for the held lock).
4507 */
4508 if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK)) {
4509 zone_free(zone);
4510 nvlist_free(rctls);
4511 return (zone_create_error(error, 0, extended_error));
4512 }
4513
4514 if (block_mounts(zone) == 0) {
4515 mutex_enter(&pp->p_lock);
4516 if (curthread != pp->p_agenttp)
4517 continuelwps(pp);
4518 mutex_exit(&pp->p_lock);
4519 zone_free(zone);
4520 nvlist_free(rctls);
4521 return (zone_create_error(error, 0, extended_error));
4522 }
4523
4524 /*
4525 * Set up credential for kernel access. After this, any errors
4526 * should go through the dance in errout rather than calling
4527 * zone_free directly.
4528 */
4529 zone->zone_kcred = crdup(kcred);
4530 crsetzone(zone->zone_kcred, zone);
4531 priv_intersect(zone->zone_privset, &CR_PPRIV(zone->zone_kcred));
4532 priv_intersect(zone->zone_privset, &CR_EPRIV(zone->zone_kcred));
4533 priv_intersect(zone->zone_privset, &CR_IPRIV(zone->zone_kcred));
4534 priv_intersect(zone->zone_privset, &CR_LPRIV(zone->zone_kcred));
4535
4536 mutex_enter(&zonehash_lock);
4537 /*
4538 * Make sure zone doesn't already exist.
4539 *
4540 * If the system and zone are labeled,
4541 * make sure no other zone exists that has the same label.
4542 */
4543 if ((ztmp = zone_find_all_by_name(zone->zone_name)) != NULL ||
4544 (insert_label_hash &&
4545 (ztmp = zone_find_all_by_label(zone->zone_slabel)) != NULL)) {
4546 zone_status_t status;
4547
4548 status = zone_status_get(ztmp);
4549 if (status == ZONE_IS_READY || status == ZONE_IS_RUNNING)
4550 error = EEXIST;
4551 else
4552 error = EBUSY;
4553
4554 if (insert_label_hash)
4555 error2 = ZE_LABELINUSE;
4556
4557 goto errout;
4558 }
4559
4560 /*
4561 * Don't allow zone creations which would cause one zone's rootpath to
4562 * be accessible from that of another (non-global) zone.
4563 */
4564 if (zone_is_nested(zone->zone_rootpath)) {
4565 error = EBUSY;
4566 goto errout;
4567 }
4568
4569 ASSERT(zonecount != 0); /* check for leaks */
4570 if (zonecount + 1 > maxzones) {
4571 error = ENOMEM;
4572 goto errout;
4573 }
4574
4575 if (zone_mount_count(zone->zone_rootpath) != 0) {
4576 error = EBUSY;
4577 error2 = ZE_AREMOUNTS;
4578 goto errout;
4579 }
4580
4581 /*
4582 * Zone is still incomplete, but we need to drop all locks while
4583 * zsched() initializes this zone's kernel process. We
4584 * optimistically add the zone to the hashtable and associated
4585 * lists so a parallel zone_create() doesn't try to create the
4586 * same zone.
4587 */
4588 zonecount++;
4589 (void) mod_hash_insert(zonehashbyid,
4590 (mod_hash_key_t)(uintptr_t)zone->zone_id,
4591 (mod_hash_val_t)(uintptr_t)zone);
4592 str = kmem_alloc(strlen(zone->zone_name) + 1, KM_SLEEP);
4593 (void) strcpy(str, zone->zone_name);
4594 (void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)str,
4595 (mod_hash_val_t)(uintptr_t)zone);
4596 if (insert_label_hash) {
4597 (void) mod_hash_insert(zonehashbylabel,
4598 (mod_hash_key_t)zone->zone_slabel, (mod_hash_val_t)zone);
4599 zone->zone_flags |= ZF_HASHED_LABEL;
4600 }
4601
4602 /*
4603 * Insert into active list. At this point there are no 'hold's
4604 * on the zone, but everyone else knows not to use it, so we can
4605 * continue to use it. zsched() will do a zone_hold() if the
4606 * newproc() is successful.
4607 */
4608 list_insert_tail(&zone_active, zone);
4609 mutex_exit(&zonehash_lock);
4610
4611 zarg.zone = zone;
4612 zarg.nvlist = rctls;
4613 /*
4614 * The process, task, and project rctls are probably wrong;
4615 * we need an interface to get the default values of all rctls,
4616 * and initialize zsched appropriately. I'm not sure that that
4617 * makes much of a difference, though.
4618 */
4619 error = newproc(zsched, (void *)&zarg, syscid, minclsyspri, NULL, 0);
4620 if (error != 0) {
4621 /*
4622 * We need to undo all globally visible state.
4623 */
4624 mutex_enter(&zonehash_lock);
4625 list_remove(&zone_active, zone);
4626 if (zone->zone_flags & ZF_HASHED_LABEL) {
4627 ASSERT(zone->zone_slabel != NULL);
4628 (void) mod_hash_destroy(zonehashbylabel,
4629 (mod_hash_key_t)zone->zone_slabel);
4630 }
4631 (void) mod_hash_destroy(zonehashbyname,
4632 (mod_hash_key_t)(uintptr_t)zone->zone_name);
4633 (void) mod_hash_destroy(zonehashbyid,
4634 (mod_hash_key_t)(uintptr_t)zone->zone_id);
4635 ASSERT(zonecount > 1);
4636 zonecount--;
4637 goto errout;
4638 }
4639
4640 /*
4641 * Zone creation can't fail from now on.
4642 */
4643
4644 /*
4645 * Create zone kstats
4646 */
4647 zone_kstat_create(zone);
4648
4649 /*
4650 * Let the other lwps continue.
4651 */
4652 mutex_enter(&pp->p_lock);
4653 if (curthread != pp->p_agenttp)
4654 continuelwps(pp);
4655 mutex_exit(&pp->p_lock);
4656
4657 /*
4658 * Wait for zsched to finish initializing the zone.
4659 */
4660 zone_status_wait(zone, ZONE_IS_READY);
4661 /*
4662 * The zone is fully visible, so we can let mounts progress.
4663 */
4664 resume_mounts(zone);
4665 nvlist_free(rctls);
4666
4667 return (zoneid);
4668
4669 errout:
4670 mutex_exit(&zonehash_lock);
4671 /*
4672 * Let the other lwps continue.
4673 */
4674 mutex_enter(&pp->p_lock);
4675 if (curthread != pp->p_agenttp)
4676 continuelwps(pp);
4677 mutex_exit(&pp->p_lock);
4678
4679 resume_mounts(zone);
4680 nvlist_free(rctls);
4681 /*
4682 * There is currently one reference to the zone, a cred_ref from
4683 * zone_kcred. To free the zone, we call crfree, which will call
4684 * zone_cred_rele, which will call zone_free.
4685 */
4686 ASSERT(zone->zone_cred_ref == 1);
4687 ASSERT(zone->zone_kcred->cr_ref == 1);
4688 ASSERT(zone->zone_ref == 0);
4689 zkcr = zone->zone_kcred;
4690 zone->zone_kcred = NULL;
4691 crfree(zkcr); /* triggers call to zone_free */
4692 return (zone_create_error(error, error2, extended_error));
4693 }
4694
4695 /*
4696 * Cause the zone to boot. This is pretty simple, since we let zoneadmd do
4697 * the heavy lifting. initname is the path to the program to launch
4698 * at the "top" of the zone; if this is NULL, we use the system default,
4699 * which is stored at zone_default_initname.
4700 */
4701 static int
4702 zone_boot(zoneid_t zoneid)
4703 {
4704 int err;
4705 zone_t *zone;
4706
4707 if (secpolicy_zone_config(CRED()) != 0)
4708 return (set_errno(EPERM));
4709 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4710 return (set_errno(EINVAL));
4711
4712 mutex_enter(&zonehash_lock);
4713 /*
4714 * Look for zone under hash lock to prevent races with calls to
4715 * zone_shutdown, zone_destroy, etc.
4716 */
4717 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4718 mutex_exit(&zonehash_lock);
4719 return (set_errno(EINVAL));
4720 }
4721
4722 mutex_enter(&zone_status_lock);
4723 if (zone_status_get(zone) != ZONE_IS_READY) {
4724 mutex_exit(&zone_status_lock);
4725 mutex_exit(&zonehash_lock);
4726 return (set_errno(EINVAL));
4727 }
4728 zone_status_set(zone, ZONE_IS_BOOTING);
4729 mutex_exit(&zone_status_lock);
4730
4731 zone_hold(zone); /* so we can use the zone_t later */
4732 mutex_exit(&zonehash_lock);
4733
4734 if (zone_status_wait_sig(zone, ZONE_IS_RUNNING) == 0) {
4735 zone_rele(zone);
4736 return (set_errno(EINTR));
4737 }
4738
4739 /*
4740 * Boot (starting init) might have failed, in which case the zone
4741 * will go to the SHUTTING_DOWN state; an appropriate errno will
4742 * be placed in zone->zone_boot_err, and so we return that.
4743 */
4744 err = zone->zone_boot_err;
4745 zone_rele(zone);
4746 return (err ? set_errno(err) : 0);
4747 }
4748
4749 /*
4750 * Kills all user processes in the zone, waiting for them all to exit
4751 * before returning.
4752 */
4753 static int
4754 zone_empty(zone_t *zone)
4755 {
4756 int waitstatus;
4757
4758 /*
4759 * We need to drop zonehash_lock before killing all
4760 * processes, otherwise we'll deadlock with zone_find_*
4761 * which can be called from the exit path.
4762 */
4763 ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
4764 while ((waitstatus = zone_status_timedwait_sig(zone,
4765 ddi_get_lbolt() + hz, ZONE_IS_EMPTY)) == -1) {
4766 killall(zone->zone_id);
4767 }
4768 /*
4769 * return EINTR if we were signaled
4770 */
4771 if (waitstatus == 0)
4772 return (EINTR);
4773 return (0);
4774 }
4775
4776 /*
4777 * This function implements the policy for zone visibility.
4778 *
4779 * In standard Solaris, a non-global zone can only see itself.
4780 *
4781 * In Trusted Extensions, a labeled zone can lookup any zone whose label
4782 * it dominates. For this test, the label of the global zone is treated as
4783 * admin_high so it is special-cased instead of being checked for dominance.
4784 *
4785 * Returns true if zone attributes are viewable, false otherwise.
4786 */
4787 static boolean_t
4788 zone_list_access(zone_t *zone)
4789 {
4790
4791 if (curproc->p_zone == global_zone ||
4792 curproc->p_zone == zone) {
4793 return (B_TRUE);
4794 } else if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) {
4795 bslabel_t *curproc_label;
4796 bslabel_t *zone_label;
4797
4798 curproc_label = label2bslabel(curproc->p_zone->zone_slabel);
4799 zone_label = label2bslabel(zone->zone_slabel);
4800
4801 if (zone->zone_id != GLOBAL_ZONEID &&
4802 bldominates(curproc_label, zone_label)) {
4803 return (B_TRUE);
4804 } else {
4805 return (B_FALSE);
4806 }
4807 } else {
4808 return (B_FALSE);
4809 }
4810 }
4811
4812 /*
4813 * Systemcall to start the zone's halt sequence. By the time this
4814 * function successfully returns, all user processes and kernel threads
4815 * executing in it will have exited, ZSD shutdown callbacks executed,
4816 * and the zone status set to ZONE_IS_DOWN.
4817 *
4818 * It is possible that the call will interrupt itself if the caller is the
4819 * parent of any process running in the zone, and doesn't have SIGCHLD blocked.
4820 */
4821 static int
4822 zone_shutdown(zoneid_t zoneid)
4823 {
4824 int error;
4825 zone_t *zone;
4826 zone_status_t status;
4827
4828 if (secpolicy_zone_config(CRED()) != 0)
4829 return (set_errno(EPERM));
4830 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4831 return (set_errno(EINVAL));
4832
4833 mutex_enter(&zonehash_lock);
4834 /*
4835 * Look for zone under hash lock to prevent races with other
4836 * calls to zone_shutdown and zone_destroy.
4837 */
4838 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4839 mutex_exit(&zonehash_lock);
4840 return (set_errno(EINVAL));
4841 }
4842
4843 /*
4844 * We have to drop zonehash_lock before calling block_mounts.
4845 * Hold the zone so we can continue to use the zone_t.
4846 */
4847 zone_hold(zone);
4848 mutex_exit(&zonehash_lock);
4849
4850 /*
4851 * Block mounts so that VFS_MOUNT() can get an accurate view of
4852 * the zone's status with regards to ZONE_IS_SHUTTING down.
4853 *
4854 * e.g. NFS can fail the mount if it determines that the zone
4855 * has already begun the shutdown sequence.
4856 *
4857 */
4858 if (block_mounts(zone) == 0) {
4859 zone_rele(zone);
4860 return (set_errno(EINTR));
4861 }
4862
4863 mutex_enter(&zonehash_lock);
4864 mutex_enter(&zone_status_lock);
4865 status = zone_status_get(zone);
4866 /*
4867 * Fail if the zone isn't fully initialized yet.
4868 */
4869 if (status < ZONE_IS_READY) {
4870 mutex_exit(&zone_status_lock);
4871 mutex_exit(&zonehash_lock);
4872 resume_mounts(zone);
4873 zone_rele(zone);
4874 return (set_errno(EINVAL));
4875 }
4876 /*
4877 * If conditions required for zone_shutdown() to return have been met,
4878 * return success.
4879 */
4880 if (status >= ZONE_IS_DOWN) {
4881 mutex_exit(&zone_status_lock);
4882 mutex_exit(&zonehash_lock);
4883 resume_mounts(zone);
4884 zone_rele(zone);
4885 return (0);
4886 }
4887 /*
4888 * If zone_shutdown() hasn't been called before, go through the motions.
4889 * If it has, there's nothing to do but wait for the kernel threads to
4890 * drain.
4891 */
4892 if (status < ZONE_IS_EMPTY) {
4893 uint_t ntasks;
4894
4895 mutex_enter(&zone->zone_lock);
4896 if ((ntasks = zone->zone_ntasks) != 1) {
4897 /*
4898 * There's still stuff running.
4899 */
4900 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
4901 }
4902 mutex_exit(&zone->zone_lock);
4903 if (ntasks == 1) {
4904 /*
4905 * The only way to create another task is through
4906 * zone_enter(), which will block until we drop
4907 * zonehash_lock. The zone is empty.
4908 */
4909 if (zone->zone_kthreads == NULL) {
4910 /*
4911 * Skip ahead to ZONE_IS_DOWN
4912 */
4913 zone_status_set(zone, ZONE_IS_DOWN);
4914 } else {
4915 zone_status_set(zone, ZONE_IS_EMPTY);
4916 }
4917 }
4918 }
4919 mutex_exit(&zone_status_lock);
4920 mutex_exit(&zonehash_lock);
4921 resume_mounts(zone);
4922
4923 if (error = zone_empty(zone)) {
4924 zone_rele(zone);
4925 return (set_errno(error));
4926 }
4927 /*
4928 * After the zone status goes to ZONE_IS_DOWN this zone will no
4929 * longer be notified of changes to the pools configuration, so
4930 * in order to not end up with a stale pool pointer, we point
4931 * ourselves at the default pool and remove all resource
4932 * visibility. This is especially important as the zone_t may
4933 * languish on the deathrow for a very long time waiting for
4934 * cred's to drain out.
4935 *
4936 * This rebinding of the zone can happen multiple times
4937 * (presumably due to interrupted or parallel systemcalls)
4938 * without any adverse effects.
4939 */
4940 if (pool_lock_intr() != 0) {
4941 zone_rele(zone);
4942 return (set_errno(EINTR));
4943 }
4944 if (pool_state == POOL_ENABLED) {
4945 mutex_enter(&cpu_lock);
4946 zone_pool_set(zone, pool_default);
4947 /*
4948 * The zone no longer needs to be able to see any cpus.
4949 */
4950 zone_pset_set(zone, ZONE_PS_INVAL);
4951 mutex_exit(&cpu_lock);
4952 }
4953 pool_unlock();
4954
4955 /*
4956 * ZSD shutdown callbacks can be executed multiple times, hence
4957 * it is safe to not be holding any locks across this call.
4958 */
4959 zone_zsd_callbacks(zone, ZSD_SHUTDOWN);
4960
4961 mutex_enter(&zone_status_lock);
4962 if (zone->zone_kthreads == NULL && zone_status_get(zone) < ZONE_IS_DOWN)
4963 zone_status_set(zone, ZONE_IS_DOWN);
4964 mutex_exit(&zone_status_lock);
4965
4966 /*
4967 * Wait for kernel threads to drain.
4968 */
4969 if (!zone_status_wait_sig(zone, ZONE_IS_DOWN)) {
4970 zone_rele(zone);
4971 return (set_errno(EINTR));
4972 }
4973
4974 /*
4975 * Zone can be become down/destroyable even if the above wait
4976 * returns EINTR, so any code added here may never execute.
4977 * (i.e. don't add code here)
4978 */
4979
4980 zone_rele(zone);
4981 return (0);
4982 }
4983
4984 /*
4985 * Log the specified zone's reference counts. The caller should not be
4986 * holding the zone's zone_lock.
4987 */
4988 static void
4989 zone_log_refcounts(zone_t *zone)
4990 {
4991 char *buffer;
4992 char *buffer_position;
4993 uint32_t buffer_size;
4994 uint32_t index;
4995 uint_t ref;
4996 uint_t cred_ref;
4997
4998 /*
4999 * Construct a string representing the subsystem-specific reference
5000 * counts. The counts are printed in ascending order by index into the
5001 * zone_t::zone_subsys_ref array. The list will be surrounded by
5002 * square brackets [] and will only contain nonzero reference counts.
5003 *
5004 * The buffer will hold two square bracket characters plus ten digits,
5005 * one colon, one space, one comma, and some characters for a
5006 * subsystem name per subsystem-specific reference count. (Unsigned 32-
5007 * bit integers have at most ten decimal digits.) The last
5008 * reference count's comma is replaced by the closing square
5009 * bracket and a NULL character to terminate the string.
5010 *
5011 * NOTE: We have to grab the zone's zone_lock to create a consistent
5012 * snapshot of the zone's reference counters.
5013 *
5014 * First, figure out how much space the string buffer will need.
5015 * The buffer's size is stored in buffer_size.
5016 */
5017 buffer_size = 2; /* for the square brackets */
5018 mutex_enter(&zone->zone_lock);
5019 zone->zone_flags |= ZF_REFCOUNTS_LOGGED;
5020 ref = zone->zone_ref;
5021 cred_ref = zone->zone_cred_ref;
5022 for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index)
5023 if (zone->zone_subsys_ref[index] != 0)
5024 buffer_size += strlen(zone_ref_subsys_names[index]) +
5025 13;
5026 if (buffer_size == 2) {
5027 /*
5028 * No subsystems had nonzero reference counts. Don't bother
5029 * with allocating a buffer; just log the general-purpose and
5030 * credential reference counts.
5031 */
5032 mutex_exit(&zone->zone_lock);
5033 (void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE,
5034 "Zone '%s' (ID: %d) is shutting down, but %u zone "
5035 "references and %u credential references are still extant",
5036 zone->zone_name, zone->zone_id, ref, cred_ref);
5037 return;
5038 }
5039
5040 /*
5041 * buffer_size contains the exact number of characters that the
5042 * buffer will need. Allocate the buffer and fill it with nonzero
5043 * subsystem-specific reference counts. Surround the results with
5044 * square brackets afterwards.
5045 */
5046 buffer = kmem_alloc(buffer_size, KM_SLEEP);
5047 buffer_position = &buffer[1];
5048 for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index) {
5049 /*
5050 * NOTE: The DDI's version of sprintf() returns a pointer to
5051 * the modified buffer rather than the number of bytes written
5052 * (as in snprintf(3C)). This is unfortunate and annoying.
5053 * Therefore, we'll use snprintf() with INT_MAX to get the
5054 * number of bytes written. Using INT_MAX is safe because
5055 * the buffer is perfectly sized for the data: we'll never
5056 * overrun the buffer.
5057 */
5058 if (zone->zone_subsys_ref[index] != 0)
5059 buffer_position += snprintf(buffer_position, INT_MAX,
5060 "%s: %u,", zone_ref_subsys_names[index],
5061 zone->zone_subsys_ref[index]);
5062 }
5063 mutex_exit(&zone->zone_lock);
5064 buffer[0] = '[';
5065 ASSERT((uintptr_t)(buffer_position - buffer) < buffer_size);
5066 ASSERT(buffer_position[0] == '\0' && buffer_position[-1] == ',');
5067 buffer_position[-1] = ']';
5068
5069 /*
5070 * Log the reference counts and free the message buffer.
5071 */
5072 (void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE,
5073 "Zone '%s' (ID: %d) is shutting down, but %u zone references and "
5074 "%u credential references are still extant %s", zone->zone_name,
5075 zone->zone_id, ref, cred_ref, buffer);
5076 kmem_free(buffer, buffer_size);
5077 }
5078
5079 /*
5080 * Systemcall entry point to finalize the zone halt process. The caller
5081 * must have already successfully called zone_shutdown().
5082 *
5083 * Upon successful completion, the zone will have been fully destroyed:
5084 * zsched will have exited, destructor callbacks executed, and the zone
5085 * removed from the list of active zones.
5086 */
5087 static int
5088 zone_destroy(zoneid_t zoneid)
5089 {
5090 uint64_t uniqid;
5091 zone_t *zone;
5092 zone_status_t status;
5093 clock_t wait_time;
5094 boolean_t log_refcounts;
5095
5096 if (secpolicy_zone_config(CRED()) != 0)
5097 return (set_errno(EPERM));
5098 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
5099 return (set_errno(EINVAL));
5100
5101 mutex_enter(&zonehash_lock);
5102 /*
5103 * Look for zone under hash lock to prevent races with other
5104 * calls to zone_destroy.
5105 */
5106 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5107 mutex_exit(&zonehash_lock);
5108 return (set_errno(EINVAL));
5109 }
5110
5111 if (zone_mount_count(zone->zone_rootpath) != 0) {
5112 mutex_exit(&zonehash_lock);
5113 return (set_errno(EBUSY));
5114 }
5115 mutex_enter(&zone_status_lock);
5116 status = zone_status_get(zone);
5117 if (status < ZONE_IS_DOWN) {
5118 mutex_exit(&zone_status_lock);
5119 mutex_exit(&zonehash_lock);
5120 return (set_errno(EBUSY));
5121 } else if (status == ZONE_IS_DOWN) {
5122 zone_status_set(zone, ZONE_IS_DYING); /* Tell zsched to exit */
5123 }
5124 mutex_exit(&zone_status_lock);
5125 zone_hold(zone);
5126 mutex_exit(&zonehash_lock);
5127
5128 /*
5129 * wait for zsched to exit
5130 */
5131 zone_status_wait(zone, ZONE_IS_DEAD);
5132 zone_zsd_callbacks(zone, ZSD_DESTROY);
5133 zone->zone_netstack = NULL;
5134 uniqid = zone->zone_uniqid;
5135 zone_rele(zone);
5136 zone = NULL; /* potentially free'd */
5137
5138 log_refcounts = B_FALSE;
5139 wait_time = SEC_TO_TICK(ZONE_DESTROY_TIMEOUT_SECS);
5140 mutex_enter(&zonehash_lock);
5141 for (; /* ever */; ) {
5142 boolean_t unref;
5143 boolean_t refs_have_been_logged;
5144
5145 if ((zone = zone_find_all_by_id(zoneid)) == NULL ||
5146 zone->zone_uniqid != uniqid) {
5147 /*
5148 * The zone has gone away. Necessary conditions
5149 * are met, so we return success.
5150 */
5151 mutex_exit(&zonehash_lock);
5152 return (0);
5153 }
5154 mutex_enter(&zone->zone_lock);
5155 unref = ZONE_IS_UNREF(zone);
5156 refs_have_been_logged = (zone->zone_flags &
5157 ZF_REFCOUNTS_LOGGED);
5158 mutex_exit(&zone->zone_lock);
5159 if (unref) {
5160 /*
5161 * There is only one reference to the zone -- that
5162 * added when the zone was added to the hashtables --
5163 * and things will remain this way until we drop
5164 * zonehash_lock... we can go ahead and cleanup the
5165 * zone.
5166 */
5167 break;
5168 }
5169
5170 /*
5171 * Wait for zone_rele_common() or zone_cred_rele() to signal
5172 * zone_destroy_cv. zone_destroy_cv is signaled only when
5173 * some zone's general-purpose reference count reaches one.
5174 * If ZONE_DESTROY_TIMEOUT_SECS seconds elapse while waiting
5175 * on zone_destroy_cv, then log the zone's reference counts and
5176 * continue to wait for zone_rele() and zone_cred_rele().
5177 */
5178 if (!refs_have_been_logged) {
5179 if (!log_refcounts) {
5180 /*
5181 * This thread hasn't timed out waiting on
5182 * zone_destroy_cv yet. Wait wait_time clock
5183 * ticks (initially ZONE_DESTROY_TIMEOUT_SECS
5184 * seconds) for the zone's references to clear.
5185 */
5186 ASSERT(wait_time > 0);
5187 wait_time = cv_reltimedwait_sig(
5188 &zone_destroy_cv, &zonehash_lock, wait_time,
5189 TR_SEC);
5190 if (wait_time > 0) {
5191 /*
5192 * A thread in zone_rele() or
5193 * zone_cred_rele() signaled
5194 * zone_destroy_cv before this thread's
5195 * wait timed out. The zone might have
5196 * only one reference left; find out!
5197 */
5198 continue;
5199 } else if (wait_time == 0) {
5200 /* The thread's process was signaled. */
5201 mutex_exit(&zonehash_lock);
5202 return (set_errno(EINTR));
5203 }
5204
5205 /*
5206 * The thread timed out while waiting on
5207 * zone_destroy_cv. Even though the thread
5208 * timed out, it has to check whether another
5209 * thread woke up from zone_destroy_cv and
5210 * destroyed the zone.
5211 *
5212 * If the zone still exists and has more than
5213 * one unreleased general-purpose reference,
5214 * then log the zone's reference counts.
5215 */
5216 log_refcounts = B_TRUE;
5217 continue;
5218 }
5219
5220 /*
5221 * The thread already timed out on zone_destroy_cv while
5222 * waiting for subsystems to release the zone's last
5223 * general-purpose references. Log the zone's reference
5224 * counts and wait indefinitely on zone_destroy_cv.
5225 */
5226 zone_log_refcounts(zone);
5227 }
5228 if (cv_wait_sig(&zone_destroy_cv, &zonehash_lock) == 0) {
5229 /* The thread's process was signaled. */
5230 mutex_exit(&zonehash_lock);
5231 return (set_errno(EINTR));
5232 }
5233 }
5234
5235 /*
5236 * Remove CPU cap for this zone now since we're not going to
5237 * fail below this point.
5238 */
5239 cpucaps_zone_remove(zone);
5240
5241 /* Get rid of the zone's kstats */
5242 zone_kstat_delete(zone);
5243
5244 /* remove the pfexecd doors */
5245 if (zone->zone_pfexecd != NULL) {
5246 klpd_freelist(&zone->zone_pfexecd);
5247 zone->zone_pfexecd = NULL;
5248 }
5249
5250 /* free brand specific data */
5251 if (ZONE_IS_BRANDED(zone))
5252 ZBROP(zone)->b_free_brand_data(zone);
5253
5254 /* Say goodbye to brand framework. */
5255 brand_unregister_zone(zone->zone_brand);
5256
5257 /*
5258 * It is now safe to let the zone be recreated; remove it from the
5259 * lists. The memory will not be freed until the last cred
5260 * reference goes away.
5261 */
5262 ASSERT(zonecount > 1); /* must be > 1; can't destroy global zone */
5263 zonecount--;
5264 /* remove from active list and hash tables */
5265 list_remove(&zone_active, zone);
5266 (void) mod_hash_destroy(zonehashbyname,
5267 (mod_hash_key_t)zone->zone_name);
5268 (void) mod_hash_destroy(zonehashbyid,
5269 (mod_hash_key_t)(uintptr_t)zone->zone_id);
5270 if (zone->zone_flags & ZF_HASHED_LABEL)
5271 (void) mod_hash_destroy(zonehashbylabel,
5272 (mod_hash_key_t)zone->zone_slabel);
5273 mutex_exit(&zonehash_lock);
5274
5275 /*
5276 * Release the root vnode; we're not using it anymore. Nor should any
5277 * other thread that might access it exist.
5278 */
5279 if (zone->zone_rootvp != NULL) {
5280 VN_RELE(zone->zone_rootvp);
5281 zone->zone_rootvp = NULL;
5282 }
5283
5284 /* add to deathrow list */
5285 mutex_enter(&zone_deathrow_lock);
5286 list_insert_tail(&zone_deathrow, zone);
5287 mutex_exit(&zone_deathrow_lock);
5288
5289 /*
5290 * Drop last reference (which was added by zsched()), this will
5291 * free the zone unless there are outstanding cred references.
5292 */
5293 zone_rele(zone);
5294 return (0);
5295 }
5296
5297 /*
5298 * Systemcall entry point for zone_getattr(2).
5299 */
5300 static ssize_t
5301 zone_getattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
5302 {
5303 size_t size;
5304 int error = 0, err;
5305 zone_t *zone;
5306 char *zonepath;
5307 char *outstr;
5308 zone_status_t zone_status;
5309 pid_t initpid;
5310 boolean_t global = (curzone == global_zone);
5311 boolean_t inzone = (curzone->zone_id == zoneid);
5312 ushort_t flags;
5313 zone_net_data_t *zbuf;
5314
5315 mutex_enter(&zonehash_lock);
5316 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5317 mutex_exit(&zonehash_lock);
5318 return (set_errno(EINVAL));
5319 }
5320 zone_status = zone_status_get(zone);
5321 if (zone_status < ZONE_IS_INITIALIZED) {
5322 mutex_exit(&zonehash_lock);
5323 return (set_errno(EINVAL));
5324 }
5325 zone_hold(zone);
5326 mutex_exit(&zonehash_lock);
5327
5328 /*
5329 * If not in the global zone, don't show information about other zones,
5330 * unless the system is labeled and the local zone's label dominates
5331 * the other zone.
5332 */
5333 if (!zone_list_access(zone)) {
5334 zone_rele(zone);
5335 return (set_errno(EINVAL));
5336 }
5337
5338 switch (attr) {
5339 case ZONE_ATTR_ROOT:
5340 if (global) {
5341 /*
5342 * Copy the path to trim the trailing "/" (except for
5343 * the global zone).
5344 */
5345 if (zone != global_zone)
5346 size = zone->zone_rootpathlen - 1;
5347 else
5348 size = zone->zone_rootpathlen;
5349 zonepath = kmem_alloc(size, KM_SLEEP);
5350 bcopy(zone->zone_rootpath, zonepath, size);
5351 zonepath[size - 1] = '\0';
5352 } else {
5353 if (inzone || !is_system_labeled()) {
5354 /*
5355 * Caller is not in the global zone.
5356 * if the query is on the current zone
5357 * or the system is not labeled,
5358 * just return faked-up path for current zone.
5359 */
5360 zonepath = "/";
5361 size = 2;
5362 } else {
5363 /*
5364 * Return related path for current zone.
5365 */
5366 int prefix_len = strlen(zone_prefix);
5367 int zname_len = strlen(zone->zone_name);
5368
5369 size = prefix_len + zname_len + 1;
5370 zonepath = kmem_alloc(size, KM_SLEEP);
5371 bcopy(zone_prefix, zonepath, prefix_len);
5372 bcopy(zone->zone_name, zonepath +
5373 prefix_len, zname_len);
5374 zonepath[size - 1] = '\0';
5375 }
5376 }
5377 if (bufsize > size)
5378 bufsize = size;
5379 if (buf != NULL) {
5380 err = copyoutstr(zonepath, buf, bufsize, NULL);
5381 if (err != 0 && err != ENAMETOOLONG)
5382 error = EFAULT;
5383 }
5384 if (global || (is_system_labeled() && !inzone))
5385 kmem_free(zonepath, size);
5386 break;
5387
5388 case ZONE_ATTR_NAME:
5389 size = strlen(zone->zone_name) + 1;
5390 if (bufsize > size)
5391 bufsize = size;
5392 if (buf != NULL) {
5393 err = copyoutstr(zone->zone_name, buf, bufsize, NULL);
5394 if (err != 0 && err != ENAMETOOLONG)
5395 error = EFAULT;
5396 }
5397 break;
5398
5399 case ZONE_ATTR_STATUS:
5400 /*
5401 * Since we're not holding zonehash_lock, the zone status
5402 * may be anything; leave it up to userland to sort it out.
5403 */
5404 size = sizeof (zone_status);
5405 if (bufsize > size)
5406 bufsize = size;
5407 zone_status = zone_status_get(zone);
5408 if (buf != NULL &&
5409 copyout(&zone_status, buf, bufsize) != 0)
5410 error = EFAULT;
5411 break;
5412 case ZONE_ATTR_FLAGS:
5413 size = sizeof (zone->zone_flags);
5414 if (bufsize > size)
5415 bufsize = size;
5416 flags = zone->zone_flags;
5417 if (buf != NULL &&
5418 copyout(&flags, buf, bufsize) != 0)
5419 error = EFAULT;
5420 break;
5421 case ZONE_ATTR_PRIVSET:
5422 size = sizeof (priv_set_t);
5423 if (bufsize > size)
5424 bufsize = size;
5425 if (buf != NULL &&
5426 copyout(zone->zone_privset, buf, bufsize) != 0)
5427 error = EFAULT;
5428 break;
5429 case ZONE_ATTR_UNIQID:
5430 size = sizeof (zone->zone_uniqid);
5431 if (bufsize > size)
5432 bufsize = size;
5433 if (buf != NULL &&
5434 copyout(&zone->zone_uniqid, buf, bufsize) != 0)
5435 error = EFAULT;
5436 break;
5437 case ZONE_ATTR_POOLID:
5438 {
5439 pool_t *pool;
5440 poolid_t poolid;
5441
5442 if (pool_lock_intr() != 0) {
5443 error = EINTR;
5444 break;
5445 }
5446 pool = zone_pool_get(zone);
5447 poolid = pool->pool_id;
5448 pool_unlock();
5449 size = sizeof (poolid);
5450 if (bufsize > size)
5451 bufsize = size;
5452 if (buf != NULL && copyout(&poolid, buf, size) != 0)
5453 error = EFAULT;
5454 }
5455 break;
5456 case ZONE_ATTR_SLBL:
5457 size = sizeof (bslabel_t);
5458 if (bufsize > size)
5459 bufsize = size;
5460 if (zone->zone_slabel == NULL)
5461 error = EINVAL;
5462 else if (buf != NULL &&
5463 copyout(label2bslabel(zone->zone_slabel), buf,
5464 bufsize) != 0)
5465 error = EFAULT;
5466 break;
5467 case ZONE_ATTR_INITPID:
5468 size = sizeof (initpid);
5469 if (bufsize > size)
5470 bufsize = size;
5471 initpid = zone->zone_proc_initpid;
5472 if (initpid == -1) {
5473 error = ESRCH;
5474 break;
5475 }
5476 if (buf != NULL &&
5477 copyout(&initpid, buf, bufsize) != 0)
5478 error = EFAULT;
5479 break;
5480 case ZONE_ATTR_BRAND:
5481 size = strlen(zone->zone_brand->b_name) + 1;
5482
5483 if (bufsize > size)
5484 bufsize = size;
5485 if (buf != NULL) {
5486 err = copyoutstr(zone->zone_brand->b_name, buf,
5487 bufsize, NULL);
5488 if (err != 0 && err != ENAMETOOLONG)
5489 error = EFAULT;
5490 }
5491 break;
5492 case ZONE_ATTR_INITNAME:
5493 size = strlen(zone->zone_initname) + 1;
5494 if (bufsize > size)
5495 bufsize = size;
5496 if (buf != NULL) {
5497 err = copyoutstr(zone->zone_initname, buf, bufsize,
5498 NULL);
5499 if (err != 0 && err != ENAMETOOLONG)
5500 error = EFAULT;
5501 }
5502 break;
5503 case ZONE_ATTR_BOOTARGS:
5504 if (zone->zone_bootargs == NULL)
5505 outstr = "";
5506 else
5507 outstr = zone->zone_bootargs;
5508 size = strlen(outstr) + 1;
5509 if (bufsize > size)
5510 bufsize = size;
5511 if (buf != NULL) {
5512 err = copyoutstr(outstr, buf, bufsize, NULL);
5513 if (err != 0 && err != ENAMETOOLONG)
5514 error = EFAULT;
5515 }
5516 break;
5517 case ZONE_ATTR_PHYS_MCAP:
5518 size = sizeof (zone->zone_phys_mcap);
5519 if (bufsize > size)
5520 bufsize = size;
5521 if (buf != NULL &&
5522 copyout(&zone->zone_phys_mcap, buf, bufsize) != 0)
5523 error = EFAULT;
5524 break;
5525 case ZONE_ATTR_SCHED_CLASS:
5526 mutex_enter(&class_lock);
5527
5528 if (zone->zone_defaultcid >= loaded_classes)
5529 outstr = "";
5530 else
5531 outstr = sclass[zone->zone_defaultcid].cl_name;
5532 size = strlen(outstr) + 1;
5533 if (bufsize > size)
5534 bufsize = size;
5535 if (buf != NULL) {
5536 err = copyoutstr(outstr, buf, bufsize, NULL);
5537 if (err != 0 && err != ENAMETOOLONG)
5538 error = EFAULT;
5539 }
5540
5541 mutex_exit(&class_lock);
5542 break;
5543 case ZONE_ATTR_HOSTID:
5544 if (zone->zone_hostid != HW_INVALID_HOSTID &&
5545 bufsize == sizeof (zone->zone_hostid)) {
5546 size = sizeof (zone->zone_hostid);
5547 if (buf != NULL && copyout(&zone->zone_hostid, buf,
5548 bufsize) != 0)
5549 error = EFAULT;
5550 } else {
5551 error = EINVAL;
5552 }
5553 break;
5554 case ZONE_ATTR_FS_ALLOWED:
5555 if (zone->zone_fs_allowed == NULL)
5556 outstr = "";
5557 else
5558 outstr = zone->zone_fs_allowed;
5559 size = strlen(outstr) + 1;
5560 if (bufsize > size)
5561 bufsize = size;
5562 if (buf != NULL) {
5563 err = copyoutstr(outstr, buf, bufsize, NULL);
5564 if (err != 0 && err != ENAMETOOLONG)
5565 error = EFAULT;
5566 }
5567 break;
5568 case ZONE_ATTR_SECFLAGS:
5569 size = sizeof (zone->zone_secflags);
5570 if (bufsize > size)
5571 bufsize = size;
5572 if ((err = copyout(&zone->zone_secflags, buf, bufsize)) != 0)
5573 error = EFAULT;
5574 break;
5575 case ZONE_ATTR_NETWORK:
5576 zbuf = kmem_alloc(bufsize, KM_SLEEP);
5577 if (copyin(buf, zbuf, bufsize) != 0) {
5578 error = EFAULT;
5579 } else {
5580 error = zone_get_network(zoneid, zbuf);
5581 if (error == 0 && copyout(zbuf, buf, bufsize) != 0)
5582 error = EFAULT;
5583 }
5584 kmem_free(zbuf, bufsize);
5585 break;
5586 default:
5587 if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone)) {
5588 size = bufsize;
5589 error = ZBROP(zone)->b_getattr(zone, attr, buf, &size);
5590 } else {
5591 error = EINVAL;
5592 }
5593 }
5594 zone_rele(zone);
5595
5596 if (error)
5597 return (set_errno(error));
5598 return ((ssize_t)size);
5599 }
5600
5601 /*
5602 * Systemcall entry point for zone_setattr(2).
5603 */
5604 /*ARGSUSED*/
5605 static int
5606 zone_setattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
5607 {
5608 zone_t *zone;
5609 zone_status_t zone_status;
5610 int err = -1;
5611 zone_net_data_t *zbuf;
5612
5613 if (secpolicy_zone_config(CRED()) != 0)
5614 return (set_errno(EPERM));
5615
5616 /*
5617 * Only the ZONE_ATTR_PHYS_MCAP attribute can be set on the
5618 * global zone.
5619 */
5620 if (zoneid == GLOBAL_ZONEID && attr != ZONE_ATTR_PHYS_MCAP) {
5621 return (set_errno(EINVAL));
5622 }
5623
5624 mutex_enter(&zonehash_lock);
5625 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5626 mutex_exit(&zonehash_lock);
5627 return (set_errno(EINVAL));
5628 }
5629 zone_hold(zone);
5630 mutex_exit(&zonehash_lock);
5631
5632 /*
5633 * At present most attributes can only be set on non-running,
5634 * non-global zones.
5635 */
5636 zone_status = zone_status_get(zone);
5637 if (attr != ZONE_ATTR_PHYS_MCAP && zone_status > ZONE_IS_READY) {
5638 err = EINVAL;
5639 goto done;
5640 }
5641
5642 switch (attr) {
5643 case ZONE_ATTR_INITNAME:
5644 err = zone_set_initname(zone, (const char *)buf);
5645 break;
5646 case ZONE_ATTR_INITNORESTART:
5647 zone->zone_restart_init = B_FALSE;
5648 err = 0;
5649 break;
5650 case ZONE_ATTR_BOOTARGS:
5651 err = zone_set_bootargs(zone, (const char *)buf);
5652 break;
5653 case ZONE_ATTR_BRAND:
5654 err = zone_set_brand(zone, (const char *)buf);
5655 break;
5656 case ZONE_ATTR_FS_ALLOWED:
5657 err = zone_set_fs_allowed(zone, (const char *)buf);
5658 break;
5659 case ZONE_ATTR_SECFLAGS:
5660 err = zone_set_secflags(zone, (psecflags_t *)buf);
5661 break;
5662 case ZONE_ATTR_PHYS_MCAP:
5663 err = zone_set_phys_mcap(zone, (const uint64_t *)buf);
5664 break;
5665 case ZONE_ATTR_SCHED_CLASS:
5666 err = zone_set_sched_class(zone, (const char *)buf);
5667 break;
5668 case ZONE_ATTR_HOSTID:
5669 if (bufsize == sizeof (zone->zone_hostid)) {
5670 if (copyin(buf, &zone->zone_hostid, bufsize) == 0)
5671 err = 0;
5672 else
5673 err = EFAULT;
5674 } else {
5675 err = EINVAL;
5676 }
5677 break;
5678 case ZONE_ATTR_NETWORK:
5679 if (bufsize > (PIPE_BUF + sizeof (zone_net_data_t))) {
5680 err = EINVAL;
5681 break;
5682 }
5683 zbuf = kmem_alloc(bufsize, KM_SLEEP);
5684 if (copyin(buf, zbuf, bufsize) != 0) {
5685 kmem_free(zbuf, bufsize);
5686 err = EFAULT;
5687 break;
5688 }
5689 err = zone_set_network(zoneid, zbuf);
5690 kmem_free(zbuf, bufsize);
5691 break;
5692 default:
5693 if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone))
5694 err = ZBROP(zone)->b_setattr(zone, attr, buf, bufsize);
5695 else
5696 err = EINVAL;
5697 }
5698
5699 done:
5700 zone_rele(zone);
5701 ASSERT(err != -1);
5702 return (err != 0 ? set_errno(err) : 0);
5703 }
5704
5705 /*
5706 * Return zero if the process has at least one vnode mapped in to its
5707 * address space which shouldn't be allowed to change zones.
5708 *
5709 * Also return zero if the process has any shared mappings which reserve
5710 * swap. This is because the counting for zone.max-swap does not allow swap
5711 * reservation to be shared between zones. zone swap reservation is counted
5712 * on zone->zone_max_swap.
5713 */
5714 static int
5715 as_can_change_zones(void)
5716 {
5717 proc_t *pp = curproc;
5718 struct seg *seg;
5719 struct as *as = pp->p_as;
5720 vnode_t *vp;
5721 int allow = 1;
5722
5723 ASSERT(pp->p_as != &kas);
5724 AS_LOCK_ENTER(as, RW_READER);
5725 for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) {
5726
5727 /*
5728 * Cannot enter zone with shared anon memory which
5729 * reserves swap. See comment above.
5730 */
5731 if (seg_can_change_zones(seg) == B_FALSE) {
5732 allow = 0;
5733 break;
5734 }
5735 /*
5736 * if we can't get a backing vnode for this segment then skip
5737 * it.
5738 */
5739 vp = NULL;
5740 if (SEGOP_GETVP(seg, seg->s_base, &vp) != 0 || vp == NULL)
5741 continue;
5742 if (!vn_can_change_zones(vp)) { /* bail on first match */
5743 allow = 0;
5744 break;
5745 }
5746 }
5747 AS_LOCK_EXIT(as);
5748 return (allow);
5749 }
5750
5751 /*
5752 * Count swap reserved by curproc's address space
5753 */
5754 static size_t
5755 as_swresv(void)
5756 {
5757 proc_t *pp = curproc;
5758 struct seg *seg;
5759 struct as *as = pp->p_as;
5760 size_t swap = 0;
5761
5762 ASSERT(pp->p_as != &kas);
5763 ASSERT(AS_WRITE_HELD(as));
5764 for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg))
5765 swap += seg_swresv(seg);
5766
5767 return (swap);
5768 }
5769
5770 /*
5771 * Systemcall entry point for zone_enter().
5772 *
5773 * The current process is injected into said zone. In the process
5774 * it will change its project membership, privileges, rootdir/cwd,
5775 * zone-wide rctls, and pool association to match those of the zone.
5776 *
5777 * The first zone_enter() called while the zone is in the ZONE_IS_READY
5778 * state will transition it to ZONE_IS_RUNNING. Processes may only
5779 * enter a zone that is "ready" or "running".
5780 */
5781 static int
5782 zone_enter(zoneid_t zoneid)
5783 {
5784 zone_t *zone;
5785 vnode_t *vp;
5786 proc_t *pp = curproc;
5787 contract_t *ct;
5788 cont_process_t *ctp;
5789 task_t *tk, *oldtk;
5790 kproject_t *zone_proj0;
5791 cred_t *cr, *newcr;
5792 pool_t *oldpool, *newpool;
5793 sess_t *sp;
5794 uid_t uid;
5795 zone_status_t status;
5796 int err = 0;
5797 rctl_entity_p_t e;
5798 size_t swap;
5799 kthread_id_t t;
5800
5801 if (secpolicy_zone_config(CRED()) != 0)
5802 return (set_errno(EPERM));
5803 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
5804 return (set_errno(EINVAL));
5805
5806 /*
5807 * Stop all lwps so we don't need to hold a lock to look at
5808 * curproc->p_zone. This needs to happen before we grab any
5809 * locks to avoid deadlock (another lwp in the process could
5810 * be waiting for the held lock).
5811 */
5812 if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK))
5813 return (set_errno(EINTR));
5814
5815 /*
5816 * Make sure we're not changing zones with files open or mapped in
5817 * to our address space which shouldn't be changing zones.
5818 */
5819 if (!files_can_change_zones()) {
5820 err = EBADF;
5821 goto out;
5822 }
5823 if (!as_can_change_zones()) {
5824 err = EFAULT;
5825 goto out;
5826 }
5827
5828 mutex_enter(&zonehash_lock);
5829 if (pp->p_zone != global_zone) {
5830 mutex_exit(&zonehash_lock);
5831 err = EINVAL;
5832 goto out;
5833 }
5834
5835 zone = zone_find_all_by_id(zoneid);
5836 if (zone == NULL) {
5837 mutex_exit(&zonehash_lock);
5838 err = EINVAL;
5839 goto out;
5840 }
5841
5842 /*
5843 * To prevent processes in a zone from holding contracts on
5844 * extrazonal resources, and to avoid process contract
5845 * memberships which span zones, contract holders and processes
5846 * which aren't the sole members of their encapsulating process
5847 * contracts are not allowed to zone_enter.
5848 */
5849 ctp = pp->p_ct_process;
5850 ct = &ctp->conp_contract;
5851 mutex_enter(&ct->ct_lock);
5852 mutex_enter(&pp->p_lock);
5853 if ((avl_numnodes(&pp->p_ct_held) != 0) || (ctp->conp_nmembers != 1)) {
5854 mutex_exit(&pp->p_lock);
5855 mutex_exit(&ct->ct_lock);
5856 mutex_exit(&zonehash_lock);
5857 err = EINVAL;
5858 goto out;
5859 }
5860
5861 /*
5862 * Moreover, we don't allow processes whose encapsulating
5863 * process contracts have inherited extrazonal contracts.
5864 * While it would be easier to eliminate all process contracts
5865 * with inherited contracts, we need to be able to give a
5866 * restarted init (or other zone-penetrating process) its
5867 * predecessor's contracts.
5868 */
5869 if (ctp->conp_ninherited != 0) {
5870 contract_t *next;
5871 for (next = list_head(&ctp->conp_inherited); next;
5872 next = list_next(&ctp->conp_inherited, next)) {
5873 if (contract_getzuniqid(next) != zone->zone_uniqid) {
5874 mutex_exit(&pp->p_lock);
5875 mutex_exit(&ct->ct_lock);
5876 mutex_exit(&zonehash_lock);
5877 err = EINVAL;
5878 goto out;
5879 }
5880 }
5881 }
5882
5883 mutex_exit(&pp->p_lock);
5884 mutex_exit(&ct->ct_lock);
5885
5886 status = zone_status_get(zone);
5887 if (status < ZONE_IS_READY || status >= ZONE_IS_SHUTTING_DOWN) {
5888 /*
5889 * Can't join
5890 */
5891 mutex_exit(&zonehash_lock);
5892 err = EINVAL;
5893 goto out;
5894 }
5895
5896 /*
5897 * Make sure new priv set is within the permitted set for caller
5898 */
5899 if (!priv_issubset(zone->zone_privset, &CR_OPPRIV(CRED()))) {
5900 mutex_exit(&zonehash_lock);
5901 err = EPERM;
5902 goto out;
5903 }
5904 /*
5905 * We want to momentarily drop zonehash_lock while we optimistically
5906 * bind curproc to the pool it should be running in. This is safe
5907 * since the zone can't disappear (we have a hold on it).
5908 */
5909 zone_hold(zone);
5910 mutex_exit(&zonehash_lock);
5911
5912 /*
5913 * Grab pool_lock to keep the pools configuration from changing
5914 * and to stop ourselves from getting rebound to another pool
5915 * until we join the zone.
5916 */
5917 if (pool_lock_intr() != 0) {
5918 zone_rele(zone);
5919 err = EINTR;
5920 goto out;
5921 }
5922 ASSERT(secpolicy_pool(CRED()) == 0);
5923 /*
5924 * Bind ourselves to the pool currently associated with the zone.
5925 */
5926 oldpool = curproc->p_pool;
5927 newpool = zone_pool_get(zone);
5928 if (pool_state == POOL_ENABLED && newpool != oldpool &&
5929 (err = pool_do_bind(newpool, P_PID, P_MYID,
5930 POOL_BIND_ALL)) != 0) {
5931 pool_unlock();
5932 zone_rele(zone);
5933 goto out;
5934 }
5935
5936 /*
5937 * Grab cpu_lock now; we'll need it later when we call
5938 * task_join().
5939 */
5940 mutex_enter(&cpu_lock);
5941 mutex_enter(&zonehash_lock);
5942 /*
5943 * Make sure the zone hasn't moved on since we dropped zonehash_lock.
5944 */
5945 if (zone_status_get(zone) >= ZONE_IS_SHUTTING_DOWN) {
5946 /*
5947 * Can't join anymore.
5948 */
5949 mutex_exit(&zonehash_lock);
5950 mutex_exit(&cpu_lock);
5951 if (pool_state == POOL_ENABLED &&
5952 newpool != oldpool)
5953 (void) pool_do_bind(oldpool, P_PID, P_MYID,
5954 POOL_BIND_ALL);
5955 pool_unlock();
5956 zone_rele(zone);
5957 err = EINVAL;
5958 goto out;
5959 }
5960
5961 /*
5962 * a_lock must be held while transfering locked memory and swap
5963 * reservation from the global zone to the non global zone because
5964 * asynchronous faults on the processes' address space can lock
5965 * memory and reserve swap via MCL_FUTURE and MAP_NORESERVE
5966 * segments respectively.
5967 */
5968 AS_LOCK_ENTER(pp->p_as, RW_WRITER);
5969 swap = as_swresv();
5970 mutex_enter(&pp->p_lock);
5971 zone_proj0 = zone->zone_zsched->p_task->tk_proj;
5972 /* verify that we do not exceed and task or lwp limits */
5973 mutex_enter(&zone->zone_nlwps_lock);
5974 /* add new lwps to zone and zone's proj0 */
5975 zone_proj0->kpj_nlwps += pp->p_lwpcnt;
5976 zone->zone_nlwps += pp->p_lwpcnt;
5977 /* add 1 task to zone's proj0 */
5978 zone_proj0->kpj_ntasks += 1;
5979
5980 zone_proj0->kpj_nprocs++;
5981 zone->zone_nprocs++;
5982 mutex_exit(&zone->zone_nlwps_lock);
5983
5984 mutex_enter(&zone->zone_mem_lock);
5985 zone->zone_locked_mem += pp->p_locked_mem;
5986 zone_proj0->kpj_data.kpd_locked_mem += pp->p_locked_mem;
5987 zone->zone_max_swap += swap;
5988 mutex_exit(&zone->zone_mem_lock);
5989
5990 mutex_enter(&(zone_proj0->kpj_data.kpd_crypto_lock));
5991 zone_proj0->kpj_data.kpd_crypto_mem += pp->p_crypto_mem;
5992 mutex_exit(&(zone_proj0->kpj_data.kpd_crypto_lock));
5993
5994 /* remove lwps and process from proc's old zone and old project */
5995 mutex_enter(&pp->p_zone->zone_nlwps_lock);
5996 pp->p_zone->zone_nlwps -= pp->p_lwpcnt;
5997 pp->p_task->tk_proj->kpj_nlwps -= pp->p_lwpcnt;
5998 pp->p_task->tk_proj->kpj_nprocs--;
5999 pp->p_zone->zone_nprocs--;
6000 mutex_exit(&pp->p_zone->zone_nlwps_lock);
6001
6002 mutex_enter(&pp->p_zone->zone_mem_lock);
6003 pp->p_zone->zone_locked_mem -= pp->p_locked_mem;
6004 pp->p_task->tk_proj->kpj_data.kpd_locked_mem -= pp->p_locked_mem;
6005 pp->p_zone->zone_max_swap -= swap;
6006 mutex_exit(&pp->p_zone->zone_mem_lock);
6007
6008 mutex_enter(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock));
6009 pp->p_task->tk_proj->kpj_data.kpd_crypto_mem -= pp->p_crypto_mem;
6010 mutex_exit(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock));
6011
6012 pp->p_flag |= SZONETOP;
6013 pp->p_zone = zone;
6014 mutex_exit(&pp->p_lock);
6015 AS_LOCK_EXIT(pp->p_as);
6016
6017 /*
6018 * Joining the zone cannot fail from now on.
6019 *
6020 * This means that a lot of the following code can be commonized and
6021 * shared with zsched().
6022 */
6023
6024 /*
6025 * If the process contract fmri was inherited, we need to
6026 * flag this so that any contract status will not leak
6027 * extra zone information, svc_fmri in this case
6028 */
6029 if (ctp->conp_svc_ctid != ct->ct_id) {
6030 mutex_enter(&ct->ct_lock);
6031 ctp->conp_svc_zone_enter = ct->ct_id;
6032 mutex_exit(&ct->ct_lock);
6033 }
6034
6035 /*
6036 * Reset the encapsulating process contract's zone.
6037 */
6038 ASSERT(ct->ct_mzuniqid == GLOBAL_ZONEUNIQID);
6039 contract_setzuniqid(ct, zone->zone_uniqid);
6040
6041 /*
6042 * Create a new task and associate the process with the project keyed
6043 * by (projid,zoneid).
6044 *
6045 * We might as well be in project 0; the global zone's projid doesn't
6046 * make much sense in a zone anyhow.
6047 *
6048 * This also increments zone_ntasks, and returns with p_lock held.
6049 */
6050 tk = task_create(0, zone);
6051 oldtk = task_join(tk, 0);
6052 mutex_exit(&cpu_lock);
6053
6054 /*
6055 * call RCTLOP_SET functions on this proc
6056 */
6057 e.rcep_p.zone = zone;
6058 e.rcep_t = RCENTITY_ZONE;
6059 (void) rctl_set_dup(NULL, NULL, pp, &e, zone->zone_rctls, NULL,
6060 RCD_CALLBACK);
6061 mutex_exit(&pp->p_lock);
6062
6063 /*
6064 * We don't need to hold any of zsched's locks here; not only do we know
6065 * the process and zone aren't going away, we know its session isn't
6066 * changing either.
6067 *
6068 * By joining zsched's session here, we mimic the behavior in the
6069 * global zone of init's sid being the pid of sched. We extend this
6070 * to all zlogin-like zone_enter()'ing processes as well.
6071 */
6072 mutex_enter(&pidlock);
6073 sp = zone->zone_zsched->p_sessp;
6074 sess_hold(zone->zone_zsched);
6075 mutex_enter(&pp->p_lock);
6076 pgexit(pp);
6077 sess_rele(pp->p_sessp, B_TRUE);
6078 pp->p_sessp = sp;
6079 pgjoin(pp, zone->zone_zsched->p_pidp);
6080
6081 /*
6082 * If any threads are scheduled to be placed on zone wait queue they
6083 * should abandon the idea since the wait queue is changing.
6084 * We need to be holding pidlock & p_lock to do this.
6085 */
6086 if ((t = pp->p_tlist) != NULL) {
6087 do {
6088 thread_lock(t);
6089 /*
6090 * Kick this thread so that he doesn't sit
6091 * on a wrong wait queue.
6092 */
6093 if (ISWAITING(t))
6094 setrun_locked(t);
6095
6096 if (t->t_schedflag & TS_ANYWAITQ)
6097 t->t_schedflag &= ~ TS_ANYWAITQ;
6098
6099 thread_unlock(t);
6100 } while ((t = t->t_forw) != pp->p_tlist);
6101 }
6102
6103 /*
6104 * If there is a default scheduling class for the zone and it is not
6105 * the class we are currently in, change all of the threads in the
6106 * process to the new class. We need to be holding pidlock & p_lock
6107 * when we call parmsset so this is a good place to do it.
6108 */
6109 if (zone->zone_defaultcid > 0 &&
6110 zone->zone_defaultcid != curthread->t_cid) {
6111 pcparms_t pcparms;
6112
6113 pcparms.pc_cid = zone->zone_defaultcid;
6114 pcparms.pc_clparms[0] = 0;
6115
6116 /*
6117 * If setting the class fails, we still want to enter the zone.
6118 */
6119 if ((t = pp->p_tlist) != NULL) {
6120 do {
6121 (void) parmsset(&pcparms, t);
6122 } while ((t = t->t_forw) != pp->p_tlist);
6123 }
6124 }
6125
6126 mutex_exit(&pp->p_lock);
6127 mutex_exit(&pidlock);
6128
6129 mutex_exit(&zonehash_lock);
6130 /*
6131 * We're firmly in the zone; let pools progress.
6132 */
6133 pool_unlock();
6134 task_rele(oldtk);
6135 /*
6136 * We don't need to retain a hold on the zone since we already
6137 * incremented zone_ntasks, so the zone isn't going anywhere.
6138 */
6139 zone_rele(zone);
6140
6141 /*
6142 * Chroot
6143 */
6144 vp = zone->zone_rootvp;
6145 zone_chdir(vp, &PTOU(pp)->u_cdir, pp);
6146 zone_chdir(vp, &PTOU(pp)->u_rdir, pp);
6147
6148 /*
6149 * Change process credentials
6150 */
6151 newcr = cralloc();
6152 mutex_enter(&pp->p_crlock);
6153 cr = pp->p_cred;
6154 crcopy_to(cr, newcr);
6155 crsetzone(newcr, zone);
6156 pp->p_cred = newcr;
6157
6158 /*
6159 * Restrict all process privilege sets to zone limit
6160 */
6161 priv_intersect(zone->zone_privset, &CR_PPRIV(newcr));
6162 priv_intersect(zone->zone_privset, &CR_EPRIV(newcr));
6163 priv_intersect(zone->zone_privset, &CR_IPRIV(newcr));
6164 priv_intersect(zone->zone_privset, &CR_LPRIV(newcr));
6165 mutex_exit(&pp->p_crlock);
6166 crset(pp, newcr);
6167
6168 /*
6169 * Adjust upcount to reflect zone entry.
6170 */
6171 uid = crgetruid(newcr);
6172 mutex_enter(&pidlock);
6173 upcount_dec(uid, GLOBAL_ZONEID);
6174 upcount_inc(uid, zoneid);
6175 mutex_exit(&pidlock);
6176
6177 /*
6178 * Set up core file path and content.
6179 */
6180 set_core_defaults();
6181
6182 out:
6183 /*
6184 * Let the other lwps continue.
6185 */
6186 mutex_enter(&pp->p_lock);
6187 if (curthread != pp->p_agenttp)
6188 continuelwps(pp);
6189 mutex_exit(&pp->p_lock);
6190
6191 return (err != 0 ? set_errno(err) : 0);
6192 }
6193
6194 /*
6195 * Systemcall entry point for zone_list(2).
6196 *
6197 * Processes running in a (non-global) zone only see themselves.
6198 * On labeled systems, they see all zones whose label they dominate.
6199 */
6200 static int
6201 zone_list(zoneid_t *zoneidlist, uint_t *numzones)
6202 {
6203 zoneid_t *zoneids;
6204 zone_t *zone, *myzone;
6205 uint_t user_nzones, real_nzones;
6206 uint_t domi_nzones;
6207 int error;
6208
6209 if (copyin(numzones, &user_nzones, sizeof (uint_t)) != 0)
6210 return (set_errno(EFAULT));
6211
6212 myzone = curproc->p_zone;
6213 if (myzone != global_zone) {
6214 bslabel_t *mybslab;
6215
6216 if (!is_system_labeled()) {
6217 /* just return current zone */
6218 real_nzones = domi_nzones = 1;
6219 zoneids = kmem_alloc(sizeof (zoneid_t), KM_SLEEP);
6220 zoneids[0] = myzone->zone_id;
6221 } else {
6222 /* return all zones that are dominated */
6223 mutex_enter(&zonehash_lock);
6224 real_nzones = zonecount;
6225 domi_nzones = 0;
6226 if (real_nzones > 0) {
6227 zoneids = kmem_alloc(real_nzones *
6228 sizeof (zoneid_t), KM_SLEEP);
6229 mybslab = label2bslabel(myzone->zone_slabel);
6230 for (zone = list_head(&zone_active);
6231 zone != NULL;
6232 zone = list_next(&zone_active, zone)) {
6233 if (zone->zone_id == GLOBAL_ZONEID)
6234 continue;
6235 if (zone != myzone &&
6236 (zone->zone_flags & ZF_IS_SCRATCH))
6237 continue;
6238 /*
6239 * Note that a label always dominates
6240 * itself, so myzone is always included
6241 * in the list.
6242 */
6243 if (bldominates(mybslab,
6244 label2bslabel(zone->zone_slabel))) {
6245 zoneids[domi_nzones++] =
6246 zone->zone_id;
6247 }
6248 }
6249 }
6250 mutex_exit(&zonehash_lock);
6251 }
6252 } else {
6253 mutex_enter(&zonehash_lock);
6254 real_nzones = zonecount;
6255 domi_nzones = 0;
6256 if (real_nzones > 0) {
6257 zoneids = kmem_alloc(real_nzones * sizeof (zoneid_t),
6258 KM_SLEEP);
6259 for (zone = list_head(&zone_active); zone != NULL;
6260 zone = list_next(&zone_active, zone))
6261 zoneids[domi_nzones++] = zone->zone_id;
6262 ASSERT(domi_nzones == real_nzones);
6263 }
6264 mutex_exit(&zonehash_lock);
6265 }
6266
6267 /*
6268 * If user has allocated space for fewer entries than we found, then
6269 * return only up to his limit. Either way, tell him exactly how many
6270 * we found.
6271 */
6272 if (domi_nzones < user_nzones)
6273 user_nzones = domi_nzones;
6274 error = 0;
6275 if (copyout(&domi_nzones, numzones, sizeof (uint_t)) != 0) {
6276 error = EFAULT;
6277 } else if (zoneidlist != NULL && user_nzones != 0) {
6278 if (copyout(zoneids, zoneidlist,
6279 user_nzones * sizeof (zoneid_t)) != 0)
6280 error = EFAULT;
6281 }
6282
6283 if (real_nzones > 0)
6284 kmem_free(zoneids, real_nzones * sizeof (zoneid_t));
6285
6286 if (error != 0)
6287 return (set_errno(error));
6288 else
6289 return (0);
6290 }
6291
6292 /*
6293 * Systemcall entry point for zone_lookup(2).
6294 *
6295 * Non-global zones are only able to see themselves and (on labeled systems)
6296 * the zones they dominate.
6297 */
6298 static zoneid_t
6299 zone_lookup(const char *zone_name)
6300 {
6301 char *kname;
6302 zone_t *zone;
6303 zoneid_t zoneid;
6304 int err;
6305
6306 if (zone_name == NULL) {
6307 /* return caller's zone id */
6308 return (getzoneid());
6309 }
6310
6311 kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
6312 if ((err = copyinstr(zone_name, kname, ZONENAME_MAX, NULL)) != 0) {
6313 kmem_free(kname, ZONENAME_MAX);
6314 return (set_errno(err));
6315 }
6316
6317 mutex_enter(&zonehash_lock);
6318 zone = zone_find_all_by_name(kname);
6319 kmem_free(kname, ZONENAME_MAX);
6320 /*
6321 * In a non-global zone, can only lookup global and own name.
6322 * In Trusted Extensions zone label dominance rules apply.
6323 */
6324 if (zone == NULL ||
6325 zone_status_get(zone) < ZONE_IS_READY ||
6326 !zone_list_access(zone)) {
6327 mutex_exit(&zonehash_lock);
6328 return (set_errno(EINVAL));
6329 } else {
6330 zoneid = zone->zone_id;
6331 mutex_exit(&zonehash_lock);
6332 return (zoneid);
6333 }
6334 }
6335
6336 static int
6337 zone_version(int *version_arg)
6338 {
6339 int version = ZONE_SYSCALL_API_VERSION;
6340
6341 if (copyout(&version, version_arg, sizeof (int)) != 0)
6342 return (set_errno(EFAULT));
6343 return (0);
6344 }
6345
6346 /* ARGSUSED */
6347 long
6348 zone(int cmd, void *arg1, void *arg2, void *arg3, void *arg4)
6349 {
6350 zone_def zs;
6351 int err;
6352
6353 switch (cmd) {
6354 case ZONE_CREATE:
6355 if (get_udatamodel() == DATAMODEL_NATIVE) {
6356 if (copyin(arg1, &zs, sizeof (zone_def))) {
6357 return (set_errno(EFAULT));
6358 }
6359 } else {
6360 #ifdef _SYSCALL32_IMPL
6361 zone_def32 zs32;
6362
6363 if (copyin(arg1, &zs32, sizeof (zone_def32))) {
6364 return (set_errno(EFAULT));
6365 }
6366 zs.zone_name =
6367 (const char *)(unsigned long)zs32.zone_name;
6368 zs.zone_root =
6369 (const char *)(unsigned long)zs32.zone_root;
6370 zs.zone_privs =
6371 (const struct priv_set *)
6372 (unsigned long)zs32.zone_privs;
6373 zs.zone_privssz = zs32.zone_privssz;
6374 zs.rctlbuf = (caddr_t)(unsigned long)zs32.rctlbuf;
6375 zs.rctlbufsz = zs32.rctlbufsz;
6376 zs.zfsbuf = (caddr_t)(unsigned long)zs32.zfsbuf;
6377 zs.zfsbufsz = zs32.zfsbufsz;
6378 zs.extended_error =
6379 (int *)(unsigned long)zs32.extended_error;
6380 zs.match = zs32.match;
6381 zs.doi = zs32.doi;
6382 zs.label = (const bslabel_t *)(uintptr_t)zs32.label;
6383 zs.flags = zs32.flags;
6384 #else
6385 panic("get_udatamodel() returned bogus result\n");
6386 #endif
6387 }
6388
6389 return (zone_create(zs.zone_name, zs.zone_root,
6390 zs.zone_privs, zs.zone_privssz,
6391 (caddr_t)zs.rctlbuf, zs.rctlbufsz,
6392 (caddr_t)zs.zfsbuf, zs.zfsbufsz,
6393 zs.extended_error, zs.match, zs.doi,
6394 zs.label, zs.flags));
6395 case ZONE_BOOT:
6396 return (zone_boot((zoneid_t)(uintptr_t)arg1));
6397 case ZONE_DESTROY:
6398 return (zone_destroy((zoneid_t)(uintptr_t)arg1));
6399 case ZONE_GETATTR:
6400 return (zone_getattr((zoneid_t)(uintptr_t)arg1,
6401 (int)(uintptr_t)arg2, arg3, (size_t)arg4));
6402 case ZONE_SETATTR:
6403 return (zone_setattr((zoneid_t)(uintptr_t)arg1,
6404 (int)(uintptr_t)arg2, arg3, (size_t)arg4));
6405 case ZONE_ENTER:
6406 return (zone_enter((zoneid_t)(uintptr_t)arg1));
6407 case ZONE_LIST:
6408 return (zone_list((zoneid_t *)arg1, (uint_t *)arg2));
6409 case ZONE_SHUTDOWN:
6410 return (zone_shutdown((zoneid_t)(uintptr_t)arg1));
6411 case ZONE_LOOKUP:
6412 return (zone_lookup((const char *)arg1));
6413 case ZONE_VERSION:
6414 return (zone_version((int *)arg1));
6415 case ZONE_ADD_DATALINK:
6416 return (zone_add_datalink((zoneid_t)(uintptr_t)arg1,
6417 (datalink_id_t)(uintptr_t)arg2));
6418 case ZONE_DEL_DATALINK:
6419 return (zone_remove_datalink((zoneid_t)(uintptr_t)arg1,
6420 (datalink_id_t)(uintptr_t)arg2));
6421 case ZONE_CHECK_DATALINK: {
6422 zoneid_t zoneid;
6423 boolean_t need_copyout;
6424
6425 if (copyin(arg1, &zoneid, sizeof (zoneid)) != 0)
6426 return (EFAULT);
6427 need_copyout = (zoneid == ALL_ZONES);
6428 err = zone_check_datalink(&zoneid,
6429 (datalink_id_t)(uintptr_t)arg2);
6430 if (err == 0 && need_copyout) {
6431 if (copyout(&zoneid, arg1, sizeof (zoneid)) != 0)
6432 err = EFAULT;
6433 }
6434 return (err == 0 ? 0 : set_errno(err));
6435 }
6436 case ZONE_LIST_DATALINK:
6437 return (zone_list_datalink((zoneid_t)(uintptr_t)arg1,
6438 (int *)arg2, (datalink_id_t *)(uintptr_t)arg3));
6439 default:
6440 return (set_errno(EINVAL));
6441 }
6442 }
6443
6444 struct zarg {
6445 zone_t *zone;
6446 zone_cmd_arg_t arg;
6447 };
6448
6449 static int
6450 zone_lookup_door(const char *zone_name, door_handle_t *doorp)
6451 {
6452 char *buf;
6453 size_t buflen;
6454 int error;
6455
6456 buflen = sizeof (ZONE_DOOR_PATH) + strlen(zone_name);
6457 buf = kmem_alloc(buflen, KM_SLEEP);
6458 (void) snprintf(buf, buflen, ZONE_DOOR_PATH, zone_name);
6459 error = door_ki_open(buf, doorp);
6460 kmem_free(buf, buflen);
6461 return (error);
6462 }
6463
6464 static void
6465 zone_release_door(door_handle_t *doorp)
6466 {
6467 door_ki_rele(*doorp);
6468 *doorp = NULL;
6469 }
6470
6471 static void
6472 zone_ki_call_zoneadmd(struct zarg *zargp)
6473 {
6474 door_handle_t door = NULL;
6475 door_arg_t darg, save_arg;
6476 char *zone_name;
6477 size_t zone_namelen;
6478 zoneid_t zoneid;
6479 zone_t *zone;
6480 zone_cmd_arg_t arg;
6481 uint64_t uniqid;
6482 size_t size;
6483 int error;
6484 int retry;
6485
6486 zone = zargp->zone;
6487 arg = zargp->arg;
6488 kmem_free(zargp, sizeof (*zargp));
6489
6490 zone_namelen = strlen(zone->zone_name) + 1;
6491 zone_name = kmem_alloc(zone_namelen, KM_SLEEP);
6492 bcopy(zone->zone_name, zone_name, zone_namelen);
6493 zoneid = zone->zone_id;
6494 uniqid = zone->zone_uniqid;
6495 /*
6496 * zoneadmd may be down, but at least we can empty out the zone.
6497 * We can ignore the return value of zone_empty() since we're called
6498 * from a kernel thread and know we won't be delivered any signals.
6499 */
6500 ASSERT(curproc == &p0);
6501 (void) zone_empty(zone);
6502 ASSERT(zone_status_get(zone) >= ZONE_IS_EMPTY);
6503 zone_rele(zone);
6504
6505 size = sizeof (arg);
6506 darg.rbuf = (char *)&arg;
6507 darg.data_ptr = (char *)&arg;
6508 darg.rsize = size;
6509 darg.data_size = size;
6510 darg.desc_ptr = NULL;
6511 darg.desc_num = 0;
6512
6513 save_arg = darg;
6514 /*
6515 * Since we're not holding a reference to the zone, any number of
6516 * things can go wrong, including the zone disappearing before we get a
6517 * chance to talk to zoneadmd.
6518 */
6519 for (retry = 0; /* forever */; retry++) {
6520 if (door == NULL &&
6521 (error = zone_lookup_door(zone_name, &door)) != 0) {
6522 goto next;
6523 }
6524 ASSERT(door != NULL);
6525
6526 if ((error = door_ki_upcall_limited(door, &darg, NULL,
6527 SIZE_MAX, 0)) == 0) {
6528 break;
6529 }
6530 switch (error) {
6531 case EINTR:
6532 /* FALLTHROUGH */
6533 case EAGAIN: /* process may be forking */
6534 /*
6535 * Back off for a bit
6536 */
6537 break;
6538 case EBADF:
6539 zone_release_door(&door);
6540 if (zone_lookup_door(zone_name, &door) != 0) {
6541 /*
6542 * zoneadmd may be dead, but it may come back to
6543 * life later.
6544 */
6545 break;
6546 }
6547 break;
6548 default:
6549 cmn_err(CE_WARN,
6550 "zone_ki_call_zoneadmd: door_ki_upcall error %d\n",
6551 error);
6552 goto out;
6553 }
6554 next:
6555 /*
6556 * If this isn't the same zone_t that we originally had in mind,
6557 * then this is the same as if two kadmin requests come in at
6558 * the same time: the first one wins. This means we lose, so we
6559 * bail.
6560 */
6561 if ((zone = zone_find_by_id(zoneid)) == NULL) {
6562 /*
6563 * Problem is solved.
6564 */
6565 break;
6566 }
6567 if (zone->zone_uniqid != uniqid) {
6568 /*
6569 * zoneid recycled
6570 */
6571 zone_rele(zone);
6572 break;
6573 }
6574 /*
6575 * We could zone_status_timedwait(), but there doesn't seem to
6576 * be much point in doing that (plus, it would mean that
6577 * zone_free() isn't called until this thread exits).
6578 */
6579 zone_rele(zone);
6580 delay(hz);
6581 darg = save_arg;
6582 }
6583 out:
6584 if (door != NULL) {
6585 zone_release_door(&door);
6586 }
6587 kmem_free(zone_name, zone_namelen);
6588 thread_exit();
6589 }
6590
6591 /*
6592 * Entry point for uadmin() to tell the zone to go away or reboot. Analog to
6593 * kadmin(). The caller is a process in the zone.
6594 *
6595 * In order to shutdown the zone, we will hand off control to zoneadmd
6596 * (running in the global zone) via a door. We do a half-hearted job at
6597 * killing all processes in the zone, create a kernel thread to contact
6598 * zoneadmd, and make note of the "uniqid" of the zone. The uniqid is
6599 * a form of generation number used to let zoneadmd (as well as
6600 * zone_destroy()) know exactly which zone they're re talking about.
6601 */
6602 int
6603 zone_kadmin(int cmd, int fcn, const char *mdep, cred_t *credp)
6604 {
6605 struct zarg *zargp;
6606 zone_cmd_t zcmd;
6607 zone_t *zone;
6608
6609 zone = curproc->p_zone;
6610 ASSERT(getzoneid() != GLOBAL_ZONEID);
6611
6612 switch (cmd) {
6613 case A_SHUTDOWN:
6614 switch (fcn) {
6615 case AD_HALT:
6616 case AD_POWEROFF:
6617 zcmd = Z_HALT;
6618 break;
6619 case AD_BOOT:
6620 zcmd = Z_REBOOT;
6621 break;
6622 case AD_IBOOT:
6623 case AD_SBOOT:
6624 case AD_SIBOOT:
6625 case AD_NOSYNC:
6626 return (ENOTSUP);
6627 default:
6628 return (EINVAL);
6629 }
6630 break;
6631 case A_REBOOT:
6632 zcmd = Z_REBOOT;
6633 break;
6634 case A_FTRACE:
6635 case A_REMOUNT:
6636 case A_FREEZE:
6637 case A_DUMP:
6638 case A_CONFIG:
6639 return (ENOTSUP);
6640 default:
6641 ASSERT(cmd != A_SWAPCTL); /* handled by uadmin() */
6642 return (EINVAL);
6643 }
6644
6645 if (secpolicy_zone_admin(credp, B_FALSE))
6646 return (EPERM);
6647 mutex_enter(&zone_status_lock);
6648
6649 /*
6650 * zone_status can't be ZONE_IS_EMPTY or higher since curproc
6651 * is in the zone.
6652 */
6653 ASSERT(zone_status_get(zone) < ZONE_IS_EMPTY);
6654 if (zone_status_get(zone) > ZONE_IS_RUNNING) {
6655 /*
6656 * This zone is already on its way down.
6657 */
6658 mutex_exit(&zone_status_lock);
6659 return (0);
6660 }
6661 /*
6662 * Prevent future zone_enter()s
6663 */
6664 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
6665 mutex_exit(&zone_status_lock);
6666
6667 /*
6668 * Kill everyone now and call zoneadmd later.
6669 * zone_ki_call_zoneadmd() will do a more thorough job of this
6670 * later.
6671 */
6672 killall(zone->zone_id);
6673 /*
6674 * Now, create the thread to contact zoneadmd and do the rest of the
6675 * work. This thread can't be created in our zone otherwise
6676 * zone_destroy() would deadlock.
6677 */
6678 zargp = kmem_zalloc(sizeof (*zargp), KM_SLEEP);
6679 zargp->arg.cmd = zcmd;
6680 zargp->arg.uniqid = zone->zone_uniqid;
6681 zargp->zone = zone;
6682 (void) strcpy(zargp->arg.locale, "C");
6683 /* mdep was already copied in for us by uadmin */
6684 if (mdep != NULL)
6685 (void) strlcpy(zargp->arg.bootbuf, mdep,
6686 sizeof (zargp->arg.bootbuf));
6687 zone_hold(zone);
6688
6689 (void) thread_create(NULL, 0, zone_ki_call_zoneadmd, zargp, 0, &p0,
6690 TS_RUN, minclsyspri);
6691 exit(CLD_EXITED, 0);
6692
6693 return (EINVAL);
6694 }
6695
6696 /*
6697 * Entry point so kadmin(A_SHUTDOWN, ...) can set the global zone's
6698 * status to ZONE_IS_SHUTTING_DOWN.
6699 *
6700 * This function also shuts down all running zones to ensure that they won't
6701 * fork new processes.
6702 */
6703 void
6704 zone_shutdown_global(void)
6705 {
6706 zone_t *current_zonep;
6707
6708 ASSERT(INGLOBALZONE(curproc));
6709 mutex_enter(&zonehash_lock);
6710 mutex_enter(&zone_status_lock);
6711
6712 /* Modify the global zone's status first. */
6713 ASSERT(zone_status_get(global_zone) == ZONE_IS_RUNNING);
6714 zone_status_set(global_zone, ZONE_IS_SHUTTING_DOWN);
6715
6716 /*
6717 * Now change the states of all running zones to ZONE_IS_SHUTTING_DOWN.
6718 * We don't mark all zones with ZONE_IS_SHUTTING_DOWN because doing so
6719 * could cause assertions to fail (e.g., assertions about a zone's
6720 * state during initialization, readying, or booting) or produce races.
6721 * We'll let threads continue to initialize and ready new zones: they'll
6722 * fail to boot the new zones when they see that the global zone is
6723 * shutting down.
6724 */
6725 for (current_zonep = list_head(&zone_active); current_zonep != NULL;
6726 current_zonep = list_next(&zone_active, current_zonep)) {
6727 if (zone_status_get(current_zonep) == ZONE_IS_RUNNING)
6728 zone_status_set(current_zonep, ZONE_IS_SHUTTING_DOWN);
6729 }
6730 mutex_exit(&zone_status_lock);
6731 mutex_exit(&zonehash_lock);
6732 }
6733
6734 /*
6735 * Returns true if the named dataset is visible in the current zone.
6736 * The 'write' parameter is set to 1 if the dataset is also writable.
6737 */
6738 int
6739 zone_dataset_visible(const char *dataset, int *write)
6740 {
6741 static int zfstype = -1;
6742 zone_dataset_t *zd;
6743 size_t len;
6744 zone_t *zone = curproc->p_zone;
6745 const char *name = NULL;
6746 vfs_t *vfsp = NULL;
6747
6748 if (dataset[0] == '\0')
6749 return (0);
6750
6751 /*
6752 * Walk the list once, looking for datasets which match exactly, or
6753 * specify a dataset underneath an exported dataset. If found, return
6754 * true and note that it is writable.
6755 */
6756 for (zd = list_head(&zone->zone_datasets); zd != NULL;
6757 zd = list_next(&zone->zone_datasets, zd)) {
6758
6759 len = strlen(zd->zd_dataset);
6760 if (strlen(dataset) >= len &&
6761 bcmp(dataset, zd->zd_dataset, len) == 0 &&
6762 (dataset[len] == '\0' || dataset[len] == '/' ||
6763 dataset[len] == '@')) {
6764 if (write)
6765 *write = 1;
6766 return (1);
6767 }
6768 }
6769
6770 /*
6771 * Walk the list a second time, searching for datasets which are parents
6772 * of exported datasets. These should be visible, but read-only.
6773 *
6774 * Note that we also have to support forms such as 'pool/dataset/', with
6775 * a trailing slash.
6776 */
6777 for (zd = list_head(&zone->zone_datasets); zd != NULL;
6778 zd = list_next(&zone->zone_datasets, zd)) {
6779
6780 len = strlen(dataset);
6781 if (dataset[len - 1] == '/')
6782 len--; /* Ignore trailing slash */
6783 if (len < strlen(zd->zd_dataset) &&
6784 bcmp(dataset, zd->zd_dataset, len) == 0 &&
6785 zd->zd_dataset[len] == '/') {
6786 if (write)
6787 *write = 0;
6788 return (1);
6789 }
6790 }
6791
6792 /*
6793 * We reach here if the given dataset is not found in the zone_dataset
6794 * list. Check if this dataset was added as a filesystem (ie. "add fs")
6795 * instead of delegation. For this we search for the dataset in the
6796 * zone_vfslist of this zone. If found, return true and note that it is
6797 * not writable.
6798 */
6799
6800 /*
6801 * Initialize zfstype if it is not initialized yet.
6802 */
6803 if (zfstype == -1) {
6804 struct vfssw *vswp = vfs_getvfssw("zfs");
6805 zfstype = vswp - vfssw;
6806 vfs_unrefvfssw(vswp);
6807 }
6808
6809 vfs_list_read_lock();
6810 vfsp = zone->zone_vfslist;
6811 do {
6812 ASSERT(vfsp);
6813 if (vfsp->vfs_fstype == zfstype) {
6814 name = refstr_value(vfsp->vfs_resource);
6815
6816 /*
6817 * Check if we have an exact match.
6818 */
6819 if (strcmp(dataset, name) == 0) {
6820 vfs_list_unlock();
6821 if (write)
6822 *write = 0;
6823 return (1);
6824 }
6825 /*
6826 * We need to check if we are looking for parents of
6827 * a dataset. These should be visible, but read-only.
6828 */
6829 len = strlen(dataset);
6830 if (dataset[len - 1] == '/')
6831 len--;
6832
6833 if (len < strlen(name) &&
6834 bcmp(dataset, name, len) == 0 && name[len] == '/') {
6835 vfs_list_unlock();
6836 if (write)
6837 *write = 0;
6838 return (1);
6839 }
6840 }
6841 vfsp = vfsp->vfs_zone_next;
6842 } while (vfsp != zone->zone_vfslist);
6843
6844 vfs_list_unlock();
6845 return (0);
6846 }
6847
6848 /*
6849 * zone_find_by_any_path() -
6850 *
6851 * kernel-private routine similar to zone_find_by_path(), but which
6852 * effectively compares against zone paths rather than zonerootpath
6853 * (i.e., the last component of zonerootpaths, which should be "root/",
6854 * are not compared.) This is done in order to accurately identify all
6855 * paths, whether zone-visible or not, including those which are parallel
6856 * to /root/, such as /dev/, /home/, etc...
6857 *
6858 * If the specified path does not fall under any zone path then global
6859 * zone is returned.
6860 *
6861 * The treat_abs parameter indicates whether the path should be treated as
6862 * an absolute path although it does not begin with "/". (This supports
6863 * nfs mount syntax such as host:any/path.)
6864 *
6865 * The caller is responsible for zone_rele of the returned zone.
6866 */
6867 zone_t *
6868 zone_find_by_any_path(const char *path, boolean_t treat_abs)
6869 {
6870 zone_t *zone;
6871 int path_offset = 0;
6872
6873 if (path == NULL) {
6874 zone_hold(global_zone);
6875 return (global_zone);
6876 }
6877
6878 if (*path != '/') {
6879 ASSERT(treat_abs);
6880 path_offset = 1;
6881 }
6882
6883 mutex_enter(&zonehash_lock);
6884 for (zone = list_head(&zone_active); zone != NULL;
6885 zone = list_next(&zone_active, zone)) {
6886 char *c;
6887 size_t pathlen;
6888 char *rootpath_start;
6889
6890 if (zone == global_zone) /* skip global zone */
6891 continue;
6892
6893 /* scan backwards to find start of last component */
6894 c = zone->zone_rootpath + zone->zone_rootpathlen - 2;
6895 do {
6896 c--;
6897 } while (*c != '/');
6898
6899 pathlen = c - zone->zone_rootpath + 1 - path_offset;
6900 rootpath_start = (zone->zone_rootpath + path_offset);
6901 if (strncmp(path, rootpath_start, pathlen) == 0)
6902 break;
6903 }
6904 if (zone == NULL)
6905 zone = global_zone;
6906 zone_hold(zone);
6907 mutex_exit(&zonehash_lock);
6908 return (zone);
6909 }
6910
6911 /*
6912 * Finds a zone_dl_t with the given linkid in the given zone. Returns the
6913 * zone_dl_t pointer if found, and NULL otherwise.
6914 */
6915 static zone_dl_t *
6916 zone_find_dl(zone_t *zone, datalink_id_t linkid)
6917 {
6918 zone_dl_t *zdl;
6919
6920 ASSERT(mutex_owned(&zone->zone_lock));
6921 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
6922 zdl = list_next(&zone->zone_dl_list, zdl)) {
6923 if (zdl->zdl_id == linkid)
6924 break;
6925 }
6926 return (zdl);
6927 }
6928
6929 static boolean_t
6930 zone_dl_exists(zone_t *zone, datalink_id_t linkid)
6931 {
6932 boolean_t exists;
6933
6934 mutex_enter(&zone->zone_lock);
6935 exists = (zone_find_dl(zone, linkid) != NULL);
6936 mutex_exit(&zone->zone_lock);
6937 return (exists);
6938 }
6939
6940 /*
6941 * Add an data link name for the zone.
6942 */
6943 static int
6944 zone_add_datalink(zoneid_t zoneid, datalink_id_t linkid)
6945 {
6946 zone_dl_t *zdl;
6947 zone_t *zone;
6948 zone_t *thiszone;
6949
6950 if ((thiszone = zone_find_by_id(zoneid)) == NULL)
6951 return (set_errno(ENXIO));
6952
6953 /* Verify that the datalink ID doesn't already belong to a zone. */
6954 mutex_enter(&zonehash_lock);
6955 for (zone = list_head(&zone_active); zone != NULL;
6956 zone = list_next(&zone_active, zone)) {
6957 if (zone_dl_exists(zone, linkid)) {
6958 mutex_exit(&zonehash_lock);
6959 zone_rele(thiszone);
6960 return (set_errno((zone == thiszone) ? EEXIST : EPERM));
6961 }
6962 }
6963
6964 zdl = kmem_zalloc(sizeof (*zdl), KM_SLEEP);
6965 zdl->zdl_id = linkid;
6966 zdl->zdl_net = NULL;
6967 mutex_enter(&thiszone->zone_lock);
6968 list_insert_head(&thiszone->zone_dl_list, zdl);
6969 mutex_exit(&thiszone->zone_lock);
6970 mutex_exit(&zonehash_lock);
6971 zone_rele(thiszone);
6972 return (0);
6973 }
6974
6975 static int
6976 zone_remove_datalink(zoneid_t zoneid, datalink_id_t linkid)
6977 {
6978 zone_dl_t *zdl;
6979 zone_t *zone;
6980 int err = 0;
6981
6982 if ((zone = zone_find_by_id(zoneid)) == NULL)
6983 return (set_errno(EINVAL));
6984
6985 mutex_enter(&zone->zone_lock);
6986 if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
6987 err = ENXIO;
6988 } else {
6989 list_remove(&zone->zone_dl_list, zdl);
6990 nvlist_free(zdl->zdl_net);
6991 kmem_free(zdl, sizeof (zone_dl_t));
6992 }
6993 mutex_exit(&zone->zone_lock);
6994 zone_rele(zone);
6995 return (err == 0 ? 0 : set_errno(err));
6996 }
6997
6998 /*
6999 * Using the zoneidp as ALL_ZONES, we can lookup which zone has been assigned
7000 * the linkid. Otherwise we just check if the specified zoneidp has been
7001 * assigned the supplied linkid.
7002 */
7003 int
7004 zone_check_datalink(zoneid_t *zoneidp, datalink_id_t linkid)
7005 {
7006 zone_t *zone;
7007 int err = ENXIO;
7008
7009 if (*zoneidp != ALL_ZONES) {
7010 if ((zone = zone_find_by_id(*zoneidp)) != NULL) {
7011 if (zone_dl_exists(zone, linkid))
7012 err = 0;
7013 zone_rele(zone);
7014 }
7015 return (err);
7016 }
7017
7018 mutex_enter(&zonehash_lock);
7019 for (zone = list_head(&zone_active); zone != NULL;
7020 zone = list_next(&zone_active, zone)) {
7021 if (zone_dl_exists(zone, linkid)) {
7022 *zoneidp = zone->zone_id;
7023 err = 0;
7024 break;
7025 }
7026 }
7027 mutex_exit(&zonehash_lock);
7028 return (err);
7029 }
7030
7031 /*
7032 * Get the list of datalink IDs assigned to a zone.
7033 *
7034 * On input, *nump is the number of datalink IDs that can fit in the supplied
7035 * idarray. Upon return, *nump is either set to the number of datalink IDs
7036 * that were placed in the array if the array was large enough, or to the
7037 * number of datalink IDs that the function needs to place in the array if the
7038 * array is too small.
7039 */
7040 static int
7041 zone_list_datalink(zoneid_t zoneid, int *nump, datalink_id_t *idarray)
7042 {
7043 uint_t num, dlcount;
7044 zone_t *zone;
7045 zone_dl_t *zdl;
7046 datalink_id_t *idptr = idarray;
7047
7048 if (copyin(nump, &dlcount, sizeof (dlcount)) != 0)
7049 return (set_errno(EFAULT));
7050 if ((zone = zone_find_by_id(zoneid)) == NULL)
7051 return (set_errno(ENXIO));
7052
7053 num = 0;
7054 mutex_enter(&zone->zone_lock);
7055 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7056 zdl = list_next(&zone->zone_dl_list, zdl)) {
7057 /*
7058 * If the list is bigger than what the caller supplied, just
7059 * count, don't do copyout.
7060 */
7061 if (++num > dlcount)
7062 continue;
7063 if (copyout(&zdl->zdl_id, idptr, sizeof (*idptr)) != 0) {
7064 mutex_exit(&zone->zone_lock);
7065 zone_rele(zone);
7066 return (set_errno(EFAULT));
7067 }
7068 idptr++;
7069 }
7070 mutex_exit(&zone->zone_lock);
7071 zone_rele(zone);
7072
7073 /* Increased or decreased, caller should be notified. */
7074 if (num != dlcount) {
7075 if (copyout(&num, nump, sizeof (num)) != 0)
7076 return (set_errno(EFAULT));
7077 }
7078 return (0);
7079 }
7080
7081 /*
7082 * Public interface for looking up a zone by zoneid. It's a customized version
7083 * for netstack_zone_create(). It can only be called from the zsd create
7084 * callbacks, since it doesn't have reference on the zone structure hence if
7085 * it is called elsewhere the zone could disappear after the zonehash_lock
7086 * is dropped.
7087 *
7088 * Furthermore it
7089 * 1. Doesn't check the status of the zone.
7090 * 2. It will be called even before zone_init is called, in that case the
7091 * address of zone0 is returned directly, and netstack_zone_create()
7092 * will only assign a value to zone0.zone_netstack, won't break anything.
7093 * 3. Returns without the zone being held.
7094 */
7095 zone_t *
7096 zone_find_by_id_nolock(zoneid_t zoneid)
7097 {
7098 zone_t *zone;
7099
7100 mutex_enter(&zonehash_lock);
7101 if (zonehashbyid == NULL)
7102 zone = &zone0;
7103 else
7104 zone = zone_find_all_by_id(zoneid);
7105 mutex_exit(&zonehash_lock);
7106 return (zone);
7107 }
7108
7109 /*
7110 * Walk the datalinks for a given zone
7111 */
7112 int
7113 zone_datalink_walk(zoneid_t zoneid, int (*cb)(datalink_id_t, void *),
7114 void *data)
7115 {
7116 zone_t *zone;
7117 zone_dl_t *zdl;
7118 datalink_id_t *idarray;
7119 uint_t idcount = 0;
7120 int i, ret = 0;
7121
7122 if ((zone = zone_find_by_id(zoneid)) == NULL)
7123 return (ENOENT);
7124
7125 /*
7126 * We first build an array of linkid's so that we can walk these and
7127 * execute the callback with the zone_lock dropped.
7128 */
7129 mutex_enter(&zone->zone_lock);
7130 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7131 zdl = list_next(&zone->zone_dl_list, zdl)) {
7132 idcount++;
7133 }
7134
7135 if (idcount == 0) {
7136 mutex_exit(&zone->zone_lock);
7137 zone_rele(zone);
7138 return (0);
7139 }
7140
7141 idarray = kmem_alloc(sizeof (datalink_id_t) * idcount, KM_NOSLEEP);
7142 if (idarray == NULL) {
7143 mutex_exit(&zone->zone_lock);
7144 zone_rele(zone);
7145 return (ENOMEM);
7146 }
7147
7148 for (i = 0, zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7149 i++, zdl = list_next(&zone->zone_dl_list, zdl)) {
7150 idarray[i] = zdl->zdl_id;
7151 }
7152
7153 mutex_exit(&zone->zone_lock);
7154
7155 for (i = 0; i < idcount && ret == 0; i++) {
7156 if ((ret = (*cb)(idarray[i], data)) != 0)
7157 break;
7158 }
7159
7160 zone_rele(zone);
7161 kmem_free(idarray, sizeof (datalink_id_t) * idcount);
7162 return (ret);
7163 }
7164
7165 static char *
7166 zone_net_type2name(int type)
7167 {
7168 switch (type) {
7169 case ZONE_NETWORK_ADDRESS:
7170 return (ZONE_NET_ADDRNAME);
7171 case ZONE_NETWORK_DEFROUTER:
7172 return (ZONE_NET_RTRNAME);
7173 default:
7174 return (NULL);
7175 }
7176 }
7177
7178 static int
7179 zone_set_network(zoneid_t zoneid, zone_net_data_t *znbuf)
7180 {
7181 zone_t *zone;
7182 zone_dl_t *zdl;
7183 nvlist_t *nvl;
7184 int err = 0;
7185 uint8_t *new = NULL;
7186 char *nvname;
7187 int bufsize;
7188 datalink_id_t linkid = znbuf->zn_linkid;
7189
7190 if (secpolicy_zone_config(CRED()) != 0)
7191 return (set_errno(EPERM));
7192
7193 if (zoneid == GLOBAL_ZONEID)
7194 return (set_errno(EINVAL));
7195
7196 nvname = zone_net_type2name(znbuf->zn_type);
7197 bufsize = znbuf->zn_len;
7198 new = znbuf->zn_val;
7199 if (nvname == NULL)
7200 return (set_errno(EINVAL));
7201
7202 if ((zone = zone_find_by_id(zoneid)) == NULL) {
7203 return (set_errno(EINVAL));
7204 }
7205
7206 mutex_enter(&zone->zone_lock);
7207 if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
7208 err = ENXIO;
7209 goto done;
7210 }
7211 if ((nvl = zdl->zdl_net) == NULL) {
7212 if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP)) {
7213 err = ENOMEM;
7214 goto done;
7215 } else {
7216 zdl->zdl_net = nvl;
7217 }
7218 }
7219 if (nvlist_exists(nvl, nvname)) {
7220 err = EINVAL;
7221 goto done;
7222 }
7223 err = nvlist_add_uint8_array(nvl, nvname, new, bufsize);
7224 ASSERT(err == 0);
7225 done:
7226 mutex_exit(&zone->zone_lock);
7227 zone_rele(zone);
7228 if (err != 0)
7229 return (set_errno(err));
7230 else
7231 return (0);
7232 }
7233
7234 static int
7235 zone_get_network(zoneid_t zoneid, zone_net_data_t *znbuf)
7236 {
7237 zone_t *zone;
7238 zone_dl_t *zdl;
7239 nvlist_t *nvl;
7240 uint8_t *ptr;
7241 uint_t psize;
7242 int err = 0;
7243 char *nvname;
7244 int bufsize;
7245 void *buf;
7246 datalink_id_t linkid = znbuf->zn_linkid;
7247
7248 if (zoneid == GLOBAL_ZONEID)
7249 return (set_errno(EINVAL));
7250
7251 nvname = zone_net_type2name(znbuf->zn_type);
7252 bufsize = znbuf->zn_len;
7253 buf = znbuf->zn_val;
7254
7255 if (nvname == NULL)
7256 return (set_errno(EINVAL));
7257 if ((zone = zone_find_by_id(zoneid)) == NULL)
7258 return (set_errno(EINVAL));
7259
7260 mutex_enter(&zone->zone_lock);
7261 if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
7262 err = ENXIO;
7263 goto done;
7264 }
7265 if ((nvl = zdl->zdl_net) == NULL || !nvlist_exists(nvl, nvname)) {
7266 err = ENOENT;
7267 goto done;
7268 }
7269 err = nvlist_lookup_uint8_array(nvl, nvname, &ptr, &psize);
7270 ASSERT(err == 0);
7271
7272 if (psize > bufsize) {
7273 err = ENOBUFS;
7274 goto done;
7275 }
7276 znbuf->zn_len = psize;
7277 bcopy(ptr, buf, psize);
7278 done:
7279 mutex_exit(&zone->zone_lock);
7280 zone_rele(zone);
7281 if (err != 0)
7282 return (set_errno(err));
7283 else
7284 return (0);
7285 }