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 rctl_qty_t
1490 zone_shmmax_usage(rctl_t *rctl, struct proc *p)
1491 {
1492 ASSERT(MUTEX_HELD(&p->p_lock));
1493 return (p->p_zone->zone_shmmax);
1494 }
1495
1496 /*ARGSUSED*/
1497 static int
1498 zone_shmmax_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1499 rctl_qty_t incr, uint_t flags)
1500 {
1501 rctl_qty_t v;
1502 ASSERT(MUTEX_HELD(&p->p_lock));
1503 ASSERT(e->rcep_t == RCENTITY_ZONE);
1504 v = e->rcep_p.zone->zone_shmmax + incr;
1505 if (v > rval->rcv_value)
1506 return (1);
1507 return (0);
1508 }
1509
1510 static rctl_ops_t zone_shmmax_ops = {
1511 rcop_no_action,
1512 zone_shmmax_usage,
1513 rcop_no_set,
1514 zone_shmmax_test
1515 };
1516
1517 /*ARGSUSED*/
1518 static rctl_qty_t
1519 zone_shmmni_usage(rctl_t *rctl, struct proc *p)
1520 {
1521 ASSERT(MUTEX_HELD(&p->p_lock));
1522 return (p->p_zone->zone_ipc.ipcq_shmmni);
1523 }
1524
1525 /*ARGSUSED*/
1526 static int
1527 zone_shmmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1528 rctl_qty_t incr, uint_t flags)
1529 {
1530 rctl_qty_t v;
1531 ASSERT(MUTEX_HELD(&p->p_lock));
1532 ASSERT(e->rcep_t == RCENTITY_ZONE);
1533 v = e->rcep_p.zone->zone_ipc.ipcq_shmmni + incr;
1534 if (v > rval->rcv_value)
1535 return (1);
1536 return (0);
1537 }
1538
1539 static rctl_ops_t zone_shmmni_ops = {
1540 rcop_no_action,
1541 zone_shmmni_usage,
1542 rcop_no_set,
1543 zone_shmmni_test
1544 };
1545
1546 /*ARGSUSED*/
1547 static rctl_qty_t
1548 zone_semmni_usage(rctl_t *rctl, struct proc *p)
1549 {
1550 ASSERT(MUTEX_HELD(&p->p_lock));
1551 return (p->p_zone->zone_ipc.ipcq_semmni);
1552 }
1553
1554 /*ARGSUSED*/
1555 static int
1556 zone_semmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1557 rctl_qty_t incr, uint_t flags)
1558 {
1559 rctl_qty_t v;
1560 ASSERT(MUTEX_HELD(&p->p_lock));
1561 ASSERT(e->rcep_t == RCENTITY_ZONE);
1562 v = e->rcep_p.zone->zone_ipc.ipcq_semmni + incr;
1563 if (v > rval->rcv_value)
1564 return (1);
1565 return (0);
1566 }
1567
1568 static rctl_ops_t zone_semmni_ops = {
1569 rcop_no_action,
1570 zone_semmni_usage,
1571 rcop_no_set,
1572 zone_semmni_test
1573 };
1574
1575 /*ARGSUSED*/
1576 static rctl_qty_t
1577 zone_msgmni_usage(rctl_t *rctl, struct proc *p)
1578 {
1579 ASSERT(MUTEX_HELD(&p->p_lock));
1580 return (p->p_zone->zone_ipc.ipcq_msgmni);
1581 }
1582
1583 /*ARGSUSED*/
1584 static int
1585 zone_msgmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1586 rctl_qty_t incr, uint_t flags)
1587 {
1588 rctl_qty_t v;
1589 ASSERT(MUTEX_HELD(&p->p_lock));
1590 ASSERT(e->rcep_t == RCENTITY_ZONE);
1591 v = e->rcep_p.zone->zone_ipc.ipcq_msgmni + incr;
1592 if (v > rval->rcv_value)
1593 return (1);
1594 return (0);
1595 }
1596
1597 static rctl_ops_t zone_msgmni_ops = {
1598 rcop_no_action,
1599 zone_msgmni_usage,
1600 rcop_no_set,
1601 zone_msgmni_test
1602 };
1603
1604 /*ARGSUSED*/
1605 static rctl_qty_t
1606 zone_locked_mem_usage(rctl_t *rctl, struct proc *p)
1607 {
1608 rctl_qty_t q;
1609 ASSERT(MUTEX_HELD(&p->p_lock));
1610 mutex_enter(&p->p_zone->zone_mem_lock);
1611 q = p->p_zone->zone_locked_mem;
1612 mutex_exit(&p->p_zone->zone_mem_lock);
1613 return (q);
1614 }
1615
1616 /*ARGSUSED*/
1617 static int
1618 zone_locked_mem_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1619 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1620 {
1621 rctl_qty_t q;
1622 zone_t *z;
1623
1624 z = e->rcep_p.zone;
1625 ASSERT(MUTEX_HELD(&p->p_lock));
1626 ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1627 q = z->zone_locked_mem;
1628 if (q + incr > rcntl->rcv_value)
1629 return (1);
1630 return (0);
1631 }
1632
1633 /*ARGSUSED*/
1634 static int
1635 zone_locked_mem_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1636 rctl_qty_t nv)
1637 {
1638 ASSERT(MUTEX_HELD(&p->p_lock));
1639 ASSERT(e->rcep_t == RCENTITY_ZONE);
1640 if (e->rcep_p.zone == NULL)
1641 return (0);
1642 e->rcep_p.zone->zone_locked_mem_ctl = nv;
1643 return (0);
1644 }
1645
1646 static rctl_ops_t zone_locked_mem_ops = {
1647 rcop_no_action,
1648 zone_locked_mem_usage,
1649 zone_locked_mem_set,
1650 zone_locked_mem_test
1651 };
1652
1653 /*ARGSUSED*/
1654 static rctl_qty_t
1655 zone_max_swap_usage(rctl_t *rctl, struct proc *p)
1656 {
1657 rctl_qty_t q;
1658 zone_t *z = p->p_zone;
1659
1660 ASSERT(MUTEX_HELD(&p->p_lock));
1661 mutex_enter(&z->zone_mem_lock);
1662 q = z->zone_max_swap;
1663 mutex_exit(&z->zone_mem_lock);
1664 return (q);
1665 }
1666
1667 /*ARGSUSED*/
1668 static int
1669 zone_max_swap_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1670 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1671 {
1672 rctl_qty_t q;
1673 zone_t *z;
1674
1675 z = e->rcep_p.zone;
1676 ASSERT(MUTEX_HELD(&p->p_lock));
1677 ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1678 q = z->zone_max_swap;
1679 if (q + incr > rcntl->rcv_value)
1680 return (1);
1681 return (0);
1682 }
1683
1684 /*ARGSUSED*/
1685 static int
1686 zone_max_swap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1687 rctl_qty_t nv)
1688 {
1689 ASSERT(MUTEX_HELD(&p->p_lock));
1690 ASSERT(e->rcep_t == RCENTITY_ZONE);
1691 if (e->rcep_p.zone == NULL)
1692 return (0);
1693 e->rcep_p.zone->zone_max_swap_ctl = nv;
1694 return (0);
1695 }
1696
1697 static rctl_ops_t zone_max_swap_ops = {
1698 rcop_no_action,
1699 zone_max_swap_usage,
1700 zone_max_swap_set,
1701 zone_max_swap_test
1702 };
1703
1704 /*ARGSUSED*/
1705 static rctl_qty_t
1706 zone_max_lofi_usage(rctl_t *rctl, struct proc *p)
1707 {
1708 rctl_qty_t q;
1709 zone_t *z = p->p_zone;
1710
1711 ASSERT(MUTEX_HELD(&p->p_lock));
1712 mutex_enter(&z->zone_rctl_lock);
1713 q = z->zone_max_lofi;
1714 mutex_exit(&z->zone_rctl_lock);
1715 return (q);
1716 }
1717
1718 /*ARGSUSED*/
1719 static int
1720 zone_max_lofi_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1721 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1722 {
1723 rctl_qty_t q;
1724 zone_t *z;
1725
1726 z = e->rcep_p.zone;
1727 ASSERT(MUTEX_HELD(&p->p_lock));
1728 ASSERT(MUTEX_HELD(&z->zone_rctl_lock));
1729 q = z->zone_max_lofi;
1730 if (q + incr > rcntl->rcv_value)
1731 return (1);
1732 return (0);
1733 }
1734
1735 /*ARGSUSED*/
1736 static int
1737 zone_max_lofi_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1738 rctl_qty_t nv)
1739 {
1740 ASSERT(MUTEX_HELD(&p->p_lock));
1741 ASSERT(e->rcep_t == RCENTITY_ZONE);
1742 if (e->rcep_p.zone == NULL)
1743 return (0);
1744 e->rcep_p.zone->zone_max_lofi_ctl = nv;
1745 return (0);
1746 }
1747
1748 static rctl_ops_t zone_max_lofi_ops = {
1749 rcop_no_action,
1750 zone_max_lofi_usage,
1751 zone_max_lofi_set,
1752 zone_max_lofi_test
1753 };
1754
1755 /*
1756 * Helper function to brand the zone with a unique ID.
1757 */
1758 static void
1759 zone_uniqid(zone_t *zone)
1760 {
1761 static uint64_t uniqid = 0;
1762
1763 ASSERT(MUTEX_HELD(&zonehash_lock));
1764 zone->zone_uniqid = uniqid++;
1765 }
1766
1767 /*
1768 * Returns a held pointer to the "kcred" for the specified zone.
1769 */
1770 struct cred *
1771 zone_get_kcred(zoneid_t zoneid)
1772 {
1773 zone_t *zone;
1774 cred_t *cr;
1775
1776 if ((zone = zone_find_by_id(zoneid)) == NULL)
1777 return (NULL);
1778 cr = zone->zone_kcred;
1779 crhold(cr);
1780 zone_rele(zone);
1781 return (cr);
1782 }
1783
1784 static int
1785 zone_lockedmem_kstat_update(kstat_t *ksp, int rw)
1786 {
1787 zone_t *zone = ksp->ks_private;
1788 zone_kstat_t *zk = ksp->ks_data;
1789
1790 if (rw == KSTAT_WRITE)
1791 return (EACCES);
1792
1793 zk->zk_usage.value.ui64 = zone->zone_locked_mem;
1794 zk->zk_value.value.ui64 = zone->zone_locked_mem_ctl;
1795 return (0);
1796 }
1797
1798 static int
1799 zone_nprocs_kstat_update(kstat_t *ksp, int rw)
1800 {
1801 zone_t *zone = ksp->ks_private;
1802 zone_kstat_t *zk = ksp->ks_data;
1803
1804 if (rw == KSTAT_WRITE)
1805 return (EACCES);
1806
1807 zk->zk_usage.value.ui64 = zone->zone_nprocs;
1808 zk->zk_value.value.ui64 = zone->zone_nprocs_ctl;
1809 return (0);
1810 }
1811
1812 static int
1813 zone_swapresv_kstat_update(kstat_t *ksp, int rw)
1814 {
1815 zone_t *zone = ksp->ks_private;
1816 zone_kstat_t *zk = ksp->ks_data;
1817
1818 if (rw == KSTAT_WRITE)
1819 return (EACCES);
1820
1821 zk->zk_usage.value.ui64 = zone->zone_max_swap;
1822 zk->zk_value.value.ui64 = zone->zone_max_swap_ctl;
1823 return (0);
1824 }
1825
1826 static kstat_t *
1827 zone_kstat_create_common(zone_t *zone, char *name,
1828 int (*updatefunc) (kstat_t *, int))
1829 {
1830 kstat_t *ksp;
1831 zone_kstat_t *zk;
1832
1833 ksp = rctl_kstat_create_zone(zone, name, KSTAT_TYPE_NAMED,
1834 sizeof (zone_kstat_t) / sizeof (kstat_named_t),
1835 KSTAT_FLAG_VIRTUAL);
1836
1837 if (ksp == NULL)
1838 return (NULL);
1839
1840 zk = ksp->ks_data = kmem_alloc(sizeof (zone_kstat_t), KM_SLEEP);
1841 ksp->ks_data_size += strlen(zone->zone_name) + 1;
1842 kstat_named_init(&zk->zk_zonename, "zonename", KSTAT_DATA_STRING);
1843 kstat_named_setstr(&zk->zk_zonename, zone->zone_name);
1844 kstat_named_init(&zk->zk_usage, "usage", KSTAT_DATA_UINT64);
1845 kstat_named_init(&zk->zk_value, "value", KSTAT_DATA_UINT64);
1846 ksp->ks_update = updatefunc;
1847 ksp->ks_private = zone;
1848 kstat_install(ksp);
1849 return (ksp);
1850 }
1851
1852
1853 static int
1854 zone_mcap_kstat_update(kstat_t *ksp, int rw)
1855 {
1856 zone_t *zone = ksp->ks_private;
1857 zone_mcap_kstat_t *zmp = ksp->ks_data;
1858
1859 if (rw == KSTAT_WRITE)
1860 return (EACCES);
1861
1862 zmp->zm_pgpgin.value.ui64 = zone->zone_pgpgin;
1863 zmp->zm_anonpgin.value.ui64 = zone->zone_anonpgin;
1864 zmp->zm_execpgin.value.ui64 = zone->zone_execpgin;
1865 zmp->zm_fspgin.value.ui64 = zone->zone_fspgin;
1866 zmp->zm_anon_alloc_fail.value.ui64 = zone->zone_anon_alloc_fail;
1867
1868 return (0);
1869 }
1870
1871 static kstat_t *
1872 zone_mcap_kstat_create(zone_t *zone)
1873 {
1874 kstat_t *ksp;
1875 zone_mcap_kstat_t *zmp;
1876
1877 if ((ksp = kstat_create_zone("memory_cap", zone->zone_id,
1878 zone->zone_name, "zone_memory_cap", KSTAT_TYPE_NAMED,
1879 sizeof (zone_mcap_kstat_t) / sizeof (kstat_named_t),
1880 KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
1881 return (NULL);
1882
1883 if (zone->zone_id != GLOBAL_ZONEID)
1884 kstat_zone_add(ksp, GLOBAL_ZONEID);
1885
1886 zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_mcap_kstat_t), KM_SLEEP);
1887 ksp->ks_data_size += strlen(zone->zone_name) + 1;
1888 ksp->ks_lock = &zone->zone_mcap_lock;
1889 zone->zone_mcap_stats = zmp;
1890
1891 /* The kstat "name" field is not large enough for a full zonename */
1892 kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
1893 kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
1894 kstat_named_init(&zmp->zm_pgpgin, "pgpgin", KSTAT_DATA_UINT64);
1895 kstat_named_init(&zmp->zm_anonpgin, "anonpgin", KSTAT_DATA_UINT64);
1896 kstat_named_init(&zmp->zm_execpgin, "execpgin", KSTAT_DATA_UINT64);
1897 kstat_named_init(&zmp->zm_fspgin, "fspgin", KSTAT_DATA_UINT64);
1898 kstat_named_init(&zmp->zm_anon_alloc_fail, "anon_alloc_fail",
1899 KSTAT_DATA_UINT64);
1900
1901 ksp->ks_update = zone_mcap_kstat_update;
1902 ksp->ks_private = zone;
1903
1904 kstat_install(ksp);
1905 return (ksp);
1906 }
1907
1908 static int
1909 zone_misc_kstat_update(kstat_t *ksp, int rw)
1910 {
1911 zone_t *zone = ksp->ks_private;
1912 zone_misc_kstat_t *zmp = ksp->ks_data;
1913 hrtime_t tmp;
1914
1915 if (rw == KSTAT_WRITE)
1916 return (EACCES);
1917
1918 tmp = zone->zone_utime;
1919 scalehrtime(&tmp);
1920 zmp->zm_utime.value.ui64 = tmp;
1921 tmp = zone->zone_stime;
1922 scalehrtime(&tmp);
1923 zmp->zm_stime.value.ui64 = tmp;
1924 tmp = zone->zone_wtime;
1925 scalehrtime(&tmp);
1926 zmp->zm_wtime.value.ui64 = tmp;
1927
1928 zmp->zm_avenrun1.value.ui32 = zone->zone_avenrun[0];
1929 zmp->zm_avenrun5.value.ui32 = zone->zone_avenrun[1];
1930 zmp->zm_avenrun15.value.ui32 = zone->zone_avenrun[2];
1931
1932 zmp->zm_ffcap.value.ui32 = zone->zone_ffcap;
1933 zmp->zm_ffnoproc.value.ui32 = zone->zone_ffnoproc;
1934 zmp->zm_ffnomem.value.ui32 = zone->zone_ffnomem;
1935 zmp->zm_ffmisc.value.ui32 = zone->zone_ffmisc;
1936
1937 zmp->zm_nested_intp.value.ui32 = zone->zone_nested_intp;
1938
1939 zmp->zm_init_pid.value.ui32 = zone->zone_proc_initpid;
1940 zmp->zm_boot_time.value.ui64 = (uint64_t)zone->zone_boot_time;
1941
1942 return (0);
1943 }
1944
1945 static kstat_t *
1946 zone_misc_kstat_create(zone_t *zone)
1947 {
1948 kstat_t *ksp;
1949 zone_misc_kstat_t *zmp;
1950
1951 if ((ksp = kstat_create_zone("zones", zone->zone_id,
1952 zone->zone_name, "zone_misc", KSTAT_TYPE_NAMED,
1953 sizeof (zone_misc_kstat_t) / sizeof (kstat_named_t),
1954 KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
1955 return (NULL);
1956
1957 if (zone->zone_id != GLOBAL_ZONEID)
1958 kstat_zone_add(ksp, GLOBAL_ZONEID);
1959
1960 zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_misc_kstat_t), KM_SLEEP);
1961 ksp->ks_data_size += strlen(zone->zone_name) + 1;
1962 ksp->ks_lock = &zone->zone_misc_lock;
1963 zone->zone_misc_stats = zmp;
1964
1965 /* The kstat "name" field is not large enough for a full zonename */
1966 kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
1967 kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
1968 kstat_named_init(&zmp->zm_utime, "nsec_user", KSTAT_DATA_UINT64);
1969 kstat_named_init(&zmp->zm_stime, "nsec_sys", KSTAT_DATA_UINT64);
1970 kstat_named_init(&zmp->zm_wtime, "nsec_waitrq", KSTAT_DATA_UINT64);
1971 kstat_named_init(&zmp->zm_avenrun1, "avenrun_1min", KSTAT_DATA_UINT32);
1972 kstat_named_init(&zmp->zm_avenrun5, "avenrun_5min", KSTAT_DATA_UINT32);
1973 kstat_named_init(&zmp->zm_avenrun15, "avenrun_15min",
1974 KSTAT_DATA_UINT32);
1975 kstat_named_init(&zmp->zm_ffcap, "forkfail_cap", KSTAT_DATA_UINT32);
1976 kstat_named_init(&zmp->zm_ffnoproc, "forkfail_noproc",
1977 KSTAT_DATA_UINT32);
1978 kstat_named_init(&zmp->zm_ffnomem, "forkfail_nomem", KSTAT_DATA_UINT32);
1979 kstat_named_init(&zmp->zm_ffmisc, "forkfail_misc", KSTAT_DATA_UINT32);
1980 kstat_named_init(&zmp->zm_nested_intp, "nested_interp",
1981 KSTAT_DATA_UINT32);
1982 kstat_named_init(&zmp->zm_init_pid, "init_pid", KSTAT_DATA_UINT32);
1983 kstat_named_init(&zmp->zm_boot_time, "boot_time", KSTAT_DATA_UINT64);
1984
1985 ksp->ks_update = zone_misc_kstat_update;
1986 ksp->ks_private = zone;
1987
1988 kstat_install(ksp);
1989 return (ksp);
1990 }
1991
1992 static void
1993 zone_kstat_create(zone_t *zone)
1994 {
1995 zone->zone_lockedmem_kstat = zone_kstat_create_common(zone,
1996 "lockedmem", zone_lockedmem_kstat_update);
1997 zone->zone_swapresv_kstat = zone_kstat_create_common(zone,
1998 "swapresv", zone_swapresv_kstat_update);
1999 zone->zone_nprocs_kstat = zone_kstat_create_common(zone,
2000 "nprocs", zone_nprocs_kstat_update);
2001
2002 if ((zone->zone_mcap_ksp = zone_mcap_kstat_create(zone)) == NULL) {
2003 zone->zone_mcap_stats = kmem_zalloc(
2004 sizeof (zone_mcap_kstat_t), KM_SLEEP);
2005 }
2006
2007 if ((zone->zone_misc_ksp = zone_misc_kstat_create(zone)) == NULL) {
2008 zone->zone_misc_stats = kmem_zalloc(
2009 sizeof (zone_misc_kstat_t), KM_SLEEP);
2010 }
2011 }
2012
2013 static void
2014 zone_kstat_delete_common(kstat_t **pkstat, size_t datasz)
2015 {
2016 void *data;
2017
2018 if (*pkstat != NULL) {
2019 data = (*pkstat)->ks_data;
2020 kstat_delete(*pkstat);
2021 kmem_free(data, datasz);
2022 *pkstat = NULL;
2023 }
2024 }
2025
2026 static void
2027 zone_kstat_delete(zone_t *zone)
2028 {
2029 zone_kstat_delete_common(&zone->zone_lockedmem_kstat,
2030 sizeof (zone_kstat_t));
2031 zone_kstat_delete_common(&zone->zone_swapresv_kstat,
2032 sizeof (zone_kstat_t));
2033 zone_kstat_delete_common(&zone->zone_nprocs_kstat,
2034 sizeof (zone_kstat_t));
2035 zone_kstat_delete_common(&zone->zone_mcap_ksp,
2036 sizeof (zone_mcap_kstat_t));
2037 zone_kstat_delete_common(&zone->zone_misc_ksp,
2038 sizeof (zone_misc_kstat_t));
2039 }
2040
2041 /*
2042 * Called very early on in boot to initialize the ZSD list so that
2043 * zone_key_create() can be called before zone_init(). It also initializes
2044 * portions of zone0 which may be used before zone_init() is called. The
2045 * variable "global_zone" will be set when zone0 is fully initialized by
2046 * zone_init().
2047 */
2048 void
2049 zone_zsd_init(void)
2050 {
2051 mutex_init(&zonehash_lock, NULL, MUTEX_DEFAULT, NULL);
2052 mutex_init(&zsd_key_lock, NULL, MUTEX_DEFAULT, NULL);
2053 list_create(&zsd_registered_keys, sizeof (struct zsd_entry),
2054 offsetof(struct zsd_entry, zsd_linkage));
2055 list_create(&zone_active, sizeof (zone_t),
2056 offsetof(zone_t, zone_linkage));
2057 list_create(&zone_deathrow, sizeof (zone_t),
2058 offsetof(zone_t, zone_linkage));
2059
2060 mutex_init(&zone0.zone_lock, NULL, MUTEX_DEFAULT, NULL);
2061 mutex_init(&zone0.zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
2062 mutex_init(&zone0.zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
2063 zone0.zone_shares = 1;
2064 zone0.zone_nlwps = 0;
2065 zone0.zone_nlwps_ctl = INT_MAX;
2066 zone0.zone_nprocs = 0;
2067 zone0.zone_nprocs_ctl = INT_MAX;
2068 zone0.zone_locked_mem = 0;
2069 zone0.zone_locked_mem_ctl = UINT64_MAX;
2070 ASSERT(zone0.zone_max_swap == 0);
2071 zone0.zone_max_swap_ctl = UINT64_MAX;
2072 zone0.zone_max_lofi = 0;
2073 zone0.zone_max_lofi_ctl = UINT64_MAX;
2074 zone0.zone_shmmax = 0;
2075 zone0.zone_ipc.ipcq_shmmni = 0;
2076 zone0.zone_ipc.ipcq_semmni = 0;
2077 zone0.zone_ipc.ipcq_msgmni = 0;
2078 zone0.zone_name = GLOBAL_ZONENAME;
2079 zone0.zone_nodename = utsname.nodename;
2080 zone0.zone_domain = srpc_domain;
2081 zone0.zone_hostid = HW_INVALID_HOSTID;
2082 zone0.zone_fs_allowed = NULL;
2083 psecflags_default(&zone0.zone_secflags);
2084 zone0.zone_ref = 1;
2085 zone0.zone_id = GLOBAL_ZONEID;
2086 zone0.zone_status = ZONE_IS_RUNNING;
2087 zone0.zone_rootpath = "/";
2088 zone0.zone_rootpathlen = 2;
2089 zone0.zone_psetid = ZONE_PS_INVAL;
2090 zone0.zone_ncpus = 0;
2091 zone0.zone_ncpus_online = 0;
2092 zone0.zone_proc_initpid = 1;
2093 zone0.zone_initname = initname;
2094 zone0.zone_lockedmem_kstat = NULL;
2095 zone0.zone_swapresv_kstat = NULL;
2096 zone0.zone_nprocs_kstat = NULL;
2097
2098 zone0.zone_stime = 0;
2099 zone0.zone_utime = 0;
2100 zone0.zone_wtime = 0;
2101
2102 list_create(&zone0.zone_ref_list, sizeof (zone_ref_t),
2103 offsetof(zone_ref_t, zref_linkage));
2104 list_create(&zone0.zone_zsd, sizeof (struct zsd_entry),
2105 offsetof(struct zsd_entry, zsd_linkage));
2106 list_insert_head(&zone_active, &zone0);
2107
2108 /*
2109 * The root filesystem is not mounted yet, so zone_rootvp cannot be set
2110 * to anything meaningful. It is assigned to be 'rootdir' in
2111 * vfs_mountroot().
2112 */
2113 zone0.zone_rootvp = NULL;
2114 zone0.zone_vfslist = NULL;
2115 zone0.zone_bootargs = initargs;
2116 zone0.zone_privset = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
2117 /*
2118 * The global zone has all privileges
2119 */
2120 priv_fillset(zone0.zone_privset);
2121 /*
2122 * Add p0 to the global zone
2123 */
2124 zone0.zone_zsched = &p0;
2125 p0.p_zone = &zone0;
2126 }
2127
2128 /*
2129 * Compute a hash value based on the contents of the label and the DOI. The
2130 * hash algorithm is somewhat arbitrary, but is based on the observation that
2131 * humans will likely pick labels that differ by amounts that work out to be
2132 * multiples of the number of hash chains, and thus stirring in some primes
2133 * should help.
2134 */
2135 static uint_t
2136 hash_bylabel(void *hdata, mod_hash_key_t key)
2137 {
2138 const ts_label_t *lab = (ts_label_t *)key;
2139 const uint32_t *up, *ue;
2140 uint_t hash;
2141 int i;
2142
2143 _NOTE(ARGUNUSED(hdata));
2144
2145 hash = lab->tsl_doi + (lab->tsl_doi << 1);
2146 /* we depend on alignment of label, but not representation */
2147 up = (const uint32_t *)&lab->tsl_label;
2148 ue = up + sizeof (lab->tsl_label) / sizeof (*up);
2149 i = 1;
2150 while (up < ue) {
2151 /* using 2^n + 1, 1 <= n <= 16 as source of many primes */
2152 hash += *up + (*up << ((i % 16) + 1));
2153 up++;
2154 i++;
2155 }
2156 return (hash);
2157 }
2158
2159 /*
2160 * All that mod_hash cares about here is zero (equal) versus non-zero (not
2161 * equal). This may need to be changed if less than / greater than is ever
2162 * needed.
2163 */
2164 static int
2165 hash_labelkey_cmp(mod_hash_key_t key1, mod_hash_key_t key2)
2166 {
2167 ts_label_t *lab1 = (ts_label_t *)key1;
2168 ts_label_t *lab2 = (ts_label_t *)key2;
2169
2170 return (label_equal(lab1, lab2) ? 0 : 1);
2171 }
2172
2173 /*
2174 * Called by main() to initialize the zones framework.
2175 */
2176 void
2177 zone_init(void)
2178 {
2179 rctl_dict_entry_t *rde;
2180 rctl_val_t *dval;
2181 rctl_set_t *set;
2182 rctl_alloc_gp_t *gp;
2183 rctl_entity_p_t e;
2184 int res;
2185
2186 ASSERT(curproc == &p0);
2187
2188 /*
2189 * Create ID space for zone IDs. ID 0 is reserved for the
2190 * global zone.
2191 */
2192 zoneid_space = id_space_create("zoneid_space", 1, MAX_ZONEID);
2193
2194 /*
2195 * Initialize generic zone resource controls, if any.
2196 */
2197 rc_zone_cpu_shares = rctl_register("zone.cpu-shares",
2198 RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_NEVER |
2199 RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT | RCTL_GLOBAL_SYSLOG_NEVER,
2200 FSS_MAXSHARES, FSS_MAXSHARES, &zone_cpu_shares_ops);
2201
2202 rc_zone_cpu_cap = rctl_register("zone.cpu-cap",
2203 RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_ALWAYS |
2204 RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |RCTL_GLOBAL_SYSLOG_NEVER |
2205 RCTL_GLOBAL_INFINITE,
2206 MAXCAP, MAXCAP, &zone_cpu_cap_ops);
2207
2208 rc_zone_nlwps = rctl_register("zone.max-lwps", RCENTITY_ZONE,
2209 RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
2210 INT_MAX, INT_MAX, &zone_lwps_ops);
2211
2212 rc_zone_nprocs = rctl_register("zone.max-processes", RCENTITY_ZONE,
2213 RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
2214 INT_MAX, INT_MAX, &zone_procs_ops);
2215
2216 /*
2217 * System V IPC resource controls
2218 */
2219 rc_zone_msgmni = rctl_register("zone.max-msg-ids",
2220 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2221 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_msgmni_ops);
2222
2223 rc_zone_semmni = rctl_register("zone.max-sem-ids",
2224 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2225 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_semmni_ops);
2226
2227 rc_zone_shmmni = rctl_register("zone.max-shm-ids",
2228 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2229 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_shmmni_ops);
2230
2231 rc_zone_shmmax = rctl_register("zone.max-shm-memory",
2232 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2233 RCTL_GLOBAL_BYTES, UINT64_MAX, UINT64_MAX, &zone_shmmax_ops);
2234
2235 /*
2236 * Create a rctl_val with PRIVILEGED, NOACTION, value = 1. Then attach
2237 * this at the head of the rctl_dict_entry for ``zone.cpu-shares''.
2238 */
2239 dval = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
2240 bzero(dval, sizeof (rctl_val_t));
2241 dval->rcv_value = 1;
2242 dval->rcv_privilege = RCPRIV_PRIVILEGED;
2243 dval->rcv_flagaction = RCTL_LOCAL_NOACTION;
2244 dval->rcv_action_recip_pid = -1;
2245
2246 rde = rctl_dict_lookup("zone.cpu-shares");
2247 (void) rctl_val_list_insert(&rde->rcd_default_value, dval);
2248
2249 rc_zone_locked_mem = rctl_register("zone.max-locked-memory",
2250 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
2251 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2252 &zone_locked_mem_ops);
2253
2254 rc_zone_max_swap = rctl_register("zone.max-swap",
2255 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
2256 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2257 &zone_max_swap_ops);
2258
2259 rc_zone_max_lofi = rctl_register("zone.max-lofi",
2260 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |
2261 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2262 &zone_max_lofi_ops);
2263
2264 /*
2265 * Initialize the ``global zone''.
2266 */
2267 set = rctl_set_create();
2268 gp = rctl_set_init_prealloc(RCENTITY_ZONE);
2269 mutex_enter(&p0.p_lock);
2270 e.rcep_p.zone = &zone0;
2271 e.rcep_t = RCENTITY_ZONE;
2272 zone0.zone_rctls = rctl_set_init(RCENTITY_ZONE, &p0, &e, set,
2273 gp);
2274
2275 zone0.zone_nlwps = p0.p_lwpcnt;
2276 zone0.zone_nprocs = 1;
2277 zone0.zone_ntasks = 1;
2278 mutex_exit(&p0.p_lock);
2279 zone0.zone_restart_init = B_TRUE;
2280 zone0.zone_brand = &native_brand;
2281 rctl_prealloc_destroy(gp);
2282 /*
2283 * pool_default hasn't been initialized yet, so we let pool_init()
2284 * take care of making sure the global zone is in the default pool.
2285 */
2286
2287 /*
2288 * Initialize global zone kstats
2289 */
2290 zone_kstat_create(&zone0);
2291
2292 /*
2293 * Initialize zone label.
2294 * mlp are initialized when tnzonecfg is loaded.
2295 */
2296 zone0.zone_slabel = l_admin_low;
2297 rw_init(&zone0.zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
2298 label_hold(l_admin_low);
2299
2300 /*
2301 * Initialise the lock for the database structure used by mntfs.
2302 */
2303 rw_init(&zone0.zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL);
2304
2305 mutex_enter(&zonehash_lock);
2306 zone_uniqid(&zone0);
2307 ASSERT(zone0.zone_uniqid == GLOBAL_ZONEUNIQID);
2308
2309 zonehashbyid = mod_hash_create_idhash("zone_by_id", zone_hash_size,
2310 mod_hash_null_valdtor);
2311 zonehashbyname = mod_hash_create_strhash("zone_by_name",
2312 zone_hash_size, mod_hash_null_valdtor);
2313 /*
2314 * maintain zonehashbylabel only for labeled systems
2315 */
2316 if (is_system_labeled())
2317 zonehashbylabel = mod_hash_create_extended("zone_by_label",
2318 zone_hash_size, mod_hash_null_keydtor,
2319 mod_hash_null_valdtor, hash_bylabel, NULL,
2320 hash_labelkey_cmp, KM_SLEEP);
2321 zonecount = 1;
2322
2323 (void) mod_hash_insert(zonehashbyid, (mod_hash_key_t)GLOBAL_ZONEID,
2324 (mod_hash_val_t)&zone0);
2325 (void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)zone0.zone_name,
2326 (mod_hash_val_t)&zone0);
2327 if (is_system_labeled()) {
2328 zone0.zone_flags |= ZF_HASHED_LABEL;
2329 (void) mod_hash_insert(zonehashbylabel,
2330 (mod_hash_key_t)zone0.zone_slabel, (mod_hash_val_t)&zone0);
2331 }
2332 mutex_exit(&zonehash_lock);
2333
2334 /*
2335 * We avoid setting zone_kcred until now, since kcred is initialized
2336 * sometime after zone_zsd_init() and before zone_init().
2337 */
2338 zone0.zone_kcred = kcred;
2339 /*
2340 * The global zone is fully initialized (except for zone_rootvp which
2341 * will be set when the root filesystem is mounted).
2342 */
2343 global_zone = &zone0;
2344
2345 /*
2346 * Setup an event channel to send zone status change notifications on
2347 */
2348 res = sysevent_evc_bind(ZONE_EVENT_CHANNEL, &zone_event_chan,
2349 EVCH_CREAT);
2350
2351 if (res)
2352 panic("Sysevent_evc_bind failed during zone setup.\n");
2353
2354 }
2355
2356 static void
2357 zone_free(zone_t *zone)
2358 {
2359 ASSERT(zone != global_zone);
2360 ASSERT(zone->zone_ntasks == 0);
2361 ASSERT(zone->zone_nlwps == 0);
2362 ASSERT(zone->zone_nprocs == 0);
2363 ASSERT(zone->zone_cred_ref == 0);
2364 ASSERT(zone->zone_kcred == NULL);
2365 ASSERT(zone_status_get(zone) == ZONE_IS_DEAD ||
2366 zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
2367 ASSERT(list_is_empty(&zone->zone_ref_list));
2368
2369 /*
2370 * Remove any zone caps.
2371 */
2372 cpucaps_zone_remove(zone);
2373
2374 ASSERT(zone->zone_cpucap == NULL);
2375
2376 /* remove from deathrow list */
2377 if (zone_status_get(zone) == ZONE_IS_DEAD) {
2378 ASSERT(zone->zone_ref == 0);
2379 mutex_enter(&zone_deathrow_lock);
2380 list_remove(&zone_deathrow, zone);
2381 mutex_exit(&zone_deathrow_lock);
2382 }
2383
2384 list_destroy(&zone->zone_ref_list);
2385 zone_free_zsd(zone);
2386 zone_free_datasets(zone);
2387 list_destroy(&zone->zone_dl_list);
2388
2389 if (zone->zone_rootvp != NULL)
2390 VN_RELE(zone->zone_rootvp);
2391 if (zone->zone_rootpath)
2392 kmem_free(zone->zone_rootpath, zone->zone_rootpathlen);
2393 if (zone->zone_name != NULL)
2394 kmem_free(zone->zone_name, ZONENAME_MAX);
2395 if (zone->zone_slabel != NULL)
2396 label_rele(zone->zone_slabel);
2397 if (zone->zone_nodename != NULL)
2398 kmem_free(zone->zone_nodename, _SYS_NMLN);
2399 if (zone->zone_domain != NULL)
2400 kmem_free(zone->zone_domain, _SYS_NMLN);
2401 if (zone->zone_privset != NULL)
2402 kmem_free(zone->zone_privset, sizeof (priv_set_t));
2403 if (zone->zone_rctls != NULL)
2404 rctl_set_free(zone->zone_rctls);
2405 if (zone->zone_bootargs != NULL)
2406 strfree(zone->zone_bootargs);
2407 if (zone->zone_initname != NULL)
2408 strfree(zone->zone_initname);
2409 if (zone->zone_fs_allowed != NULL)
2410 strfree(zone->zone_fs_allowed);
2411 if (zone->zone_pfexecd != NULL)
2412 klpd_freelist(&zone->zone_pfexecd);
2413 id_free(zoneid_space, zone->zone_id);
2414 mutex_destroy(&zone->zone_lock);
2415 cv_destroy(&zone->zone_cv);
2416 rw_destroy(&zone->zone_mlps.mlpl_rwlock);
2417 rw_destroy(&zone->zone_mntfs_db_lock);
2418 kmem_free(zone, sizeof (zone_t));
2419 }
2420
2421 /*
2422 * See block comment at the top of this file for information about zone
2423 * status values.
2424 */
2425 /*
2426 * Convenience function for setting zone status.
2427 */
2428 static void
2429 zone_status_set(zone_t *zone, zone_status_t status)
2430 {
2431
2432 nvlist_t *nvl = NULL;
2433 ASSERT(MUTEX_HELD(&zone_status_lock));
2434 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE &&
2435 status >= zone_status_get(zone));
2436
2437 if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) ||
2438 nvlist_add_string(nvl, ZONE_CB_NAME, zone->zone_name) ||
2439 nvlist_add_string(nvl, ZONE_CB_NEWSTATE,
2440 zone_status_table[status]) ||
2441 nvlist_add_string(nvl, ZONE_CB_OLDSTATE,
2442 zone_status_table[zone->zone_status]) ||
2443 nvlist_add_int32(nvl, ZONE_CB_ZONEID, zone->zone_id) ||
2444 nvlist_add_uint64(nvl, ZONE_CB_TIMESTAMP, (uint64_t)gethrtime()) ||
2445 sysevent_evc_publish(zone_event_chan, ZONE_EVENT_STATUS_CLASS,
2446 ZONE_EVENT_STATUS_SUBCLASS, "sun.com", "kernel", nvl, EVCH_SLEEP)) {
2447 #ifdef DEBUG
2448 (void) printf(
2449 "Failed to allocate and send zone state change event.\n");
2450 #endif
2451 }
2452 nvlist_free(nvl);
2453
2454 zone->zone_status = status;
2455
2456 cv_broadcast(&zone->zone_cv);
2457 }
2458
2459 /*
2460 * Public function to retrieve the zone status. The zone status may
2461 * change after it is retrieved.
2462 */
2463 zone_status_t
2464 zone_status_get(zone_t *zone)
2465 {
2466 return (zone->zone_status);
2467 }
2468
2469 static int
2470 zone_set_bootargs(zone_t *zone, const char *zone_bootargs)
2471 {
2472 char *buf = kmem_zalloc(BOOTARGS_MAX, KM_SLEEP);
2473 int err = 0;
2474
2475 ASSERT(zone != global_zone);
2476 if ((err = copyinstr(zone_bootargs, buf, BOOTARGS_MAX, NULL)) != 0)
2477 goto done; /* EFAULT or ENAMETOOLONG */
2478
2479 if (zone->zone_bootargs != NULL)
2480 strfree(zone->zone_bootargs);
2481
2482 zone->zone_bootargs = strdup(buf);
2483
2484 done:
2485 kmem_free(buf, BOOTARGS_MAX);
2486 return (err);
2487 }
2488
2489 static int
2490 zone_set_brand(zone_t *zone, const char *brand)
2491 {
2492 struct brand_attr *attrp;
2493 brand_t *bp;
2494
2495 attrp = kmem_alloc(sizeof (struct brand_attr), KM_SLEEP);
2496 if (copyin(brand, attrp, sizeof (struct brand_attr)) != 0) {
2497 kmem_free(attrp, sizeof (struct brand_attr));
2498 return (EFAULT);
2499 }
2500
2501 bp = brand_register_zone(attrp);
2502 kmem_free(attrp, sizeof (struct brand_attr));
2503 if (bp == NULL)
2504 return (EINVAL);
2505
2506 /*
2507 * This is the only place where a zone can change it's brand.
2508 * We already need to hold zone_status_lock to check the zone
2509 * status, so we'll just use that lock to serialize zone
2510 * branding requests as well.
2511 */
2512 mutex_enter(&zone_status_lock);
2513
2514 /* Re-Branding is not allowed and the zone can't be booted yet */
2515 if ((ZONE_IS_BRANDED(zone)) ||
2516 (zone_status_get(zone) >= ZONE_IS_BOOTING)) {
2517 mutex_exit(&zone_status_lock);
2518 brand_unregister_zone(bp);
2519 return (EINVAL);
2520 }
2521
2522 /* set up the brand specific data */
2523 zone->zone_brand = bp;
2524 ZBROP(zone)->b_init_brand_data(zone);
2525
2526 mutex_exit(&zone_status_lock);
2527 return (0);
2528 }
2529
2530 static int
2531 zone_set_secflags(zone_t *zone, const psecflags_t *zone_secflags)
2532 {
2533 int err = 0;
2534 psecflags_t psf;
2535
2536 ASSERT(zone != global_zone);
2537
2538 if ((err = copyin(zone_secflags, &psf, sizeof (psf))) != 0)
2539 return (err);
2540
2541 if (zone_status_get(zone) > ZONE_IS_READY)
2542 return (EINVAL);
2543
2544 if (!psecflags_validate(&psf))
2545 return (EINVAL);
2546
2547 (void) memcpy(&zone->zone_secflags, &psf, sizeof (psf));
2548
2549 /* Set security flags on the zone's zsched */
2550 (void) memcpy(&zone->zone_zsched->p_secflags, &zone->zone_secflags,
2551 sizeof (zone->zone_zsched->p_secflags));
2552
2553 return (0);
2554 }
2555
2556 static int
2557 zone_set_fs_allowed(zone_t *zone, const char *zone_fs_allowed)
2558 {
2559 char *buf = kmem_zalloc(ZONE_FS_ALLOWED_MAX, KM_SLEEP);
2560 int err = 0;
2561
2562 ASSERT(zone != global_zone);
2563 if ((err = copyinstr(zone_fs_allowed, buf,
2564 ZONE_FS_ALLOWED_MAX, NULL)) != 0)
2565 goto done;
2566
2567 if (zone->zone_fs_allowed != NULL)
2568 strfree(zone->zone_fs_allowed);
2569
2570 zone->zone_fs_allowed = strdup(buf);
2571
2572 done:
2573 kmem_free(buf, ZONE_FS_ALLOWED_MAX);
2574 return (err);
2575 }
2576
2577 static int
2578 zone_set_initname(zone_t *zone, const char *zone_initname)
2579 {
2580 char initname[INITNAME_SZ];
2581 size_t len;
2582 int err = 0;
2583
2584 ASSERT(zone != global_zone);
2585 if ((err = copyinstr(zone_initname, initname, INITNAME_SZ, &len)) != 0)
2586 return (err); /* EFAULT or ENAMETOOLONG */
2587
2588 if (zone->zone_initname != NULL)
2589 strfree(zone->zone_initname);
2590
2591 zone->zone_initname = kmem_alloc(strlen(initname) + 1, KM_SLEEP);
2592 (void) strcpy(zone->zone_initname, initname);
2593 return (0);
2594 }
2595
2596 static int
2597 zone_set_phys_mcap(zone_t *zone, const uint64_t *zone_mcap)
2598 {
2599 uint64_t mcap;
2600 int err = 0;
2601
2602 if ((err = copyin(zone_mcap, &mcap, sizeof (uint64_t))) == 0)
2603 zone->zone_phys_mcap = mcap;
2604
2605 return (err);
2606 }
2607
2608 static int
2609 zone_set_sched_class(zone_t *zone, const char *new_class)
2610 {
2611 char sched_class[PC_CLNMSZ];
2612 id_t classid;
2613 int err;
2614
2615 ASSERT(zone != global_zone);
2616 if ((err = copyinstr(new_class, sched_class, PC_CLNMSZ, NULL)) != 0)
2617 return (err); /* EFAULT or ENAMETOOLONG */
2618
2619 if (getcid(sched_class, &classid) != 0 || CLASS_KERNEL(classid))
2620 return (set_errno(EINVAL));
2621 zone->zone_defaultcid = classid;
2622 ASSERT(zone->zone_defaultcid > 0 &&
2623 zone->zone_defaultcid < loaded_classes);
2624
2625 return (0);
2626 }
2627
2628 /*
2629 * Block indefinitely waiting for (zone_status >= status)
2630 */
2631 void
2632 zone_status_wait(zone_t *zone, zone_status_t status)
2633 {
2634 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2635
2636 mutex_enter(&zone_status_lock);
2637 while (zone->zone_status < status) {
2638 cv_wait(&zone->zone_cv, &zone_status_lock);
2639 }
2640 mutex_exit(&zone_status_lock);
2641 }
2642
2643 /*
2644 * Private CPR-safe version of zone_status_wait().
2645 */
2646 static void
2647 zone_status_wait_cpr(zone_t *zone, zone_status_t status, char *str)
2648 {
2649 callb_cpr_t cprinfo;
2650
2651 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2652
2653 CALLB_CPR_INIT(&cprinfo, &zone_status_lock, callb_generic_cpr,
2654 str);
2655 mutex_enter(&zone_status_lock);
2656 while (zone->zone_status < status) {
2657 CALLB_CPR_SAFE_BEGIN(&cprinfo);
2658 cv_wait(&zone->zone_cv, &zone_status_lock);
2659 CALLB_CPR_SAFE_END(&cprinfo, &zone_status_lock);
2660 }
2661 /*
2662 * zone_status_lock is implicitly released by the following.
2663 */
2664 CALLB_CPR_EXIT(&cprinfo);
2665 }
2666
2667 /*
2668 * Block until zone enters requested state or signal is received. Return (0)
2669 * if signaled, non-zero otherwise.
2670 */
2671 int
2672 zone_status_wait_sig(zone_t *zone, zone_status_t status)
2673 {
2674 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2675
2676 mutex_enter(&zone_status_lock);
2677 while (zone->zone_status < status) {
2678 if (!cv_wait_sig(&zone->zone_cv, &zone_status_lock)) {
2679 mutex_exit(&zone_status_lock);
2680 return (0);
2681 }
2682 }
2683 mutex_exit(&zone_status_lock);
2684 return (1);
2685 }
2686
2687 /*
2688 * Block until the zone enters the requested state or the timeout expires,
2689 * whichever happens first. Return (-1) if operation timed out, time remaining
2690 * otherwise.
2691 */
2692 clock_t
2693 zone_status_timedwait(zone_t *zone, clock_t tim, zone_status_t status)
2694 {
2695 clock_t timeleft = 0;
2696
2697 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2698
2699 mutex_enter(&zone_status_lock);
2700 while (zone->zone_status < status && timeleft != -1) {
2701 timeleft = cv_timedwait(&zone->zone_cv, &zone_status_lock, tim);
2702 }
2703 mutex_exit(&zone_status_lock);
2704 return (timeleft);
2705 }
2706
2707 /*
2708 * Block until the zone enters the requested state, the current process is
2709 * signaled, or the timeout expires, whichever happens first. Return (-1) if
2710 * operation timed out, 0 if signaled, time remaining otherwise.
2711 */
2712 clock_t
2713 zone_status_timedwait_sig(zone_t *zone, clock_t tim, zone_status_t status)
2714 {
2715 clock_t timeleft = tim - ddi_get_lbolt();
2716
2717 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2718
2719 mutex_enter(&zone_status_lock);
2720 while (zone->zone_status < status) {
2721 timeleft = cv_timedwait_sig(&zone->zone_cv, &zone_status_lock,
2722 tim);
2723 if (timeleft <= 0)
2724 break;
2725 }
2726 mutex_exit(&zone_status_lock);
2727 return (timeleft);
2728 }
2729
2730 /*
2731 * Zones have two reference counts: one for references from credential
2732 * structures (zone_cred_ref), and one (zone_ref) for everything else.
2733 * This is so we can allow a zone to be rebooted while there are still
2734 * outstanding cred references, since certain drivers cache dblks (which
2735 * implicitly results in cached creds). We wait for zone_ref to drop to
2736 * 0 (actually 1), but not zone_cred_ref. The zone structure itself is
2737 * later freed when the zone_cred_ref drops to 0, though nothing other
2738 * than the zone id and privilege set should be accessed once the zone
2739 * is "dead".
2740 *
2741 * A debugging flag, zone_wait_for_cred, can be set to a non-zero value
2742 * to force halt/reboot to block waiting for the zone_cred_ref to drop
2743 * to 0. This can be useful to flush out other sources of cached creds
2744 * that may be less innocuous than the driver case.
2745 *
2746 * Zones also provide a tracked reference counting mechanism in which zone
2747 * references are represented by "crumbs" (zone_ref structures). Crumbs help
2748 * debuggers determine the sources of leaked zone references. See
2749 * zone_hold_ref() and zone_rele_ref() below for more information.
2750 */
2751
2752 int zone_wait_for_cred = 0;
2753
2754 static void
2755 zone_hold_locked(zone_t *z)
2756 {
2757 ASSERT(MUTEX_HELD(&z->zone_lock));
2758 z->zone_ref++;
2759 ASSERT(z->zone_ref != 0);
2760 }
2761
2762 /*
2763 * Increment the specified zone's reference count. The zone's zone_t structure
2764 * will not be freed as long as the zone's reference count is nonzero.
2765 * Decrement the zone's reference count via zone_rele().
2766 *
2767 * NOTE: This function should only be used to hold zones for short periods of
2768 * time. Use zone_hold_ref() if the zone must be held for a long time.
2769 */
2770 void
2771 zone_hold(zone_t *z)
2772 {
2773 mutex_enter(&z->zone_lock);
2774 zone_hold_locked(z);
2775 mutex_exit(&z->zone_lock);
2776 }
2777
2778 /*
2779 * If the non-cred ref count drops to 1 and either the cred ref count
2780 * is 0 or we aren't waiting for cred references, the zone is ready to
2781 * be destroyed.
2782 */
2783 #define ZONE_IS_UNREF(zone) ((zone)->zone_ref == 1 && \
2784 (!zone_wait_for_cred || (zone)->zone_cred_ref == 0))
2785
2786 /*
2787 * Common zone reference release function invoked by zone_rele() and
2788 * zone_rele_ref(). If subsys is ZONE_REF_NUM_SUBSYS, then the specified
2789 * zone's subsystem-specific reference counters are not affected by the
2790 * release. If ref is not NULL, then the zone_ref_t to which it refers is
2791 * removed from the specified zone's reference list. ref must be non-NULL iff
2792 * subsys is not ZONE_REF_NUM_SUBSYS.
2793 */
2794 static void
2795 zone_rele_common(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
2796 {
2797 boolean_t wakeup;
2798
2799 mutex_enter(&z->zone_lock);
2800 ASSERT(z->zone_ref != 0);
2801 z->zone_ref--;
2802 if (subsys != ZONE_REF_NUM_SUBSYS) {
2803 ASSERT(z->zone_subsys_ref[subsys] != 0);
2804 z->zone_subsys_ref[subsys]--;
2805 list_remove(&z->zone_ref_list, ref);
2806 }
2807 if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
2808 /* no more refs, free the structure */
2809 mutex_exit(&z->zone_lock);
2810 zone_free(z);
2811 return;
2812 }
2813 /* signal zone_destroy so the zone can finish halting */
2814 wakeup = (ZONE_IS_UNREF(z) && zone_status_get(z) >= ZONE_IS_DEAD);
2815 mutex_exit(&z->zone_lock);
2816
2817 if (wakeup) {
2818 /*
2819 * Grabbing zonehash_lock here effectively synchronizes with
2820 * zone_destroy() to avoid missed signals.
2821 */
2822 mutex_enter(&zonehash_lock);
2823 cv_broadcast(&zone_destroy_cv);
2824 mutex_exit(&zonehash_lock);
2825 }
2826 }
2827
2828 /*
2829 * Decrement the specified zone's reference count. The specified zone will
2830 * cease to exist after this function returns if the reference count drops to
2831 * zero. This function should be paired with zone_hold().
2832 */
2833 void
2834 zone_rele(zone_t *z)
2835 {
2836 zone_rele_common(z, NULL, ZONE_REF_NUM_SUBSYS);
2837 }
2838
2839 /*
2840 * Initialize a zone reference structure. This function must be invoked for
2841 * a reference structure before the structure is passed to zone_hold_ref().
2842 */
2843 void
2844 zone_init_ref(zone_ref_t *ref)
2845 {
2846 ref->zref_zone = NULL;
2847 list_link_init(&ref->zref_linkage);
2848 }
2849
2850 /*
2851 * Acquire a reference to zone z. The caller must specify the
2852 * zone_ref_subsys_t constant associated with its subsystem. The specified
2853 * zone_ref_t structure will represent a reference to the specified zone. Use
2854 * zone_rele_ref() to release the reference.
2855 *
2856 * The referenced zone_t structure will not be freed as long as the zone_t's
2857 * zone_status field is not ZONE_IS_DEAD and the zone has outstanding
2858 * references.
2859 *
2860 * NOTE: The zone_ref_t structure must be initialized before it is used.
2861 * See zone_init_ref() above.
2862 */
2863 void
2864 zone_hold_ref(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
2865 {
2866 ASSERT(subsys >= 0 && subsys < ZONE_REF_NUM_SUBSYS);
2867
2868 /*
2869 * Prevent consumers from reusing a reference structure before
2870 * releasing it.
2871 */
2872 VERIFY(ref->zref_zone == NULL);
2873
2874 ref->zref_zone = z;
2875 mutex_enter(&z->zone_lock);
2876 zone_hold_locked(z);
2877 z->zone_subsys_ref[subsys]++;
2878 ASSERT(z->zone_subsys_ref[subsys] != 0);
2879 list_insert_head(&z->zone_ref_list, ref);
2880 mutex_exit(&z->zone_lock);
2881 }
2882
2883 /*
2884 * Release the zone reference represented by the specified zone_ref_t.
2885 * The reference is invalid after it's released; however, the zone_ref_t
2886 * structure can be reused without having to invoke zone_init_ref().
2887 * subsys should be the same value that was passed to zone_hold_ref()
2888 * when the reference was acquired.
2889 */
2890 void
2891 zone_rele_ref(zone_ref_t *ref, zone_ref_subsys_t subsys)
2892 {
2893 zone_rele_common(ref->zref_zone, ref, subsys);
2894
2895 /*
2896 * Set the zone_ref_t's zref_zone field to NULL to generate panics
2897 * when consumers dereference the reference. This helps us catch
2898 * consumers who use released references. Furthermore, this lets
2899 * consumers reuse the zone_ref_t structure without having to
2900 * invoke zone_init_ref().
2901 */
2902 ref->zref_zone = NULL;
2903 }
2904
2905 void
2906 zone_cred_hold(zone_t *z)
2907 {
2908 mutex_enter(&z->zone_lock);
2909 z->zone_cred_ref++;
2910 ASSERT(z->zone_cred_ref != 0);
2911 mutex_exit(&z->zone_lock);
2912 }
2913
2914 void
2915 zone_cred_rele(zone_t *z)
2916 {
2917 boolean_t wakeup;
2918
2919 mutex_enter(&z->zone_lock);
2920 ASSERT(z->zone_cred_ref != 0);
2921 z->zone_cred_ref--;
2922 if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
2923 /* no more refs, free the structure */
2924 mutex_exit(&z->zone_lock);
2925 zone_free(z);
2926 return;
2927 }
2928 /*
2929 * If zone_destroy is waiting for the cred references to drain
2930 * out, and they have, signal it.
2931 */
2932 wakeup = (zone_wait_for_cred && ZONE_IS_UNREF(z) &&
2933 zone_status_get(z) >= ZONE_IS_DEAD);
2934 mutex_exit(&z->zone_lock);
2935
2936 if (wakeup) {
2937 /*
2938 * Grabbing zonehash_lock here effectively synchronizes with
2939 * zone_destroy() to avoid missed signals.
2940 */
2941 mutex_enter(&zonehash_lock);
2942 cv_broadcast(&zone_destroy_cv);
2943 mutex_exit(&zonehash_lock);
2944 }
2945 }
2946
2947 void
2948 zone_task_hold(zone_t *z)
2949 {
2950 mutex_enter(&z->zone_lock);
2951 z->zone_ntasks++;
2952 ASSERT(z->zone_ntasks != 0);
2953 mutex_exit(&z->zone_lock);
2954 }
2955
2956 void
2957 zone_task_rele(zone_t *zone)
2958 {
2959 uint_t refcnt;
2960
2961 mutex_enter(&zone->zone_lock);
2962 ASSERT(zone->zone_ntasks != 0);
2963 refcnt = --zone->zone_ntasks;
2964 if (refcnt > 1) { /* Common case */
2965 mutex_exit(&zone->zone_lock);
2966 return;
2967 }
2968 zone_hold_locked(zone); /* so we can use the zone_t later */
2969 mutex_exit(&zone->zone_lock);
2970 if (refcnt == 1) {
2971 /*
2972 * See if the zone is shutting down.
2973 */
2974 mutex_enter(&zone_status_lock);
2975 if (zone_status_get(zone) != ZONE_IS_SHUTTING_DOWN) {
2976 goto out;
2977 }
2978
2979 /*
2980 * Make sure the ntasks didn't change since we
2981 * dropped zone_lock.
2982 */
2983 mutex_enter(&zone->zone_lock);
2984 if (refcnt != zone->zone_ntasks) {
2985 mutex_exit(&zone->zone_lock);
2986 goto out;
2987 }
2988 mutex_exit(&zone->zone_lock);
2989
2990 /*
2991 * No more user processes in the zone. The zone is empty.
2992 */
2993 zone_status_set(zone, ZONE_IS_EMPTY);
2994 goto out;
2995 }
2996
2997 ASSERT(refcnt == 0);
2998 /*
2999 * zsched has exited; the zone is dead.
3000 */
3001 zone->zone_zsched = NULL; /* paranoia */
3002 mutex_enter(&zone_status_lock);
3003 zone_status_set(zone, ZONE_IS_DEAD);
3004 out:
3005 mutex_exit(&zone_status_lock);
3006 zone_rele(zone);
3007 }
3008
3009 zoneid_t
3010 getzoneid(void)
3011 {
3012 return (curproc->p_zone->zone_id);
3013 }
3014
3015 /*
3016 * Internal versions of zone_find_by_*(). These don't zone_hold() or
3017 * check the validity of a zone's state.
3018 */
3019 static zone_t *
3020 zone_find_all_by_id(zoneid_t zoneid)
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(zonehashbyid,
3028 (mod_hash_key_t)(uintptr_t)zoneid, &hv) == 0)
3029 zone = (zone_t *)hv;
3030 return (zone);
3031 }
3032
3033 static zone_t *
3034 zone_find_all_by_label(const ts_label_t *label)
3035 {
3036 mod_hash_val_t hv;
3037 zone_t *zone = NULL;
3038
3039 ASSERT(MUTEX_HELD(&zonehash_lock));
3040
3041 /*
3042 * zonehashbylabel is not maintained for unlabeled systems
3043 */
3044 if (!is_system_labeled())
3045 return (NULL);
3046 if (mod_hash_find(zonehashbylabel, (mod_hash_key_t)label, &hv) == 0)
3047 zone = (zone_t *)hv;
3048 return (zone);
3049 }
3050
3051 static zone_t *
3052 zone_find_all_by_name(char *name)
3053 {
3054 mod_hash_val_t hv;
3055 zone_t *zone = NULL;
3056
3057 ASSERT(MUTEX_HELD(&zonehash_lock));
3058
3059 if (mod_hash_find(zonehashbyname, (mod_hash_key_t)name, &hv) == 0)
3060 zone = (zone_t *)hv;
3061 return (zone);
3062 }
3063
3064 /*
3065 * Public interface for looking up a zone by zoneid. Only returns the zone if
3066 * it is fully initialized, and has not yet begun the zone_destroy() sequence.
3067 * Caller must call zone_rele() once it is done with the zone.
3068 *
3069 * The zone may begin the zone_destroy() sequence immediately after this
3070 * function returns, but may be safely used until zone_rele() is called.
3071 */
3072 zone_t *
3073 zone_find_by_id(zoneid_t zoneid)
3074 {
3075 zone_t *zone;
3076 zone_status_t status;
3077
3078 mutex_enter(&zonehash_lock);
3079 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
3080 mutex_exit(&zonehash_lock);
3081 return (NULL);
3082 }
3083 status = zone_status_get(zone);
3084 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3085 /*
3086 * For all practical purposes the zone doesn't exist.
3087 */
3088 mutex_exit(&zonehash_lock);
3089 return (NULL);
3090 }
3091 zone_hold(zone);
3092 mutex_exit(&zonehash_lock);
3093 return (zone);
3094 }
3095
3096 /*
3097 * Similar to zone_find_by_id, but using zone label as the key.
3098 */
3099 zone_t *
3100 zone_find_by_label(const ts_label_t *label)
3101 {
3102 zone_t *zone;
3103 zone_status_t status;
3104
3105 mutex_enter(&zonehash_lock);
3106 if ((zone = zone_find_all_by_label(label)) == NULL) {
3107 mutex_exit(&zonehash_lock);
3108 return (NULL);
3109 }
3110
3111 status = zone_status_get(zone);
3112 if (status > ZONE_IS_DOWN) {
3113 /*
3114 * For all practical purposes the zone doesn't exist.
3115 */
3116 mutex_exit(&zonehash_lock);
3117 return (NULL);
3118 }
3119 zone_hold(zone);
3120 mutex_exit(&zonehash_lock);
3121 return (zone);
3122 }
3123
3124 /*
3125 * Similar to zone_find_by_id, but using zone name as the key.
3126 */
3127 zone_t *
3128 zone_find_by_name(char *name)
3129 {
3130 zone_t *zone;
3131 zone_status_t status;
3132
3133 mutex_enter(&zonehash_lock);
3134 if ((zone = zone_find_all_by_name(name)) == NULL) {
3135 mutex_exit(&zonehash_lock);
3136 return (NULL);
3137 }
3138 status = zone_status_get(zone);
3139 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3140 /*
3141 * For all practical purposes the zone doesn't exist.
3142 */
3143 mutex_exit(&zonehash_lock);
3144 return (NULL);
3145 }
3146 zone_hold(zone);
3147 mutex_exit(&zonehash_lock);
3148 return (zone);
3149 }
3150
3151 /*
3152 * Similar to zone_find_by_id(), using the path as a key. For instance,
3153 * if there is a zone "foo" rooted at /foo/root, and the path argument
3154 * is "/foo/root/proc", it will return the held zone_t corresponding to
3155 * zone "foo".
3156 *
3157 * zone_find_by_path() always returns a non-NULL value, since at the
3158 * very least every path will be contained in the global zone.
3159 *
3160 * As with the other zone_find_by_*() functions, the caller is
3161 * responsible for zone_rele()ing the return value of this function.
3162 */
3163 zone_t *
3164 zone_find_by_path(const char *path)
3165 {
3166 zone_t *zone;
3167 zone_t *zret = NULL;
3168 zone_status_t status;
3169
3170 if (path == NULL) {
3171 /*
3172 * Call from rootconf().
3173 */
3174 zone_hold(global_zone);
3175 return (global_zone);
3176 }
3177 ASSERT(*path == '/');
3178 mutex_enter(&zonehash_lock);
3179 for (zone = list_head(&zone_active); zone != NULL;
3180 zone = list_next(&zone_active, zone)) {
3181 if (ZONE_PATH_VISIBLE(path, zone))
3182 zret = zone;
3183 }
3184 ASSERT(zret != NULL);
3185 status = zone_status_get(zret);
3186 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3187 /*
3188 * Zone practically doesn't exist.
3189 */
3190 zret = global_zone;
3191 }
3192 zone_hold(zret);
3193 mutex_exit(&zonehash_lock);
3194 return (zret);
3195 }
3196
3197 /*
3198 * Public interface for updating per-zone load averages. Called once per
3199 * second.
3200 *
3201 * Based on loadavg_update(), genloadavg() and calcloadavg() from clock.c.
3202 */
3203 void
3204 zone_loadavg_update()
3205 {
3206 zone_t *zp;
3207 zone_status_t status;
3208 struct loadavg_s *lavg;
3209 hrtime_t zone_total;
3210 int i;
3211 hrtime_t hr_avg;
3212 int nrun;
3213 static int64_t f[3] = { 135, 27, 9 };
3214 int64_t q, r;
3215
3216 mutex_enter(&zonehash_lock);
3217 for (zp = list_head(&zone_active); zp != NULL;
3218 zp = list_next(&zone_active, zp)) {
3219 mutex_enter(&zp->zone_lock);
3220
3221 /* Skip zones that are on the way down or not yet up */
3222 status = zone_status_get(zp);
3223 if (status < ZONE_IS_READY || status >= ZONE_IS_DOWN) {
3224 /* For all practical purposes the zone doesn't exist. */
3225 mutex_exit(&zp->zone_lock);
3226 continue;
3227 }
3228
3229 /*
3230 * Update the 10 second moving average data in zone_loadavg.
3231 */
3232 lavg = &zp->zone_loadavg;
3233
3234 zone_total = zp->zone_utime + zp->zone_stime + zp->zone_wtime;
3235 scalehrtime(&zone_total);
3236
3237 /* The zone_total should always be increasing. */
3238 lavg->lg_loads[lavg->lg_cur] = (zone_total > lavg->lg_total) ?
3239 zone_total - lavg->lg_total : 0;
3240 lavg->lg_cur = (lavg->lg_cur + 1) % S_LOADAVG_SZ;
3241 /* lg_total holds the prev. 1 sec. total */
3242 lavg->lg_total = zone_total;
3243
3244 /*
3245 * To simplify the calculation, we don't calculate the load avg.
3246 * until the zone has been up for at least 10 seconds and our
3247 * moving average is thus full.
3248 */
3249 if ((lavg->lg_len + 1) < S_LOADAVG_SZ) {
3250 lavg->lg_len++;
3251 mutex_exit(&zp->zone_lock);
3252 continue;
3253 }
3254
3255 /* Now calculate the 1min, 5min, 15 min load avg. */
3256 hr_avg = 0;
3257 for (i = 0; i < S_LOADAVG_SZ; i++)
3258 hr_avg += lavg->lg_loads[i];
3259 hr_avg = hr_avg / S_LOADAVG_SZ;
3260 nrun = hr_avg / (NANOSEC / LGRP_LOADAVG_IN_THREAD_MAX);
3261
3262 /* Compute load avg. See comment in calcloadavg() */
3263 for (i = 0; i < 3; i++) {
3264 q = (zp->zone_hp_avenrun[i] >> 16) << 7;
3265 r = (zp->zone_hp_avenrun[i] & 0xffff) << 7;
3266 zp->zone_hp_avenrun[i] +=
3267 ((nrun - q) * f[i] - ((r * f[i]) >> 16)) >> 4;
3268
3269 /* avenrun[] can only hold 31 bits of load avg. */
3270 if (zp->zone_hp_avenrun[i] <
3271 ((uint64_t)1<<(31+16-FSHIFT)))
3272 zp->zone_avenrun[i] = (int32_t)
3273 (zp->zone_hp_avenrun[i] >> (16 - FSHIFT));
3274 else
3275 zp->zone_avenrun[i] = 0x7fffffff;
3276 }
3277
3278 mutex_exit(&zp->zone_lock);
3279 }
3280 mutex_exit(&zonehash_lock);
3281 }
3282
3283 /*
3284 * Get the number of cpus visible to this zone. The system-wide global
3285 * 'ncpus' is returned if pools are disabled, the caller is in the
3286 * global zone, or a NULL zone argument is passed in.
3287 */
3288 int
3289 zone_ncpus_get(zone_t *zone)
3290 {
3291 int myncpus = zone == NULL ? 0 : zone->zone_ncpus;
3292
3293 return (myncpus != 0 ? myncpus : ncpus);
3294 }
3295
3296 /*
3297 * Get the number of online cpus visible to this zone. The system-wide
3298 * global 'ncpus_online' is returned if pools are disabled, the caller
3299 * is in the global zone, or a NULL zone argument is passed in.
3300 */
3301 int
3302 zone_ncpus_online_get(zone_t *zone)
3303 {
3304 int myncpus_online = zone == NULL ? 0 : zone->zone_ncpus_online;
3305
3306 return (myncpus_online != 0 ? myncpus_online : ncpus_online);
3307 }
3308
3309 /*
3310 * Return the pool to which the zone is currently bound.
3311 */
3312 pool_t *
3313 zone_pool_get(zone_t *zone)
3314 {
3315 ASSERT(pool_lock_held());
3316
3317 return (zone->zone_pool);
3318 }
3319
3320 /*
3321 * Set the zone's pool pointer and update the zone's visibility to match
3322 * the resources in the new pool.
3323 */
3324 void
3325 zone_pool_set(zone_t *zone, pool_t *pool)
3326 {
3327 ASSERT(pool_lock_held());
3328 ASSERT(MUTEX_HELD(&cpu_lock));
3329
3330 zone->zone_pool = pool;
3331 zone_pset_set(zone, pool->pool_pset->pset_id);
3332 }
3333
3334 /*
3335 * Return the cached value of the id of the processor set to which the
3336 * zone is currently bound. The value will be ZONE_PS_INVAL if the pools
3337 * facility is disabled.
3338 */
3339 psetid_t
3340 zone_pset_get(zone_t *zone)
3341 {
3342 ASSERT(MUTEX_HELD(&cpu_lock));
3343
3344 return (zone->zone_psetid);
3345 }
3346
3347 /*
3348 * Set the cached value of the id of the processor set to which the zone
3349 * is currently bound. Also update the zone's visibility to match the
3350 * resources in the new processor set.
3351 */
3352 void
3353 zone_pset_set(zone_t *zone, psetid_t newpsetid)
3354 {
3355 psetid_t oldpsetid;
3356
3357 ASSERT(MUTEX_HELD(&cpu_lock));
3358 oldpsetid = zone_pset_get(zone);
3359
3360 if (oldpsetid == newpsetid)
3361 return;
3362 /*
3363 * Global zone sees all.
3364 */
3365 if (zone != global_zone) {
3366 zone->zone_psetid = newpsetid;
3367 if (newpsetid != ZONE_PS_INVAL)
3368 pool_pset_visibility_add(newpsetid, zone);
3369 if (oldpsetid != ZONE_PS_INVAL)
3370 pool_pset_visibility_remove(oldpsetid, zone);
3371 }
3372 /*
3373 * Disabling pools, so we should start using the global values
3374 * for ncpus and ncpus_online.
3375 */
3376 if (newpsetid == ZONE_PS_INVAL) {
3377 zone->zone_ncpus = 0;
3378 zone->zone_ncpus_online = 0;
3379 }
3380 }
3381
3382 /*
3383 * Walk the list of active zones and issue the provided callback for
3384 * each of them.
3385 *
3386 * Caller must not be holding any locks that may be acquired under
3387 * zonehash_lock. See comment at the beginning of the file for a list of
3388 * common locks and their interactions with zones.
3389 */
3390 int
3391 zone_walk(int (*cb)(zone_t *, void *), void *data)
3392 {
3393 zone_t *zone;
3394 int ret = 0;
3395 zone_status_t status;
3396
3397 mutex_enter(&zonehash_lock);
3398 for (zone = list_head(&zone_active); zone != NULL;
3399 zone = list_next(&zone_active, zone)) {
3400 /*
3401 * Skip zones that shouldn't be externally visible.
3402 */
3403 status = zone_status_get(zone);
3404 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN)
3405 continue;
3406 /*
3407 * Bail immediately if any callback invocation returns a
3408 * non-zero value.
3409 */
3410 ret = (*cb)(zone, data);
3411 if (ret != 0)
3412 break;
3413 }
3414 mutex_exit(&zonehash_lock);
3415 return (ret);
3416 }
3417
3418 static int
3419 zone_set_root(zone_t *zone, const char *upath)
3420 {
3421 vnode_t *vp;
3422 int trycount;
3423 int error = 0;
3424 char *path;
3425 struct pathname upn, pn;
3426 size_t pathlen;
3427
3428 if ((error = pn_get((char *)upath, UIO_USERSPACE, &upn)) != 0)
3429 return (error);
3430
3431 pn_alloc(&pn);
3432
3433 /* prevent infinite loop */
3434 trycount = 10;
3435 for (;;) {
3436 if (--trycount <= 0) {
3437 error = ESTALE;
3438 goto out;
3439 }
3440
3441 if ((error = lookuppn(&upn, &pn, FOLLOW, NULLVPP, &vp)) == 0) {
3442 /*
3443 * VOP_ACCESS() may cover 'vp' with a new
3444 * filesystem, if 'vp' is an autoFS vnode.
3445 * Get the new 'vp' if so.
3446 */
3447 if ((error =
3448 VOP_ACCESS(vp, VEXEC, 0, CRED(), NULL)) == 0 &&
3449 (!vn_ismntpt(vp) ||
3450 (error = traverse(&vp)) == 0)) {
3451 pathlen = pn.pn_pathlen + 2;
3452 path = kmem_alloc(pathlen, KM_SLEEP);
3453 (void) strncpy(path, pn.pn_path,
3454 pn.pn_pathlen + 1);
3455 path[pathlen - 2] = '/';
3456 path[pathlen - 1] = '\0';
3457 pn_free(&pn);
3458 pn_free(&upn);
3459
3460 /* Success! */
3461 break;
3462 }
3463 VN_RELE(vp);
3464 }
3465 if (error != ESTALE)
3466 goto out;
3467 }
3468
3469 ASSERT(error == 0);
3470 zone->zone_rootvp = vp; /* we hold a reference to vp */
3471 zone->zone_rootpath = path;
3472 zone->zone_rootpathlen = pathlen;
3473 if (pathlen > 5 && strcmp(path + pathlen - 5, "/lu/") == 0)
3474 zone->zone_flags |= ZF_IS_SCRATCH;
3475 return (0);
3476
3477 out:
3478 pn_free(&pn);
3479 pn_free(&upn);
3480 return (error);
3481 }
3482
3483 #define isalnum(c) (((c) >= '0' && (c) <= '9') || \
3484 ((c) >= 'a' && (c) <= 'z') || \
3485 ((c) >= 'A' && (c) <= 'Z'))
3486
3487 static int
3488 zone_set_name(zone_t *zone, const char *uname)
3489 {
3490 char *kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
3491 size_t len;
3492 int i, err;
3493
3494 if ((err = copyinstr(uname, kname, ZONENAME_MAX, &len)) != 0) {
3495 kmem_free(kname, ZONENAME_MAX);
3496 return (err); /* EFAULT or ENAMETOOLONG */
3497 }
3498
3499 /* must be less than ZONENAME_MAX */
3500 if (len == ZONENAME_MAX && kname[ZONENAME_MAX - 1] != '\0') {
3501 kmem_free(kname, ZONENAME_MAX);
3502 return (EINVAL);
3503 }
3504
3505 /*
3506 * Name must start with an alphanumeric and must contain only
3507 * alphanumerics, '-', '_' and '.'.
3508 */
3509 if (!isalnum(kname[0])) {
3510 kmem_free(kname, ZONENAME_MAX);
3511 return (EINVAL);
3512 }
3513 for (i = 1; i < len - 1; i++) {
3514 if (!isalnum(kname[i]) && kname[i] != '-' && kname[i] != '_' &&
3515 kname[i] != '.') {
3516 kmem_free(kname, ZONENAME_MAX);
3517 return (EINVAL);
3518 }
3519 }
3520
3521 zone->zone_name = kname;
3522 return (0);
3523 }
3524
3525 /*
3526 * Gets the 32-bit hostid of the specified zone as an unsigned int. If 'zonep'
3527 * is NULL or it points to a zone with no hostid emulation, then the machine's
3528 * hostid (i.e., the global zone's hostid) is returned. This function returns
3529 * zero if neither the zone nor the host machine (global zone) have hostids. It
3530 * returns HW_INVALID_HOSTID if the function attempts to return the machine's
3531 * hostid and the machine's hostid is invalid.
3532 */
3533 uint32_t
3534 zone_get_hostid(zone_t *zonep)
3535 {
3536 unsigned long machine_hostid;
3537
3538 if (zonep == NULL || zonep->zone_hostid == HW_INVALID_HOSTID) {
3539 if (ddi_strtoul(hw_serial, NULL, 10, &machine_hostid) != 0)
3540 return (HW_INVALID_HOSTID);
3541 return ((uint32_t)machine_hostid);
3542 }
3543 return (zonep->zone_hostid);
3544 }
3545
3546 /*
3547 * Similar to thread_create(), but makes sure the thread is in the appropriate
3548 * zone's zsched process (curproc->p_zone->zone_zsched) before returning.
3549 */
3550 /*ARGSUSED*/
3551 kthread_t *
3552 zthread_create(
3553 caddr_t stk,
3554 size_t stksize,
3555 void (*proc)(),
3556 void *arg,
3557 size_t len,
3558 pri_t pri)
3559 {
3560 kthread_t *t;
3561 zone_t *zone = curproc->p_zone;
3562 proc_t *pp = zone->zone_zsched;
3563
3564 zone_hold(zone); /* Reference to be dropped when thread exits */
3565
3566 /*
3567 * No-one should be trying to create threads if the zone is shutting
3568 * down and there aren't any kernel threads around. See comment
3569 * in zthread_exit().
3570 */
3571 ASSERT(!(zone->zone_kthreads == NULL &&
3572 zone_status_get(zone) >= ZONE_IS_EMPTY));
3573 /*
3574 * Create a thread, but don't let it run until we've finished setting
3575 * things up.
3576 */
3577 t = thread_create(stk, stksize, proc, arg, len, pp, TS_STOPPED, pri);
3578 ASSERT(t->t_forw == NULL);
3579 mutex_enter(&zone_status_lock);
3580 if (zone->zone_kthreads == NULL) {
3581 t->t_forw = t->t_back = t;
3582 } else {
3583 kthread_t *tx = zone->zone_kthreads;
3584
3585 t->t_forw = tx;
3586 t->t_back = tx->t_back;
3587 tx->t_back->t_forw = t;
3588 tx->t_back = t;
3589 }
3590 zone->zone_kthreads = t;
3591 mutex_exit(&zone_status_lock);
3592
3593 mutex_enter(&pp->p_lock);
3594 t->t_proc_flag |= TP_ZTHREAD;
3595 project_rele(t->t_proj);
3596 t->t_proj = project_hold(pp->p_task->tk_proj);
3597
3598 /*
3599 * Setup complete, let it run.
3600 */
3601 thread_lock(t);
3602 t->t_schedflag |= TS_ALLSTART;
3603 setrun_locked(t);
3604 thread_unlock(t);
3605
3606 mutex_exit(&pp->p_lock);
3607
3608 return (t);
3609 }
3610
3611 /*
3612 * Similar to thread_exit(). Must be called by threads created via
3613 * zthread_exit().
3614 */
3615 void
3616 zthread_exit(void)
3617 {
3618 kthread_t *t = curthread;
3619 proc_t *pp = curproc;
3620 zone_t *zone = pp->p_zone;
3621
3622 mutex_enter(&zone_status_lock);
3623
3624 /*
3625 * Reparent to p0
3626 */
3627 kpreempt_disable();
3628 mutex_enter(&pp->p_lock);
3629 t->t_proc_flag &= ~TP_ZTHREAD;
3630 t->t_procp = &p0;
3631 hat_thread_exit(t);
3632 mutex_exit(&pp->p_lock);
3633 kpreempt_enable();
3634
3635 if (t->t_back == t) {
3636 ASSERT(t->t_forw == t);
3637 /*
3638 * If the zone is empty, once the thread count
3639 * goes to zero no further kernel threads can be
3640 * created. This is because if the creator is a process
3641 * in the zone, then it must have exited before the zone
3642 * state could be set to ZONE_IS_EMPTY.
3643 * Otherwise, if the creator is a kernel thread in the
3644 * zone, the thread count is non-zero.
3645 *
3646 * This really means that non-zone kernel threads should
3647 * not create zone kernel threads.
3648 */
3649 zone->zone_kthreads = NULL;
3650 if (zone_status_get(zone) == ZONE_IS_EMPTY) {
3651 zone_status_set(zone, ZONE_IS_DOWN);
3652 /*
3653 * Remove any CPU caps on this zone.
3654 */
3655 cpucaps_zone_remove(zone);
3656 }
3657 } else {
3658 t->t_forw->t_back = t->t_back;
3659 t->t_back->t_forw = t->t_forw;
3660 if (zone->zone_kthreads == t)
3661 zone->zone_kthreads = t->t_forw;
3662 }
3663 mutex_exit(&zone_status_lock);
3664 zone_rele(zone);
3665 thread_exit();
3666 /* NOTREACHED */
3667 }
3668
3669 static void
3670 zone_chdir(vnode_t *vp, vnode_t **vpp, proc_t *pp)
3671 {
3672 vnode_t *oldvp;
3673
3674 /* we're going to hold a reference here to the directory */
3675 VN_HOLD(vp);
3676
3677 /* update abs cwd/root path see c2/audit.c */
3678 if (AU_AUDITING())
3679 audit_chdirec(vp, vpp);
3680
3681 mutex_enter(&pp->p_lock);
3682 oldvp = *vpp;
3683 *vpp = vp;
3684 mutex_exit(&pp->p_lock);
3685 if (oldvp != NULL)
3686 VN_RELE(oldvp);
3687 }
3688
3689 /*
3690 * Convert an rctl value represented by an nvlist_t into an rctl_val_t.
3691 */
3692 static int
3693 nvlist2rctlval(nvlist_t *nvl, rctl_val_t *rv)
3694 {
3695 nvpair_t *nvp = NULL;
3696 boolean_t priv_set = B_FALSE;
3697 boolean_t limit_set = B_FALSE;
3698 boolean_t action_set = B_FALSE;
3699
3700 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
3701 const char *name;
3702 uint64_t ui64;
3703
3704 name = nvpair_name(nvp);
3705 if (nvpair_type(nvp) != DATA_TYPE_UINT64)
3706 return (EINVAL);
3707 (void) nvpair_value_uint64(nvp, &ui64);
3708 if (strcmp(name, "privilege") == 0) {
3709 /*
3710 * Currently only privileged values are allowed, but
3711 * this may change in the future.
3712 */
3713 if (ui64 != RCPRIV_PRIVILEGED)
3714 return (EINVAL);
3715 rv->rcv_privilege = ui64;
3716 priv_set = B_TRUE;
3717 } else if (strcmp(name, "limit") == 0) {
3718 rv->rcv_value = ui64;
3719 limit_set = B_TRUE;
3720 } else if (strcmp(name, "action") == 0) {
3721 if (ui64 != RCTL_LOCAL_NOACTION &&
3722 ui64 != RCTL_LOCAL_DENY)
3723 return (EINVAL);
3724 rv->rcv_flagaction = ui64;
3725 action_set = B_TRUE;
3726 } else {
3727 return (EINVAL);
3728 }
3729 }
3730
3731 if (!(priv_set && limit_set && action_set))
3732 return (EINVAL);
3733 rv->rcv_action_signal = 0;
3734 rv->rcv_action_recipient = NULL;
3735 rv->rcv_action_recip_pid = -1;
3736 rv->rcv_firing_time = 0;
3737
3738 return (0);
3739 }
3740
3741 /*
3742 * Non-global zone version of start_init.
3743 */
3744 void
3745 zone_start_init(void)
3746 {
3747 proc_t *p = ttoproc(curthread);
3748 zone_t *z = p->p_zone;
3749
3750 ASSERT(!INGLOBALZONE(curproc));
3751
3752 /*
3753 * For all purposes (ZONE_ATTR_INITPID and restart_init),
3754 * storing just the pid of init is sufficient.
3755 */
3756 z->zone_proc_initpid = p->p_pid;
3757
3758 /*
3759 * We maintain zone_boot_err so that we can return the cause of the
3760 * failure back to the caller of the zone_boot syscall.
3761 */
3762 p->p_zone->zone_boot_err = start_init_common();
3763
3764 /*
3765 * We will prevent booting zones from becoming running zones if the
3766 * global zone is shutting down.
3767 */
3768 mutex_enter(&zone_status_lock);
3769 if (z->zone_boot_err != 0 || zone_status_get(global_zone) >=
3770 ZONE_IS_SHUTTING_DOWN) {
3771 /*
3772 * Make sure we are still in the booting state-- we could have
3773 * raced and already be shutting down, or even further along.
3774 */
3775 if (zone_status_get(z) == ZONE_IS_BOOTING) {
3776 zone_status_set(z, ZONE_IS_SHUTTING_DOWN);
3777 }
3778 mutex_exit(&zone_status_lock);
3779 /* It's gone bad, dispose of the process */
3780 if (proc_exit(CLD_EXITED, z->zone_boot_err) != 0) {
3781 mutex_enter(&p->p_lock);
3782 ASSERT(p->p_flag & SEXITLWPS);
3783 lwp_exit();
3784 }
3785 } else {
3786 if (zone_status_get(z) == ZONE_IS_BOOTING)
3787 zone_status_set(z, ZONE_IS_RUNNING);
3788 mutex_exit(&zone_status_lock);
3789 /* cause the process to return to userland. */
3790 lwp_rtt();
3791 }
3792 }
3793
3794 struct zsched_arg {
3795 zone_t *zone;
3796 nvlist_t *nvlist;
3797 };
3798
3799 /*
3800 * Per-zone "sched" workalike. The similarity to "sched" doesn't have
3801 * anything to do with scheduling, but rather with the fact that
3802 * per-zone kernel threads are parented to zsched, just like regular
3803 * kernel threads are parented to sched (p0).
3804 *
3805 * zsched is also responsible for launching init for the zone.
3806 */
3807 static void
3808 zsched(void *arg)
3809 {
3810 struct zsched_arg *za = arg;
3811 proc_t *pp = curproc;
3812 proc_t *initp = proc_init;
3813 zone_t *zone = za->zone;
3814 cred_t *cr, *oldcred;
3815 rctl_set_t *set;
3816 rctl_alloc_gp_t *gp;
3817 contract_t *ct = NULL;
3818 task_t *tk, *oldtk;
3819 rctl_entity_p_t e;
3820 kproject_t *pj;
3821
3822 nvlist_t *nvl = za->nvlist;
3823 nvpair_t *nvp = NULL;
3824
3825 bcopy("zsched", PTOU(pp)->u_psargs, sizeof ("zsched"));
3826 bcopy("zsched", PTOU(pp)->u_comm, sizeof ("zsched"));
3827 PTOU(pp)->u_argc = 0;
3828 PTOU(pp)->u_argv = NULL;
3829 PTOU(pp)->u_envp = NULL;
3830 closeall(P_FINFO(pp));
3831
3832 /*
3833 * We are this zone's "zsched" process. As the zone isn't generally
3834 * visible yet we don't need to grab any locks before initializing its
3835 * zone_proc pointer.
3836 */
3837 zone_hold(zone); /* this hold is released by zone_destroy() */
3838 zone->zone_zsched = pp;
3839 mutex_enter(&pp->p_lock);
3840 pp->p_zone = zone;
3841 mutex_exit(&pp->p_lock);
3842
3843 /*
3844 * Disassociate process from its 'parent'; parent ourselves to init
3845 * (pid 1) and change other values as needed.
3846 */
3847 sess_create();
3848
3849 mutex_enter(&pidlock);
3850 proc_detach(pp);
3851 pp->p_ppid = 1;
3852 pp->p_flag |= SZONETOP;
3853 pp->p_ancpid = 1;
3854 pp->p_parent = initp;
3855 pp->p_psibling = NULL;
3856 if (initp->p_child)
3857 initp->p_child->p_psibling = pp;
3858 pp->p_sibling = initp->p_child;
3859 initp->p_child = pp;
3860
3861 /* Decrement what newproc() incremented. */
3862 upcount_dec(crgetruid(CRED()), GLOBAL_ZONEID);
3863 /*
3864 * Our credentials are about to become kcred-like, so we don't care
3865 * about the caller's ruid.
3866 */
3867 upcount_inc(crgetruid(kcred), zone->zone_id);
3868 mutex_exit(&pidlock);
3869
3870 /*
3871 * getting out of global zone, so decrement lwp and process counts
3872 */
3873 pj = pp->p_task->tk_proj;
3874 mutex_enter(&global_zone->zone_nlwps_lock);
3875 pj->kpj_nlwps -= pp->p_lwpcnt;
3876 global_zone->zone_nlwps -= pp->p_lwpcnt;
3877 pj->kpj_nprocs--;
3878 global_zone->zone_nprocs--;
3879 mutex_exit(&global_zone->zone_nlwps_lock);
3880
3881 /*
3882 * Decrement locked memory counts on old zone and project.
3883 */
3884 mutex_enter(&global_zone->zone_mem_lock);
3885 global_zone->zone_locked_mem -= pp->p_locked_mem;
3886 pj->kpj_data.kpd_locked_mem -= pp->p_locked_mem;
3887 mutex_exit(&global_zone->zone_mem_lock);
3888
3889 /*
3890 * Create and join a new task in project '0' of this zone.
3891 *
3892 * We don't need to call holdlwps() since we know we're the only lwp in
3893 * this process.
3894 *
3895 * task_join() returns with p_lock held.
3896 */
3897 tk = task_create(0, zone);
3898 mutex_enter(&cpu_lock);
3899 oldtk = task_join(tk, 0);
3900
3901 pj = pp->p_task->tk_proj;
3902
3903 mutex_enter(&zone->zone_mem_lock);
3904 zone->zone_locked_mem += pp->p_locked_mem;
3905 pj->kpj_data.kpd_locked_mem += pp->p_locked_mem;
3906 mutex_exit(&zone->zone_mem_lock);
3907
3908 /*
3909 * add lwp and process counts to zsched's zone, and increment
3910 * project's task and process count due to the task created in
3911 * the above task_create.
3912 */
3913 mutex_enter(&zone->zone_nlwps_lock);
3914 pj->kpj_nlwps += pp->p_lwpcnt;
3915 pj->kpj_ntasks += 1;
3916 zone->zone_nlwps += pp->p_lwpcnt;
3917 pj->kpj_nprocs++;
3918 zone->zone_nprocs++;
3919 mutex_exit(&zone->zone_nlwps_lock);
3920
3921 mutex_exit(&curproc->p_lock);
3922 mutex_exit(&cpu_lock);
3923 task_rele(oldtk);
3924
3925 /*
3926 * The process was created by a process in the global zone, hence the
3927 * credentials are wrong. We might as well have kcred-ish credentials.
3928 */
3929 cr = zone->zone_kcred;
3930 crhold(cr);
3931 mutex_enter(&pp->p_crlock);
3932 oldcred = pp->p_cred;
3933 pp->p_cred = cr;
3934 mutex_exit(&pp->p_crlock);
3935 crfree(oldcred);
3936
3937 /*
3938 * Hold credentials again (for thread)
3939 */
3940 crhold(cr);
3941
3942 /*
3943 * p_lwpcnt can't change since this is a kernel process.
3944 */
3945 crset(pp, cr);
3946
3947 /*
3948 * Chroot
3949 */
3950 zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_cdir, pp);
3951 zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_rdir, pp);
3952
3953 /*
3954 * Initialize zone's rctl set.
3955 */
3956 set = rctl_set_create();
3957 gp = rctl_set_init_prealloc(RCENTITY_ZONE);
3958 mutex_enter(&pp->p_lock);
3959 e.rcep_p.zone = zone;
3960 e.rcep_t = RCENTITY_ZONE;
3961 zone->zone_rctls = rctl_set_init(RCENTITY_ZONE, pp, &e, set, gp);
3962 mutex_exit(&pp->p_lock);
3963 rctl_prealloc_destroy(gp);
3964
3965 /*
3966 * Apply the rctls passed in to zone_create(). This is basically a list
3967 * assignment: all of the old values are removed and the new ones
3968 * inserted. That is, if an empty list is passed in, all values are
3969 * removed.
3970 */
3971 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
3972 rctl_dict_entry_t *rde;
3973 rctl_hndl_t hndl;
3974 char *name;
3975 nvlist_t **nvlarray;
3976 uint_t i, nelem;
3977 int error; /* For ASSERT()s */
3978
3979 name = nvpair_name(nvp);
3980 hndl = rctl_hndl_lookup(name);
3981 ASSERT(hndl != -1);
3982 rde = rctl_dict_lookup_hndl(hndl);
3983 ASSERT(rde != NULL);
3984
3985 for (; /* ever */; ) {
3986 rctl_val_t oval;
3987
3988 mutex_enter(&pp->p_lock);
3989 error = rctl_local_get(hndl, NULL, &oval, pp);
3990 mutex_exit(&pp->p_lock);
3991 ASSERT(error == 0); /* Can't fail for RCTL_FIRST */
3992 ASSERT(oval.rcv_privilege != RCPRIV_BASIC);
3993 if (oval.rcv_privilege == RCPRIV_SYSTEM)
3994 break;
3995 mutex_enter(&pp->p_lock);
3996 error = rctl_local_delete(hndl, &oval, pp);
3997 mutex_exit(&pp->p_lock);
3998 ASSERT(error == 0);
3999 }
4000 error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
4001 ASSERT(error == 0);
4002 for (i = 0; i < nelem; i++) {
4003 rctl_val_t *nvalp;
4004
4005 nvalp = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
4006 error = nvlist2rctlval(nvlarray[i], nvalp);
4007 ASSERT(error == 0);
4008 /*
4009 * rctl_local_insert can fail if the value being
4010 * inserted is a duplicate; this is OK.
4011 */
4012 mutex_enter(&pp->p_lock);
4013 if (rctl_local_insert(hndl, nvalp, pp) != 0)
4014 kmem_cache_free(rctl_val_cache, nvalp);
4015 mutex_exit(&pp->p_lock);
4016 }
4017 }
4018
4019 /*
4020 * Tell the world that we're done setting up.
4021 *
4022 * At this point we want to set the zone status to ZONE_IS_INITIALIZED
4023 * and atomically set the zone's processor set visibility. Once
4024 * we drop pool_lock() this zone will automatically get updated
4025 * to reflect any future changes to the pools configuration.
4026 *
4027 * Note that after we drop the locks below (zonehash_lock in
4028 * particular) other operations such as a zone_getattr call can
4029 * now proceed and observe the zone. That is the reason for doing a
4030 * state transition to the INITIALIZED state.
4031 */
4032 pool_lock();
4033 mutex_enter(&cpu_lock);
4034 mutex_enter(&zonehash_lock);
4035 zone_uniqid(zone);
4036 zone_zsd_configure(zone);
4037 if (pool_state == POOL_ENABLED)
4038 zone_pset_set(zone, pool_default->pool_pset->pset_id);
4039 mutex_enter(&zone_status_lock);
4040 ASSERT(zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
4041 zone_status_set(zone, ZONE_IS_INITIALIZED);
4042 mutex_exit(&zone_status_lock);
4043 mutex_exit(&zonehash_lock);
4044 mutex_exit(&cpu_lock);
4045 pool_unlock();
4046
4047 /* Now call the create callback for this key */
4048 zsd_apply_all_keys(zsd_apply_create, zone);
4049
4050 /* The callbacks are complete. Mark ZONE_IS_READY */
4051 mutex_enter(&zone_status_lock);
4052 ASSERT(zone_status_get(zone) == ZONE_IS_INITIALIZED);
4053 zone_status_set(zone, ZONE_IS_READY);
4054 mutex_exit(&zone_status_lock);
4055
4056 /*
4057 * Once we see the zone transition to the ZONE_IS_BOOTING state,
4058 * we launch init, and set the state to running.
4059 */
4060 zone_status_wait_cpr(zone, ZONE_IS_BOOTING, "zsched");
4061
4062 if (zone_status_get(zone) == ZONE_IS_BOOTING) {
4063 id_t cid;
4064
4065 /*
4066 * Ok, this is a little complicated. We need to grab the
4067 * zone's pool's scheduling class ID; note that by now, we
4068 * are already bound to a pool if we need to be (zoneadmd
4069 * will have done that to us while we're in the READY
4070 * state). *But* the scheduling class for the zone's 'init'
4071 * must be explicitly passed to newproc, which doesn't
4072 * respect pool bindings.
4073 *
4074 * We hold the pool_lock across the call to newproc() to
4075 * close the obvious race: the pool's scheduling class
4076 * could change before we manage to create the LWP with
4077 * classid 'cid'.
4078 */
4079 pool_lock();
4080 if (zone->zone_defaultcid > 0)
4081 cid = zone->zone_defaultcid;
4082 else
4083 cid = pool_get_class(zone->zone_pool);
4084 if (cid == -1)
4085 cid = defaultcid;
4086
4087 /*
4088 * If this fails, zone_boot will ultimately fail. The
4089 * state of the zone will be set to SHUTTING_DOWN-- userland
4090 * will have to tear down the zone, and fail, or try again.
4091 */
4092 if ((zone->zone_boot_err = newproc(zone_start_init, NULL, cid,
4093 minclsyspri - 1, &ct, 0)) != 0) {
4094 mutex_enter(&zone_status_lock);
4095 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
4096 mutex_exit(&zone_status_lock);
4097 } else {
4098 zone->zone_boot_time = gethrestime_sec();
4099 }
4100
4101 pool_unlock();
4102 }
4103
4104 /*
4105 * Wait for zone_destroy() to be called. This is what we spend
4106 * most of our life doing.
4107 */
4108 zone_status_wait_cpr(zone, ZONE_IS_DYING, "zsched");
4109
4110 if (ct)
4111 /*
4112 * At this point the process contract should be empty.
4113 * (Though if it isn't, it's not the end of the world.)
4114 */
4115 VERIFY(contract_abandon(ct, curproc, B_TRUE) == 0);
4116
4117 /*
4118 * Allow kcred to be freed when all referring processes
4119 * (including this one) go away. We can't just do this in
4120 * zone_free because we need to wait for the zone_cred_ref to
4121 * drop to 0 before calling zone_free, and the existence of
4122 * zone_kcred will prevent that. Thus, we call crfree here to
4123 * balance the crdup in zone_create. The crhold calls earlier
4124 * in zsched will be dropped when the thread and process exit.
4125 */
4126 crfree(zone->zone_kcred);
4127 zone->zone_kcred = NULL;
4128
4129 exit(CLD_EXITED, 0);
4130 }
4131
4132 /*
4133 * Helper function to determine if there are any submounts of the
4134 * provided path. Used to make sure the zone doesn't "inherit" any
4135 * mounts from before it is created.
4136 */
4137 static uint_t
4138 zone_mount_count(const char *rootpath)
4139 {
4140 vfs_t *vfsp;
4141 uint_t count = 0;
4142 size_t rootpathlen = strlen(rootpath);
4143
4144 /*
4145 * Holding zonehash_lock prevents race conditions with
4146 * vfs_list_add()/vfs_list_remove() since we serialize with
4147 * zone_find_by_path().
4148 */
4149 ASSERT(MUTEX_HELD(&zonehash_lock));
4150 /*
4151 * The rootpath must end with a '/'
4152 */
4153 ASSERT(rootpath[rootpathlen - 1] == '/');
4154
4155 /*
4156 * This intentionally does not count the rootpath itself if that
4157 * happens to be a mount point.
4158 */
4159 vfs_list_read_lock();
4160 vfsp = rootvfs;
4161 do {
4162 if (strncmp(rootpath, refstr_value(vfsp->vfs_mntpt),
4163 rootpathlen) == 0)
4164 count++;
4165 vfsp = vfsp->vfs_next;
4166 } while (vfsp != rootvfs);
4167 vfs_list_unlock();
4168 return (count);
4169 }
4170
4171 /*
4172 * Helper function to make sure that a zone created on 'rootpath'
4173 * wouldn't end up containing other zones' rootpaths.
4174 */
4175 static boolean_t
4176 zone_is_nested(const char *rootpath)
4177 {
4178 zone_t *zone;
4179 size_t rootpathlen = strlen(rootpath);
4180 size_t len;
4181
4182 ASSERT(MUTEX_HELD(&zonehash_lock));
4183
4184 /*
4185 * zone_set_root() appended '/' and '\0' at the end of rootpath
4186 */
4187 if ((rootpathlen <= 3) && (rootpath[0] == '/') &&
4188 (rootpath[1] == '/') && (rootpath[2] == '\0'))
4189 return (B_TRUE);
4190
4191 for (zone = list_head(&zone_active); zone != NULL;
4192 zone = list_next(&zone_active, zone)) {
4193 if (zone == global_zone)
4194 continue;
4195 len = strlen(zone->zone_rootpath);
4196 if (strncmp(rootpath, zone->zone_rootpath,
4197 MIN(rootpathlen, len)) == 0)
4198 return (B_TRUE);
4199 }
4200 return (B_FALSE);
4201 }
4202
4203 static int
4204 zone_set_privset(zone_t *zone, const priv_set_t *zone_privs,
4205 size_t zone_privssz)
4206 {
4207 priv_set_t *privs;
4208
4209 if (zone_privssz < sizeof (priv_set_t))
4210 return (ENOMEM);
4211
4212 privs = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
4213
4214 if (copyin(zone_privs, privs, sizeof (priv_set_t))) {
4215 kmem_free(privs, sizeof (priv_set_t));
4216 return (EFAULT);
4217 }
4218
4219 zone->zone_privset = privs;
4220 return (0);
4221 }
4222
4223 /*
4224 * We make creative use of nvlists to pass in rctls from userland. The list is
4225 * a list of the following structures:
4226 *
4227 * (name = rctl_name, value = nvpair_list_array)
4228 *
4229 * Where each element of the nvpair_list_array is of the form:
4230 *
4231 * [(name = "privilege", value = RCPRIV_PRIVILEGED),
4232 * (name = "limit", value = uint64_t),
4233 * (name = "action", value = (RCTL_LOCAL_NOACTION || RCTL_LOCAL_DENY))]
4234 */
4235 static int
4236 parse_rctls(caddr_t ubuf, size_t buflen, nvlist_t **nvlp)
4237 {
4238 nvpair_t *nvp = NULL;
4239 nvlist_t *nvl = NULL;
4240 char *kbuf;
4241 int error;
4242 rctl_val_t rv;
4243
4244 *nvlp = NULL;
4245
4246 if (buflen == 0)
4247 return (0);
4248
4249 if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL)
4250 return (ENOMEM);
4251 if (copyin(ubuf, kbuf, buflen)) {
4252 error = EFAULT;
4253 goto out;
4254 }
4255 if (nvlist_unpack(kbuf, buflen, &nvl, KM_SLEEP) != 0) {
4256 /*
4257 * nvl may have been allocated/free'd, but the value set to
4258 * non-NULL, so we reset it here.
4259 */
4260 nvl = NULL;
4261 error = EINVAL;
4262 goto out;
4263 }
4264 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
4265 rctl_dict_entry_t *rde;
4266 rctl_hndl_t hndl;
4267 nvlist_t **nvlarray;
4268 uint_t i, nelem;
4269 char *name;
4270
4271 error = EINVAL;
4272 name = nvpair_name(nvp);
4273 if (strncmp(nvpair_name(nvp), "zone.", sizeof ("zone.") - 1)
4274 != 0 || nvpair_type(nvp) != DATA_TYPE_NVLIST_ARRAY) {
4275 goto out;
4276 }
4277 if ((hndl = rctl_hndl_lookup(name)) == -1) {
4278 goto out;
4279 }
4280 rde = rctl_dict_lookup_hndl(hndl);
4281 error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
4282 ASSERT(error == 0);
4283 for (i = 0; i < nelem; i++) {
4284 if (error = nvlist2rctlval(nvlarray[i], &rv))
4285 goto out;
4286 }
4287 if (rctl_invalid_value(rde, &rv)) {
4288 error = EINVAL;
4289 goto out;
4290 }
4291 }
4292 error = 0;
4293 *nvlp = nvl;
4294 out:
4295 kmem_free(kbuf, buflen);
4296 if (error && nvl != NULL)
4297 nvlist_free(nvl);
4298 return (error);
4299 }
4300
4301 int
4302 zone_create_error(int er_error, int er_ext, int *er_out)
4303 {
4304 if (er_out != NULL) {
4305 if (copyout(&er_ext, er_out, sizeof (int))) {
4306 return (set_errno(EFAULT));
4307 }
4308 }
4309 return (set_errno(er_error));
4310 }
4311
4312 static int
4313 zone_set_label(zone_t *zone, const bslabel_t *lab, uint32_t doi)
4314 {
4315 ts_label_t *tsl;
4316 bslabel_t blab;
4317
4318 /* Get label from user */
4319 if (copyin(lab, &blab, sizeof (blab)) != 0)
4320 return (EFAULT);
4321 tsl = labelalloc(&blab, doi, KM_NOSLEEP);
4322 if (tsl == NULL)
4323 return (ENOMEM);
4324
4325 zone->zone_slabel = tsl;
4326 return (0);
4327 }
4328
4329 /*
4330 * Parses a comma-separated list of ZFS datasets into a per-zone dictionary.
4331 */
4332 static int
4333 parse_zfs(zone_t *zone, caddr_t ubuf, size_t buflen)
4334 {
4335 char *kbuf;
4336 char *dataset, *next;
4337 zone_dataset_t *zd;
4338 size_t len;
4339
4340 if (ubuf == NULL || buflen == 0)
4341 return (0);
4342
4343 if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL)
4344 return (ENOMEM);
4345
4346 if (copyin(ubuf, kbuf, buflen) != 0) {
4347 kmem_free(kbuf, buflen);
4348 return (EFAULT);
4349 }
4350
4351 dataset = next = kbuf;
4352 for (;;) {
4353 zd = kmem_alloc(sizeof (zone_dataset_t), KM_SLEEP);
4354
4355 next = strchr(dataset, ',');
4356
4357 if (next == NULL)
4358 len = strlen(dataset);
4359 else
4360 len = next - dataset;
4361
4362 zd->zd_dataset = kmem_alloc(len + 1, KM_SLEEP);
4363 bcopy(dataset, zd->zd_dataset, len);
4364 zd->zd_dataset[len] = '\0';
4365
4366 list_insert_head(&zone->zone_datasets, zd);
4367
4368 if (next == NULL)
4369 break;
4370
4371 dataset = next + 1;
4372 }
4373
4374 kmem_free(kbuf, buflen);
4375 return (0);
4376 }
4377
4378 /*
4379 * System call to create/initialize a new zone named 'zone_name', rooted
4380 * at 'zone_root', with a zone-wide privilege limit set of 'zone_privs',
4381 * and initialized with the zone-wide rctls described in 'rctlbuf', and
4382 * with labeling set by 'match', 'doi', and 'label'.
4383 *
4384 * If extended error is non-null, we may use it to return more detailed
4385 * error information.
4386 */
4387 static zoneid_t
4388 zone_create(const char *zone_name, const char *zone_root,
4389 const priv_set_t *zone_privs, size_t zone_privssz,
4390 caddr_t rctlbuf, size_t rctlbufsz,
4391 caddr_t zfsbuf, size_t zfsbufsz, int *extended_error,
4392 int match, uint32_t doi, const bslabel_t *label,
4393 int flags)
4394 {
4395 struct zsched_arg zarg;
4396 nvlist_t *rctls = NULL;
4397 proc_t *pp = curproc;
4398 zone_t *zone, *ztmp;
4399 zoneid_t zoneid;
4400 int error;
4401 int error2 = 0;
4402 char *str;
4403 cred_t *zkcr;
4404 boolean_t insert_label_hash;
4405
4406 if (secpolicy_zone_config(CRED()) != 0)
4407 return (set_errno(EPERM));
4408
4409 /* can't boot zone from within chroot environment */
4410 if (PTOU(pp)->u_rdir != NULL && PTOU(pp)->u_rdir != rootdir)
4411 return (zone_create_error(ENOTSUP, ZE_CHROOTED,
4412 extended_error));
4413
4414 zone = kmem_zalloc(sizeof (zone_t), KM_SLEEP);
4415 zoneid = zone->zone_id = id_alloc(zoneid_space);
4416 zone->zone_status = ZONE_IS_UNINITIALIZED;
4417 zone->zone_pool = pool_default;
4418 zone->zone_pool_mod = gethrtime();
4419 zone->zone_psetid = ZONE_PS_INVAL;
4420 zone->zone_ncpus = 0;
4421 zone->zone_ncpus_online = 0;
4422 zone->zone_restart_init = B_TRUE;
4423 zone->zone_brand = &native_brand;
4424 zone->zone_initname = NULL;
4425 mutex_init(&zone->zone_lock, NULL, MUTEX_DEFAULT, NULL);
4426 mutex_init(&zone->zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
4427 mutex_init(&zone->zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
4428 cv_init(&zone->zone_cv, NULL, CV_DEFAULT, NULL);
4429 list_create(&zone->zone_ref_list, sizeof (zone_ref_t),
4430 offsetof(zone_ref_t, zref_linkage));
4431 list_create(&zone->zone_zsd, sizeof (struct zsd_entry),
4432 offsetof(struct zsd_entry, zsd_linkage));
4433 list_create(&zone->zone_datasets, sizeof (zone_dataset_t),
4434 offsetof(zone_dataset_t, zd_linkage));
4435 list_create(&zone->zone_dl_list, sizeof (zone_dl_t),
4436 offsetof(zone_dl_t, zdl_linkage));
4437 rw_init(&zone->zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
4438 rw_init(&zone->zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL);
4439
4440 if (flags & ZCF_NET_EXCL) {
4441 zone->zone_flags |= ZF_NET_EXCL;
4442 }
4443
4444 if ((error = zone_set_name(zone, zone_name)) != 0) {
4445 zone_free(zone);
4446 return (zone_create_error(error, 0, extended_error));
4447 }
4448
4449 if ((error = zone_set_root(zone, zone_root)) != 0) {
4450 zone_free(zone);
4451 return (zone_create_error(error, 0, extended_error));
4452 }
4453 if ((error = zone_set_privset(zone, zone_privs, zone_privssz)) != 0) {
4454 zone_free(zone);
4455 return (zone_create_error(error, 0, extended_error));
4456 }
4457
4458 /* initialize node name to be the same as zone name */
4459 zone->zone_nodename = kmem_alloc(_SYS_NMLN, KM_SLEEP);
4460 (void) strncpy(zone->zone_nodename, zone->zone_name, _SYS_NMLN);
4461 zone->zone_nodename[_SYS_NMLN - 1] = '\0';
4462
4463 zone->zone_domain = kmem_alloc(_SYS_NMLN, KM_SLEEP);
4464 zone->zone_domain[0] = '\0';
4465 zone->zone_hostid = HW_INVALID_HOSTID;
4466 zone->zone_shares = 1;
4467 zone->zone_shmmax = 0;
4468 zone->zone_ipc.ipcq_shmmni = 0;
4469 zone->zone_ipc.ipcq_semmni = 0;
4470 zone->zone_ipc.ipcq_msgmni = 0;
4471 zone->zone_bootargs = NULL;
4472 zone->zone_fs_allowed = NULL;
4473
4474 secflags_zero(&zone0.zone_secflags.psf_lower);
4475 secflags_zero(&zone0.zone_secflags.psf_effective);
4476 secflags_zero(&zone0.zone_secflags.psf_inherit);
4477 secflags_fullset(&zone0.zone_secflags.psf_upper);
4478
4479 zone->zone_initname =
4480 kmem_alloc(strlen(zone_default_initname) + 1, KM_SLEEP);
4481 (void) strcpy(zone->zone_initname, zone_default_initname);
4482 zone->zone_nlwps = 0;
4483 zone->zone_nlwps_ctl = INT_MAX;
4484 zone->zone_nprocs = 0;
4485 zone->zone_nprocs_ctl = INT_MAX;
4486 zone->zone_locked_mem = 0;
4487 zone->zone_locked_mem_ctl = UINT64_MAX;
4488 zone->zone_max_swap = 0;
4489 zone->zone_max_swap_ctl = UINT64_MAX;
4490 zone->zone_max_lofi = 0;
4491 zone->zone_max_lofi_ctl = UINT64_MAX;
4492 zone0.zone_lockedmem_kstat = NULL;
4493 zone0.zone_swapresv_kstat = NULL;
4494
4495 /*
4496 * Zsched initializes the rctls.
4497 */
4498 zone->zone_rctls = NULL;
4499
4500 if ((error = parse_rctls(rctlbuf, rctlbufsz, &rctls)) != 0) {
4501 zone_free(zone);
4502 return (zone_create_error(error, 0, extended_error));
4503 }
4504
4505 if ((error = parse_zfs(zone, zfsbuf, zfsbufsz)) != 0) {
4506 zone_free(zone);
4507 return (set_errno(error));
4508 }
4509
4510 /*
4511 * Read in the trusted system parameters:
4512 * match flag and sensitivity label.
4513 */
4514 zone->zone_match = match;
4515 if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) {
4516 /* Fail if requested to set doi to anything but system's doi */
4517 if (doi != 0 && doi != default_doi) {
4518 zone_free(zone);
4519 return (set_errno(EINVAL));
4520 }
4521 /* Always apply system's doi to the zone */
4522 error = zone_set_label(zone, label, default_doi);
4523 if (error != 0) {
4524 zone_free(zone);
4525 return (set_errno(error));
4526 }
4527 insert_label_hash = B_TRUE;
4528 } else {
4529 /* all zones get an admin_low label if system is not labeled */
4530 zone->zone_slabel = l_admin_low;
4531 label_hold(l_admin_low);
4532 insert_label_hash = B_FALSE;
4533 }
4534
4535 /*
4536 * Stop all lwps since that's what normally happens as part of fork().
4537 * This needs to happen before we grab any locks to avoid deadlock
4538 * (another lwp in the process could be waiting for the held lock).
4539 */
4540 if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK)) {
4541 zone_free(zone);
4542 nvlist_free(rctls);
4543 return (zone_create_error(error, 0, extended_error));
4544 }
4545
4546 if (block_mounts(zone) == 0) {
4547 mutex_enter(&pp->p_lock);
4548 if (curthread != pp->p_agenttp)
4549 continuelwps(pp);
4550 mutex_exit(&pp->p_lock);
4551 zone_free(zone);
4552 nvlist_free(rctls);
4553 return (zone_create_error(error, 0, extended_error));
4554 }
4555
4556 /*
4557 * Set up credential for kernel access. After this, any errors
4558 * should go through the dance in errout rather than calling
4559 * zone_free directly.
4560 */
4561 zone->zone_kcred = crdup(kcred);
4562 crsetzone(zone->zone_kcred, zone);
4563 priv_intersect(zone->zone_privset, &CR_PPRIV(zone->zone_kcred));
4564 priv_intersect(zone->zone_privset, &CR_EPRIV(zone->zone_kcred));
4565 priv_intersect(zone->zone_privset, &CR_IPRIV(zone->zone_kcred));
4566 priv_intersect(zone->zone_privset, &CR_LPRIV(zone->zone_kcred));
4567
4568 mutex_enter(&zonehash_lock);
4569 /*
4570 * Make sure zone doesn't already exist.
4571 *
4572 * If the system and zone are labeled,
4573 * make sure no other zone exists that has the same label.
4574 */
4575 if ((ztmp = zone_find_all_by_name(zone->zone_name)) != NULL ||
4576 (insert_label_hash &&
4577 (ztmp = zone_find_all_by_label(zone->zone_slabel)) != NULL)) {
4578 zone_status_t status;
4579
4580 status = zone_status_get(ztmp);
4581 if (status == ZONE_IS_READY || status == ZONE_IS_RUNNING)
4582 error = EEXIST;
4583 else
4584 error = EBUSY;
4585
4586 if (insert_label_hash)
4587 error2 = ZE_LABELINUSE;
4588
4589 goto errout;
4590 }
4591
4592 /*
4593 * Don't allow zone creations which would cause one zone's rootpath to
4594 * be accessible from that of another (non-global) zone.
4595 */
4596 if (zone_is_nested(zone->zone_rootpath)) {
4597 error = EBUSY;
4598 goto errout;
4599 }
4600
4601 ASSERT(zonecount != 0); /* check for leaks */
4602 if (zonecount + 1 > maxzones) {
4603 error = ENOMEM;
4604 goto errout;
4605 }
4606
4607 if (zone_mount_count(zone->zone_rootpath) != 0) {
4608 error = EBUSY;
4609 error2 = ZE_AREMOUNTS;
4610 goto errout;
4611 }
4612
4613 /*
4614 * Zone is still incomplete, but we need to drop all locks while
4615 * zsched() initializes this zone's kernel process. We
4616 * optimistically add the zone to the hashtable and associated
4617 * lists so a parallel zone_create() doesn't try to create the
4618 * same zone.
4619 */
4620 zonecount++;
4621 (void) mod_hash_insert(zonehashbyid,
4622 (mod_hash_key_t)(uintptr_t)zone->zone_id,
4623 (mod_hash_val_t)(uintptr_t)zone);
4624 str = kmem_alloc(strlen(zone->zone_name) + 1, KM_SLEEP);
4625 (void) strcpy(str, zone->zone_name);
4626 (void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)str,
4627 (mod_hash_val_t)(uintptr_t)zone);
4628 if (insert_label_hash) {
4629 (void) mod_hash_insert(zonehashbylabel,
4630 (mod_hash_key_t)zone->zone_slabel, (mod_hash_val_t)zone);
4631 zone->zone_flags |= ZF_HASHED_LABEL;
4632 }
4633
4634 /*
4635 * Insert into active list. At this point there are no 'hold's
4636 * on the zone, but everyone else knows not to use it, so we can
4637 * continue to use it. zsched() will do a zone_hold() if the
4638 * newproc() is successful.
4639 */
4640 list_insert_tail(&zone_active, zone);
4641 mutex_exit(&zonehash_lock);
4642
4643 zarg.zone = zone;
4644 zarg.nvlist = rctls;
4645 /*
4646 * The process, task, and project rctls are probably wrong;
4647 * we need an interface to get the default values of all rctls,
4648 * and initialize zsched appropriately. I'm not sure that that
4649 * makes much of a difference, though.
4650 */
4651 error = newproc(zsched, (void *)&zarg, syscid, minclsyspri, NULL, 0);
4652 if (error != 0) {
4653 /*
4654 * We need to undo all globally visible state.
4655 */
4656 mutex_enter(&zonehash_lock);
4657 list_remove(&zone_active, zone);
4658 if (zone->zone_flags & ZF_HASHED_LABEL) {
4659 ASSERT(zone->zone_slabel != NULL);
4660 (void) mod_hash_destroy(zonehashbylabel,
4661 (mod_hash_key_t)zone->zone_slabel);
4662 }
4663 (void) mod_hash_destroy(zonehashbyname,
4664 (mod_hash_key_t)(uintptr_t)zone->zone_name);
4665 (void) mod_hash_destroy(zonehashbyid,
4666 (mod_hash_key_t)(uintptr_t)zone->zone_id);
4667 ASSERT(zonecount > 1);
4668 zonecount--;
4669 goto errout;
4670 }
4671
4672 /*
4673 * Zone creation can't fail from now on.
4674 */
4675
4676 /*
4677 * Create zone kstats
4678 */
4679 zone_kstat_create(zone);
4680
4681 /*
4682 * Let the other lwps continue.
4683 */
4684 mutex_enter(&pp->p_lock);
4685 if (curthread != pp->p_agenttp)
4686 continuelwps(pp);
4687 mutex_exit(&pp->p_lock);
4688
4689 /*
4690 * Wait for zsched to finish initializing the zone.
4691 */
4692 zone_status_wait(zone, ZONE_IS_READY);
4693 /*
4694 * The zone is fully visible, so we can let mounts progress.
4695 */
4696 resume_mounts(zone);
4697 nvlist_free(rctls);
4698
4699 return (zoneid);
4700
4701 errout:
4702 mutex_exit(&zonehash_lock);
4703 /*
4704 * Let the other lwps continue.
4705 */
4706 mutex_enter(&pp->p_lock);
4707 if (curthread != pp->p_agenttp)
4708 continuelwps(pp);
4709 mutex_exit(&pp->p_lock);
4710
4711 resume_mounts(zone);
4712 nvlist_free(rctls);
4713 /*
4714 * There is currently one reference to the zone, a cred_ref from
4715 * zone_kcred. To free the zone, we call crfree, which will call
4716 * zone_cred_rele, which will call zone_free.
4717 */
4718 ASSERT(zone->zone_cred_ref == 1);
4719 ASSERT(zone->zone_kcred->cr_ref == 1);
4720 ASSERT(zone->zone_ref == 0);
4721 zkcr = zone->zone_kcred;
4722 zone->zone_kcred = NULL;
4723 crfree(zkcr); /* triggers call to zone_free */
4724 return (zone_create_error(error, error2, extended_error));
4725 }
4726
4727 /*
4728 * Cause the zone to boot. This is pretty simple, since we let zoneadmd do
4729 * the heavy lifting. initname is the path to the program to launch
4730 * at the "top" of the zone; if this is NULL, we use the system default,
4731 * which is stored at zone_default_initname.
4732 */
4733 static int
4734 zone_boot(zoneid_t zoneid)
4735 {
4736 int err;
4737 zone_t *zone;
4738
4739 if (secpolicy_zone_config(CRED()) != 0)
4740 return (set_errno(EPERM));
4741 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4742 return (set_errno(EINVAL));
4743
4744 mutex_enter(&zonehash_lock);
4745 /*
4746 * Look for zone under hash lock to prevent races with calls to
4747 * zone_shutdown, zone_destroy, etc.
4748 */
4749 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4750 mutex_exit(&zonehash_lock);
4751 return (set_errno(EINVAL));
4752 }
4753
4754 mutex_enter(&zone_status_lock);
4755 if (zone_status_get(zone) != ZONE_IS_READY) {
4756 mutex_exit(&zone_status_lock);
4757 mutex_exit(&zonehash_lock);
4758 return (set_errno(EINVAL));
4759 }
4760 zone_status_set(zone, ZONE_IS_BOOTING);
4761 mutex_exit(&zone_status_lock);
4762
4763 zone_hold(zone); /* so we can use the zone_t later */
4764 mutex_exit(&zonehash_lock);
4765
4766 if (zone_status_wait_sig(zone, ZONE_IS_RUNNING) == 0) {
4767 zone_rele(zone);
4768 return (set_errno(EINTR));
4769 }
4770
4771 /*
4772 * Boot (starting init) might have failed, in which case the zone
4773 * will go to the SHUTTING_DOWN state; an appropriate errno will
4774 * be placed in zone->zone_boot_err, and so we return that.
4775 */
4776 err = zone->zone_boot_err;
4777 zone_rele(zone);
4778 return (err ? set_errno(err) : 0);
4779 }
4780
4781 /*
4782 * Kills all user processes in the zone, waiting for them all to exit
4783 * before returning.
4784 */
4785 static int
4786 zone_empty(zone_t *zone)
4787 {
4788 int waitstatus;
4789
4790 /*
4791 * We need to drop zonehash_lock before killing all
4792 * processes, otherwise we'll deadlock with zone_find_*
4793 * which can be called from the exit path.
4794 */
4795 ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
4796 while ((waitstatus = zone_status_timedwait_sig(zone,
4797 ddi_get_lbolt() + hz, ZONE_IS_EMPTY)) == -1) {
4798 killall(zone->zone_id);
4799 }
4800 /*
4801 * return EINTR if we were signaled
4802 */
4803 if (waitstatus == 0)
4804 return (EINTR);
4805 return (0);
4806 }
4807
4808 /*
4809 * This function implements the policy for zone visibility.
4810 *
4811 * In standard Solaris, a non-global zone can only see itself.
4812 *
4813 * In Trusted Extensions, a labeled zone can lookup any zone whose label
4814 * it dominates. For this test, the label of the global zone is treated as
4815 * admin_high so it is special-cased instead of being checked for dominance.
4816 *
4817 * Returns true if zone attributes are viewable, false otherwise.
4818 */
4819 static boolean_t
4820 zone_list_access(zone_t *zone)
4821 {
4822
4823 if (curproc->p_zone == global_zone ||
4824 curproc->p_zone == zone) {
4825 return (B_TRUE);
4826 } else if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) {
4827 bslabel_t *curproc_label;
4828 bslabel_t *zone_label;
4829
4830 curproc_label = label2bslabel(curproc->p_zone->zone_slabel);
4831 zone_label = label2bslabel(zone->zone_slabel);
4832
4833 if (zone->zone_id != GLOBAL_ZONEID &&
4834 bldominates(curproc_label, zone_label)) {
4835 return (B_TRUE);
4836 } else {
4837 return (B_FALSE);
4838 }
4839 } else {
4840 return (B_FALSE);
4841 }
4842 }
4843
4844 /*
4845 * Systemcall to start the zone's halt sequence. By the time this
4846 * function successfully returns, all user processes and kernel threads
4847 * executing in it will have exited, ZSD shutdown callbacks executed,
4848 * and the zone status set to ZONE_IS_DOWN.
4849 *
4850 * It is possible that the call will interrupt itself if the caller is the
4851 * parent of any process running in the zone, and doesn't have SIGCHLD blocked.
4852 */
4853 static int
4854 zone_shutdown(zoneid_t zoneid)
4855 {
4856 int error;
4857 zone_t *zone;
4858 zone_status_t status;
4859
4860 if (secpolicy_zone_config(CRED()) != 0)
4861 return (set_errno(EPERM));
4862 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4863 return (set_errno(EINVAL));
4864
4865 mutex_enter(&zonehash_lock);
4866 /*
4867 * Look for zone under hash lock to prevent races with other
4868 * calls to zone_shutdown and zone_destroy.
4869 */
4870 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4871 mutex_exit(&zonehash_lock);
4872 return (set_errno(EINVAL));
4873 }
4874
4875 /*
4876 * We have to drop zonehash_lock before calling block_mounts.
4877 * Hold the zone so we can continue to use the zone_t.
4878 */
4879 zone_hold(zone);
4880 mutex_exit(&zonehash_lock);
4881
4882 /*
4883 * Block mounts so that VFS_MOUNT() can get an accurate view of
4884 * the zone's status with regards to ZONE_IS_SHUTTING down.
4885 *
4886 * e.g. NFS can fail the mount if it determines that the zone
4887 * has already begun the shutdown sequence.
4888 *
4889 */
4890 if (block_mounts(zone) == 0) {
4891 zone_rele(zone);
4892 return (set_errno(EINTR));
4893 }
4894
4895 mutex_enter(&zonehash_lock);
4896 mutex_enter(&zone_status_lock);
4897 status = zone_status_get(zone);
4898 /*
4899 * Fail if the zone isn't fully initialized yet.
4900 */
4901 if (status < ZONE_IS_READY) {
4902 mutex_exit(&zone_status_lock);
4903 mutex_exit(&zonehash_lock);
4904 resume_mounts(zone);
4905 zone_rele(zone);
4906 return (set_errno(EINVAL));
4907 }
4908 /*
4909 * If conditions required for zone_shutdown() to return have been met,
4910 * return success.
4911 */
4912 if (status >= ZONE_IS_DOWN) {
4913 mutex_exit(&zone_status_lock);
4914 mutex_exit(&zonehash_lock);
4915 resume_mounts(zone);
4916 zone_rele(zone);
4917 return (0);
4918 }
4919 /*
4920 * If zone_shutdown() hasn't been called before, go through the motions.
4921 * If it has, there's nothing to do but wait for the kernel threads to
4922 * drain.
4923 */
4924 if (status < ZONE_IS_EMPTY) {
4925 uint_t ntasks;
4926
4927 mutex_enter(&zone->zone_lock);
4928 if ((ntasks = zone->zone_ntasks) != 1) {
4929 /*
4930 * There's still stuff running.
4931 */
4932 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
4933 }
4934 mutex_exit(&zone->zone_lock);
4935 if (ntasks == 1) {
4936 /*
4937 * The only way to create another task is through
4938 * zone_enter(), which will block until we drop
4939 * zonehash_lock. The zone is empty.
4940 */
4941 if (zone->zone_kthreads == NULL) {
4942 /*
4943 * Skip ahead to ZONE_IS_DOWN
4944 */
4945 zone_status_set(zone, ZONE_IS_DOWN);
4946 } else {
4947 zone_status_set(zone, ZONE_IS_EMPTY);
4948 }
4949 }
4950 }
4951 mutex_exit(&zone_status_lock);
4952 mutex_exit(&zonehash_lock);
4953 resume_mounts(zone);
4954
4955 if (error = zone_empty(zone)) {
4956 zone_rele(zone);
4957 return (set_errno(error));
4958 }
4959 /*
4960 * After the zone status goes to ZONE_IS_DOWN this zone will no
4961 * longer be notified of changes to the pools configuration, so
4962 * in order to not end up with a stale pool pointer, we point
4963 * ourselves at the default pool and remove all resource
4964 * visibility. This is especially important as the zone_t may
4965 * languish on the deathrow for a very long time waiting for
4966 * cred's to drain out.
4967 *
4968 * This rebinding of the zone can happen multiple times
4969 * (presumably due to interrupted or parallel systemcalls)
4970 * without any adverse effects.
4971 */
4972 if (pool_lock_intr() != 0) {
4973 zone_rele(zone);
4974 return (set_errno(EINTR));
4975 }
4976 if (pool_state == POOL_ENABLED) {
4977 mutex_enter(&cpu_lock);
4978 zone_pool_set(zone, pool_default);
4979 /*
4980 * The zone no longer needs to be able to see any cpus.
4981 */
4982 zone_pset_set(zone, ZONE_PS_INVAL);
4983 mutex_exit(&cpu_lock);
4984 }
4985 pool_unlock();
4986
4987 /*
4988 * ZSD shutdown callbacks can be executed multiple times, hence
4989 * it is safe to not be holding any locks across this call.
4990 */
4991 zone_zsd_callbacks(zone, ZSD_SHUTDOWN);
4992
4993 mutex_enter(&zone_status_lock);
4994 if (zone->zone_kthreads == NULL && zone_status_get(zone) < ZONE_IS_DOWN)
4995 zone_status_set(zone, ZONE_IS_DOWN);
4996 mutex_exit(&zone_status_lock);
4997
4998 /*
4999 * Wait for kernel threads to drain.
5000 */
5001 if (!zone_status_wait_sig(zone, ZONE_IS_DOWN)) {
5002 zone_rele(zone);
5003 return (set_errno(EINTR));
5004 }
5005
5006 /*
5007 * Zone can be become down/destroyable even if the above wait
5008 * returns EINTR, so any code added here may never execute.
5009 * (i.e. don't add code here)
5010 */
5011
5012 zone_rele(zone);
5013 return (0);
5014 }
5015
5016 /*
5017 * Log the specified zone's reference counts. The caller should not be
5018 * holding the zone's zone_lock.
5019 */
5020 static void
5021 zone_log_refcounts(zone_t *zone)
5022 {
5023 char *buffer;
5024 char *buffer_position;
5025 uint32_t buffer_size;
5026 uint32_t index;
5027 uint_t ref;
5028 uint_t cred_ref;
5029
5030 /*
5031 * Construct a string representing the subsystem-specific reference
5032 * counts. The counts are printed in ascending order by index into the
5033 * zone_t::zone_subsys_ref array. The list will be surrounded by
5034 * square brackets [] and will only contain nonzero reference counts.
5035 *
5036 * The buffer will hold two square bracket characters plus ten digits,
5037 * one colon, one space, one comma, and some characters for a
5038 * subsystem name per subsystem-specific reference count. (Unsigned 32-
5039 * bit integers have at most ten decimal digits.) The last
5040 * reference count's comma is replaced by the closing square
5041 * bracket and a NULL character to terminate the string.
5042 *
5043 * NOTE: We have to grab the zone's zone_lock to create a consistent
5044 * snapshot of the zone's reference counters.
5045 *
5046 * First, figure out how much space the string buffer will need.
5047 * The buffer's size is stored in buffer_size.
5048 */
5049 buffer_size = 2; /* for the square brackets */
5050 mutex_enter(&zone->zone_lock);
5051 zone->zone_flags |= ZF_REFCOUNTS_LOGGED;
5052 ref = zone->zone_ref;
5053 cred_ref = zone->zone_cred_ref;
5054 for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index)
5055 if (zone->zone_subsys_ref[index] != 0)
5056 buffer_size += strlen(zone_ref_subsys_names[index]) +
5057 13;
5058 if (buffer_size == 2) {
5059 /*
5060 * No subsystems had nonzero reference counts. Don't bother
5061 * with allocating a buffer; just log the general-purpose and
5062 * credential reference counts.
5063 */
5064 mutex_exit(&zone->zone_lock);
5065 (void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE,
5066 "Zone '%s' (ID: %d) is shutting down, but %u zone "
5067 "references and %u credential references are still extant",
5068 zone->zone_name, zone->zone_id, ref, cred_ref);
5069 return;
5070 }
5071
5072 /*
5073 * buffer_size contains the exact number of characters that the
5074 * buffer will need. Allocate the buffer and fill it with nonzero
5075 * subsystem-specific reference counts. Surround the results with
5076 * square brackets afterwards.
5077 */
5078 buffer = kmem_alloc(buffer_size, KM_SLEEP);
5079 buffer_position = &buffer[1];
5080 for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index) {
5081 /*
5082 * NOTE: The DDI's version of sprintf() returns a pointer to
5083 * the modified buffer rather than the number of bytes written
5084 * (as in snprintf(3C)). This is unfortunate and annoying.
5085 * Therefore, we'll use snprintf() with INT_MAX to get the
5086 * number of bytes written. Using INT_MAX is safe because
5087 * the buffer is perfectly sized for the data: we'll never
5088 * overrun the buffer.
5089 */
5090 if (zone->zone_subsys_ref[index] != 0)
5091 buffer_position += snprintf(buffer_position, INT_MAX,
5092 "%s: %u,", zone_ref_subsys_names[index],
5093 zone->zone_subsys_ref[index]);
5094 }
5095 mutex_exit(&zone->zone_lock);
5096 buffer[0] = '[';
5097 ASSERT((uintptr_t)(buffer_position - buffer) < buffer_size);
5098 ASSERT(buffer_position[0] == '\0' && buffer_position[-1] == ',');
5099 buffer_position[-1] = ']';
5100
5101 /*
5102 * Log the reference counts and free the message buffer.
5103 */
5104 (void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE,
5105 "Zone '%s' (ID: %d) is shutting down, but %u zone references and "
5106 "%u credential references are still extant %s", zone->zone_name,
5107 zone->zone_id, ref, cred_ref, buffer);
5108 kmem_free(buffer, buffer_size);
5109 }
5110
5111 /*
5112 * Systemcall entry point to finalize the zone halt process. The caller
5113 * must have already successfully called zone_shutdown().
5114 *
5115 * Upon successful completion, the zone will have been fully destroyed:
5116 * zsched will have exited, destructor callbacks executed, and the zone
5117 * removed from the list of active zones.
5118 */
5119 static int
5120 zone_destroy(zoneid_t zoneid)
5121 {
5122 uint64_t uniqid;
5123 zone_t *zone;
5124 zone_status_t status;
5125 clock_t wait_time;
5126 boolean_t log_refcounts;
5127
5128 if (secpolicy_zone_config(CRED()) != 0)
5129 return (set_errno(EPERM));
5130 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
5131 return (set_errno(EINVAL));
5132
5133 mutex_enter(&zonehash_lock);
5134 /*
5135 * Look for zone under hash lock to prevent races with other
5136 * calls to zone_destroy.
5137 */
5138 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5139 mutex_exit(&zonehash_lock);
5140 return (set_errno(EINVAL));
5141 }
5142
5143 if (zone_mount_count(zone->zone_rootpath) != 0) {
5144 mutex_exit(&zonehash_lock);
5145 return (set_errno(EBUSY));
5146 }
5147 mutex_enter(&zone_status_lock);
5148 status = zone_status_get(zone);
5149 if (status < ZONE_IS_DOWN) {
5150 mutex_exit(&zone_status_lock);
5151 mutex_exit(&zonehash_lock);
5152 return (set_errno(EBUSY));
5153 } else if (status == ZONE_IS_DOWN) {
5154 zone_status_set(zone, ZONE_IS_DYING); /* Tell zsched to exit */
5155 }
5156 mutex_exit(&zone_status_lock);
5157 zone_hold(zone);
5158 mutex_exit(&zonehash_lock);
5159
5160 /*
5161 * wait for zsched to exit
5162 */
5163 zone_status_wait(zone, ZONE_IS_DEAD);
5164 zone_zsd_callbacks(zone, ZSD_DESTROY);
5165 zone->zone_netstack = NULL;
5166 uniqid = zone->zone_uniqid;
5167 zone_rele(zone);
5168 zone = NULL; /* potentially free'd */
5169
5170 log_refcounts = B_FALSE;
5171 wait_time = SEC_TO_TICK(ZONE_DESTROY_TIMEOUT_SECS);
5172 mutex_enter(&zonehash_lock);
5173 for (; /* ever */; ) {
5174 boolean_t unref;
5175 boolean_t refs_have_been_logged;
5176
5177 if ((zone = zone_find_all_by_id(zoneid)) == NULL ||
5178 zone->zone_uniqid != uniqid) {
5179 /*
5180 * The zone has gone away. Necessary conditions
5181 * are met, so we return success.
5182 */
5183 mutex_exit(&zonehash_lock);
5184 return (0);
5185 }
5186 mutex_enter(&zone->zone_lock);
5187 unref = ZONE_IS_UNREF(zone);
5188 refs_have_been_logged = (zone->zone_flags &
5189 ZF_REFCOUNTS_LOGGED);
5190 mutex_exit(&zone->zone_lock);
5191 if (unref) {
5192 /*
5193 * There is only one reference to the zone -- that
5194 * added when the zone was added to the hashtables --
5195 * and things will remain this way until we drop
5196 * zonehash_lock... we can go ahead and cleanup the
5197 * zone.
5198 */
5199 break;
5200 }
5201
5202 /*
5203 * Wait for zone_rele_common() or zone_cred_rele() to signal
5204 * zone_destroy_cv. zone_destroy_cv is signaled only when
5205 * some zone's general-purpose reference count reaches one.
5206 * If ZONE_DESTROY_TIMEOUT_SECS seconds elapse while waiting
5207 * on zone_destroy_cv, then log the zone's reference counts and
5208 * continue to wait for zone_rele() and zone_cred_rele().
5209 */
5210 if (!refs_have_been_logged) {
5211 if (!log_refcounts) {
5212 /*
5213 * This thread hasn't timed out waiting on
5214 * zone_destroy_cv yet. Wait wait_time clock
5215 * ticks (initially ZONE_DESTROY_TIMEOUT_SECS
5216 * seconds) for the zone's references to clear.
5217 */
5218 ASSERT(wait_time > 0);
5219 wait_time = cv_reltimedwait_sig(
5220 &zone_destroy_cv, &zonehash_lock, wait_time,
5221 TR_SEC);
5222 if (wait_time > 0) {
5223 /*
5224 * A thread in zone_rele() or
5225 * zone_cred_rele() signaled
5226 * zone_destroy_cv before this thread's
5227 * wait timed out. The zone might have
5228 * only one reference left; find out!
5229 */
5230 continue;
5231 } else if (wait_time == 0) {
5232 /* The thread's process was signaled. */
5233 mutex_exit(&zonehash_lock);
5234 return (set_errno(EINTR));
5235 }
5236
5237 /*
5238 * The thread timed out while waiting on
5239 * zone_destroy_cv. Even though the thread
5240 * timed out, it has to check whether another
5241 * thread woke up from zone_destroy_cv and
5242 * destroyed the zone.
5243 *
5244 * If the zone still exists and has more than
5245 * one unreleased general-purpose reference,
5246 * then log the zone's reference counts.
5247 */
5248 log_refcounts = B_TRUE;
5249 continue;
5250 }
5251
5252 /*
5253 * The thread already timed out on zone_destroy_cv while
5254 * waiting for subsystems to release the zone's last
5255 * general-purpose references. Log the zone's reference
5256 * counts and wait indefinitely on zone_destroy_cv.
5257 */
5258 zone_log_refcounts(zone);
5259 }
5260 if (cv_wait_sig(&zone_destroy_cv, &zonehash_lock) == 0) {
5261 /* The thread's process was signaled. */
5262 mutex_exit(&zonehash_lock);
5263 return (set_errno(EINTR));
5264 }
5265 }
5266
5267 /*
5268 * Remove CPU cap for this zone now since we're not going to
5269 * fail below this point.
5270 */
5271 cpucaps_zone_remove(zone);
5272
5273 /* Get rid of the zone's kstats */
5274 zone_kstat_delete(zone);
5275
5276 /* remove the pfexecd doors */
5277 if (zone->zone_pfexecd != NULL) {
5278 klpd_freelist(&zone->zone_pfexecd);
5279 zone->zone_pfexecd = NULL;
5280 }
5281
5282 /* free brand specific data */
5283 if (ZONE_IS_BRANDED(zone))
5284 ZBROP(zone)->b_free_brand_data(zone);
5285
5286 /* Say goodbye to brand framework. */
5287 brand_unregister_zone(zone->zone_brand);
5288
5289 /*
5290 * It is now safe to let the zone be recreated; remove it from the
5291 * lists. The memory will not be freed until the last cred
5292 * reference goes away.
5293 */
5294 ASSERT(zonecount > 1); /* must be > 1; can't destroy global zone */
5295 zonecount--;
5296 /* remove from active list and hash tables */
5297 list_remove(&zone_active, zone);
5298 (void) mod_hash_destroy(zonehashbyname,
5299 (mod_hash_key_t)zone->zone_name);
5300 (void) mod_hash_destroy(zonehashbyid,
5301 (mod_hash_key_t)(uintptr_t)zone->zone_id);
5302 if (zone->zone_flags & ZF_HASHED_LABEL)
5303 (void) mod_hash_destroy(zonehashbylabel,
5304 (mod_hash_key_t)zone->zone_slabel);
5305 mutex_exit(&zonehash_lock);
5306
5307 /*
5308 * Release the root vnode; we're not using it anymore. Nor should any
5309 * other thread that might access it exist.
5310 */
5311 if (zone->zone_rootvp != NULL) {
5312 VN_RELE(zone->zone_rootvp);
5313 zone->zone_rootvp = NULL;
5314 }
5315
5316 /* add to deathrow list */
5317 mutex_enter(&zone_deathrow_lock);
5318 list_insert_tail(&zone_deathrow, zone);
5319 mutex_exit(&zone_deathrow_lock);
5320
5321 /*
5322 * Drop last reference (which was added by zsched()), this will
5323 * free the zone unless there are outstanding cred references.
5324 */
5325 zone_rele(zone);
5326 return (0);
5327 }
5328
5329 /*
5330 * Systemcall entry point for zone_getattr(2).
5331 */
5332 static ssize_t
5333 zone_getattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
5334 {
5335 size_t size;
5336 int error = 0, err;
5337 zone_t *zone;
5338 char *zonepath;
5339 char *outstr;
5340 zone_status_t zone_status;
5341 pid_t initpid;
5342 boolean_t global = (curzone == global_zone);
5343 boolean_t inzone = (curzone->zone_id == zoneid);
5344 ushort_t flags;
5345 zone_net_data_t *zbuf;
5346
5347 mutex_enter(&zonehash_lock);
5348 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5349 mutex_exit(&zonehash_lock);
5350 return (set_errno(EINVAL));
5351 }
5352 zone_status = zone_status_get(zone);
5353 if (zone_status < ZONE_IS_INITIALIZED) {
5354 mutex_exit(&zonehash_lock);
5355 return (set_errno(EINVAL));
5356 }
5357 zone_hold(zone);
5358 mutex_exit(&zonehash_lock);
5359
5360 /*
5361 * If not in the global zone, don't show information about other zones,
5362 * unless the system is labeled and the local zone's label dominates
5363 * the other zone.
5364 */
5365 if (!zone_list_access(zone)) {
5366 zone_rele(zone);
5367 return (set_errno(EINVAL));
5368 }
5369
5370 switch (attr) {
5371 case ZONE_ATTR_ROOT:
5372 if (global) {
5373 /*
5374 * Copy the path to trim the trailing "/" (except for
5375 * the global zone).
5376 */
5377 if (zone != global_zone)
5378 size = zone->zone_rootpathlen - 1;
5379 else
5380 size = zone->zone_rootpathlen;
5381 zonepath = kmem_alloc(size, KM_SLEEP);
5382 bcopy(zone->zone_rootpath, zonepath, size);
5383 zonepath[size - 1] = '\0';
5384 } else {
5385 if (inzone || !is_system_labeled()) {
5386 /*
5387 * Caller is not in the global zone.
5388 * if the query is on the current zone
5389 * or the system is not labeled,
5390 * just return faked-up path for current zone.
5391 */
5392 zonepath = "/";
5393 size = 2;
5394 } else {
5395 /*
5396 * Return related path for current zone.
5397 */
5398 int prefix_len = strlen(zone_prefix);
5399 int zname_len = strlen(zone->zone_name);
5400
5401 size = prefix_len + zname_len + 1;
5402 zonepath = kmem_alloc(size, KM_SLEEP);
5403 bcopy(zone_prefix, zonepath, prefix_len);
5404 bcopy(zone->zone_name, zonepath +
5405 prefix_len, zname_len);
5406 zonepath[size - 1] = '\0';
5407 }
5408 }
5409 if (bufsize > size)
5410 bufsize = size;
5411 if (buf != NULL) {
5412 err = copyoutstr(zonepath, buf, bufsize, NULL);
5413 if (err != 0 && err != ENAMETOOLONG)
5414 error = EFAULT;
5415 }
5416 if (global || (is_system_labeled() && !inzone))
5417 kmem_free(zonepath, size);
5418 break;
5419
5420 case ZONE_ATTR_NAME:
5421 size = strlen(zone->zone_name) + 1;
5422 if (bufsize > size)
5423 bufsize = size;
5424 if (buf != NULL) {
5425 err = copyoutstr(zone->zone_name, buf, bufsize, NULL);
5426 if (err != 0 && err != ENAMETOOLONG)
5427 error = EFAULT;
5428 }
5429 break;
5430
5431 case ZONE_ATTR_STATUS:
5432 /*
5433 * Since we're not holding zonehash_lock, the zone status
5434 * may be anything; leave it up to userland to sort it out.
5435 */
5436 size = sizeof (zone_status);
5437 if (bufsize > size)
5438 bufsize = size;
5439 zone_status = zone_status_get(zone);
5440 if (buf != NULL &&
5441 copyout(&zone_status, buf, bufsize) != 0)
5442 error = EFAULT;
5443 break;
5444 case ZONE_ATTR_FLAGS:
5445 size = sizeof (zone->zone_flags);
5446 if (bufsize > size)
5447 bufsize = size;
5448 flags = zone->zone_flags;
5449 if (buf != NULL &&
5450 copyout(&flags, buf, bufsize) != 0)
5451 error = EFAULT;
5452 break;
5453 case ZONE_ATTR_PRIVSET:
5454 size = sizeof (priv_set_t);
5455 if (bufsize > size)
5456 bufsize = size;
5457 if (buf != NULL &&
5458 copyout(zone->zone_privset, buf, bufsize) != 0)
5459 error = EFAULT;
5460 break;
5461 case ZONE_ATTR_UNIQID:
5462 size = sizeof (zone->zone_uniqid);
5463 if (bufsize > size)
5464 bufsize = size;
5465 if (buf != NULL &&
5466 copyout(&zone->zone_uniqid, buf, bufsize) != 0)
5467 error = EFAULT;
5468 break;
5469 case ZONE_ATTR_POOLID:
5470 {
5471 pool_t *pool;
5472 poolid_t poolid;
5473
5474 if (pool_lock_intr() != 0) {
5475 error = EINTR;
5476 break;
5477 }
5478 pool = zone_pool_get(zone);
5479 poolid = pool->pool_id;
5480 pool_unlock();
5481 size = sizeof (poolid);
5482 if (bufsize > size)
5483 bufsize = size;
5484 if (buf != NULL && copyout(&poolid, buf, size) != 0)
5485 error = EFAULT;
5486 }
5487 break;
5488 case ZONE_ATTR_SLBL:
5489 size = sizeof (bslabel_t);
5490 if (bufsize > size)
5491 bufsize = size;
5492 if (zone->zone_slabel == NULL)
5493 error = EINVAL;
5494 else if (buf != NULL &&
5495 copyout(label2bslabel(zone->zone_slabel), buf,
5496 bufsize) != 0)
5497 error = EFAULT;
5498 break;
5499 case ZONE_ATTR_INITPID:
5500 size = sizeof (initpid);
5501 if (bufsize > size)
5502 bufsize = size;
5503 initpid = zone->zone_proc_initpid;
5504 if (initpid == -1) {
5505 error = ESRCH;
5506 break;
5507 }
5508 if (buf != NULL &&
5509 copyout(&initpid, buf, bufsize) != 0)
5510 error = EFAULT;
5511 break;
5512 case ZONE_ATTR_BRAND:
5513 size = strlen(zone->zone_brand->b_name) + 1;
5514
5515 if (bufsize > size)
5516 bufsize = size;
5517 if (buf != NULL) {
5518 err = copyoutstr(zone->zone_brand->b_name, buf,
5519 bufsize, NULL);
5520 if (err != 0 && err != ENAMETOOLONG)
5521 error = EFAULT;
5522 }
5523 break;
5524 case ZONE_ATTR_INITNAME:
5525 size = strlen(zone->zone_initname) + 1;
5526 if (bufsize > size)
5527 bufsize = size;
5528 if (buf != NULL) {
5529 err = copyoutstr(zone->zone_initname, buf, bufsize,
5530 NULL);
5531 if (err != 0 && err != ENAMETOOLONG)
5532 error = EFAULT;
5533 }
5534 break;
5535 case ZONE_ATTR_BOOTARGS:
5536 if (zone->zone_bootargs == NULL)
5537 outstr = "";
5538 else
5539 outstr = zone->zone_bootargs;
5540 size = strlen(outstr) + 1;
5541 if (bufsize > size)
5542 bufsize = size;
5543 if (buf != NULL) {
5544 err = copyoutstr(outstr, buf, bufsize, NULL);
5545 if (err != 0 && err != ENAMETOOLONG)
5546 error = EFAULT;
5547 }
5548 break;
5549 case ZONE_ATTR_PHYS_MCAP:
5550 size = sizeof (zone->zone_phys_mcap);
5551 if (bufsize > size)
5552 bufsize = size;
5553 if (buf != NULL &&
5554 copyout(&zone->zone_phys_mcap, buf, bufsize) != 0)
5555 error = EFAULT;
5556 break;
5557 case ZONE_ATTR_SCHED_CLASS:
5558 mutex_enter(&class_lock);
5559
5560 if (zone->zone_defaultcid >= loaded_classes)
5561 outstr = "";
5562 else
5563 outstr = sclass[zone->zone_defaultcid].cl_name;
5564 size = strlen(outstr) + 1;
5565 if (bufsize > size)
5566 bufsize = size;
5567 if (buf != NULL) {
5568 err = copyoutstr(outstr, buf, bufsize, NULL);
5569 if (err != 0 && err != ENAMETOOLONG)
5570 error = EFAULT;
5571 }
5572
5573 mutex_exit(&class_lock);
5574 break;
5575 case ZONE_ATTR_HOSTID:
5576 if (zone->zone_hostid != HW_INVALID_HOSTID &&
5577 bufsize == sizeof (zone->zone_hostid)) {
5578 size = sizeof (zone->zone_hostid);
5579 if (buf != NULL && copyout(&zone->zone_hostid, buf,
5580 bufsize) != 0)
5581 error = EFAULT;
5582 } else {
5583 error = EINVAL;
5584 }
5585 break;
5586 case ZONE_ATTR_FS_ALLOWED:
5587 if (zone->zone_fs_allowed == NULL)
5588 outstr = "";
5589 else
5590 outstr = zone->zone_fs_allowed;
5591 size = strlen(outstr) + 1;
5592 if (bufsize > size)
5593 bufsize = size;
5594 if (buf != NULL) {
5595 err = copyoutstr(outstr, buf, bufsize, NULL);
5596 if (err != 0 && err != ENAMETOOLONG)
5597 error = EFAULT;
5598 }
5599 break;
5600 case ZONE_ATTR_SECFLAGS:
5601 size = sizeof (zone->zone_secflags);
5602 if (bufsize > size)
5603 bufsize = size;
5604 if ((err = copyout(&zone->zone_secflags, buf, bufsize)) != 0)
5605 error = EFAULT;
5606 break;
5607 case ZONE_ATTR_NETWORK:
5608 zbuf = kmem_alloc(bufsize, KM_SLEEP);
5609 if (copyin(buf, zbuf, bufsize) != 0) {
5610 error = EFAULT;
5611 } else {
5612 error = zone_get_network(zoneid, zbuf);
5613 if (error == 0 && copyout(zbuf, buf, bufsize) != 0)
5614 error = EFAULT;
5615 }
5616 kmem_free(zbuf, bufsize);
5617 break;
5618 default:
5619 if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone)) {
5620 size = bufsize;
5621 error = ZBROP(zone)->b_getattr(zone, attr, buf, &size);
5622 } else {
5623 error = EINVAL;
5624 }
5625 }
5626 zone_rele(zone);
5627
5628 if (error)
5629 return (set_errno(error));
5630 return ((ssize_t)size);
5631 }
5632
5633 /*
5634 * Systemcall entry point for zone_setattr(2).
5635 */
5636 /*ARGSUSED*/
5637 static int
5638 zone_setattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
5639 {
5640 zone_t *zone;
5641 zone_status_t zone_status;
5642 int err = -1;
5643 zone_net_data_t *zbuf;
5644
5645 if (secpolicy_zone_config(CRED()) != 0)
5646 return (set_errno(EPERM));
5647
5648 /*
5649 * Only the ZONE_ATTR_PHYS_MCAP attribute can be set on the
5650 * global zone.
5651 */
5652 if (zoneid == GLOBAL_ZONEID && attr != ZONE_ATTR_PHYS_MCAP) {
5653 return (set_errno(EINVAL));
5654 }
5655
5656 mutex_enter(&zonehash_lock);
5657 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5658 mutex_exit(&zonehash_lock);
5659 return (set_errno(EINVAL));
5660 }
5661 zone_hold(zone);
5662 mutex_exit(&zonehash_lock);
5663
5664 /*
5665 * At present most attributes can only be set on non-running,
5666 * non-global zones.
5667 */
5668 zone_status = zone_status_get(zone);
5669 if (attr != ZONE_ATTR_PHYS_MCAP && zone_status > ZONE_IS_READY) {
5670 err = EINVAL;
5671 goto done;
5672 }
5673
5674 switch (attr) {
5675 case ZONE_ATTR_INITNAME:
5676 err = zone_set_initname(zone, (const char *)buf);
5677 break;
5678 case ZONE_ATTR_INITNORESTART:
5679 zone->zone_restart_init = B_FALSE;
5680 err = 0;
5681 break;
5682 case ZONE_ATTR_BOOTARGS:
5683 err = zone_set_bootargs(zone, (const char *)buf);
5684 break;
5685 case ZONE_ATTR_BRAND:
5686 err = zone_set_brand(zone, (const char *)buf);
5687 break;
5688 case ZONE_ATTR_FS_ALLOWED:
5689 err = zone_set_fs_allowed(zone, (const char *)buf);
5690 break;
5691 case ZONE_ATTR_SECFLAGS:
5692 err = zone_set_secflags(zone, (psecflags_t *)buf);
5693 break;
5694 case ZONE_ATTR_PHYS_MCAP:
5695 err = zone_set_phys_mcap(zone, (const uint64_t *)buf);
5696 break;
5697 case ZONE_ATTR_SCHED_CLASS:
5698 err = zone_set_sched_class(zone, (const char *)buf);
5699 break;
5700 case ZONE_ATTR_HOSTID:
5701 if (bufsize == sizeof (zone->zone_hostid)) {
5702 if (copyin(buf, &zone->zone_hostid, bufsize) == 0)
5703 err = 0;
5704 else
5705 err = EFAULT;
5706 } else {
5707 err = EINVAL;
5708 }
5709 break;
5710 case ZONE_ATTR_NETWORK:
5711 if (bufsize > (PIPE_BUF + sizeof (zone_net_data_t))) {
5712 err = EINVAL;
5713 break;
5714 }
5715 zbuf = kmem_alloc(bufsize, KM_SLEEP);
5716 if (copyin(buf, zbuf, bufsize) != 0) {
5717 kmem_free(zbuf, bufsize);
5718 err = EFAULT;
5719 break;
5720 }
5721 err = zone_set_network(zoneid, zbuf);
5722 kmem_free(zbuf, bufsize);
5723 break;
5724 default:
5725 if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone))
5726 err = ZBROP(zone)->b_setattr(zone, attr, buf, bufsize);
5727 else
5728 err = EINVAL;
5729 }
5730
5731 done:
5732 zone_rele(zone);
5733 ASSERT(err != -1);
5734 return (err != 0 ? set_errno(err) : 0);
5735 }
5736
5737 /*
5738 * Return zero if the process has at least one vnode mapped in to its
5739 * address space which shouldn't be allowed to change zones.
5740 *
5741 * Also return zero if the process has any shared mappings which reserve
5742 * swap. This is because the counting for zone.max-swap does not allow swap
5743 * reservation to be shared between zones. zone swap reservation is counted
5744 * on zone->zone_max_swap.
5745 */
5746 static int
5747 as_can_change_zones(void)
5748 {
5749 proc_t *pp = curproc;
5750 struct seg *seg;
5751 struct as *as = pp->p_as;
5752 vnode_t *vp;
5753 int allow = 1;
5754
5755 ASSERT(pp->p_as != &kas);
5756 AS_LOCK_ENTER(as, RW_READER);
5757 for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) {
5758
5759 /*
5760 * Cannot enter zone with shared anon memory which
5761 * reserves swap. See comment above.
5762 */
5763 if (seg_can_change_zones(seg) == B_FALSE) {
5764 allow = 0;
5765 break;
5766 }
5767 /*
5768 * if we can't get a backing vnode for this segment then skip
5769 * it.
5770 */
5771 vp = NULL;
5772 if (SEGOP_GETVP(seg, seg->s_base, &vp) != 0 || vp == NULL)
5773 continue;
5774 if (!vn_can_change_zones(vp)) { /* bail on first match */
5775 allow = 0;
5776 break;
5777 }
5778 }
5779 AS_LOCK_EXIT(as);
5780 return (allow);
5781 }
5782
5783 /*
5784 * Count swap reserved by curproc's address space
5785 */
5786 static size_t
5787 as_swresv(void)
5788 {
5789 proc_t *pp = curproc;
5790 struct seg *seg;
5791 struct as *as = pp->p_as;
5792 size_t swap = 0;
5793
5794 ASSERT(pp->p_as != &kas);
5795 ASSERT(AS_WRITE_HELD(as));
5796 for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg))
5797 swap += seg_swresv(seg);
5798
5799 return (swap);
5800 }
5801
5802 /*
5803 * Systemcall entry point for zone_enter().
5804 *
5805 * The current process is injected into said zone. In the process
5806 * it will change its project membership, privileges, rootdir/cwd,
5807 * zone-wide rctls, and pool association to match those of the zone.
5808 *
5809 * The first zone_enter() called while the zone is in the ZONE_IS_READY
5810 * state will transition it to ZONE_IS_RUNNING. Processes may only
5811 * enter a zone that is "ready" or "running".
5812 */
5813 static int
5814 zone_enter(zoneid_t zoneid)
5815 {
5816 zone_t *zone;
5817 vnode_t *vp;
5818 proc_t *pp = curproc;
5819 contract_t *ct;
5820 cont_process_t *ctp;
5821 task_t *tk, *oldtk;
5822 kproject_t *zone_proj0;
5823 cred_t *cr, *newcr;
5824 pool_t *oldpool, *newpool;
5825 sess_t *sp;
5826 uid_t uid;
5827 zone_status_t status;
5828 int err = 0;
5829 rctl_entity_p_t e;
5830 size_t swap;
5831 kthread_id_t t;
5832
5833 if (secpolicy_zone_config(CRED()) != 0)
5834 return (set_errno(EPERM));
5835 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
5836 return (set_errno(EINVAL));
5837
5838 /*
5839 * Stop all lwps so we don't need to hold a lock to look at
5840 * curproc->p_zone. This needs to happen before we grab any
5841 * locks to avoid deadlock (another lwp in the process could
5842 * be waiting for the held lock).
5843 */
5844 if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK))
5845 return (set_errno(EINTR));
5846
5847 /*
5848 * Make sure we're not changing zones with files open or mapped in
5849 * to our address space which shouldn't be changing zones.
5850 */
5851 if (!files_can_change_zones()) {
5852 err = EBADF;
5853 goto out;
5854 }
5855 if (!as_can_change_zones()) {
5856 err = EFAULT;
5857 goto out;
5858 }
5859
5860 mutex_enter(&zonehash_lock);
5861 if (pp->p_zone != global_zone) {
5862 mutex_exit(&zonehash_lock);
5863 err = EINVAL;
5864 goto out;
5865 }
5866
5867 zone = zone_find_all_by_id(zoneid);
5868 if (zone == NULL) {
5869 mutex_exit(&zonehash_lock);
5870 err = EINVAL;
5871 goto out;
5872 }
5873
5874 /*
5875 * To prevent processes in a zone from holding contracts on
5876 * extrazonal resources, and to avoid process contract
5877 * memberships which span zones, contract holders and processes
5878 * which aren't the sole members of their encapsulating process
5879 * contracts are not allowed to zone_enter.
5880 */
5881 ctp = pp->p_ct_process;
5882 ct = &ctp->conp_contract;
5883 mutex_enter(&ct->ct_lock);
5884 mutex_enter(&pp->p_lock);
5885 if ((avl_numnodes(&pp->p_ct_held) != 0) || (ctp->conp_nmembers != 1)) {
5886 mutex_exit(&pp->p_lock);
5887 mutex_exit(&ct->ct_lock);
5888 mutex_exit(&zonehash_lock);
5889 err = EINVAL;
5890 goto out;
5891 }
5892
5893 /*
5894 * Moreover, we don't allow processes whose encapsulating
5895 * process contracts have inherited extrazonal contracts.
5896 * While it would be easier to eliminate all process contracts
5897 * with inherited contracts, we need to be able to give a
5898 * restarted init (or other zone-penetrating process) its
5899 * predecessor's contracts.
5900 */
5901 if (ctp->conp_ninherited != 0) {
5902 contract_t *next;
5903 for (next = list_head(&ctp->conp_inherited); next;
5904 next = list_next(&ctp->conp_inherited, next)) {
5905 if (contract_getzuniqid(next) != zone->zone_uniqid) {
5906 mutex_exit(&pp->p_lock);
5907 mutex_exit(&ct->ct_lock);
5908 mutex_exit(&zonehash_lock);
5909 err = EINVAL;
5910 goto out;
5911 }
5912 }
5913 }
5914
5915 mutex_exit(&pp->p_lock);
5916 mutex_exit(&ct->ct_lock);
5917
5918 status = zone_status_get(zone);
5919 if (status < ZONE_IS_READY || status >= ZONE_IS_SHUTTING_DOWN) {
5920 /*
5921 * Can't join
5922 */
5923 mutex_exit(&zonehash_lock);
5924 err = EINVAL;
5925 goto out;
5926 }
5927
5928 /*
5929 * Make sure new priv set is within the permitted set for caller
5930 */
5931 if (!priv_issubset(zone->zone_privset, &CR_OPPRIV(CRED()))) {
5932 mutex_exit(&zonehash_lock);
5933 err = EPERM;
5934 goto out;
5935 }
5936 /*
5937 * We want to momentarily drop zonehash_lock while we optimistically
5938 * bind curproc to the pool it should be running in. This is safe
5939 * since the zone can't disappear (we have a hold on it).
5940 */
5941 zone_hold(zone);
5942 mutex_exit(&zonehash_lock);
5943
5944 /*
5945 * Grab pool_lock to keep the pools configuration from changing
5946 * and to stop ourselves from getting rebound to another pool
5947 * until we join the zone.
5948 */
5949 if (pool_lock_intr() != 0) {
5950 zone_rele(zone);
5951 err = EINTR;
5952 goto out;
5953 }
5954 ASSERT(secpolicy_pool(CRED()) == 0);
5955 /*
5956 * Bind ourselves to the pool currently associated with the zone.
5957 */
5958 oldpool = curproc->p_pool;
5959 newpool = zone_pool_get(zone);
5960 if (pool_state == POOL_ENABLED && newpool != oldpool &&
5961 (err = pool_do_bind(newpool, P_PID, P_MYID,
5962 POOL_BIND_ALL)) != 0) {
5963 pool_unlock();
5964 zone_rele(zone);
5965 goto out;
5966 }
5967
5968 /*
5969 * Grab cpu_lock now; we'll need it later when we call
5970 * task_join().
5971 */
5972 mutex_enter(&cpu_lock);
5973 mutex_enter(&zonehash_lock);
5974 /*
5975 * Make sure the zone hasn't moved on since we dropped zonehash_lock.
5976 */
5977 if (zone_status_get(zone) >= ZONE_IS_SHUTTING_DOWN) {
5978 /*
5979 * Can't join anymore.
5980 */
5981 mutex_exit(&zonehash_lock);
5982 mutex_exit(&cpu_lock);
5983 if (pool_state == POOL_ENABLED &&
5984 newpool != oldpool)
5985 (void) pool_do_bind(oldpool, P_PID, P_MYID,
5986 POOL_BIND_ALL);
5987 pool_unlock();
5988 zone_rele(zone);
5989 err = EINVAL;
5990 goto out;
5991 }
5992
5993 /*
5994 * a_lock must be held while transfering locked memory and swap
5995 * reservation from the global zone to the non global zone because
5996 * asynchronous faults on the processes' address space can lock
5997 * memory and reserve swap via MCL_FUTURE and MAP_NORESERVE
5998 * segments respectively.
5999 */
6000 AS_LOCK_ENTER(pp->p_as, RW_WRITER);
6001 swap = as_swresv();
6002 mutex_enter(&pp->p_lock);
6003 zone_proj0 = zone->zone_zsched->p_task->tk_proj;
6004 /* verify that we do not exceed and task or lwp limits */
6005 mutex_enter(&zone->zone_nlwps_lock);
6006 /* add new lwps to zone and zone's proj0 */
6007 zone_proj0->kpj_nlwps += pp->p_lwpcnt;
6008 zone->zone_nlwps += pp->p_lwpcnt;
6009 /* add 1 task to zone's proj0 */
6010 zone_proj0->kpj_ntasks += 1;
6011
6012 zone_proj0->kpj_nprocs++;
6013 zone->zone_nprocs++;
6014 mutex_exit(&zone->zone_nlwps_lock);
6015
6016 mutex_enter(&zone->zone_mem_lock);
6017 zone->zone_locked_mem += pp->p_locked_mem;
6018 zone_proj0->kpj_data.kpd_locked_mem += pp->p_locked_mem;
6019 zone->zone_max_swap += swap;
6020 mutex_exit(&zone->zone_mem_lock);
6021
6022 mutex_enter(&(zone_proj0->kpj_data.kpd_crypto_lock));
6023 zone_proj0->kpj_data.kpd_crypto_mem += pp->p_crypto_mem;
6024 mutex_exit(&(zone_proj0->kpj_data.kpd_crypto_lock));
6025
6026 /* remove lwps and process from proc's old zone and old project */
6027 mutex_enter(&pp->p_zone->zone_nlwps_lock);
6028 pp->p_zone->zone_nlwps -= pp->p_lwpcnt;
6029 pp->p_task->tk_proj->kpj_nlwps -= pp->p_lwpcnt;
6030 pp->p_task->tk_proj->kpj_nprocs--;
6031 pp->p_zone->zone_nprocs--;
6032 mutex_exit(&pp->p_zone->zone_nlwps_lock);
6033
6034 mutex_enter(&pp->p_zone->zone_mem_lock);
6035 pp->p_zone->zone_locked_mem -= pp->p_locked_mem;
6036 pp->p_task->tk_proj->kpj_data.kpd_locked_mem -= pp->p_locked_mem;
6037 pp->p_zone->zone_max_swap -= swap;
6038 mutex_exit(&pp->p_zone->zone_mem_lock);
6039
6040 mutex_enter(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock));
6041 pp->p_task->tk_proj->kpj_data.kpd_crypto_mem -= pp->p_crypto_mem;
6042 mutex_exit(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock));
6043
6044 pp->p_flag |= SZONETOP;
6045 pp->p_zone = zone;
6046 mutex_exit(&pp->p_lock);
6047 AS_LOCK_EXIT(pp->p_as);
6048
6049 /*
6050 * Joining the zone cannot fail from now on.
6051 *
6052 * This means that a lot of the following code can be commonized and
6053 * shared with zsched().
6054 */
6055
6056 /*
6057 * If the process contract fmri was inherited, we need to
6058 * flag this so that any contract status will not leak
6059 * extra zone information, svc_fmri in this case
6060 */
6061 if (ctp->conp_svc_ctid != ct->ct_id) {
6062 mutex_enter(&ct->ct_lock);
6063 ctp->conp_svc_zone_enter = ct->ct_id;
6064 mutex_exit(&ct->ct_lock);
6065 }
6066
6067 /*
6068 * Reset the encapsulating process contract's zone.
6069 */
6070 ASSERT(ct->ct_mzuniqid == GLOBAL_ZONEUNIQID);
6071 contract_setzuniqid(ct, zone->zone_uniqid);
6072
6073 /*
6074 * Create a new task and associate the process with the project keyed
6075 * by (projid,zoneid).
6076 *
6077 * We might as well be in project 0; the global zone's projid doesn't
6078 * make much sense in a zone anyhow.
6079 *
6080 * This also increments zone_ntasks, and returns with p_lock held.
6081 */
6082 tk = task_create(0, zone);
6083 oldtk = task_join(tk, 0);
6084 mutex_exit(&cpu_lock);
6085
6086 /*
6087 * call RCTLOP_SET functions on this proc
6088 */
6089 e.rcep_p.zone = zone;
6090 e.rcep_t = RCENTITY_ZONE;
6091 (void) rctl_set_dup(NULL, NULL, pp, &e, zone->zone_rctls, NULL,
6092 RCD_CALLBACK);
6093 mutex_exit(&pp->p_lock);
6094
6095 /*
6096 * We don't need to hold any of zsched's locks here; not only do we know
6097 * the process and zone aren't going away, we know its session isn't
6098 * changing either.
6099 *
6100 * By joining zsched's session here, we mimic the behavior in the
6101 * global zone of init's sid being the pid of sched. We extend this
6102 * to all zlogin-like zone_enter()'ing processes as well.
6103 */
6104 mutex_enter(&pidlock);
6105 sp = zone->zone_zsched->p_sessp;
6106 sess_hold(zone->zone_zsched);
6107 mutex_enter(&pp->p_lock);
6108 pgexit(pp);
6109 sess_rele(pp->p_sessp, B_TRUE);
6110 pp->p_sessp = sp;
6111 pgjoin(pp, zone->zone_zsched->p_pidp);
6112
6113 /*
6114 * If any threads are scheduled to be placed on zone wait queue they
6115 * should abandon the idea since the wait queue is changing.
6116 * We need to be holding pidlock & p_lock to do this.
6117 */
6118 if ((t = pp->p_tlist) != NULL) {
6119 do {
6120 thread_lock(t);
6121 /*
6122 * Kick this thread so that he doesn't sit
6123 * on a wrong wait queue.
6124 */
6125 if (ISWAITING(t))
6126 setrun_locked(t);
6127
6128 if (t->t_schedflag & TS_ANYWAITQ)
6129 t->t_schedflag &= ~ TS_ANYWAITQ;
6130
6131 thread_unlock(t);
6132 } while ((t = t->t_forw) != pp->p_tlist);
6133 }
6134
6135 /*
6136 * If there is a default scheduling class for the zone and it is not
6137 * the class we are currently in, change all of the threads in the
6138 * process to the new class. We need to be holding pidlock & p_lock
6139 * when we call parmsset so this is a good place to do it.
6140 */
6141 if (zone->zone_defaultcid > 0 &&
6142 zone->zone_defaultcid != curthread->t_cid) {
6143 pcparms_t pcparms;
6144
6145 pcparms.pc_cid = zone->zone_defaultcid;
6146 pcparms.pc_clparms[0] = 0;
6147
6148 /*
6149 * If setting the class fails, we still want to enter the zone.
6150 */
6151 if ((t = pp->p_tlist) != NULL) {
6152 do {
6153 (void) parmsset(&pcparms, t);
6154 } while ((t = t->t_forw) != pp->p_tlist);
6155 }
6156 }
6157
6158 mutex_exit(&pp->p_lock);
6159 mutex_exit(&pidlock);
6160
6161 mutex_exit(&zonehash_lock);
6162 /*
6163 * We're firmly in the zone; let pools progress.
6164 */
6165 pool_unlock();
6166 task_rele(oldtk);
6167 /*
6168 * We don't need to retain a hold on the zone since we already
6169 * incremented zone_ntasks, so the zone isn't going anywhere.
6170 */
6171 zone_rele(zone);
6172
6173 /*
6174 * Chroot
6175 */
6176 vp = zone->zone_rootvp;
6177 zone_chdir(vp, &PTOU(pp)->u_cdir, pp);
6178 zone_chdir(vp, &PTOU(pp)->u_rdir, pp);
6179
6180 /*
6181 * Change process credentials
6182 */
6183 newcr = cralloc();
6184 mutex_enter(&pp->p_crlock);
6185 cr = pp->p_cred;
6186 crcopy_to(cr, newcr);
6187 crsetzone(newcr, zone);
6188 pp->p_cred = newcr;
6189
6190 /*
6191 * Restrict all process privilege sets to zone limit
6192 */
6193 priv_intersect(zone->zone_privset, &CR_PPRIV(newcr));
6194 priv_intersect(zone->zone_privset, &CR_EPRIV(newcr));
6195 priv_intersect(zone->zone_privset, &CR_IPRIV(newcr));
6196 priv_intersect(zone->zone_privset, &CR_LPRIV(newcr));
6197 mutex_exit(&pp->p_crlock);
6198 crset(pp, newcr);
6199
6200 /*
6201 * Adjust upcount to reflect zone entry.
6202 */
6203 uid = crgetruid(newcr);
6204 mutex_enter(&pidlock);
6205 upcount_dec(uid, GLOBAL_ZONEID);
6206 upcount_inc(uid, zoneid);
6207 mutex_exit(&pidlock);
6208
6209 /*
6210 * Set up core file path and content.
6211 */
6212 set_core_defaults();
6213
6214 out:
6215 /*
6216 * Let the other lwps continue.
6217 */
6218 mutex_enter(&pp->p_lock);
6219 if (curthread != pp->p_agenttp)
6220 continuelwps(pp);
6221 mutex_exit(&pp->p_lock);
6222
6223 return (err != 0 ? set_errno(err) : 0);
6224 }
6225
6226 /*
6227 * Systemcall entry point for zone_list(2).
6228 *
6229 * Processes running in a (non-global) zone only see themselves.
6230 * On labeled systems, they see all zones whose label they dominate.
6231 */
6232 static int
6233 zone_list(zoneid_t *zoneidlist, uint_t *numzones)
6234 {
6235 zoneid_t *zoneids;
6236 zone_t *zone, *myzone;
6237 uint_t user_nzones, real_nzones;
6238 uint_t domi_nzones;
6239 int error;
6240
6241 if (copyin(numzones, &user_nzones, sizeof (uint_t)) != 0)
6242 return (set_errno(EFAULT));
6243
6244 myzone = curproc->p_zone;
6245 if (myzone != global_zone) {
6246 bslabel_t *mybslab;
6247
6248 if (!is_system_labeled()) {
6249 /* just return current zone */
6250 real_nzones = domi_nzones = 1;
6251 zoneids = kmem_alloc(sizeof (zoneid_t), KM_SLEEP);
6252 zoneids[0] = myzone->zone_id;
6253 } else {
6254 /* return all zones that are dominated */
6255 mutex_enter(&zonehash_lock);
6256 real_nzones = zonecount;
6257 domi_nzones = 0;
6258 if (real_nzones > 0) {
6259 zoneids = kmem_alloc(real_nzones *
6260 sizeof (zoneid_t), KM_SLEEP);
6261 mybslab = label2bslabel(myzone->zone_slabel);
6262 for (zone = list_head(&zone_active);
6263 zone != NULL;
6264 zone = list_next(&zone_active, zone)) {
6265 if (zone->zone_id == GLOBAL_ZONEID)
6266 continue;
6267 if (zone != myzone &&
6268 (zone->zone_flags & ZF_IS_SCRATCH))
6269 continue;
6270 /*
6271 * Note that a label always dominates
6272 * itself, so myzone is always included
6273 * in the list.
6274 */
6275 if (bldominates(mybslab,
6276 label2bslabel(zone->zone_slabel))) {
6277 zoneids[domi_nzones++] =
6278 zone->zone_id;
6279 }
6280 }
6281 }
6282 mutex_exit(&zonehash_lock);
6283 }
6284 } else {
6285 mutex_enter(&zonehash_lock);
6286 real_nzones = zonecount;
6287 domi_nzones = 0;
6288 if (real_nzones > 0) {
6289 zoneids = kmem_alloc(real_nzones * sizeof (zoneid_t),
6290 KM_SLEEP);
6291 for (zone = list_head(&zone_active); zone != NULL;
6292 zone = list_next(&zone_active, zone))
6293 zoneids[domi_nzones++] = zone->zone_id;
6294 ASSERT(domi_nzones == real_nzones);
6295 }
6296 mutex_exit(&zonehash_lock);
6297 }
6298
6299 /*
6300 * If user has allocated space for fewer entries than we found, then
6301 * return only up to his limit. Either way, tell him exactly how many
6302 * we found.
6303 */
6304 if (domi_nzones < user_nzones)
6305 user_nzones = domi_nzones;
6306 error = 0;
6307 if (copyout(&domi_nzones, numzones, sizeof (uint_t)) != 0) {
6308 error = EFAULT;
6309 } else if (zoneidlist != NULL && user_nzones != 0) {
6310 if (copyout(zoneids, zoneidlist,
6311 user_nzones * sizeof (zoneid_t)) != 0)
6312 error = EFAULT;
6313 }
6314
6315 if (real_nzones > 0)
6316 kmem_free(zoneids, real_nzones * sizeof (zoneid_t));
6317
6318 if (error != 0)
6319 return (set_errno(error));
6320 else
6321 return (0);
6322 }
6323
6324 /*
6325 * Systemcall entry point for zone_lookup(2).
6326 *
6327 * Non-global zones are only able to see themselves and (on labeled systems)
6328 * the zones they dominate.
6329 */
6330 static zoneid_t
6331 zone_lookup(const char *zone_name)
6332 {
6333 char *kname;
6334 zone_t *zone;
6335 zoneid_t zoneid;
6336 int err;
6337
6338 if (zone_name == NULL) {
6339 /* return caller's zone id */
6340 return (getzoneid());
6341 }
6342
6343 kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
6344 if ((err = copyinstr(zone_name, kname, ZONENAME_MAX, NULL)) != 0) {
6345 kmem_free(kname, ZONENAME_MAX);
6346 return (set_errno(err));
6347 }
6348
6349 mutex_enter(&zonehash_lock);
6350 zone = zone_find_all_by_name(kname);
6351 kmem_free(kname, ZONENAME_MAX);
6352 /*
6353 * In a non-global zone, can only lookup global and own name.
6354 * In Trusted Extensions zone label dominance rules apply.
6355 */
6356 if (zone == NULL ||
6357 zone_status_get(zone) < ZONE_IS_READY ||
6358 !zone_list_access(zone)) {
6359 mutex_exit(&zonehash_lock);
6360 return (set_errno(EINVAL));
6361 } else {
6362 zoneid = zone->zone_id;
6363 mutex_exit(&zonehash_lock);
6364 return (zoneid);
6365 }
6366 }
6367
6368 static int
6369 zone_version(int *version_arg)
6370 {
6371 int version = ZONE_SYSCALL_API_VERSION;
6372
6373 if (copyout(&version, version_arg, sizeof (int)) != 0)
6374 return (set_errno(EFAULT));
6375 return (0);
6376 }
6377
6378 /* ARGSUSED */
6379 long
6380 zone(int cmd, void *arg1, void *arg2, void *arg3, void *arg4)
6381 {
6382 zone_def zs;
6383 int err;
6384
6385 switch (cmd) {
6386 case ZONE_CREATE:
6387 if (get_udatamodel() == DATAMODEL_NATIVE) {
6388 if (copyin(arg1, &zs, sizeof (zone_def))) {
6389 return (set_errno(EFAULT));
6390 }
6391 } else {
6392 #ifdef _SYSCALL32_IMPL
6393 zone_def32 zs32;
6394
6395 if (copyin(arg1, &zs32, sizeof (zone_def32))) {
6396 return (set_errno(EFAULT));
6397 }
6398 zs.zone_name =
6399 (const char *)(unsigned long)zs32.zone_name;
6400 zs.zone_root =
6401 (const char *)(unsigned long)zs32.zone_root;
6402 zs.zone_privs =
6403 (const struct priv_set *)
6404 (unsigned long)zs32.zone_privs;
6405 zs.zone_privssz = zs32.zone_privssz;
6406 zs.rctlbuf = (caddr_t)(unsigned long)zs32.rctlbuf;
6407 zs.rctlbufsz = zs32.rctlbufsz;
6408 zs.zfsbuf = (caddr_t)(unsigned long)zs32.zfsbuf;
6409 zs.zfsbufsz = zs32.zfsbufsz;
6410 zs.extended_error =
6411 (int *)(unsigned long)zs32.extended_error;
6412 zs.match = zs32.match;
6413 zs.doi = zs32.doi;
6414 zs.label = (const bslabel_t *)(uintptr_t)zs32.label;
6415 zs.flags = zs32.flags;
6416 #else
6417 panic("get_udatamodel() returned bogus result\n");
6418 #endif
6419 }
6420
6421 return (zone_create(zs.zone_name, zs.zone_root,
6422 zs.zone_privs, zs.zone_privssz,
6423 (caddr_t)zs.rctlbuf, zs.rctlbufsz,
6424 (caddr_t)zs.zfsbuf, zs.zfsbufsz,
6425 zs.extended_error, zs.match, zs.doi,
6426 zs.label, zs.flags));
6427 case ZONE_BOOT:
6428 return (zone_boot((zoneid_t)(uintptr_t)arg1));
6429 case ZONE_DESTROY:
6430 return (zone_destroy((zoneid_t)(uintptr_t)arg1));
6431 case ZONE_GETATTR:
6432 return (zone_getattr((zoneid_t)(uintptr_t)arg1,
6433 (int)(uintptr_t)arg2, arg3, (size_t)arg4));
6434 case ZONE_SETATTR:
6435 return (zone_setattr((zoneid_t)(uintptr_t)arg1,
6436 (int)(uintptr_t)arg2, arg3, (size_t)arg4));
6437 case ZONE_ENTER:
6438 return (zone_enter((zoneid_t)(uintptr_t)arg1));
6439 case ZONE_LIST:
6440 return (zone_list((zoneid_t *)arg1, (uint_t *)arg2));
6441 case ZONE_SHUTDOWN:
6442 return (zone_shutdown((zoneid_t)(uintptr_t)arg1));
6443 case ZONE_LOOKUP:
6444 return (zone_lookup((const char *)arg1));
6445 case ZONE_VERSION:
6446 return (zone_version((int *)arg1));
6447 case ZONE_ADD_DATALINK:
6448 return (zone_add_datalink((zoneid_t)(uintptr_t)arg1,
6449 (datalink_id_t)(uintptr_t)arg2));
6450 case ZONE_DEL_DATALINK:
6451 return (zone_remove_datalink((zoneid_t)(uintptr_t)arg1,
6452 (datalink_id_t)(uintptr_t)arg2));
6453 case ZONE_CHECK_DATALINK: {
6454 zoneid_t zoneid;
6455 boolean_t need_copyout;
6456
6457 if (copyin(arg1, &zoneid, sizeof (zoneid)) != 0)
6458 return (EFAULT);
6459 need_copyout = (zoneid == ALL_ZONES);
6460 err = zone_check_datalink(&zoneid,
6461 (datalink_id_t)(uintptr_t)arg2);
6462 if (err == 0 && need_copyout) {
6463 if (copyout(&zoneid, arg1, sizeof (zoneid)) != 0)
6464 err = EFAULT;
6465 }
6466 return (err == 0 ? 0 : set_errno(err));
6467 }
6468 case ZONE_LIST_DATALINK:
6469 return (zone_list_datalink((zoneid_t)(uintptr_t)arg1,
6470 (int *)arg2, (datalink_id_t *)(uintptr_t)arg3));
6471 default:
6472 return (set_errno(EINVAL));
6473 }
6474 }
6475
6476 struct zarg {
6477 zone_t *zone;
6478 zone_cmd_arg_t arg;
6479 };
6480
6481 static int
6482 zone_lookup_door(const char *zone_name, door_handle_t *doorp)
6483 {
6484 char *buf;
6485 size_t buflen;
6486 int error;
6487
6488 buflen = sizeof (ZONE_DOOR_PATH) + strlen(zone_name);
6489 buf = kmem_alloc(buflen, KM_SLEEP);
6490 (void) snprintf(buf, buflen, ZONE_DOOR_PATH, zone_name);
6491 error = door_ki_open(buf, doorp);
6492 kmem_free(buf, buflen);
6493 return (error);
6494 }
6495
6496 static void
6497 zone_release_door(door_handle_t *doorp)
6498 {
6499 door_ki_rele(*doorp);
6500 *doorp = NULL;
6501 }
6502
6503 static void
6504 zone_ki_call_zoneadmd(struct zarg *zargp)
6505 {
6506 door_handle_t door = NULL;
6507 door_arg_t darg, save_arg;
6508 char *zone_name;
6509 size_t zone_namelen;
6510 zoneid_t zoneid;
6511 zone_t *zone;
6512 zone_cmd_arg_t arg;
6513 uint64_t uniqid;
6514 size_t size;
6515 int error;
6516 int retry;
6517
6518 zone = zargp->zone;
6519 arg = zargp->arg;
6520 kmem_free(zargp, sizeof (*zargp));
6521
6522 zone_namelen = strlen(zone->zone_name) + 1;
6523 zone_name = kmem_alloc(zone_namelen, KM_SLEEP);
6524 bcopy(zone->zone_name, zone_name, zone_namelen);
6525 zoneid = zone->zone_id;
6526 uniqid = zone->zone_uniqid;
6527 /*
6528 * zoneadmd may be down, but at least we can empty out the zone.
6529 * We can ignore the return value of zone_empty() since we're called
6530 * from a kernel thread and know we won't be delivered any signals.
6531 */
6532 ASSERT(curproc == &p0);
6533 (void) zone_empty(zone);
6534 ASSERT(zone_status_get(zone) >= ZONE_IS_EMPTY);
6535 zone_rele(zone);
6536
6537 size = sizeof (arg);
6538 darg.rbuf = (char *)&arg;
6539 darg.data_ptr = (char *)&arg;
6540 darg.rsize = size;
6541 darg.data_size = size;
6542 darg.desc_ptr = NULL;
6543 darg.desc_num = 0;
6544
6545 save_arg = darg;
6546 /*
6547 * Since we're not holding a reference to the zone, any number of
6548 * things can go wrong, including the zone disappearing before we get a
6549 * chance to talk to zoneadmd.
6550 */
6551 for (retry = 0; /* forever */; retry++) {
6552 if (door == NULL &&
6553 (error = zone_lookup_door(zone_name, &door)) != 0) {
6554 goto next;
6555 }
6556 ASSERT(door != NULL);
6557
6558 if ((error = door_ki_upcall_limited(door, &darg, NULL,
6559 SIZE_MAX, 0)) == 0) {
6560 break;
6561 }
6562 switch (error) {
6563 case EINTR:
6564 /* FALLTHROUGH */
6565 case EAGAIN: /* process may be forking */
6566 /*
6567 * Back off for a bit
6568 */
6569 break;
6570 case EBADF:
6571 zone_release_door(&door);
6572 if (zone_lookup_door(zone_name, &door) != 0) {
6573 /*
6574 * zoneadmd may be dead, but it may come back to
6575 * life later.
6576 */
6577 break;
6578 }
6579 break;
6580 default:
6581 cmn_err(CE_WARN,
6582 "zone_ki_call_zoneadmd: door_ki_upcall error %d\n",
6583 error);
6584 goto out;
6585 }
6586 next:
6587 /*
6588 * If this isn't the same zone_t that we originally had in mind,
6589 * then this is the same as if two kadmin requests come in at
6590 * the same time: the first one wins. This means we lose, so we
6591 * bail.
6592 */
6593 if ((zone = zone_find_by_id(zoneid)) == NULL) {
6594 /*
6595 * Problem is solved.
6596 */
6597 break;
6598 }
6599 if (zone->zone_uniqid != uniqid) {
6600 /*
6601 * zoneid recycled
6602 */
6603 zone_rele(zone);
6604 break;
6605 }
6606 /*
6607 * We could zone_status_timedwait(), but there doesn't seem to
6608 * be much point in doing that (plus, it would mean that
6609 * zone_free() isn't called until this thread exits).
6610 */
6611 zone_rele(zone);
6612 delay(hz);
6613 darg = save_arg;
6614 }
6615 out:
6616 if (door != NULL) {
6617 zone_release_door(&door);
6618 }
6619 kmem_free(zone_name, zone_namelen);
6620 thread_exit();
6621 }
6622
6623 /*
6624 * Entry point for uadmin() to tell the zone to go away or reboot. Analog to
6625 * kadmin(). The caller is a process in the zone.
6626 *
6627 * In order to shutdown the zone, we will hand off control to zoneadmd
6628 * (running in the global zone) via a door. We do a half-hearted job at
6629 * killing all processes in the zone, create a kernel thread to contact
6630 * zoneadmd, and make note of the "uniqid" of the zone. The uniqid is
6631 * a form of generation number used to let zoneadmd (as well as
6632 * zone_destroy()) know exactly which zone they're re talking about.
6633 */
6634 int
6635 zone_kadmin(int cmd, int fcn, const char *mdep, cred_t *credp)
6636 {
6637 struct zarg *zargp;
6638 zone_cmd_t zcmd;
6639 zone_t *zone;
6640
6641 zone = curproc->p_zone;
6642 ASSERT(getzoneid() != GLOBAL_ZONEID);
6643
6644 switch (cmd) {
6645 case A_SHUTDOWN:
6646 switch (fcn) {
6647 case AD_HALT:
6648 case AD_POWEROFF:
6649 zcmd = Z_HALT;
6650 break;
6651 case AD_BOOT:
6652 zcmd = Z_REBOOT;
6653 break;
6654 case AD_IBOOT:
6655 case AD_SBOOT:
6656 case AD_SIBOOT:
6657 case AD_NOSYNC:
6658 return (ENOTSUP);
6659 default:
6660 return (EINVAL);
6661 }
6662 break;
6663 case A_REBOOT:
6664 zcmd = Z_REBOOT;
6665 break;
6666 case A_FTRACE:
6667 case A_REMOUNT:
6668 case A_FREEZE:
6669 case A_DUMP:
6670 case A_CONFIG:
6671 return (ENOTSUP);
6672 default:
6673 ASSERT(cmd != A_SWAPCTL); /* handled by uadmin() */
6674 return (EINVAL);
6675 }
6676
6677 if (secpolicy_zone_admin(credp, B_FALSE))
6678 return (EPERM);
6679 mutex_enter(&zone_status_lock);
6680
6681 /*
6682 * zone_status can't be ZONE_IS_EMPTY or higher since curproc
6683 * is in the zone.
6684 */
6685 ASSERT(zone_status_get(zone) < ZONE_IS_EMPTY);
6686 if (zone_status_get(zone) > ZONE_IS_RUNNING) {
6687 /*
6688 * This zone is already on its way down.
6689 */
6690 mutex_exit(&zone_status_lock);
6691 return (0);
6692 }
6693 /*
6694 * Prevent future zone_enter()s
6695 */
6696 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
6697 mutex_exit(&zone_status_lock);
6698
6699 /*
6700 * Kill everyone now and call zoneadmd later.
6701 * zone_ki_call_zoneadmd() will do a more thorough job of this
6702 * later.
6703 */
6704 killall(zone->zone_id);
6705 /*
6706 * Now, create the thread to contact zoneadmd and do the rest of the
6707 * work. This thread can't be created in our zone otherwise
6708 * zone_destroy() would deadlock.
6709 */
6710 zargp = kmem_zalloc(sizeof (*zargp), KM_SLEEP);
6711 zargp->arg.cmd = zcmd;
6712 zargp->arg.uniqid = zone->zone_uniqid;
6713 zargp->zone = zone;
6714 (void) strcpy(zargp->arg.locale, "C");
6715 /* mdep was already copied in for us by uadmin */
6716 if (mdep != NULL)
6717 (void) strlcpy(zargp->arg.bootbuf, mdep,
6718 sizeof (zargp->arg.bootbuf));
6719 zone_hold(zone);
6720
6721 (void) thread_create(NULL, 0, zone_ki_call_zoneadmd, zargp, 0, &p0,
6722 TS_RUN, minclsyspri);
6723 exit(CLD_EXITED, 0);
6724
6725 return (EINVAL);
6726 }
6727
6728 /*
6729 * Entry point so kadmin(A_SHUTDOWN, ...) can set the global zone's
6730 * status to ZONE_IS_SHUTTING_DOWN.
6731 *
6732 * This function also shuts down all running zones to ensure that they won't
6733 * fork new processes.
6734 */
6735 void
6736 zone_shutdown_global(void)
6737 {
6738 zone_t *current_zonep;
6739
6740 ASSERT(INGLOBALZONE(curproc));
6741 mutex_enter(&zonehash_lock);
6742 mutex_enter(&zone_status_lock);
6743
6744 /* Modify the global zone's status first. */
6745 ASSERT(zone_status_get(global_zone) == ZONE_IS_RUNNING);
6746 zone_status_set(global_zone, ZONE_IS_SHUTTING_DOWN);
6747
6748 /*
6749 * Now change the states of all running zones to ZONE_IS_SHUTTING_DOWN.
6750 * We don't mark all zones with ZONE_IS_SHUTTING_DOWN because doing so
6751 * could cause assertions to fail (e.g., assertions about a zone's
6752 * state during initialization, readying, or booting) or produce races.
6753 * We'll let threads continue to initialize and ready new zones: they'll
6754 * fail to boot the new zones when they see that the global zone is
6755 * shutting down.
6756 */
6757 for (current_zonep = list_head(&zone_active); current_zonep != NULL;
6758 current_zonep = list_next(&zone_active, current_zonep)) {
6759 if (zone_status_get(current_zonep) == ZONE_IS_RUNNING)
6760 zone_status_set(current_zonep, ZONE_IS_SHUTTING_DOWN);
6761 }
6762 mutex_exit(&zone_status_lock);
6763 mutex_exit(&zonehash_lock);
6764 }
6765
6766 /*
6767 * Returns true if the named dataset is visible in the current zone.
6768 * The 'write' parameter is set to 1 if the dataset is also writable.
6769 */
6770 int
6771 zone_dataset_visible(const char *dataset, int *write)
6772 {
6773 static int zfstype = -1;
6774 zone_dataset_t *zd;
6775 size_t len;
6776 zone_t *zone = curproc->p_zone;
6777 const char *name = NULL;
6778 vfs_t *vfsp = NULL;
6779
6780 if (dataset[0] == '\0')
6781 return (0);
6782
6783 /*
6784 * Walk the list once, looking for datasets which match exactly, or
6785 * specify a dataset underneath an exported dataset. If found, return
6786 * true and note that it is writable.
6787 */
6788 for (zd = list_head(&zone->zone_datasets); zd != NULL;
6789 zd = list_next(&zone->zone_datasets, zd)) {
6790
6791 len = strlen(zd->zd_dataset);
6792 if (strlen(dataset) >= len &&
6793 bcmp(dataset, zd->zd_dataset, len) == 0 &&
6794 (dataset[len] == '\0' || dataset[len] == '/' ||
6795 dataset[len] == '@')) {
6796 if (write)
6797 *write = 1;
6798 return (1);
6799 }
6800 }
6801
6802 /*
6803 * Walk the list a second time, searching for datasets which are parents
6804 * of exported datasets. These should be visible, but read-only.
6805 *
6806 * Note that we also have to support forms such as 'pool/dataset/', with
6807 * a trailing slash.
6808 */
6809 for (zd = list_head(&zone->zone_datasets); zd != NULL;
6810 zd = list_next(&zone->zone_datasets, zd)) {
6811
6812 len = strlen(dataset);
6813 if (dataset[len - 1] == '/')
6814 len--; /* Ignore trailing slash */
6815 if (len < strlen(zd->zd_dataset) &&
6816 bcmp(dataset, zd->zd_dataset, len) == 0 &&
6817 zd->zd_dataset[len] == '/') {
6818 if (write)
6819 *write = 0;
6820 return (1);
6821 }
6822 }
6823
6824 /*
6825 * We reach here if the given dataset is not found in the zone_dataset
6826 * list. Check if this dataset was added as a filesystem (ie. "add fs")
6827 * instead of delegation. For this we search for the dataset in the
6828 * zone_vfslist of this zone. If found, return true and note that it is
6829 * not writable.
6830 */
6831
6832 /*
6833 * Initialize zfstype if it is not initialized yet.
6834 */
6835 if (zfstype == -1) {
6836 struct vfssw *vswp = vfs_getvfssw("zfs");
6837 zfstype = vswp - vfssw;
6838 vfs_unrefvfssw(vswp);
6839 }
6840
6841 vfs_list_read_lock();
6842 vfsp = zone->zone_vfslist;
6843 do {
6844 ASSERT(vfsp);
6845 if (vfsp->vfs_fstype == zfstype) {
6846 name = refstr_value(vfsp->vfs_resource);
6847
6848 /*
6849 * Check if we have an exact match.
6850 */
6851 if (strcmp(dataset, name) == 0) {
6852 vfs_list_unlock();
6853 if (write)
6854 *write = 0;
6855 return (1);
6856 }
6857 /*
6858 * We need to check if we are looking for parents of
6859 * a dataset. These should be visible, but read-only.
6860 */
6861 len = strlen(dataset);
6862 if (dataset[len - 1] == '/')
6863 len--;
6864
6865 if (len < strlen(name) &&
6866 bcmp(dataset, name, len) == 0 && name[len] == '/') {
6867 vfs_list_unlock();
6868 if (write)
6869 *write = 0;
6870 return (1);
6871 }
6872 }
6873 vfsp = vfsp->vfs_zone_next;
6874 } while (vfsp != zone->zone_vfslist);
6875
6876 vfs_list_unlock();
6877 return (0);
6878 }
6879
6880 /*
6881 * zone_find_by_any_path() -
6882 *
6883 * kernel-private routine similar to zone_find_by_path(), but which
6884 * effectively compares against zone paths rather than zonerootpath
6885 * (i.e., the last component of zonerootpaths, which should be "root/",
6886 * are not compared.) This is done in order to accurately identify all
6887 * paths, whether zone-visible or not, including those which are parallel
6888 * to /root/, such as /dev/, /home/, etc...
6889 *
6890 * If the specified path does not fall under any zone path then global
6891 * zone is returned.
6892 *
6893 * The treat_abs parameter indicates whether the path should be treated as
6894 * an absolute path although it does not begin with "/". (This supports
6895 * nfs mount syntax such as host:any/path.)
6896 *
6897 * The caller is responsible for zone_rele of the returned zone.
6898 */
6899 zone_t *
6900 zone_find_by_any_path(const char *path, boolean_t treat_abs)
6901 {
6902 zone_t *zone;
6903 int path_offset = 0;
6904
6905 if (path == NULL) {
6906 zone_hold(global_zone);
6907 return (global_zone);
6908 }
6909
6910 if (*path != '/') {
6911 ASSERT(treat_abs);
6912 path_offset = 1;
6913 }
6914
6915 mutex_enter(&zonehash_lock);
6916 for (zone = list_head(&zone_active); zone != NULL;
6917 zone = list_next(&zone_active, zone)) {
6918 char *c;
6919 size_t pathlen;
6920 char *rootpath_start;
6921
6922 if (zone == global_zone) /* skip global zone */
6923 continue;
6924
6925 /* scan backwards to find start of last component */
6926 c = zone->zone_rootpath + zone->zone_rootpathlen - 2;
6927 do {
6928 c--;
6929 } while (*c != '/');
6930
6931 pathlen = c - zone->zone_rootpath + 1 - path_offset;
6932 rootpath_start = (zone->zone_rootpath + path_offset);
6933 if (strncmp(path, rootpath_start, pathlen) == 0)
6934 break;
6935 }
6936 if (zone == NULL)
6937 zone = global_zone;
6938 zone_hold(zone);
6939 mutex_exit(&zonehash_lock);
6940 return (zone);
6941 }
6942
6943 /*
6944 * Finds a zone_dl_t with the given linkid in the given zone. Returns the
6945 * zone_dl_t pointer if found, and NULL otherwise.
6946 */
6947 static zone_dl_t *
6948 zone_find_dl(zone_t *zone, datalink_id_t linkid)
6949 {
6950 zone_dl_t *zdl;
6951
6952 ASSERT(mutex_owned(&zone->zone_lock));
6953 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
6954 zdl = list_next(&zone->zone_dl_list, zdl)) {
6955 if (zdl->zdl_id == linkid)
6956 break;
6957 }
6958 return (zdl);
6959 }
6960
6961 static boolean_t
6962 zone_dl_exists(zone_t *zone, datalink_id_t linkid)
6963 {
6964 boolean_t exists;
6965
6966 mutex_enter(&zone->zone_lock);
6967 exists = (zone_find_dl(zone, linkid) != NULL);
6968 mutex_exit(&zone->zone_lock);
6969 return (exists);
6970 }
6971
6972 /*
6973 * Add an data link name for the zone.
6974 */
6975 static int
6976 zone_add_datalink(zoneid_t zoneid, datalink_id_t linkid)
6977 {
6978 zone_dl_t *zdl;
6979 zone_t *zone;
6980 zone_t *thiszone;
6981
6982 if ((thiszone = zone_find_by_id(zoneid)) == NULL)
6983 return (set_errno(ENXIO));
6984
6985 /* Verify that the datalink ID doesn't already belong to a zone. */
6986 mutex_enter(&zonehash_lock);
6987 for (zone = list_head(&zone_active); zone != NULL;
6988 zone = list_next(&zone_active, zone)) {
6989 if (zone_dl_exists(zone, linkid)) {
6990 mutex_exit(&zonehash_lock);
6991 zone_rele(thiszone);
6992 return (set_errno((zone == thiszone) ? EEXIST : EPERM));
6993 }
6994 }
6995
6996 zdl = kmem_zalloc(sizeof (*zdl), KM_SLEEP);
6997 zdl->zdl_id = linkid;
6998 zdl->zdl_net = NULL;
6999 mutex_enter(&thiszone->zone_lock);
7000 list_insert_head(&thiszone->zone_dl_list, zdl);
7001 mutex_exit(&thiszone->zone_lock);
7002 mutex_exit(&zonehash_lock);
7003 zone_rele(thiszone);
7004 return (0);
7005 }
7006
7007 static int
7008 zone_remove_datalink(zoneid_t zoneid, datalink_id_t linkid)
7009 {
7010 zone_dl_t *zdl;
7011 zone_t *zone;
7012 int err = 0;
7013
7014 if ((zone = zone_find_by_id(zoneid)) == NULL)
7015 return (set_errno(EINVAL));
7016
7017 mutex_enter(&zone->zone_lock);
7018 if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
7019 err = ENXIO;
7020 } else {
7021 list_remove(&zone->zone_dl_list, zdl);
7022 nvlist_free(zdl->zdl_net);
7023 kmem_free(zdl, sizeof (zone_dl_t));
7024 }
7025 mutex_exit(&zone->zone_lock);
7026 zone_rele(zone);
7027 return (err == 0 ? 0 : set_errno(err));
7028 }
7029
7030 /*
7031 * Using the zoneidp as ALL_ZONES, we can lookup which zone has been assigned
7032 * the linkid. Otherwise we just check if the specified zoneidp has been
7033 * assigned the supplied linkid.
7034 */
7035 int
7036 zone_check_datalink(zoneid_t *zoneidp, datalink_id_t linkid)
7037 {
7038 zone_t *zone;
7039 int err = ENXIO;
7040
7041 if (*zoneidp != ALL_ZONES) {
7042 if ((zone = zone_find_by_id(*zoneidp)) != NULL) {
7043 if (zone_dl_exists(zone, linkid))
7044 err = 0;
7045 zone_rele(zone);
7046 }
7047 return (err);
7048 }
7049
7050 mutex_enter(&zonehash_lock);
7051 for (zone = list_head(&zone_active); zone != NULL;
7052 zone = list_next(&zone_active, zone)) {
7053 if (zone_dl_exists(zone, linkid)) {
7054 *zoneidp = zone->zone_id;
7055 err = 0;
7056 break;
7057 }
7058 }
7059 mutex_exit(&zonehash_lock);
7060 return (err);
7061 }
7062
7063 /*
7064 * Get the list of datalink IDs assigned to a zone.
7065 *
7066 * On input, *nump is the number of datalink IDs that can fit in the supplied
7067 * idarray. Upon return, *nump is either set to the number of datalink IDs
7068 * that were placed in the array if the array was large enough, or to the
7069 * number of datalink IDs that the function needs to place in the array if the
7070 * array is too small.
7071 */
7072 static int
7073 zone_list_datalink(zoneid_t zoneid, int *nump, datalink_id_t *idarray)
7074 {
7075 uint_t num, dlcount;
7076 zone_t *zone;
7077 zone_dl_t *zdl;
7078 datalink_id_t *idptr = idarray;
7079
7080 if (copyin(nump, &dlcount, sizeof (dlcount)) != 0)
7081 return (set_errno(EFAULT));
7082 if ((zone = zone_find_by_id(zoneid)) == NULL)
7083 return (set_errno(ENXIO));
7084
7085 num = 0;
7086 mutex_enter(&zone->zone_lock);
7087 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7088 zdl = list_next(&zone->zone_dl_list, zdl)) {
7089 /*
7090 * If the list is bigger than what the caller supplied, just
7091 * count, don't do copyout.
7092 */
7093 if (++num > dlcount)
7094 continue;
7095 if (copyout(&zdl->zdl_id, idptr, sizeof (*idptr)) != 0) {
7096 mutex_exit(&zone->zone_lock);
7097 zone_rele(zone);
7098 return (set_errno(EFAULT));
7099 }
7100 idptr++;
7101 }
7102 mutex_exit(&zone->zone_lock);
7103 zone_rele(zone);
7104
7105 /* Increased or decreased, caller should be notified. */
7106 if (num != dlcount) {
7107 if (copyout(&num, nump, sizeof (num)) != 0)
7108 return (set_errno(EFAULT));
7109 }
7110 return (0);
7111 }
7112
7113 /*
7114 * Public interface for looking up a zone by zoneid. It's a customized version
7115 * for netstack_zone_create(). It can only be called from the zsd create
7116 * callbacks, since it doesn't have reference on the zone structure hence if
7117 * it is called elsewhere the zone could disappear after the zonehash_lock
7118 * is dropped.
7119 *
7120 * Furthermore it
7121 * 1. Doesn't check the status of the zone.
7122 * 2. It will be called even before zone_init is called, in that case the
7123 * address of zone0 is returned directly, and netstack_zone_create()
7124 * will only assign a value to zone0.zone_netstack, won't break anything.
7125 * 3. Returns without the zone being held.
7126 */
7127 zone_t *
7128 zone_find_by_id_nolock(zoneid_t zoneid)
7129 {
7130 zone_t *zone;
7131
7132 mutex_enter(&zonehash_lock);
7133 if (zonehashbyid == NULL)
7134 zone = &zone0;
7135 else
7136 zone = zone_find_all_by_id(zoneid);
7137 mutex_exit(&zonehash_lock);
7138 return (zone);
7139 }
7140
7141 /*
7142 * Walk the datalinks for a given zone
7143 */
7144 int
7145 zone_datalink_walk(zoneid_t zoneid, int (*cb)(datalink_id_t, void *),
7146 void *data)
7147 {
7148 zone_t *zone;
7149 zone_dl_t *zdl;
7150 datalink_id_t *idarray;
7151 uint_t idcount = 0;
7152 int i, ret = 0;
7153
7154 if ((zone = zone_find_by_id(zoneid)) == NULL)
7155 return (ENOENT);
7156
7157 /*
7158 * We first build an array of linkid's so that we can walk these and
7159 * execute the callback with the zone_lock dropped.
7160 */
7161 mutex_enter(&zone->zone_lock);
7162 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7163 zdl = list_next(&zone->zone_dl_list, zdl)) {
7164 idcount++;
7165 }
7166
7167 if (idcount == 0) {
7168 mutex_exit(&zone->zone_lock);
7169 zone_rele(zone);
7170 return (0);
7171 }
7172
7173 idarray = kmem_alloc(sizeof (datalink_id_t) * idcount, KM_NOSLEEP);
7174 if (idarray == NULL) {
7175 mutex_exit(&zone->zone_lock);
7176 zone_rele(zone);
7177 return (ENOMEM);
7178 }
7179
7180 for (i = 0, zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7181 i++, zdl = list_next(&zone->zone_dl_list, zdl)) {
7182 idarray[i] = zdl->zdl_id;
7183 }
7184
7185 mutex_exit(&zone->zone_lock);
7186
7187 for (i = 0; i < idcount && ret == 0; i++) {
7188 if ((ret = (*cb)(idarray[i], data)) != 0)
7189 break;
7190 }
7191
7192 zone_rele(zone);
7193 kmem_free(idarray, sizeof (datalink_id_t) * idcount);
7194 return (ret);
7195 }
7196
7197 static char *
7198 zone_net_type2name(int type)
7199 {
7200 switch (type) {
7201 case ZONE_NETWORK_ADDRESS:
7202 return (ZONE_NET_ADDRNAME);
7203 case ZONE_NETWORK_DEFROUTER:
7204 return (ZONE_NET_RTRNAME);
7205 default:
7206 return (NULL);
7207 }
7208 }
7209
7210 static int
7211 zone_set_network(zoneid_t zoneid, zone_net_data_t *znbuf)
7212 {
7213 zone_t *zone;
7214 zone_dl_t *zdl;
7215 nvlist_t *nvl;
7216 int err = 0;
7217 uint8_t *new = NULL;
7218 char *nvname;
7219 int bufsize;
7220 datalink_id_t linkid = znbuf->zn_linkid;
7221
7222 if (secpolicy_zone_config(CRED()) != 0)
7223 return (set_errno(EPERM));
7224
7225 if (zoneid == GLOBAL_ZONEID)
7226 return (set_errno(EINVAL));
7227
7228 nvname = zone_net_type2name(znbuf->zn_type);
7229 bufsize = znbuf->zn_len;
7230 new = znbuf->zn_val;
7231 if (nvname == NULL)
7232 return (set_errno(EINVAL));
7233
7234 if ((zone = zone_find_by_id(zoneid)) == NULL) {
7235 return (set_errno(EINVAL));
7236 }
7237
7238 mutex_enter(&zone->zone_lock);
7239 if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
7240 err = ENXIO;
7241 goto done;
7242 }
7243 if ((nvl = zdl->zdl_net) == NULL) {
7244 if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP)) {
7245 err = ENOMEM;
7246 goto done;
7247 } else {
7248 zdl->zdl_net = nvl;
7249 }
7250 }
7251 if (nvlist_exists(nvl, nvname)) {
7252 err = EINVAL;
7253 goto done;
7254 }
7255 err = nvlist_add_uint8_array(nvl, nvname, new, bufsize);
7256 ASSERT(err == 0);
7257 done:
7258 mutex_exit(&zone->zone_lock);
7259 zone_rele(zone);
7260 if (err != 0)
7261 return (set_errno(err));
7262 else
7263 return (0);
7264 }
7265
7266 static int
7267 zone_get_network(zoneid_t zoneid, zone_net_data_t *znbuf)
7268 {
7269 zone_t *zone;
7270 zone_dl_t *zdl;
7271 nvlist_t *nvl;
7272 uint8_t *ptr;
7273 uint_t psize;
7274 int err = 0;
7275 char *nvname;
7276 int bufsize;
7277 void *buf;
7278 datalink_id_t linkid = znbuf->zn_linkid;
7279
7280 if (zoneid == GLOBAL_ZONEID)
7281 return (set_errno(EINVAL));
7282
7283 nvname = zone_net_type2name(znbuf->zn_type);
7284 bufsize = znbuf->zn_len;
7285 buf = znbuf->zn_val;
7286
7287 if (nvname == NULL)
7288 return (set_errno(EINVAL));
7289 if ((zone = zone_find_by_id(zoneid)) == NULL)
7290 return (set_errno(EINVAL));
7291
7292 mutex_enter(&zone->zone_lock);
7293 if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
7294 err = ENXIO;
7295 goto done;
7296 }
7297 if ((nvl = zdl->zdl_net) == NULL || !nvlist_exists(nvl, nvname)) {
7298 err = ENOENT;
7299 goto done;
7300 }
7301 err = nvlist_lookup_uint8_array(nvl, nvname, &ptr, &psize);
7302 ASSERT(err == 0);
7303
7304 if (psize > bufsize) {
7305 err = ENOBUFS;
7306 goto done;
7307 }
7308 znbuf->zn_len = psize;
7309 bcopy(ptr, buf, psize);
7310 done:
7311 mutex_exit(&zone->zone_lock);
7312 zone_rele(zone);
7313 if (err != 0)
7314 return (set_errno(err));
7315 else
7316 return (0);
7317 }