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 2013, Joyent Inc. All rights reserved.
25 */
26
27 /*
28 * Zones
29 *
30 * A zone is a named collection of processes, namespace constraints,
31 * and other system resources which comprise a secure and manageable
32 * application containment facility.
33 *
34 * Zones (represented by the reference counted zone_t) are tracked in
35 * the kernel in the zonehash. Elsewhere in the kernel, Zone IDs
36 * (zoneid_t) are used to track zone association. Zone IDs are
37 * dynamically generated when the zone is created; if a persistent
38 * identifier is needed (core files, accounting logs, audit trail,
39 * etc.), the zone name should be used.
40 *
41 *
42 * Global Zone:
43 *
44 * The global zone (zoneid 0) is automatically associated with all
45 * system resources that have not been bound to a user-created zone.
46 * This means that even systems where zones are not in active use
47 * have a global zone, and all processes, mounts, etc. are
48 * associated with that zone. The global zone is generally
49 * unconstrained in terms of privileges and access, though the usual
50 * credential and privilege based restrictions apply.
51 *
52 *
53 * Zone States:
54 *
55 * The states in which a zone may be in and the transitions are as
56 * follows:
57 *
58 * ZONE_IS_UNINITIALIZED: primordial state for a zone. The partially
59 * initialized zone is added to the list of active zones on the system but
60 * isn't accessible.
61 *
62 * ZONE_IS_INITIALIZED: Initialization complete except the ZSD callbacks are
63 * not yet completed. Not possible to enter the zone, but attributes can
64 * be retrieved.
65 *
66 * ZONE_IS_READY: zsched (the kernel dummy process for a zone) is
67 * ready. The zone is made visible after the ZSD constructor callbacks are
68 * executed. A zone remains in this state until it transitions into
69 * the ZONE_IS_BOOTING state as a result of a call to zone_boot().
70 *
71 * ZONE_IS_BOOTING: in this shortlived-state, zsched attempts to start
72 * init. Should that fail, the zone proceeds to the ZONE_IS_SHUTTING_DOWN
73 * state.
74 *
75 * ZONE_IS_RUNNING: The zone is open for business: zsched has
76 * successfully started init. A zone remains in this state until
77 * zone_shutdown() is called.
78 *
79 * ZONE_IS_SHUTTING_DOWN: zone_shutdown() has been called, the system is
80 * killing all processes running in the zone. The zone remains
81 * in this state until there are no more user processes running in the zone.
82 * zone_create(), zone_enter(), and zone_destroy() on this zone will fail.
83 * Since zone_shutdown() is restartable, it may be called successfully
84 * multiple times for the same zone_t. Setting of the zone's state to
85 * ZONE_IS_SHUTTING_DOWN is synchronized with mounts, so VOP_MOUNT() may check
86 * the zone's status without worrying about it being a moving target.
87 *
88 * ZONE_IS_EMPTY: zone_shutdown() has been called, and there
89 * are no more user processes in the zone. The zone remains in this
90 * state until there are no more kernel threads associated with the
91 * zone. zone_create(), zone_enter(), and zone_destroy() on this zone will
92 * fail.
93 *
94 * ZONE_IS_DOWN: All kernel threads doing work on behalf of the zone
95 * have exited. zone_shutdown() returns. Henceforth it is not possible to
96 * join the zone or create kernel threads therein.
97 *
98 * ZONE_IS_DYING: zone_destroy() has been called on the zone; zone
99 * remains in this state until zsched exits. Calls to zone_find_by_*()
100 * return NULL from now on.
101 *
102 * ZONE_IS_DEAD: zsched has exited (zone_ntasks == 0). There are no
103 * processes or threads doing work on behalf of the zone. The zone is
104 * removed from the list of active zones. zone_destroy() returns, and
105 * the zone can be recreated.
106 *
107 * ZONE_IS_FREE (internal state): zone_ref goes to 0, ZSD destructor
108 * callbacks are executed, and all memory associated with the zone is
109 * freed.
110 *
111 * Threads can wait for the zone to enter a requested state by using
112 * zone_status_wait() or zone_status_timedwait() with the desired
113 * state passed in as an argument. Zone state transitions are
114 * uni-directional; it is not possible to move back to an earlier state.
115 *
116 *
117 * Zone-Specific Data:
118 *
119 * Subsystems needing to maintain zone-specific data can store that
120 * data using the ZSD mechanism. This provides a zone-specific data
121 * store, similar to thread-specific data (see pthread_getspecific(3C)
122 * or the TSD code in uts/common/disp/thread.c. Also, ZSD can be used
123 * to register callbacks to be invoked when a zone is created, shut
124 * down, or destroyed. This can be used to initialize zone-specific
125 * data for new zones and to clean up when zones go away.
126 *
127 *
128 * Data Structures:
129 *
130 * The per-zone structure (zone_t) is reference counted, and freed
131 * when all references are released. zone_hold and zone_rele can be
132 * used to adjust the reference count. In addition, reference counts
133 * associated with the cred_t structure are tracked separately using
134 * zone_cred_hold and zone_cred_rele.
135 *
136 * Pointers to active zone_t's are stored in two hash tables; one
137 * for searching by id, the other for searching by name. Lookups
138 * can be performed on either basis, using zone_find_by_id and
139 * zone_find_by_name. Both return zone_t pointers with the zone
140 * held, so zone_rele should be called when the pointer is no longer
141 * needed. Zones can also be searched by path; zone_find_by_path
142 * returns the zone with which a path name is associated (global
143 * zone if the path is not within some other zone's file system
144 * hierarchy). This currently requires iterating through each zone,
145 * so it is slower than an id or name search via a hash table.
146 *
147 *
148 * Locking:
149 *
150 * zonehash_lock: This is a top-level global lock used to protect the
151 * zone hash tables and lists. Zones cannot be created or destroyed
152 * while this lock is held.
153 * zone_status_lock: This is a global lock protecting zone state.
154 * Zones cannot change state while this lock is held. It also
155 * protects the list of kernel threads associated with a zone.
156 * zone_lock: This is a per-zone lock used to protect several fields of
157 * the zone_t (see <sys/zone.h> for details). In addition, holding
158 * this lock means that the zone cannot go away.
159 * zone_nlwps_lock: This is a per-zone lock used to protect the fields
160 * related to the zone.max-lwps rctl.
161 * zone_mem_lock: This is a per-zone lock used to protect the fields
162 * related to the zone.max-locked-memory and zone.max-swap rctls.
163 * zone_rctl_lock: This is a per-zone lock used to protect other rctls,
164 * currently just max_lofi
165 * zsd_key_lock: This is a global lock protecting the key state for ZSD.
166 * zone_deathrow_lock: This is a global lock protecting the "deathrow"
167 * list (a list of zones in the ZONE_IS_DEAD state).
168 *
169 * Ordering requirements:
170 * pool_lock --> cpu_lock --> zonehash_lock --> zone_status_lock -->
171 * zone_lock --> zsd_key_lock --> pidlock --> p_lock
172 *
173 * When taking zone_mem_lock or zone_nlwps_lock, the lock ordering is:
174 * zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_mem_lock
175 * zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_nlwps_lock
176 *
177 * Blocking memory allocations are permitted while holding any of the
178 * zone locks.
179 *
180 *
181 * System Call Interface:
182 *
183 * The zone subsystem can be managed and queried from user level with
184 * the following system calls (all subcodes of the primary "zone"
185 * system call):
186 * - zone_create: creates a zone with selected attributes (name,
187 * root path, privileges, resource controls, ZFS datasets)
188 * - zone_enter: allows the current process to enter a zone
189 * - zone_getattr: reports attributes of a zone
190 * - zone_setattr: set attributes of a zone
191 * - zone_boot: set 'init' running for the zone
192 * - zone_list: lists all zones active in the system
193 * - zone_lookup: looks up zone id based on name
194 * - zone_shutdown: initiates shutdown process (see states above)
195 * - zone_destroy: completes shutdown process (see states above)
196 *
197 */
198
199 #include <sys/priv_impl.h>
200 #include <sys/cred.h>
201 #include <c2/audit.h>
202 #include <sys/debug.h>
203 #include <sys/file.h>
204 #include <sys/kmem.h>
205 #include <sys/kstat.h>
206 #include <sys/mutex.h>
207 #include <sys/note.h>
208 #include <sys/pathname.h>
209 #include <sys/proc.h>
210 #include <sys/project.h>
211 #include <sys/sysevent.h>
212 #include <sys/task.h>
213 #include <sys/systm.h>
214 #include <sys/types.h>
215 #include <sys/utsname.h>
216 #include <sys/vnode.h>
217 #include <sys/vfs.h>
218 #include <sys/systeminfo.h>
219 #include <sys/policy.h>
220 #include <sys/cred_impl.h>
221 #include <sys/contract_impl.h>
222 #include <sys/contract/process_impl.h>
223 #include <sys/class.h>
224 #include <sys/pool.h>
225 #include <sys/pool_pset.h>
226 #include <sys/pset.h>
227 #include <sys/strlog.h>
228 #include <sys/sysmacros.h>
229 #include <sys/callb.h>
230 #include <sys/vmparam.h>
231 #include <sys/corectl.h>
232 #include <sys/ipc_impl.h>
233 #include <sys/klpd.h>
234
235 #include <sys/door.h>
236 #include <sys/cpuvar.h>
237 #include <sys/sdt.h>
238
239 #include <sys/uadmin.h>
240 #include <sys/session.h>
241 #include <sys/cmn_err.h>
242 #include <sys/modhash.h>
243 #include <sys/sunddi.h>
244 #include <sys/nvpair.h>
245 #include <sys/rctl.h>
246 #include <sys/fss.h>
247 #include <sys/brand.h>
248 #include <sys/zone.h>
249 #include <net/if.h>
250 #include <sys/cpucaps.h>
251 #include <vm/seg.h>
252 #include <sys/mac.h>
253
254 /*
255 * This constant specifies the number of seconds that threads waiting for
256 * subsystems to release a zone's general-purpose references will wait before
257 * they log the zone's reference counts. The constant's value shouldn't
258 * be so small that reference counts are unnecessarily reported for zones
259 * whose references are slowly released. On the other hand, it shouldn't be so
260 * large that users reboot their systems out of frustration over hung zones
261 * before the system logs the zones' reference counts.
262 */
263 #define ZONE_DESTROY_TIMEOUT_SECS 60
264
265 /* List of data link IDs which are accessible from the zone */
266 typedef struct zone_dl {
267 datalink_id_t zdl_id;
268 nvlist_t *zdl_net;
269 list_node_t zdl_linkage;
270 } zone_dl_t;
271
272 /*
273 * cv used to signal that all references to the zone have been released. This
274 * needs to be global since there may be multiple waiters, and the first to
275 * wake up will free the zone_t, hence we cannot use zone->zone_cv.
276 */
277 static kcondvar_t zone_destroy_cv;
278 /*
279 * Lock used to serialize access to zone_cv. This could have been per-zone,
280 * but then we'd need another lock for zone_destroy_cv, and why bother?
281 */
282 static kmutex_t zone_status_lock;
283
284 /*
285 * ZSD-related global variables.
286 */
287 static kmutex_t zsd_key_lock; /* protects the following two */
288 /*
289 * The next caller of zone_key_create() will be assigned a key of ++zsd_keyval.
290 */
291 static zone_key_t zsd_keyval = 0;
292 /*
293 * Global list of registered keys. We use this when a new zone is created.
294 */
295 static list_t zsd_registered_keys;
296
297 int zone_hash_size = 256;
298 static mod_hash_t *zonehashbyname, *zonehashbyid, *zonehashbylabel;
299 static kmutex_t zonehash_lock;
300 static uint_t zonecount;
301 static id_space_t *zoneid_space;
302
303 /*
304 * The global zone (aka zone0) is the all-seeing, all-knowing zone in which the
305 * kernel proper runs, and which manages all other zones.
306 *
307 * Although not declared as static, the variable "zone0" should not be used
308 * except for by code that needs to reference the global zone early on in boot,
309 * before it is fully initialized. All other consumers should use
310 * 'global_zone'.
311 */
312 zone_t zone0;
313 zone_t *global_zone = NULL; /* Set when the global zone is initialized */
314
315 /*
316 * List of active zones, protected by zonehash_lock.
317 */
318 static list_t zone_active;
319
320 /*
321 * List of destroyed zones that still have outstanding cred references.
322 * Used for debugging. Uses a separate lock to avoid lock ordering
323 * problems in zone_free.
324 */
325 static list_t zone_deathrow;
326 static kmutex_t zone_deathrow_lock;
327
328 /* number of zones is limited by virtual interface limit in IP */
329 uint_t maxzones = 8192;
330
331 /* Event channel to sent zone state change notifications */
332 evchan_t *zone_event_chan;
333
334 /*
335 * This table holds the mapping from kernel zone states to
336 * states visible in the state notification API.
337 * The idea is that we only expose "obvious" states and
338 * do not expose states which are just implementation details.
339 */
340 const char *zone_status_table[] = {
341 ZONE_EVENT_UNINITIALIZED, /* uninitialized */
342 ZONE_EVENT_INITIALIZED, /* initialized */
343 ZONE_EVENT_READY, /* ready */
344 ZONE_EVENT_READY, /* booting */
345 ZONE_EVENT_RUNNING, /* running */
346 ZONE_EVENT_SHUTTING_DOWN, /* shutting_down */
347 ZONE_EVENT_SHUTTING_DOWN, /* empty */
348 ZONE_EVENT_SHUTTING_DOWN, /* down */
349 ZONE_EVENT_SHUTTING_DOWN, /* dying */
350 ZONE_EVENT_UNINITIALIZED, /* dead */
351 };
352
353 /*
354 * This array contains the names of the subsystems listed in zone_ref_subsys_t
355 * (see sys/zone.h).
356 */
357 static char *zone_ref_subsys_names[] = {
358 "NFS", /* ZONE_REF_NFS */
359 "NFSv4", /* ZONE_REF_NFSV4 */
360 "SMBFS", /* ZONE_REF_SMBFS */
361 "MNTFS", /* ZONE_REF_MNTFS */
362 "LOFI", /* ZONE_REF_LOFI */
363 "VFS", /* ZONE_REF_VFS */
364 "IPC" /* ZONE_REF_IPC */
365 };
366
367 /*
368 * This isn't static so lint doesn't complain.
369 */
370 rctl_hndl_t rc_zone_cpu_shares;
371 rctl_hndl_t rc_zone_locked_mem;
372 rctl_hndl_t rc_zone_max_swap;
373 rctl_hndl_t rc_zone_max_lofi;
374 rctl_hndl_t rc_zone_cpu_cap;
375 rctl_hndl_t rc_zone_nlwps;
376 rctl_hndl_t rc_zone_nprocs;
377 rctl_hndl_t rc_zone_shmmax;
378 rctl_hndl_t rc_zone_shmmni;
379 rctl_hndl_t rc_zone_semmni;
380 rctl_hndl_t rc_zone_msgmni;
381
382 const char * const zone_default_initname = "/sbin/init";
383 static char * const zone_prefix = "/zone/";
384 static int zone_shutdown(zoneid_t zoneid);
385 static int zone_add_datalink(zoneid_t, datalink_id_t);
386 static int zone_remove_datalink(zoneid_t, datalink_id_t);
387 static int zone_list_datalink(zoneid_t, int *, datalink_id_t *);
388 static int zone_set_network(zoneid_t, zone_net_data_t *);
389 static int zone_get_network(zoneid_t, zone_net_data_t *);
390
391 typedef boolean_t zsd_applyfn_t(kmutex_t *, boolean_t, zone_t *, zone_key_t);
392
393 static void zsd_apply_all_zones(zsd_applyfn_t *, zone_key_t);
394 static void zsd_apply_all_keys(zsd_applyfn_t *, zone_t *);
395 static boolean_t zsd_apply_create(kmutex_t *, boolean_t, zone_t *, zone_key_t);
396 static boolean_t zsd_apply_shutdown(kmutex_t *, boolean_t, zone_t *,
397 zone_key_t);
398 static boolean_t zsd_apply_destroy(kmutex_t *, boolean_t, zone_t *, zone_key_t);
399 static boolean_t zsd_wait_for_creator(zone_t *, struct zsd_entry *,
400 kmutex_t *);
401 static boolean_t zsd_wait_for_inprogress(zone_t *, struct zsd_entry *,
402 kmutex_t *);
403
404 /*
405 * Bump this number when you alter the zone syscall interfaces; this is
406 * because we need to have support for previous API versions in libc
407 * to support patching; libc calls into the kernel to determine this number.
408 *
409 * Version 1 of the API is the version originally shipped with Solaris 10
410 * Version 2 alters the zone_create system call in order to support more
411 * arguments by moving the args into a structure; and to do better
412 * error reporting when zone_create() fails.
413 * Version 3 alters the zone_create system call in order to support the
414 * import of ZFS datasets to zones.
415 * Version 4 alters the zone_create system call in order to support
416 * Trusted Extensions.
417 * Version 5 alters the zone_boot system call, and converts its old
418 * bootargs parameter to be set by the zone_setattr API instead.
419 * Version 6 adds the flag argument to zone_create.
420 */
421 static const int ZONE_SYSCALL_API_VERSION = 6;
422
423 /*
424 * Certain filesystems (such as NFS and autofs) need to know which zone
425 * the mount is being placed in. Because of this, we need to be able to
426 * ensure that a zone isn't in the process of being created/destroyed such
427 * that nfs_mount() thinks it is in the global/NGZ zone, while by the time
428 * it gets added the list of mounted zones, it ends up on the wrong zone's
429 * mount list. Since a zone can't reside on an NFS file system, we don't
430 * have to worry about the zonepath itself.
431 *
432 * The following functions: block_mounts()/resume_mounts() and
433 * mount_in_progress()/mount_completed() are used by zones and the VFS
434 * layer (respectively) to synchronize zone state transitions and new
435 * mounts within a zone. This syncronization is on a per-zone basis, so
436 * activity for one zone will not interfere with activity for another zone.
437 *
438 * The semantics are like a reader-reader lock such that there may
439 * either be multiple mounts (or zone state transitions, if that weren't
440 * serialized by zonehash_lock) in progress at the same time, but not
441 * both.
442 *
443 * We use cv's so the user can ctrl-C out of the operation if it's
444 * taking too long.
445 *
446 * The semantics are such that there is unfair bias towards the
447 * "current" operation. This means that zone halt may starve if
448 * there is a rapid succession of new mounts coming in to the zone.
449 */
450 /*
451 * Prevent new mounts from progressing to the point of calling
452 * VFS_MOUNT(). If there are already mounts in this "region", wait for
453 * them to complete.
454 */
455 static int
456 block_mounts(zone_t *zp)
457 {
458 int retval = 0;
459
460 /*
461 * Since it may block for a long time, block_mounts() shouldn't be
462 * called with zonehash_lock held.
463 */
464 ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
465 mutex_enter(&zp->zone_mount_lock);
466 while (zp->zone_mounts_in_progress > 0) {
467 if (cv_wait_sig(&zp->zone_mount_cv, &zp->zone_mount_lock) == 0)
468 goto signaled;
469 }
470 /*
471 * A negative value of mounts_in_progress indicates that mounts
472 * have been blocked by (-mounts_in_progress) different callers
473 * (remotely possible if two threads enter zone_shutdown at the same
474 * time).
475 */
476 zp->zone_mounts_in_progress--;
477 retval = 1;
478 signaled:
479 mutex_exit(&zp->zone_mount_lock);
480 return (retval);
481 }
482
483 /*
484 * The VFS layer may progress with new mounts as far as we're concerned.
485 * Allow them to progress if we were the last obstacle.
486 */
487 static void
488 resume_mounts(zone_t *zp)
489 {
490 mutex_enter(&zp->zone_mount_lock);
491 if (++zp->zone_mounts_in_progress == 0)
492 cv_broadcast(&zp->zone_mount_cv);
493 mutex_exit(&zp->zone_mount_lock);
494 }
495
496 /*
497 * The VFS layer is busy with a mount; this zone should wait until all
498 * of its mounts are completed to progress.
499 */
500 void
501 mount_in_progress(zone_t *zp)
502 {
503 mutex_enter(&zp->zone_mount_lock);
504 while (zp->zone_mounts_in_progress < 0)
505 cv_wait(&zp->zone_mount_cv, &zp->zone_mount_lock);
506 zp->zone_mounts_in_progress++;
507 mutex_exit(&zp->zone_mount_lock);
508 }
509
510 /*
511 * VFS is done with one mount; wake up any waiting block_mounts()
512 * callers if this is the last mount.
513 */
514 void
515 mount_completed(zone_t *zp)
516 {
517 mutex_enter(&zp->zone_mount_lock);
518 if (--zp->zone_mounts_in_progress == 0)
519 cv_broadcast(&zp->zone_mount_cv);
520 mutex_exit(&zp->zone_mount_lock);
521 }
522
523 /*
524 * ZSD routines.
525 *
526 * Zone Specific Data (ZSD) is modeled after Thread Specific Data as
527 * defined by the pthread_key_create() and related interfaces.
528 *
529 * Kernel subsystems may register one or more data items and/or
530 * callbacks to be executed when a zone is created, shutdown, or
531 * destroyed.
532 *
533 * Unlike the thread counterpart, destructor callbacks will be executed
534 * even if the data pointer is NULL and/or there are no constructor
535 * callbacks, so it is the responsibility of such callbacks to check for
536 * NULL data values if necessary.
537 *
538 * The locking strategy and overall picture is as follows:
539 *
540 * When someone calls zone_key_create(), a template ZSD entry is added to the
541 * global list "zsd_registered_keys", protected by zsd_key_lock. While
542 * holding that lock all the existing zones are marked as
543 * ZSD_CREATE_NEEDED and a copy of the ZSD entry added to the per-zone
544 * zone_zsd list (protected by zone_lock). The global list is updated first
545 * (under zone_key_lock) to make sure that newly created zones use the
546 * most recent list of keys. Then under zonehash_lock we walk the zones
547 * and mark them. Similar locking is used in zone_key_delete().
548 *
549 * The actual create, shutdown, and destroy callbacks are done without
550 * holding any lock. And zsd_flags are used to ensure that the operations
551 * completed so that when zone_key_create (and zone_create) is done, as well as
552 * zone_key_delete (and zone_destroy) is done, all the necessary callbacks
553 * are completed.
554 *
555 * When new zones are created constructor callbacks for all registered ZSD
556 * entries will be called. That also uses the above two phases of marking
557 * what needs to be done, and then running the callbacks without holding
558 * any locks.
559 *
560 * The framework does not provide any locking around zone_getspecific() and
561 * zone_setspecific() apart from that needed for internal consistency, so
562 * callers interested in atomic "test-and-set" semantics will need to provide
563 * their own locking.
564 */
565
566 /*
567 * Helper function to find the zsd_entry associated with the key in the
568 * given list.
569 */
570 static struct zsd_entry *
571 zsd_find(list_t *l, zone_key_t key)
572 {
573 struct zsd_entry *zsd;
574
575 for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
576 if (zsd->zsd_key == key) {
577 return (zsd);
578 }
579 }
580 return (NULL);
581 }
582
583 /*
584 * Helper function to find the zsd_entry associated with the key in the
585 * given list. Move it to the front of the list.
586 */
587 static struct zsd_entry *
588 zsd_find_mru(list_t *l, zone_key_t key)
589 {
590 struct zsd_entry *zsd;
591
592 for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
593 if (zsd->zsd_key == key) {
594 /*
595 * Move to head of list to keep list in MRU order.
596 */
597 if (zsd != list_head(l)) {
598 list_remove(l, zsd);
599 list_insert_head(l, zsd);
600 }
601 return (zsd);
602 }
603 }
604 return (NULL);
605 }
606
607 void
608 zone_key_create(zone_key_t *keyp, void *(*create)(zoneid_t),
609 void (*shutdown)(zoneid_t, void *), void (*destroy)(zoneid_t, void *))
610 {
611 struct zsd_entry *zsdp;
612 struct zsd_entry *t;
613 struct zone *zone;
614 zone_key_t key;
615
616 zsdp = kmem_zalloc(sizeof (*zsdp), KM_SLEEP);
617 zsdp->zsd_data = NULL;
618 zsdp->zsd_create = create;
619 zsdp->zsd_shutdown = shutdown;
620 zsdp->zsd_destroy = destroy;
621
622 /*
623 * Insert in global list of callbacks. Makes future zone creations
624 * see it.
625 */
626 mutex_enter(&zsd_key_lock);
627 key = zsdp->zsd_key = ++zsd_keyval;
628 ASSERT(zsd_keyval != 0);
629 list_insert_tail(&zsd_registered_keys, zsdp);
630 mutex_exit(&zsd_key_lock);
631
632 /*
633 * Insert for all existing zones and mark them as needing
634 * a create callback.
635 */
636 mutex_enter(&zonehash_lock); /* stop the world */
637 for (zone = list_head(&zone_active); zone != NULL;
638 zone = list_next(&zone_active, zone)) {
639 zone_status_t status;
640
641 mutex_enter(&zone->zone_lock);
642
643 /* Skip zones that are on the way down or not yet up */
644 status = zone_status_get(zone);
645 if (status >= ZONE_IS_DOWN ||
646 status == ZONE_IS_UNINITIALIZED) {
647 mutex_exit(&zone->zone_lock);
648 continue;
649 }
650
651 t = zsd_find_mru(&zone->zone_zsd, key);
652 if (t != NULL) {
653 /*
654 * A zsd_configure already inserted it after
655 * we dropped zsd_key_lock above.
656 */
657 mutex_exit(&zone->zone_lock);
658 continue;
659 }
660 t = kmem_zalloc(sizeof (*t), KM_SLEEP);
661 t->zsd_key = key;
662 t->zsd_create = create;
663 t->zsd_shutdown = shutdown;
664 t->zsd_destroy = destroy;
665 if (create != NULL) {
666 t->zsd_flags = ZSD_CREATE_NEEDED;
667 DTRACE_PROBE2(zsd__create__needed,
668 zone_t *, zone, zone_key_t, key);
669 }
670 list_insert_tail(&zone->zone_zsd, t);
671 mutex_exit(&zone->zone_lock);
672 }
673 mutex_exit(&zonehash_lock);
674
675 if (create != NULL) {
676 /* Now call the create callback for this key */
677 zsd_apply_all_zones(zsd_apply_create, key);
678 }
679 /*
680 * It is safe for consumers to use the key now, make it
681 * globally visible. Specifically zone_getspecific() will
682 * always successfully return the zone specific data associated
683 * with the key.
684 */
685 *keyp = key;
686
687 }
688
689 /*
690 * Function called when a module is being unloaded, or otherwise wishes
691 * to unregister its ZSD key and callbacks.
692 *
693 * Remove from the global list and determine the functions that need to
694 * be called under a global lock. Then call the functions without
695 * holding any locks. Finally free up the zone_zsd entries. (The apply
696 * functions need to access the zone_zsd entries to find zsd_data etc.)
697 */
698 int
699 zone_key_delete(zone_key_t key)
700 {
701 struct zsd_entry *zsdp = NULL;
702 zone_t *zone;
703
704 mutex_enter(&zsd_key_lock);
705 zsdp = zsd_find_mru(&zsd_registered_keys, key);
706 if (zsdp == NULL) {
707 mutex_exit(&zsd_key_lock);
708 return (-1);
709 }
710 list_remove(&zsd_registered_keys, zsdp);
711 mutex_exit(&zsd_key_lock);
712
713 mutex_enter(&zonehash_lock);
714 for (zone = list_head(&zone_active); zone != NULL;
715 zone = list_next(&zone_active, zone)) {
716 struct zsd_entry *del;
717
718 mutex_enter(&zone->zone_lock);
719 del = zsd_find_mru(&zone->zone_zsd, key);
720 if (del == NULL) {
721 /*
722 * Somebody else got here first e.g the zone going
723 * away.
724 */
725 mutex_exit(&zone->zone_lock);
726 continue;
727 }
728 ASSERT(del->zsd_shutdown == zsdp->zsd_shutdown);
729 ASSERT(del->zsd_destroy == zsdp->zsd_destroy);
730 if (del->zsd_shutdown != NULL &&
731 (del->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
732 del->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
733 DTRACE_PROBE2(zsd__shutdown__needed,
734 zone_t *, zone, zone_key_t, key);
735 }
736 if (del->zsd_destroy != NULL &&
737 (del->zsd_flags & ZSD_DESTROY_ALL) == 0) {
738 del->zsd_flags |= ZSD_DESTROY_NEEDED;
739 DTRACE_PROBE2(zsd__destroy__needed,
740 zone_t *, zone, zone_key_t, key);
741 }
742 mutex_exit(&zone->zone_lock);
743 }
744 mutex_exit(&zonehash_lock);
745 kmem_free(zsdp, sizeof (*zsdp));
746
747 /* Now call the shutdown and destroy callback for this key */
748 zsd_apply_all_zones(zsd_apply_shutdown, key);
749 zsd_apply_all_zones(zsd_apply_destroy, key);
750
751 /* Now we can free up the zsdp structures in each zone */
752 mutex_enter(&zonehash_lock);
753 for (zone = list_head(&zone_active); zone != NULL;
754 zone = list_next(&zone_active, zone)) {
755 struct zsd_entry *del;
756
757 mutex_enter(&zone->zone_lock);
758 del = zsd_find(&zone->zone_zsd, key);
759 if (del != NULL) {
760 list_remove(&zone->zone_zsd, del);
761 ASSERT(!(del->zsd_flags & ZSD_ALL_INPROGRESS));
762 kmem_free(del, sizeof (*del));
763 }
764 mutex_exit(&zone->zone_lock);
765 }
766 mutex_exit(&zonehash_lock);
767
768 return (0);
769 }
770
771 /*
772 * ZSD counterpart of pthread_setspecific().
773 *
774 * Since all zsd callbacks, including those with no create function,
775 * have an entry in zone_zsd, if the key is registered it is part of
776 * the zone_zsd list.
777 * Return an error if the key wasn't registerd.
778 */
779 int
780 zone_setspecific(zone_key_t key, zone_t *zone, const void *data)
781 {
782 struct zsd_entry *t;
783
784 mutex_enter(&zone->zone_lock);
785 t = zsd_find_mru(&zone->zone_zsd, key);
786 if (t != NULL) {
787 /*
788 * Replace old value with new
789 */
790 t->zsd_data = (void *)data;
791 mutex_exit(&zone->zone_lock);
792 return (0);
793 }
794 mutex_exit(&zone->zone_lock);
795 return (-1);
796 }
797
798 /*
799 * ZSD counterpart of pthread_getspecific().
800 */
801 void *
802 zone_getspecific(zone_key_t key, zone_t *zone)
803 {
804 struct zsd_entry *t;
805 void *data;
806
807 mutex_enter(&zone->zone_lock);
808 t = zsd_find_mru(&zone->zone_zsd, key);
809 data = (t == NULL ? NULL : t->zsd_data);
810 mutex_exit(&zone->zone_lock);
811 return (data);
812 }
813
814 /*
815 * Function used to initialize a zone's list of ZSD callbacks and data
816 * when the zone is being created. The callbacks are initialized from
817 * the template list (zsd_registered_keys). The constructor callback is
818 * executed later (once the zone exists and with locks dropped).
819 */
820 static void
821 zone_zsd_configure(zone_t *zone)
822 {
823 struct zsd_entry *zsdp;
824 struct zsd_entry *t;
825
826 ASSERT(MUTEX_HELD(&zonehash_lock));
827 ASSERT(list_head(&zone->zone_zsd) == NULL);
828 mutex_enter(&zone->zone_lock);
829 mutex_enter(&zsd_key_lock);
830 for (zsdp = list_head(&zsd_registered_keys); zsdp != NULL;
831 zsdp = list_next(&zsd_registered_keys, zsdp)) {
832 /*
833 * Since this zone is ZONE_IS_UNCONFIGURED, zone_key_create
834 * should not have added anything to it.
835 */
836 ASSERT(zsd_find(&zone->zone_zsd, zsdp->zsd_key) == NULL);
837
838 t = kmem_zalloc(sizeof (*t), KM_SLEEP);
839 t->zsd_key = zsdp->zsd_key;
840 t->zsd_create = zsdp->zsd_create;
841 t->zsd_shutdown = zsdp->zsd_shutdown;
842 t->zsd_destroy = zsdp->zsd_destroy;
843 if (zsdp->zsd_create != NULL) {
844 t->zsd_flags = ZSD_CREATE_NEEDED;
845 DTRACE_PROBE2(zsd__create__needed,
846 zone_t *, zone, zone_key_t, zsdp->zsd_key);
847 }
848 list_insert_tail(&zone->zone_zsd, t);
849 }
850 mutex_exit(&zsd_key_lock);
851 mutex_exit(&zone->zone_lock);
852 }
853
854 enum zsd_callback_type { ZSD_CREATE, ZSD_SHUTDOWN, ZSD_DESTROY };
855
856 /*
857 * Helper function to execute shutdown or destructor callbacks.
858 */
859 static void
860 zone_zsd_callbacks(zone_t *zone, enum zsd_callback_type ct)
861 {
862 struct zsd_entry *t;
863
864 ASSERT(ct == ZSD_SHUTDOWN || ct == ZSD_DESTROY);
865 ASSERT(ct != ZSD_SHUTDOWN || zone_status_get(zone) >= ZONE_IS_EMPTY);
866 ASSERT(ct != ZSD_DESTROY || zone_status_get(zone) >= ZONE_IS_DOWN);
867
868 /*
869 * Run the callback solely based on what is registered for the zone
870 * in zone_zsd. The global list can change independently of this
871 * as keys are registered and unregistered and we don't register new
872 * callbacks for a zone that is in the process of going away.
873 */
874 mutex_enter(&zone->zone_lock);
875 for (t = list_head(&zone->zone_zsd); t != NULL;
876 t = list_next(&zone->zone_zsd, t)) {
877 zone_key_t key = t->zsd_key;
878
879 /* Skip if no callbacks registered */
880
881 if (ct == ZSD_SHUTDOWN) {
882 if (t->zsd_shutdown != NULL &&
883 (t->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
884 t->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
885 DTRACE_PROBE2(zsd__shutdown__needed,
886 zone_t *, zone, zone_key_t, key);
887 }
888 } else {
889 if (t->zsd_destroy != NULL &&
890 (t->zsd_flags & ZSD_DESTROY_ALL) == 0) {
891 t->zsd_flags |= ZSD_DESTROY_NEEDED;
892 DTRACE_PROBE2(zsd__destroy__needed,
893 zone_t *, zone, zone_key_t, key);
894 }
895 }
896 }
897 mutex_exit(&zone->zone_lock);
898
899 /* Now call the shutdown and destroy callback for this key */
900 zsd_apply_all_keys(zsd_apply_shutdown, zone);
901 zsd_apply_all_keys(zsd_apply_destroy, zone);
902
903 }
904
905 /*
906 * Called when the zone is going away; free ZSD-related memory, and
907 * destroy the zone_zsd list.
908 */
909 static void
910 zone_free_zsd(zone_t *zone)
911 {
912 struct zsd_entry *t, *next;
913
914 /*
915 * Free all the zsd_entry's we had on this zone.
916 */
917 mutex_enter(&zone->zone_lock);
918 for (t = list_head(&zone->zone_zsd); t != NULL; t = next) {
919 next = list_next(&zone->zone_zsd, t);
920 list_remove(&zone->zone_zsd, t);
921 ASSERT(!(t->zsd_flags & ZSD_ALL_INPROGRESS));
922 kmem_free(t, sizeof (*t));
923 }
924 list_destroy(&zone->zone_zsd);
925 mutex_exit(&zone->zone_lock);
926
927 }
928
929 /*
930 * Apply a function to all zones for particular key value.
931 *
932 * The applyfn has to drop zonehash_lock if it does some work, and
933 * then reacquire it before it returns.
934 * When the lock is dropped we don't follow list_next even
935 * if it is possible to do so without any hazards. This is
936 * because we want the design to allow for the list of zones
937 * to change in any arbitrary way during the time the
938 * lock was dropped.
939 *
940 * It is safe to restart the loop at list_head since the applyfn
941 * changes the zsd_flags as it does work, so a subsequent
942 * pass through will have no effect in applyfn, hence the loop will terminate
943 * in at worst O(N^2).
944 */
945 static void
946 zsd_apply_all_zones(zsd_applyfn_t *applyfn, zone_key_t key)
947 {
948 zone_t *zone;
949
950 mutex_enter(&zonehash_lock);
951 zone = list_head(&zone_active);
952 while (zone != NULL) {
953 if ((applyfn)(&zonehash_lock, B_FALSE, zone, key)) {
954 /* Lock dropped - restart at head */
955 zone = list_head(&zone_active);
956 } else {
957 zone = list_next(&zone_active, zone);
958 }
959 }
960 mutex_exit(&zonehash_lock);
961 }
962
963 /*
964 * Apply a function to all keys for a particular zone.
965 *
966 * The applyfn has to drop zonehash_lock if it does some work, and
967 * then reacquire it before it returns.
968 * When the lock is dropped we don't follow list_next even
969 * if it is possible to do so without any hazards. This is
970 * because we want the design to allow for the list of zsd callbacks
971 * to change in any arbitrary way during the time the
972 * lock was dropped.
973 *
974 * It is safe to restart the loop at list_head since the applyfn
975 * changes the zsd_flags as it does work, so a subsequent
976 * pass through will have no effect in applyfn, hence the loop will terminate
977 * in at worst O(N^2).
978 */
979 static void
980 zsd_apply_all_keys(zsd_applyfn_t *applyfn, zone_t *zone)
981 {
982 struct zsd_entry *t;
983
984 mutex_enter(&zone->zone_lock);
985 t = list_head(&zone->zone_zsd);
986 while (t != NULL) {
987 if ((applyfn)(NULL, B_TRUE, zone, t->zsd_key)) {
988 /* Lock dropped - restart at head */
989 t = list_head(&zone->zone_zsd);
990 } else {
991 t = list_next(&zone->zone_zsd, t);
992 }
993 }
994 mutex_exit(&zone->zone_lock);
995 }
996
997 /*
998 * Call the create function for the zone and key if CREATE_NEEDED
999 * is set.
1000 * If some other thread gets here first and sets CREATE_INPROGRESS, then
1001 * we wait for that thread to complete so that we can ensure that
1002 * all the callbacks are done when we've looped over all zones/keys.
1003 *
1004 * When we call the create function, we drop the global held by the
1005 * caller, and return true to tell the caller it needs to re-evalute the
1006 * state.
1007 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1008 * remains held on exit.
1009 */
1010 static boolean_t
1011 zsd_apply_create(kmutex_t *lockp, boolean_t zone_lock_held,
1012 zone_t *zone, zone_key_t key)
1013 {
1014 void *result;
1015 struct zsd_entry *t;
1016 boolean_t dropped;
1017
1018 if (lockp != NULL) {
1019 ASSERT(MUTEX_HELD(lockp));
1020 }
1021 if (zone_lock_held) {
1022 ASSERT(MUTEX_HELD(&zone->zone_lock));
1023 } else {
1024 mutex_enter(&zone->zone_lock);
1025 }
1026
1027 t = zsd_find(&zone->zone_zsd, key);
1028 if (t == NULL) {
1029 /*
1030 * Somebody else got here first e.g the zone going
1031 * away.
1032 */
1033 if (!zone_lock_held)
1034 mutex_exit(&zone->zone_lock);
1035 return (B_FALSE);
1036 }
1037 dropped = B_FALSE;
1038 if (zsd_wait_for_inprogress(zone, t, lockp))
1039 dropped = B_TRUE;
1040
1041 if (t->zsd_flags & ZSD_CREATE_NEEDED) {
1042 t->zsd_flags &= ~ZSD_CREATE_NEEDED;
1043 t->zsd_flags |= ZSD_CREATE_INPROGRESS;
1044 DTRACE_PROBE2(zsd__create__inprogress,
1045 zone_t *, zone, zone_key_t, key);
1046 mutex_exit(&zone->zone_lock);
1047 if (lockp != NULL)
1048 mutex_exit(lockp);
1049
1050 dropped = B_TRUE;
1051 ASSERT(t->zsd_create != NULL);
1052 DTRACE_PROBE2(zsd__create__start,
1053 zone_t *, zone, zone_key_t, key);
1054
1055 result = (*t->zsd_create)(zone->zone_id);
1056
1057 DTRACE_PROBE2(zsd__create__end,
1058 zone_t *, zone, voidn *, result);
1059
1060 ASSERT(result != NULL);
1061 if (lockp != NULL)
1062 mutex_enter(lockp);
1063 mutex_enter(&zone->zone_lock);
1064 t->zsd_data = result;
1065 t->zsd_flags &= ~ZSD_CREATE_INPROGRESS;
1066 t->zsd_flags |= ZSD_CREATE_COMPLETED;
1067 cv_broadcast(&t->zsd_cv);
1068 DTRACE_PROBE2(zsd__create__completed,
1069 zone_t *, zone, zone_key_t, key);
1070 }
1071 if (!zone_lock_held)
1072 mutex_exit(&zone->zone_lock);
1073 return (dropped);
1074 }
1075
1076 /*
1077 * Call the shutdown function for the zone and key if SHUTDOWN_NEEDED
1078 * is set.
1079 * If some other thread gets here first and sets *_INPROGRESS, then
1080 * we wait for that thread to complete so that we can ensure that
1081 * all the callbacks are done when we've looped over all zones/keys.
1082 *
1083 * When we call the shutdown function, we drop the global held by the
1084 * caller, and return true to tell the caller it needs to re-evalute the
1085 * state.
1086 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1087 * remains held on exit.
1088 */
1089 static boolean_t
1090 zsd_apply_shutdown(kmutex_t *lockp, boolean_t zone_lock_held,
1091 zone_t *zone, zone_key_t key)
1092 {
1093 struct zsd_entry *t;
1094 void *data;
1095 boolean_t dropped;
1096
1097 if (lockp != NULL) {
1098 ASSERT(MUTEX_HELD(lockp));
1099 }
1100 if (zone_lock_held) {
1101 ASSERT(MUTEX_HELD(&zone->zone_lock));
1102 } else {
1103 mutex_enter(&zone->zone_lock);
1104 }
1105
1106 t = zsd_find(&zone->zone_zsd, key);
1107 if (t == NULL) {
1108 /*
1109 * Somebody else got here first e.g the zone going
1110 * away.
1111 */
1112 if (!zone_lock_held)
1113 mutex_exit(&zone->zone_lock);
1114 return (B_FALSE);
1115 }
1116 dropped = B_FALSE;
1117 if (zsd_wait_for_creator(zone, t, lockp))
1118 dropped = B_TRUE;
1119
1120 if (zsd_wait_for_inprogress(zone, t, lockp))
1121 dropped = B_TRUE;
1122
1123 if (t->zsd_flags & ZSD_SHUTDOWN_NEEDED) {
1124 t->zsd_flags &= ~ZSD_SHUTDOWN_NEEDED;
1125 t->zsd_flags |= ZSD_SHUTDOWN_INPROGRESS;
1126 DTRACE_PROBE2(zsd__shutdown__inprogress,
1127 zone_t *, zone, zone_key_t, key);
1128 mutex_exit(&zone->zone_lock);
1129 if (lockp != NULL)
1130 mutex_exit(lockp);
1131 dropped = B_TRUE;
1132
1133 ASSERT(t->zsd_shutdown != NULL);
1134 data = t->zsd_data;
1135
1136 DTRACE_PROBE2(zsd__shutdown__start,
1137 zone_t *, zone, zone_key_t, key);
1138
1139 (t->zsd_shutdown)(zone->zone_id, data);
1140 DTRACE_PROBE2(zsd__shutdown__end,
1141 zone_t *, zone, zone_key_t, key);
1142
1143 if (lockp != NULL)
1144 mutex_enter(lockp);
1145 mutex_enter(&zone->zone_lock);
1146 t->zsd_flags &= ~ZSD_SHUTDOWN_INPROGRESS;
1147 t->zsd_flags |= ZSD_SHUTDOWN_COMPLETED;
1148 cv_broadcast(&t->zsd_cv);
1149 DTRACE_PROBE2(zsd__shutdown__completed,
1150 zone_t *, zone, zone_key_t, key);
1151 }
1152 if (!zone_lock_held)
1153 mutex_exit(&zone->zone_lock);
1154 return (dropped);
1155 }
1156
1157 /*
1158 * Call the destroy function for the zone and key if DESTROY_NEEDED
1159 * is set.
1160 * If some other thread gets here first and sets *_INPROGRESS, then
1161 * we wait for that thread to complete so that we can ensure that
1162 * all the callbacks are done when we've looped over all zones/keys.
1163 *
1164 * When we call the destroy function, we drop the global held by the
1165 * caller, and return true to tell the caller it needs to re-evalute the
1166 * state.
1167 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1168 * remains held on exit.
1169 */
1170 static boolean_t
1171 zsd_apply_destroy(kmutex_t *lockp, boolean_t zone_lock_held,
1172 zone_t *zone, zone_key_t key)
1173 {
1174 struct zsd_entry *t;
1175 void *data;
1176 boolean_t dropped;
1177
1178 if (lockp != NULL) {
1179 ASSERT(MUTEX_HELD(lockp));
1180 }
1181 if (zone_lock_held) {
1182 ASSERT(MUTEX_HELD(&zone->zone_lock));
1183 } else {
1184 mutex_enter(&zone->zone_lock);
1185 }
1186
1187 t = zsd_find(&zone->zone_zsd, key);
1188 if (t == NULL) {
1189 /*
1190 * Somebody else got here first e.g the zone going
1191 * away.
1192 */
1193 if (!zone_lock_held)
1194 mutex_exit(&zone->zone_lock);
1195 return (B_FALSE);
1196 }
1197 dropped = B_FALSE;
1198 if (zsd_wait_for_creator(zone, t, lockp))
1199 dropped = B_TRUE;
1200
1201 if (zsd_wait_for_inprogress(zone, t, lockp))
1202 dropped = B_TRUE;
1203
1204 if (t->zsd_flags & ZSD_DESTROY_NEEDED) {
1205 t->zsd_flags &= ~ZSD_DESTROY_NEEDED;
1206 t->zsd_flags |= ZSD_DESTROY_INPROGRESS;
1207 DTRACE_PROBE2(zsd__destroy__inprogress,
1208 zone_t *, zone, zone_key_t, key);
1209 mutex_exit(&zone->zone_lock);
1210 if (lockp != NULL)
1211 mutex_exit(lockp);
1212 dropped = B_TRUE;
1213
1214 ASSERT(t->zsd_destroy != NULL);
1215 data = t->zsd_data;
1216 DTRACE_PROBE2(zsd__destroy__start,
1217 zone_t *, zone, zone_key_t, key);
1218
1219 (t->zsd_destroy)(zone->zone_id, data);
1220 DTRACE_PROBE2(zsd__destroy__end,
1221 zone_t *, zone, zone_key_t, key);
1222
1223 if (lockp != NULL)
1224 mutex_enter(lockp);
1225 mutex_enter(&zone->zone_lock);
1226 t->zsd_data = NULL;
1227 t->zsd_flags &= ~ZSD_DESTROY_INPROGRESS;
1228 t->zsd_flags |= ZSD_DESTROY_COMPLETED;
1229 cv_broadcast(&t->zsd_cv);
1230 DTRACE_PROBE2(zsd__destroy__completed,
1231 zone_t *, zone, zone_key_t, key);
1232 }
1233 if (!zone_lock_held)
1234 mutex_exit(&zone->zone_lock);
1235 return (dropped);
1236 }
1237
1238 /*
1239 * Wait for any CREATE_NEEDED flag to be cleared.
1240 * Returns true if lockp was temporarily dropped while waiting.
1241 */
1242 static boolean_t
1243 zsd_wait_for_creator(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
1244 {
1245 boolean_t dropped = B_FALSE;
1246
1247 while (t->zsd_flags & ZSD_CREATE_NEEDED) {
1248 DTRACE_PROBE2(zsd__wait__for__creator,
1249 zone_t *, zone, struct zsd_entry *, t);
1250 if (lockp != NULL) {
1251 dropped = B_TRUE;
1252 mutex_exit(lockp);
1253 }
1254 cv_wait(&t->zsd_cv, &zone->zone_lock);
1255 if (lockp != NULL) {
1256 /* First drop zone_lock to preserve order */
1257 mutex_exit(&zone->zone_lock);
1258 mutex_enter(lockp);
1259 mutex_enter(&zone->zone_lock);
1260 }
1261 }
1262 return (dropped);
1263 }
1264
1265 /*
1266 * Wait for any INPROGRESS flag to be cleared.
1267 * Returns true if lockp was temporarily dropped while waiting.
1268 */
1269 static boolean_t
1270 zsd_wait_for_inprogress(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
1271 {
1272 boolean_t dropped = B_FALSE;
1273
1274 while (t->zsd_flags & ZSD_ALL_INPROGRESS) {
1275 DTRACE_PROBE2(zsd__wait__for__inprogress,
1276 zone_t *, zone, struct zsd_entry *, t);
1277 if (lockp != NULL) {
1278 dropped = B_TRUE;
1279 mutex_exit(lockp);
1280 }
1281 cv_wait(&t->zsd_cv, &zone->zone_lock);
1282 if (lockp != NULL) {
1283 /* First drop zone_lock to preserve order */
1284 mutex_exit(&zone->zone_lock);
1285 mutex_enter(lockp);
1286 mutex_enter(&zone->zone_lock);
1287 }
1288 }
1289 return (dropped);
1290 }
1291
1292 /*
1293 * Frees memory associated with the zone dataset list.
1294 */
1295 static void
1296 zone_free_datasets(zone_t *zone)
1297 {
1298 zone_dataset_t *t, *next;
1299
1300 for (t = list_head(&zone->zone_datasets); t != NULL; t = next) {
1301 next = list_next(&zone->zone_datasets, t);
1302 list_remove(&zone->zone_datasets, t);
1303 kmem_free(t->zd_dataset, strlen(t->zd_dataset) + 1);
1304 kmem_free(t, sizeof (*t));
1305 }
1306 list_destroy(&zone->zone_datasets);
1307 }
1308
1309 /*
1310 * zone.cpu-shares resource control support.
1311 */
1312 /*ARGSUSED*/
1313 static rctl_qty_t
1314 zone_cpu_shares_usage(rctl_t *rctl, struct proc *p)
1315 {
1316 ASSERT(MUTEX_HELD(&p->p_lock));
1317 return (p->p_zone->zone_shares);
1318 }
1319
1320 /*ARGSUSED*/
1321 static int
1322 zone_cpu_shares_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1323 rctl_qty_t nv)
1324 {
1325 ASSERT(MUTEX_HELD(&p->p_lock));
1326 ASSERT(e->rcep_t == RCENTITY_ZONE);
1327 if (e->rcep_p.zone == NULL)
1328 return (0);
1329
1330 e->rcep_p.zone->zone_shares = nv;
1331 return (0);
1332 }
1333
1334 static rctl_ops_t zone_cpu_shares_ops = {
1335 rcop_no_action,
1336 zone_cpu_shares_usage,
1337 zone_cpu_shares_set,
1338 rcop_no_test
1339 };
1340
1341 /*
1342 * zone.cpu-cap resource control support.
1343 */
1344 /*ARGSUSED*/
1345 static rctl_qty_t
1346 zone_cpu_cap_get(rctl_t *rctl, struct proc *p)
1347 {
1348 ASSERT(MUTEX_HELD(&p->p_lock));
1349 return (cpucaps_zone_get(p->p_zone));
1350 }
1351
1352 /*ARGSUSED*/
1353 static int
1354 zone_cpu_cap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1355 rctl_qty_t nv)
1356 {
1357 zone_t *zone = e->rcep_p.zone;
1358
1359 ASSERT(MUTEX_HELD(&p->p_lock));
1360 ASSERT(e->rcep_t == RCENTITY_ZONE);
1361
1362 if (zone == NULL)
1363 return (0);
1364
1365 /*
1366 * set cap to the new value.
1367 */
1368 return (cpucaps_zone_set(zone, nv));
1369 }
1370
1371 static rctl_ops_t zone_cpu_cap_ops = {
1372 rcop_no_action,
1373 zone_cpu_cap_get,
1374 zone_cpu_cap_set,
1375 rcop_no_test
1376 };
1377
1378 /*ARGSUSED*/
1379 static rctl_qty_t
1380 zone_lwps_usage(rctl_t *r, proc_t *p)
1381 {
1382 rctl_qty_t nlwps;
1383 zone_t *zone = p->p_zone;
1384
1385 ASSERT(MUTEX_HELD(&p->p_lock));
1386
1387 mutex_enter(&zone->zone_nlwps_lock);
1388 nlwps = zone->zone_nlwps;
1389 mutex_exit(&zone->zone_nlwps_lock);
1390
1391 return (nlwps);
1392 }
1393
1394 /*ARGSUSED*/
1395 static int
1396 zone_lwps_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
1397 rctl_qty_t incr, uint_t flags)
1398 {
1399 rctl_qty_t nlwps;
1400
1401 ASSERT(MUTEX_HELD(&p->p_lock));
1402 ASSERT(e->rcep_t == RCENTITY_ZONE);
1403 if (e->rcep_p.zone == NULL)
1404 return (0);
1405 ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
1406 nlwps = e->rcep_p.zone->zone_nlwps;
1407
1408 if (nlwps + incr > rcntl->rcv_value)
1409 return (1);
1410
1411 return (0);
1412 }
1413
1414 /*ARGSUSED*/
1415 static int
1416 zone_lwps_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv)
1417 {
1418 ASSERT(MUTEX_HELD(&p->p_lock));
1419 ASSERT(e->rcep_t == RCENTITY_ZONE);
1420 if (e->rcep_p.zone == NULL)
1421 return (0);
1422 e->rcep_p.zone->zone_nlwps_ctl = nv;
1423 return (0);
1424 }
1425
1426 static rctl_ops_t zone_lwps_ops = {
1427 rcop_no_action,
1428 zone_lwps_usage,
1429 zone_lwps_set,
1430 zone_lwps_test,
1431 };
1432
1433 /*ARGSUSED*/
1434 static rctl_qty_t
1435 zone_procs_usage(rctl_t *r, proc_t *p)
1436 {
1437 rctl_qty_t nprocs;
1438 zone_t *zone = p->p_zone;
1439
1440 ASSERT(MUTEX_HELD(&p->p_lock));
1441
1442 mutex_enter(&zone->zone_nlwps_lock);
1443 nprocs = zone->zone_nprocs;
1444 mutex_exit(&zone->zone_nlwps_lock);
1445
1446 return (nprocs);
1447 }
1448
1449 /*ARGSUSED*/
1450 static int
1451 zone_procs_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
1452 rctl_qty_t incr, uint_t flags)
1453 {
1454 rctl_qty_t nprocs;
1455
1456 ASSERT(MUTEX_HELD(&p->p_lock));
1457 ASSERT(e->rcep_t == RCENTITY_ZONE);
1458 if (e->rcep_p.zone == NULL)
1459 return (0);
1460 ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
1461 nprocs = e->rcep_p.zone->zone_nprocs;
1462
1463 if (nprocs + incr > rcntl->rcv_value)
1464 return (1);
1465
1466 return (0);
1467 }
1468
1469 /*ARGSUSED*/
1470 static int
1471 zone_procs_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv)
1472 {
1473 ASSERT(MUTEX_HELD(&p->p_lock));
1474 ASSERT(e->rcep_t == RCENTITY_ZONE);
1475 if (e->rcep_p.zone == NULL)
1476 return (0);
1477 e->rcep_p.zone->zone_nprocs_ctl = nv;
1478 return (0);
1479 }
1480
1481 static rctl_ops_t zone_procs_ops = {
1482 rcop_no_action,
1483 zone_procs_usage,
1484 zone_procs_set,
1485 zone_procs_test,
1486 };
1487
1488 /*ARGSUSED*/
1489 static int
1490 zone_shmmax_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1491 rctl_qty_t incr, uint_t flags)
1492 {
1493 rctl_qty_t v;
1494 ASSERT(MUTEX_HELD(&p->p_lock));
1495 ASSERT(e->rcep_t == RCENTITY_ZONE);
1496 v = e->rcep_p.zone->zone_shmmax + incr;
1497 if (v > rval->rcv_value)
1498 return (1);
1499 return (0);
1500 }
1501
1502 static rctl_ops_t zone_shmmax_ops = {
1503 rcop_no_action,
1504 rcop_no_usage,
1505 rcop_no_set,
1506 zone_shmmax_test
1507 };
1508
1509 /*ARGSUSED*/
1510 static int
1511 zone_shmmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1512 rctl_qty_t incr, uint_t flags)
1513 {
1514 rctl_qty_t v;
1515 ASSERT(MUTEX_HELD(&p->p_lock));
1516 ASSERT(e->rcep_t == RCENTITY_ZONE);
1517 v = e->rcep_p.zone->zone_ipc.ipcq_shmmni + incr;
1518 if (v > rval->rcv_value)
1519 return (1);
1520 return (0);
1521 }
1522
1523 static rctl_ops_t zone_shmmni_ops = {
1524 rcop_no_action,
1525 rcop_no_usage,
1526 rcop_no_set,
1527 zone_shmmni_test
1528 };
1529
1530 /*ARGSUSED*/
1531 static int
1532 zone_semmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1533 rctl_qty_t incr, uint_t flags)
1534 {
1535 rctl_qty_t v;
1536 ASSERT(MUTEX_HELD(&p->p_lock));
1537 ASSERT(e->rcep_t == RCENTITY_ZONE);
1538 v = e->rcep_p.zone->zone_ipc.ipcq_semmni + incr;
1539 if (v > rval->rcv_value)
1540 return (1);
1541 return (0);
1542 }
1543
1544 static rctl_ops_t zone_semmni_ops = {
1545 rcop_no_action,
1546 rcop_no_usage,
1547 rcop_no_set,
1548 zone_semmni_test
1549 };
1550
1551 /*ARGSUSED*/
1552 static int
1553 zone_msgmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1554 rctl_qty_t incr, uint_t flags)
1555 {
1556 rctl_qty_t v;
1557 ASSERT(MUTEX_HELD(&p->p_lock));
1558 ASSERT(e->rcep_t == RCENTITY_ZONE);
1559 v = e->rcep_p.zone->zone_ipc.ipcq_msgmni + incr;
1560 if (v > rval->rcv_value)
1561 return (1);
1562 return (0);
1563 }
1564
1565 static rctl_ops_t zone_msgmni_ops = {
1566 rcop_no_action,
1567 rcop_no_usage,
1568 rcop_no_set,
1569 zone_msgmni_test
1570 };
1571
1572 /*ARGSUSED*/
1573 static rctl_qty_t
1574 zone_locked_mem_usage(rctl_t *rctl, struct proc *p)
1575 {
1576 rctl_qty_t q;
1577 ASSERT(MUTEX_HELD(&p->p_lock));
1578 mutex_enter(&p->p_zone->zone_mem_lock);
1579 q = p->p_zone->zone_locked_mem;
1580 mutex_exit(&p->p_zone->zone_mem_lock);
1581 return (q);
1582 }
1583
1584 /*ARGSUSED*/
1585 static int
1586 zone_locked_mem_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1587 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1588 {
1589 rctl_qty_t q;
1590 zone_t *z;
1591
1592 z = e->rcep_p.zone;
1593 ASSERT(MUTEX_HELD(&p->p_lock));
1594 ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1595 q = z->zone_locked_mem;
1596 if (q + incr > rcntl->rcv_value)
1597 return (1);
1598 return (0);
1599 }
1600
1601 /*ARGSUSED*/
1602 static int
1603 zone_locked_mem_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1604 rctl_qty_t nv)
1605 {
1606 ASSERT(MUTEX_HELD(&p->p_lock));
1607 ASSERT(e->rcep_t == RCENTITY_ZONE);
1608 if (e->rcep_p.zone == NULL)
1609 return (0);
1610 e->rcep_p.zone->zone_locked_mem_ctl = nv;
1611 return (0);
1612 }
1613
1614 static rctl_ops_t zone_locked_mem_ops = {
1615 rcop_no_action,
1616 zone_locked_mem_usage,
1617 zone_locked_mem_set,
1618 zone_locked_mem_test
1619 };
1620
1621 /*ARGSUSED*/
1622 static rctl_qty_t
1623 zone_max_swap_usage(rctl_t *rctl, struct proc *p)
1624 {
1625 rctl_qty_t q;
1626 zone_t *z = p->p_zone;
1627
1628 ASSERT(MUTEX_HELD(&p->p_lock));
1629 mutex_enter(&z->zone_mem_lock);
1630 q = z->zone_max_swap;
1631 mutex_exit(&z->zone_mem_lock);
1632 return (q);
1633 }
1634
1635 /*ARGSUSED*/
1636 static int
1637 zone_max_swap_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1638 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1639 {
1640 rctl_qty_t q;
1641 zone_t *z;
1642
1643 z = e->rcep_p.zone;
1644 ASSERT(MUTEX_HELD(&p->p_lock));
1645 ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1646 q = z->zone_max_swap;
1647 if (q + incr > rcntl->rcv_value)
1648 return (1);
1649 return (0);
1650 }
1651
1652 /*ARGSUSED*/
1653 static int
1654 zone_max_swap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1655 rctl_qty_t nv)
1656 {
1657 ASSERT(MUTEX_HELD(&p->p_lock));
1658 ASSERT(e->rcep_t == RCENTITY_ZONE);
1659 if (e->rcep_p.zone == NULL)
1660 return (0);
1661 e->rcep_p.zone->zone_max_swap_ctl = nv;
1662 return (0);
1663 }
1664
1665 static rctl_ops_t zone_max_swap_ops = {
1666 rcop_no_action,
1667 zone_max_swap_usage,
1668 zone_max_swap_set,
1669 zone_max_swap_test
1670 };
1671
1672 /*ARGSUSED*/
1673 static rctl_qty_t
1674 zone_max_lofi_usage(rctl_t *rctl, struct proc *p)
1675 {
1676 rctl_qty_t q;
1677 zone_t *z = p->p_zone;
1678
1679 ASSERT(MUTEX_HELD(&p->p_lock));
1680 mutex_enter(&z->zone_rctl_lock);
1681 q = z->zone_max_lofi;
1682 mutex_exit(&z->zone_rctl_lock);
1683 return (q);
1684 }
1685
1686 /*ARGSUSED*/
1687 static int
1688 zone_max_lofi_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1689 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1690 {
1691 rctl_qty_t q;
1692 zone_t *z;
1693
1694 z = e->rcep_p.zone;
1695 ASSERT(MUTEX_HELD(&p->p_lock));
1696 ASSERT(MUTEX_HELD(&z->zone_rctl_lock));
1697 q = z->zone_max_lofi;
1698 if (q + incr > rcntl->rcv_value)
1699 return (1);
1700 return (0);
1701 }
1702
1703 /*ARGSUSED*/
1704 static int
1705 zone_max_lofi_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1706 rctl_qty_t nv)
1707 {
1708 ASSERT(MUTEX_HELD(&p->p_lock));
1709 ASSERT(e->rcep_t == RCENTITY_ZONE);
1710 if (e->rcep_p.zone == NULL)
1711 return (0);
1712 e->rcep_p.zone->zone_max_lofi_ctl = nv;
1713 return (0);
1714 }
1715
1716 static rctl_ops_t zone_max_lofi_ops = {
1717 rcop_no_action,
1718 zone_max_lofi_usage,
1719 zone_max_lofi_set,
1720 zone_max_lofi_test
1721 };
1722
1723 /*
1724 * Helper function to brand the zone with a unique ID.
1725 */
1726 static void
1727 zone_uniqid(zone_t *zone)
1728 {
1729 static uint64_t uniqid = 0;
1730
1731 ASSERT(MUTEX_HELD(&zonehash_lock));
1732 zone->zone_uniqid = uniqid++;
1733 }
1734
1735 /*
1736 * Returns a held pointer to the "kcred" for the specified zone.
1737 */
1738 struct cred *
1739 zone_get_kcred(zoneid_t zoneid)
1740 {
1741 zone_t *zone;
1742 cred_t *cr;
1743
1744 if ((zone = zone_find_by_id(zoneid)) == NULL)
1745 return (NULL);
1746 cr = zone->zone_kcred;
1747 crhold(cr);
1748 zone_rele(zone);
1749 return (cr);
1750 }
1751
1752 static int
1753 zone_lockedmem_kstat_update(kstat_t *ksp, int rw)
1754 {
1755 zone_t *zone = ksp->ks_private;
1756 zone_kstat_t *zk = ksp->ks_data;
1757
1758 if (rw == KSTAT_WRITE)
1759 return (EACCES);
1760
1761 zk->zk_usage.value.ui64 = zone->zone_locked_mem;
1762 zk->zk_value.value.ui64 = zone->zone_locked_mem_ctl;
1763 return (0);
1764 }
1765
1766 static int
1767 zone_nprocs_kstat_update(kstat_t *ksp, int rw)
1768 {
1769 zone_t *zone = ksp->ks_private;
1770 zone_kstat_t *zk = ksp->ks_data;
1771
1772 if (rw == KSTAT_WRITE)
1773 return (EACCES);
1774
1775 zk->zk_usage.value.ui64 = zone->zone_nprocs;
1776 zk->zk_value.value.ui64 = zone->zone_nprocs_ctl;
1777 return (0);
1778 }
1779
1780 static int
1781 zone_swapresv_kstat_update(kstat_t *ksp, int rw)
1782 {
1783 zone_t *zone = ksp->ks_private;
1784 zone_kstat_t *zk = ksp->ks_data;
1785
1786 if (rw == KSTAT_WRITE)
1787 return (EACCES);
1788
1789 zk->zk_usage.value.ui64 = zone->zone_max_swap;
1790 zk->zk_value.value.ui64 = zone->zone_max_swap_ctl;
1791 return (0);
1792 }
1793
1794 static kstat_t *
1795 zone_kstat_create_common(zone_t *zone, char *name,
1796 int (*updatefunc) (kstat_t *, int))
1797 {
1798 kstat_t *ksp;
1799 zone_kstat_t *zk;
1800
1801 ksp = rctl_kstat_create_zone(zone, name, KSTAT_TYPE_NAMED,
1802 sizeof (zone_kstat_t) / sizeof (kstat_named_t),
1803 KSTAT_FLAG_VIRTUAL);
1804
1805 if (ksp == NULL)
1806 return (NULL);
1807
1808 zk = ksp->ks_data = kmem_alloc(sizeof (zone_kstat_t), KM_SLEEP);
1809 ksp->ks_data_size += strlen(zone->zone_name) + 1;
1810 kstat_named_init(&zk->zk_zonename, "zonename", KSTAT_DATA_STRING);
1811 kstat_named_setstr(&zk->zk_zonename, zone->zone_name);
1812 kstat_named_init(&zk->zk_usage, "usage", KSTAT_DATA_UINT64);
1813 kstat_named_init(&zk->zk_value, "value", KSTAT_DATA_UINT64);
1814 ksp->ks_update = updatefunc;
1815 ksp->ks_private = zone;
1816 kstat_install(ksp);
1817 return (ksp);
1818 }
1819
1820 static int
1821 zone_misc_kstat_update(kstat_t *ksp, int rw)
1822 {
1823 zone_t *zone = ksp->ks_private;
1824 zone_misc_kstat_t *zmp = ksp->ks_data;
1825 hrtime_t tmp;
1826
1827 if (rw == KSTAT_WRITE)
1828 return (EACCES);
1829
1830 tmp = zone->zone_utime;
1831 scalehrtime(&tmp);
1832 zmp->zm_utime.value.ui64 = tmp;
1833 tmp = zone->zone_stime;
1834 scalehrtime(&tmp);
1835 zmp->zm_stime.value.ui64 = tmp;
1836 tmp = zone->zone_wtime;
1837 scalehrtime(&tmp);
1838 zmp->zm_wtime.value.ui64 = tmp;
1839
1840 zmp->zm_avenrun1.value.ui32 = zone->zone_avenrun[0];
1841 zmp->zm_avenrun5.value.ui32 = zone->zone_avenrun[1];
1842 zmp->zm_avenrun15.value.ui32 = zone->zone_avenrun[2];
1843
1844 zmp->zm_ffcap.value.ui32 = zone->zone_ffcap;
1845 zmp->zm_ffnoproc.value.ui32 = zone->zone_ffnoproc;
1846 zmp->zm_ffnomem.value.ui32 = zone->zone_ffnomem;
1847 zmp->zm_ffmisc.value.ui32 = zone->zone_ffmisc;
1848
1849 return (0);
1850 }
1851
1852 static kstat_t *
1853 zone_misc_kstat_create(zone_t *zone)
1854 {
1855 kstat_t *ksp;
1856 zone_misc_kstat_t *zmp;
1857
1858 if ((ksp = kstat_create_zone("zones", zone->zone_id,
1859 zone->zone_name, "zone_misc", KSTAT_TYPE_NAMED,
1860 sizeof (zone_misc_kstat_t) / sizeof (kstat_named_t),
1861 KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
1862 return (NULL);
1863
1864 if (zone->zone_id != GLOBAL_ZONEID)
1865 kstat_zone_add(ksp, GLOBAL_ZONEID);
1866
1867 zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_misc_kstat_t), KM_SLEEP);
1868 ksp->ks_data_size += strlen(zone->zone_name) + 1;
1869 ksp->ks_lock = &zone->zone_misc_lock;
1870 zone->zone_misc_stats = zmp;
1871
1872 /* The kstat "name" field is not large enough for a full zonename */
1873 kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
1874 kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
1875 kstat_named_init(&zmp->zm_utime, "nsec_user", KSTAT_DATA_UINT64);
1876 kstat_named_init(&zmp->zm_stime, "nsec_sys", KSTAT_DATA_UINT64);
1877 kstat_named_init(&zmp->zm_wtime, "nsec_waitrq", KSTAT_DATA_UINT64);
1878 kstat_named_init(&zmp->zm_avenrun1, "avenrun_1min", KSTAT_DATA_UINT32);
1879 kstat_named_init(&zmp->zm_avenrun5, "avenrun_5min", KSTAT_DATA_UINT32);
1880 kstat_named_init(&zmp->zm_avenrun15, "avenrun_15min",
1881 KSTAT_DATA_UINT32);
1882 kstat_named_init(&zmp->zm_ffcap, "forkfail_cap", KSTAT_DATA_UINT32);
1883 kstat_named_init(&zmp->zm_ffnoproc, "forkfail_noproc",
1884 KSTAT_DATA_UINT32);
1885 kstat_named_init(&zmp->zm_ffnomem, "forkfail_nomem", KSTAT_DATA_UINT32);
1886 kstat_named_init(&zmp->zm_ffmisc, "forkfail_misc", KSTAT_DATA_UINT32);
1887
1888
1889 ksp->ks_update = zone_misc_kstat_update;
1890 ksp->ks_private = zone;
1891
1892 kstat_install(ksp);
1893 return (ksp);
1894 }
1895
1896 static void
1897 zone_kstat_create(zone_t *zone)
1898 {
1899 zone->zone_lockedmem_kstat = zone_kstat_create_common(zone,
1900 "lockedmem", zone_lockedmem_kstat_update);
1901 zone->zone_swapresv_kstat = zone_kstat_create_common(zone,
1902 "swapresv", zone_swapresv_kstat_update);
1903 zone->zone_nprocs_kstat = zone_kstat_create_common(zone,
1904 "nprocs", zone_nprocs_kstat_update);
1905
1906 if ((zone->zone_misc_ksp = zone_misc_kstat_create(zone)) == NULL) {
1907 zone->zone_misc_stats = kmem_zalloc(
1908 sizeof (zone_misc_kstat_t), KM_SLEEP);
1909 }
1910 }
1911
1912 static void
1913 zone_kstat_delete_common(kstat_t **pkstat, size_t datasz)
1914 {
1915 void *data;
1916
1917 if (*pkstat != NULL) {
1918 data = (*pkstat)->ks_data;
1919 kstat_delete(*pkstat);
1920 kmem_free(data, datasz);
1921 *pkstat = NULL;
1922 }
1923 }
1924
1925 static void
1926 zone_kstat_delete(zone_t *zone)
1927 {
1928 zone_kstat_delete_common(&zone->zone_lockedmem_kstat,
1929 sizeof (zone_kstat_t));
1930 zone_kstat_delete_common(&zone->zone_swapresv_kstat,
1931 sizeof (zone_kstat_t));
1932 zone_kstat_delete_common(&zone->zone_nprocs_kstat,
1933 sizeof (zone_kstat_t));
1934 zone_kstat_delete_common(&zone->zone_misc_ksp,
1935 sizeof (zone_misc_kstat_t));
1936 }
1937
1938 /*
1939 * Called very early on in boot to initialize the ZSD list so that
1940 * zone_key_create() can be called before zone_init(). It also initializes
1941 * portions of zone0 which may be used before zone_init() is called. The
1942 * variable "global_zone" will be set when zone0 is fully initialized by
1943 * zone_init().
1944 */
1945 void
1946 zone_zsd_init(void)
1947 {
1948 mutex_init(&zonehash_lock, NULL, MUTEX_DEFAULT, NULL);
1949 mutex_init(&zsd_key_lock, NULL, MUTEX_DEFAULT, NULL);
1950 list_create(&zsd_registered_keys, sizeof (struct zsd_entry),
1951 offsetof(struct zsd_entry, zsd_linkage));
1952 list_create(&zone_active, sizeof (zone_t),
1953 offsetof(zone_t, zone_linkage));
1954 list_create(&zone_deathrow, sizeof (zone_t),
1955 offsetof(zone_t, zone_linkage));
1956
1957 mutex_init(&zone0.zone_lock, NULL, MUTEX_DEFAULT, NULL);
1958 mutex_init(&zone0.zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
1959 mutex_init(&zone0.zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
1960 zone0.zone_shares = 1;
1961 zone0.zone_nlwps = 0;
1962 zone0.zone_nlwps_ctl = INT_MAX;
1963 zone0.zone_nprocs = 0;
1964 zone0.zone_nprocs_ctl = INT_MAX;
1965 zone0.zone_locked_mem = 0;
1966 zone0.zone_locked_mem_ctl = UINT64_MAX;
1967 ASSERT(zone0.zone_max_swap == 0);
1968 zone0.zone_max_swap_ctl = UINT64_MAX;
1969 zone0.zone_max_lofi = 0;
1970 zone0.zone_max_lofi_ctl = UINT64_MAX;
1971 zone0.zone_shmmax = 0;
1972 zone0.zone_ipc.ipcq_shmmni = 0;
1973 zone0.zone_ipc.ipcq_semmni = 0;
1974 zone0.zone_ipc.ipcq_msgmni = 0;
1975 zone0.zone_name = GLOBAL_ZONENAME;
1976 zone0.zone_nodename = utsname.nodename;
1977 zone0.zone_domain = srpc_domain;
1978 zone0.zone_hostid = HW_INVALID_HOSTID;
1979 zone0.zone_fs_allowed = NULL;
1980 zone0.zone_ref = 1;
1981 zone0.zone_id = GLOBAL_ZONEID;
1982 zone0.zone_status = ZONE_IS_RUNNING;
1983 zone0.zone_rootpath = "/";
1984 zone0.zone_rootpathlen = 2;
1985 zone0.zone_psetid = ZONE_PS_INVAL;
1986 zone0.zone_ncpus = 0;
1987 zone0.zone_ncpus_online = 0;
1988 zone0.zone_proc_initpid = 1;
1989 zone0.zone_initname = initname;
1990 zone0.zone_lockedmem_kstat = NULL;
1991 zone0.zone_swapresv_kstat = NULL;
1992 zone0.zone_nprocs_kstat = NULL;
1993
1994 zone0.zone_stime = 0;
1995 zone0.zone_utime = 0;
1996 zone0.zone_wtime = 0;
1997
1998 list_create(&zone0.zone_ref_list, sizeof (zone_ref_t),
1999 offsetof(zone_ref_t, zref_linkage));
2000 list_create(&zone0.zone_zsd, sizeof (struct zsd_entry),
2001 offsetof(struct zsd_entry, zsd_linkage));
2002 list_insert_head(&zone_active, &zone0);
2003
2004 /*
2005 * The root filesystem is not mounted yet, so zone_rootvp cannot be set
2006 * to anything meaningful. It is assigned to be 'rootdir' in
2007 * vfs_mountroot().
2008 */
2009 zone0.zone_rootvp = NULL;
2010 zone0.zone_vfslist = NULL;
2011 zone0.zone_bootargs = initargs;
2012 zone0.zone_privset = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
2013 /*
2014 * The global zone has all privileges
2015 */
2016 priv_fillset(zone0.zone_privset);
2017 /*
2018 * Add p0 to the global zone
2019 */
2020 zone0.zone_zsched = &p0;
2021 p0.p_zone = &zone0;
2022 }
2023
2024 /*
2025 * Compute a hash value based on the contents of the label and the DOI. The
2026 * hash algorithm is somewhat arbitrary, but is based on the observation that
2027 * humans will likely pick labels that differ by amounts that work out to be
2028 * multiples of the number of hash chains, and thus stirring in some primes
2029 * should help.
2030 */
2031 static uint_t
2032 hash_bylabel(void *hdata, mod_hash_key_t key)
2033 {
2034 const ts_label_t *lab = (ts_label_t *)key;
2035 const uint32_t *up, *ue;
2036 uint_t hash;
2037 int i;
2038
2039 _NOTE(ARGUNUSED(hdata));
2040
2041 hash = lab->tsl_doi + (lab->tsl_doi << 1);
2042 /* we depend on alignment of label, but not representation */
2043 up = (const uint32_t *)&lab->tsl_label;
2044 ue = up + sizeof (lab->tsl_label) / sizeof (*up);
2045 i = 1;
2046 while (up < ue) {
2047 /* using 2^n + 1, 1 <= n <= 16 as source of many primes */
2048 hash += *up + (*up << ((i % 16) + 1));
2049 up++;
2050 i++;
2051 }
2052 return (hash);
2053 }
2054
2055 /*
2056 * All that mod_hash cares about here is zero (equal) versus non-zero (not
2057 * equal). This may need to be changed if less than / greater than is ever
2058 * needed.
2059 */
2060 static int
2061 hash_labelkey_cmp(mod_hash_key_t key1, mod_hash_key_t key2)
2062 {
2063 ts_label_t *lab1 = (ts_label_t *)key1;
2064 ts_label_t *lab2 = (ts_label_t *)key2;
2065
2066 return (label_equal(lab1, lab2) ? 0 : 1);
2067 }
2068
2069 /*
2070 * Called by main() to initialize the zones framework.
2071 */
2072 void
2073 zone_init(void)
2074 {
2075 rctl_dict_entry_t *rde;
2076 rctl_val_t *dval;
2077 rctl_set_t *set;
2078 rctl_alloc_gp_t *gp;
2079 rctl_entity_p_t e;
2080 int res;
2081
2082 ASSERT(curproc == &p0);
2083
2084 /*
2085 * Create ID space for zone IDs. ID 0 is reserved for the
2086 * global zone.
2087 */
2088 zoneid_space = id_space_create("zoneid_space", 1, MAX_ZONEID);
2089
2090 /*
2091 * Initialize generic zone resource controls, if any.
2092 */
2093 rc_zone_cpu_shares = rctl_register("zone.cpu-shares",
2094 RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_NEVER |
2095 RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT | RCTL_GLOBAL_SYSLOG_NEVER,
2096 FSS_MAXSHARES, FSS_MAXSHARES, &zone_cpu_shares_ops);
2097
2098 rc_zone_cpu_cap = rctl_register("zone.cpu-cap",
2099 RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_ALWAYS |
2100 RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |RCTL_GLOBAL_SYSLOG_NEVER |
2101 RCTL_GLOBAL_INFINITE,
2102 MAXCAP, MAXCAP, &zone_cpu_cap_ops);
2103
2104 rc_zone_nlwps = rctl_register("zone.max-lwps", RCENTITY_ZONE,
2105 RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
2106 INT_MAX, INT_MAX, &zone_lwps_ops);
2107
2108 rc_zone_nprocs = rctl_register("zone.max-processes", RCENTITY_ZONE,
2109 RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
2110 INT_MAX, INT_MAX, &zone_procs_ops);
2111
2112 /*
2113 * System V IPC resource controls
2114 */
2115 rc_zone_msgmni = rctl_register("zone.max-msg-ids",
2116 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2117 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_msgmni_ops);
2118
2119 rc_zone_semmni = rctl_register("zone.max-sem-ids",
2120 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2121 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_semmni_ops);
2122
2123 rc_zone_shmmni = rctl_register("zone.max-shm-ids",
2124 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2125 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_shmmni_ops);
2126
2127 rc_zone_shmmax = rctl_register("zone.max-shm-memory",
2128 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2129 RCTL_GLOBAL_BYTES, UINT64_MAX, UINT64_MAX, &zone_shmmax_ops);
2130
2131 /*
2132 * Create a rctl_val with PRIVILEGED, NOACTION, value = 1. Then attach
2133 * this at the head of the rctl_dict_entry for ``zone.cpu-shares''.
2134 */
2135 dval = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
2136 bzero(dval, sizeof (rctl_val_t));
2137 dval->rcv_value = 1;
2138 dval->rcv_privilege = RCPRIV_PRIVILEGED;
2139 dval->rcv_flagaction = RCTL_LOCAL_NOACTION;
2140 dval->rcv_action_recip_pid = -1;
2141
2142 rde = rctl_dict_lookup("zone.cpu-shares");
2143 (void) rctl_val_list_insert(&rde->rcd_default_value, dval);
2144
2145 rc_zone_locked_mem = rctl_register("zone.max-locked-memory",
2146 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
2147 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2148 &zone_locked_mem_ops);
2149
2150 rc_zone_max_swap = rctl_register("zone.max-swap",
2151 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
2152 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2153 &zone_max_swap_ops);
2154
2155 rc_zone_max_lofi = rctl_register("zone.max-lofi",
2156 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |
2157 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2158 &zone_max_lofi_ops);
2159
2160 /*
2161 * Initialize the ``global zone''.
2162 */
2163 set = rctl_set_create();
2164 gp = rctl_set_init_prealloc(RCENTITY_ZONE);
2165 mutex_enter(&p0.p_lock);
2166 e.rcep_p.zone = &zone0;
2167 e.rcep_t = RCENTITY_ZONE;
2168 zone0.zone_rctls = rctl_set_init(RCENTITY_ZONE, &p0, &e, set,
2169 gp);
2170
2171 zone0.zone_nlwps = p0.p_lwpcnt;
2172 zone0.zone_nprocs = 1;
2173 zone0.zone_ntasks = 1;
2174 mutex_exit(&p0.p_lock);
2175 zone0.zone_restart_init = B_TRUE;
2176 zone0.zone_brand = &native_brand;
2177 rctl_prealloc_destroy(gp);
2178 /*
2179 * pool_default hasn't been initialized yet, so we let pool_init()
2180 * take care of making sure the global zone is in the default pool.
2181 */
2182
2183 /*
2184 * Initialize global zone kstats
2185 */
2186 zone_kstat_create(&zone0);
2187
2188 /*
2189 * Initialize zone label.
2190 * mlp are initialized when tnzonecfg is loaded.
2191 */
2192 zone0.zone_slabel = l_admin_low;
2193 rw_init(&zone0.zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
2194 label_hold(l_admin_low);
2195
2196 /*
2197 * Initialise the lock for the database structure used by mntfs.
2198 */
2199 rw_init(&zone0.zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL);
2200
2201 mutex_enter(&zonehash_lock);
2202 zone_uniqid(&zone0);
2203 ASSERT(zone0.zone_uniqid == GLOBAL_ZONEUNIQID);
2204
2205 zonehashbyid = mod_hash_create_idhash("zone_by_id", zone_hash_size,
2206 mod_hash_null_valdtor);
2207 zonehashbyname = mod_hash_create_strhash("zone_by_name",
2208 zone_hash_size, mod_hash_null_valdtor);
2209 /*
2210 * maintain zonehashbylabel only for labeled systems
2211 */
2212 if (is_system_labeled())
2213 zonehashbylabel = mod_hash_create_extended("zone_by_label",
2214 zone_hash_size, mod_hash_null_keydtor,
2215 mod_hash_null_valdtor, hash_bylabel, NULL,
2216 hash_labelkey_cmp, KM_SLEEP);
2217 zonecount = 1;
2218
2219 (void) mod_hash_insert(zonehashbyid, (mod_hash_key_t)GLOBAL_ZONEID,
2220 (mod_hash_val_t)&zone0);
2221 (void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)zone0.zone_name,
2222 (mod_hash_val_t)&zone0);
2223 if (is_system_labeled()) {
2224 zone0.zone_flags |= ZF_HASHED_LABEL;
2225 (void) mod_hash_insert(zonehashbylabel,
2226 (mod_hash_key_t)zone0.zone_slabel, (mod_hash_val_t)&zone0);
2227 }
2228 mutex_exit(&zonehash_lock);
2229
2230 /*
2231 * We avoid setting zone_kcred until now, since kcred is initialized
2232 * sometime after zone_zsd_init() and before zone_init().
2233 */
2234 zone0.zone_kcred = kcred;
2235 /*
2236 * The global zone is fully initialized (except for zone_rootvp which
2237 * will be set when the root filesystem is mounted).
2238 */
2239 global_zone = &zone0;
2240
2241 /*
2242 * Setup an event channel to send zone status change notifications on
2243 */
2244 res = sysevent_evc_bind(ZONE_EVENT_CHANNEL, &zone_event_chan,
2245 EVCH_CREAT);
2246
2247 if (res)
2248 panic("Sysevent_evc_bind failed during zone setup.\n");
2249
2250 }
2251
2252 static void
2253 zone_free(zone_t *zone)
2254 {
2255 zone_dl_t *zdl;
2256
2257 ASSERT(zone != global_zone);
2258 ASSERT(zone->zone_ntasks == 0);
2259 ASSERT(zone->zone_nlwps == 0);
2260 ASSERT(zone->zone_nprocs == 0);
2261 ASSERT(zone->zone_cred_ref == 0);
2262 ASSERT(zone->zone_kcred == NULL);
2263 ASSERT(zone_status_get(zone) == ZONE_IS_DEAD ||
2264 zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
2265 ASSERT(list_is_empty(&zone->zone_ref_list));
2266
2267 /*
2268 * Remove any zone caps.
2269 */
2270 cpucaps_zone_remove(zone);
2271
2272 ASSERT(zone->zone_cpucap == NULL);
2273
2274 /* remove from deathrow list */
2275 if (zone_status_get(zone) == ZONE_IS_DEAD) {
2276 ASSERT(zone->zone_ref == 0);
2277 mutex_enter(&zone_deathrow_lock);
2278 list_remove(&zone_deathrow, zone);
2279 mutex_exit(&zone_deathrow_lock);
2280 }
2281
2282 list_destroy(&zone->zone_ref_list);
2283 zone_free_zsd(zone);
2284 zone_free_datasets(zone);
2285
2286 /*
2287 * While dlmgmtd should have removed all of these, it could have left
2288 * something behind or crashed. In which case it's not safe for us to
2289 * assume that the list is empty which list_destroy() will ASSERT. We
2290 * clean up for our userland comrades which may have crashed, or worse,
2291 * been disabled by SMF.
2292 */
2293 while ((zdl = list_remove_head(&zone->zone_dl_list)) != NULL) {
2294 if (zdl->zdl_net != NULL)
2295 nvlist_free(zdl->zdl_net);
2296 kmem_free(zdl, sizeof (zone_dl_t));
2297 }
2298 list_destroy(&zone->zone_dl_list);
2299
2300 if (zone->zone_rootvp != NULL)
2301 VN_RELE(zone->zone_rootvp);
2302 if (zone->zone_rootpath)
2303 kmem_free(zone->zone_rootpath, zone->zone_rootpathlen);
2304 if (zone->zone_name != NULL)
2305 kmem_free(zone->zone_name, ZONENAME_MAX);
2306 if (zone->zone_slabel != NULL)
2307 label_rele(zone->zone_slabel);
2308 if (zone->zone_nodename != NULL)
2309 kmem_free(zone->zone_nodename, _SYS_NMLN);
2310 if (zone->zone_domain != NULL)
2311 kmem_free(zone->zone_domain, _SYS_NMLN);
2312 if (zone->zone_privset != NULL)
2313 kmem_free(zone->zone_privset, sizeof (priv_set_t));
2314 if (zone->zone_rctls != NULL)
2315 rctl_set_free(zone->zone_rctls);
2316 if (zone->zone_bootargs != NULL)
2317 strfree(zone->zone_bootargs);
2318 if (zone->zone_initname != NULL)
2319 strfree(zone->zone_initname);
2320 if (zone->zone_fs_allowed != NULL)
2321 strfree(zone->zone_fs_allowed);
2322 if (zone->zone_pfexecd != NULL)
2323 klpd_freelist(&zone->zone_pfexecd);
2324 id_free(zoneid_space, zone->zone_id);
2325 mutex_destroy(&zone->zone_lock);
2326 cv_destroy(&zone->zone_cv);
2327 rw_destroy(&zone->zone_mlps.mlpl_rwlock);
2328 rw_destroy(&zone->zone_mntfs_db_lock);
2329 kmem_free(zone, sizeof (zone_t));
2330 }
2331
2332 /*
2333 * See block comment at the top of this file for information about zone
2334 * status values.
2335 */
2336 /*
2337 * Convenience function for setting zone status.
2338 */
2339 static void
2340 zone_status_set(zone_t *zone, zone_status_t status)
2341 {
2342
2343 nvlist_t *nvl = NULL;
2344 ASSERT(MUTEX_HELD(&zone_status_lock));
2345 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE &&
2346 status >= zone_status_get(zone));
2347
2348 if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) ||
2349 nvlist_add_string(nvl, ZONE_CB_NAME, zone->zone_name) ||
2350 nvlist_add_string(nvl, ZONE_CB_NEWSTATE,
2351 zone_status_table[status]) ||
2352 nvlist_add_string(nvl, ZONE_CB_OLDSTATE,
2353 zone_status_table[zone->zone_status]) ||
2354 nvlist_add_int32(nvl, ZONE_CB_ZONEID, zone->zone_id) ||
2355 nvlist_add_uint64(nvl, ZONE_CB_TIMESTAMP, (uint64_t)gethrtime()) ||
2356 sysevent_evc_publish(zone_event_chan, ZONE_EVENT_STATUS_CLASS,
2357 ZONE_EVENT_STATUS_SUBCLASS, "sun.com", "kernel", nvl, EVCH_SLEEP)) {
2358 #ifdef DEBUG
2359 (void) printf(
2360 "Failed to allocate and send zone state change event.\n");
2361 #endif
2362 }
2363 nvlist_free(nvl);
2364
2365 zone->zone_status = status;
2366
2367 cv_broadcast(&zone->zone_cv);
2368 }
2369
2370 /*
2371 * Public function to retrieve the zone status. The zone status may
2372 * change after it is retrieved.
2373 */
2374 zone_status_t
2375 zone_status_get(zone_t *zone)
2376 {
2377 return (zone->zone_status);
2378 }
2379
2380 static int
2381 zone_set_bootargs(zone_t *zone, const char *zone_bootargs)
2382 {
2383 char *buf = kmem_zalloc(BOOTARGS_MAX, KM_SLEEP);
2384 int err = 0;
2385
2386 ASSERT(zone != global_zone);
2387 if ((err = copyinstr(zone_bootargs, buf, BOOTARGS_MAX, NULL)) != 0)
2388 goto done; /* EFAULT or ENAMETOOLONG */
2389
2390 if (zone->zone_bootargs != NULL)
2391 strfree(zone->zone_bootargs);
2392
2393 zone->zone_bootargs = strdup(buf);
2394
2395 done:
2396 kmem_free(buf, BOOTARGS_MAX);
2397 return (err);
2398 }
2399
2400 static int
2401 zone_set_brand(zone_t *zone, const char *brand)
2402 {
2403 struct brand_attr *attrp;
2404 brand_t *bp;
2405
2406 attrp = kmem_alloc(sizeof (struct brand_attr), KM_SLEEP);
2407 if (copyin(brand, attrp, sizeof (struct brand_attr)) != 0) {
2408 kmem_free(attrp, sizeof (struct brand_attr));
2409 return (EFAULT);
2410 }
2411
2412 bp = brand_register_zone(attrp);
2413 kmem_free(attrp, sizeof (struct brand_attr));
2414 if (bp == NULL)
2415 return (EINVAL);
2416
2417 /*
2418 * This is the only place where a zone can change it's brand.
2419 * We already need to hold zone_status_lock to check the zone
2420 * status, so we'll just use that lock to serialize zone
2421 * branding requests as well.
2422 */
2423 mutex_enter(&zone_status_lock);
2424
2425 /* Re-Branding is not allowed and the zone can't be booted yet */
2426 if ((ZONE_IS_BRANDED(zone)) ||
2427 (zone_status_get(zone) >= ZONE_IS_BOOTING)) {
2428 mutex_exit(&zone_status_lock);
2429 brand_unregister_zone(bp);
2430 return (EINVAL);
2431 }
2432
2433 /* set up the brand specific data */
2434 zone->zone_brand = bp;
2435 ZBROP(zone)->b_init_brand_data(zone);
2436
2437 mutex_exit(&zone_status_lock);
2438 return (0);
2439 }
2440
2441 static int
2442 zone_set_fs_allowed(zone_t *zone, const char *zone_fs_allowed)
2443 {
2444 char *buf = kmem_zalloc(ZONE_FS_ALLOWED_MAX, KM_SLEEP);
2445 int err = 0;
2446
2447 ASSERT(zone != global_zone);
2448 if ((err = copyinstr(zone_fs_allowed, buf,
2449 ZONE_FS_ALLOWED_MAX, NULL)) != 0)
2450 goto done;
2451
2452 if (zone->zone_fs_allowed != NULL)
2453 strfree(zone->zone_fs_allowed);
2454
2455 zone->zone_fs_allowed = strdup(buf);
2456
2457 done:
2458 kmem_free(buf, ZONE_FS_ALLOWED_MAX);
2459 return (err);
2460 }
2461
2462 static int
2463 zone_set_initname(zone_t *zone, const char *zone_initname)
2464 {
2465 char initname[INITNAME_SZ];
2466 size_t len;
2467 int err = 0;
2468
2469 ASSERT(zone != global_zone);
2470 if ((err = copyinstr(zone_initname, initname, INITNAME_SZ, &len)) != 0)
2471 return (err); /* EFAULT or ENAMETOOLONG */
2472
2473 if (zone->zone_initname != NULL)
2474 strfree(zone->zone_initname);
2475
2476 zone->zone_initname = kmem_alloc(strlen(initname) + 1, KM_SLEEP);
2477 (void) strcpy(zone->zone_initname, initname);
2478 return (0);
2479 }
2480
2481 static int
2482 zone_set_phys_mcap(zone_t *zone, const uint64_t *zone_mcap)
2483 {
2484 uint64_t mcap;
2485 int err = 0;
2486
2487 if ((err = copyin(zone_mcap, &mcap, sizeof (uint64_t))) == 0)
2488 zone->zone_phys_mcap = mcap;
2489
2490 return (err);
2491 }
2492
2493 static int
2494 zone_set_sched_class(zone_t *zone, const char *new_class)
2495 {
2496 char sched_class[PC_CLNMSZ];
2497 id_t classid;
2498 int err;
2499
2500 ASSERT(zone != global_zone);
2501 if ((err = copyinstr(new_class, sched_class, PC_CLNMSZ, NULL)) != 0)
2502 return (err); /* EFAULT or ENAMETOOLONG */
2503
2504 if (getcid(sched_class, &classid) != 0 || CLASS_KERNEL(classid))
2505 return (set_errno(EINVAL));
2506 zone->zone_defaultcid = classid;
2507 ASSERT(zone->zone_defaultcid > 0 &&
2508 zone->zone_defaultcid < loaded_classes);
2509
2510 return (0);
2511 }
2512
2513 /*
2514 * Block indefinitely waiting for (zone_status >= status)
2515 */
2516 void
2517 zone_status_wait(zone_t *zone, zone_status_t status)
2518 {
2519 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2520
2521 mutex_enter(&zone_status_lock);
2522 while (zone->zone_status < status) {
2523 cv_wait(&zone->zone_cv, &zone_status_lock);
2524 }
2525 mutex_exit(&zone_status_lock);
2526 }
2527
2528 /*
2529 * Private CPR-safe version of zone_status_wait().
2530 */
2531 static void
2532 zone_status_wait_cpr(zone_t *zone, zone_status_t status, char *str)
2533 {
2534 callb_cpr_t cprinfo;
2535
2536 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2537
2538 CALLB_CPR_INIT(&cprinfo, &zone_status_lock, callb_generic_cpr,
2539 str);
2540 mutex_enter(&zone_status_lock);
2541 while (zone->zone_status < status) {
2542 CALLB_CPR_SAFE_BEGIN(&cprinfo);
2543 cv_wait(&zone->zone_cv, &zone_status_lock);
2544 CALLB_CPR_SAFE_END(&cprinfo, &zone_status_lock);
2545 }
2546 /*
2547 * zone_status_lock is implicitly released by the following.
2548 */
2549 CALLB_CPR_EXIT(&cprinfo);
2550 }
2551
2552 /*
2553 * Block until zone enters requested state or signal is received. Return (0)
2554 * if signaled, non-zero otherwise.
2555 */
2556 int
2557 zone_status_wait_sig(zone_t *zone, zone_status_t status)
2558 {
2559 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2560
2561 mutex_enter(&zone_status_lock);
2562 while (zone->zone_status < status) {
2563 if (!cv_wait_sig(&zone->zone_cv, &zone_status_lock)) {
2564 mutex_exit(&zone_status_lock);
2565 return (0);
2566 }
2567 }
2568 mutex_exit(&zone_status_lock);
2569 return (1);
2570 }
2571
2572 /*
2573 * Block until the zone enters the requested state or the timeout expires,
2574 * whichever happens first. Return (-1) if operation timed out, time remaining
2575 * otherwise.
2576 */
2577 clock_t
2578 zone_status_timedwait(zone_t *zone, clock_t tim, zone_status_t status)
2579 {
2580 clock_t timeleft = 0;
2581
2582 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2583
2584 mutex_enter(&zone_status_lock);
2585 while (zone->zone_status < status && timeleft != -1) {
2586 timeleft = cv_timedwait(&zone->zone_cv, &zone_status_lock, tim);
2587 }
2588 mutex_exit(&zone_status_lock);
2589 return (timeleft);
2590 }
2591
2592 /*
2593 * Block until the zone enters the requested state, the current process is
2594 * signaled, or the timeout expires, whichever happens first. Return (-1) if
2595 * operation timed out, 0 if signaled, time remaining otherwise.
2596 */
2597 clock_t
2598 zone_status_timedwait_sig(zone_t *zone, clock_t tim, zone_status_t status)
2599 {
2600 clock_t timeleft = tim - ddi_get_lbolt();
2601
2602 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2603
2604 mutex_enter(&zone_status_lock);
2605 while (zone->zone_status < status) {
2606 timeleft = cv_timedwait_sig(&zone->zone_cv, &zone_status_lock,
2607 tim);
2608 if (timeleft <= 0)
2609 break;
2610 }
2611 mutex_exit(&zone_status_lock);
2612 return (timeleft);
2613 }
2614
2615 /*
2616 * Zones have two reference counts: one for references from credential
2617 * structures (zone_cred_ref), and one (zone_ref) for everything else.
2618 * This is so we can allow a zone to be rebooted while there are still
2619 * outstanding cred references, since certain drivers cache dblks (which
2620 * implicitly results in cached creds). We wait for zone_ref to drop to
2621 * 0 (actually 1), but not zone_cred_ref. The zone structure itself is
2622 * later freed when the zone_cred_ref drops to 0, though nothing other
2623 * than the zone id and privilege set should be accessed once the zone
2624 * is "dead".
2625 *
2626 * A debugging flag, zone_wait_for_cred, can be set to a non-zero value
2627 * to force halt/reboot to block waiting for the zone_cred_ref to drop
2628 * to 0. This can be useful to flush out other sources of cached creds
2629 * that may be less innocuous than the driver case.
2630 *
2631 * Zones also provide a tracked reference counting mechanism in which zone
2632 * references are represented by "crumbs" (zone_ref structures). Crumbs help
2633 * debuggers determine the sources of leaked zone references. See
2634 * zone_hold_ref() and zone_rele_ref() below for more information.
2635 */
2636
2637 int zone_wait_for_cred = 0;
2638
2639 static void
2640 zone_hold_locked(zone_t *z)
2641 {
2642 ASSERT(MUTEX_HELD(&z->zone_lock));
2643 z->zone_ref++;
2644 ASSERT(z->zone_ref != 0);
2645 }
2646
2647 /*
2648 * Increment the specified zone's reference count. The zone's zone_t structure
2649 * will not be freed as long as the zone's reference count is nonzero.
2650 * Decrement the zone's reference count via zone_rele().
2651 *
2652 * NOTE: This function should only be used to hold zones for short periods of
2653 * time. Use zone_hold_ref() if the zone must be held for a long time.
2654 */
2655 void
2656 zone_hold(zone_t *z)
2657 {
2658 mutex_enter(&z->zone_lock);
2659 zone_hold_locked(z);
2660 mutex_exit(&z->zone_lock);
2661 }
2662
2663 /*
2664 * If the non-cred ref count drops to 1 and either the cred ref count
2665 * is 0 or we aren't waiting for cred references, the zone is ready to
2666 * be destroyed.
2667 */
2668 #define ZONE_IS_UNREF(zone) ((zone)->zone_ref == 1 && \
2669 (!zone_wait_for_cred || (zone)->zone_cred_ref == 0))
2670
2671 /*
2672 * Common zone reference release function invoked by zone_rele() and
2673 * zone_rele_ref(). If subsys is ZONE_REF_NUM_SUBSYS, then the specified
2674 * zone's subsystem-specific reference counters are not affected by the
2675 * release. If ref is not NULL, then the zone_ref_t to which it refers is
2676 * removed from the specified zone's reference list. ref must be non-NULL iff
2677 * subsys is not ZONE_REF_NUM_SUBSYS.
2678 */
2679 static void
2680 zone_rele_common(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
2681 {
2682 boolean_t wakeup;
2683
2684 mutex_enter(&z->zone_lock);
2685 ASSERT(z->zone_ref != 0);
2686 z->zone_ref--;
2687 if (subsys != ZONE_REF_NUM_SUBSYS) {
2688 ASSERT(z->zone_subsys_ref[subsys] != 0);
2689 z->zone_subsys_ref[subsys]--;
2690 list_remove(&z->zone_ref_list, ref);
2691 }
2692 if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
2693 /* no more refs, free the structure */
2694 mutex_exit(&z->zone_lock);
2695 zone_free(z);
2696 return;
2697 }
2698 /* signal zone_destroy so the zone can finish halting */
2699 wakeup = (ZONE_IS_UNREF(z) && zone_status_get(z) >= ZONE_IS_DEAD);
2700 mutex_exit(&z->zone_lock);
2701
2702 if (wakeup) {
2703 /*
2704 * Grabbing zonehash_lock here effectively synchronizes with
2705 * zone_destroy() to avoid missed signals.
2706 */
2707 mutex_enter(&zonehash_lock);
2708 cv_broadcast(&zone_destroy_cv);
2709 mutex_exit(&zonehash_lock);
2710 }
2711 }
2712
2713 /*
2714 * Decrement the specified zone's reference count. The specified zone will
2715 * cease to exist after this function returns if the reference count drops to
2716 * zero. This function should be paired with zone_hold().
2717 */
2718 void
2719 zone_rele(zone_t *z)
2720 {
2721 zone_rele_common(z, NULL, ZONE_REF_NUM_SUBSYS);
2722 }
2723
2724 /*
2725 * Initialize a zone reference structure. This function must be invoked for
2726 * a reference structure before the structure is passed to zone_hold_ref().
2727 */
2728 void
2729 zone_init_ref(zone_ref_t *ref)
2730 {
2731 ref->zref_zone = NULL;
2732 list_link_init(&ref->zref_linkage);
2733 }
2734
2735 /*
2736 * Acquire a reference to zone z. The caller must specify the
2737 * zone_ref_subsys_t constant associated with its subsystem. The specified
2738 * zone_ref_t structure will represent a reference to the specified zone. Use
2739 * zone_rele_ref() to release the reference.
2740 *
2741 * The referenced zone_t structure will not be freed as long as the zone_t's
2742 * zone_status field is not ZONE_IS_DEAD and the zone has outstanding
2743 * references.
2744 *
2745 * NOTE: The zone_ref_t structure must be initialized before it is used.
2746 * See zone_init_ref() above.
2747 */
2748 void
2749 zone_hold_ref(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
2750 {
2751 ASSERT(subsys >= 0 && subsys < ZONE_REF_NUM_SUBSYS);
2752
2753 /*
2754 * Prevent consumers from reusing a reference structure before
2755 * releasing it.
2756 */
2757 VERIFY(ref->zref_zone == NULL);
2758
2759 ref->zref_zone = z;
2760 mutex_enter(&z->zone_lock);
2761 zone_hold_locked(z);
2762 z->zone_subsys_ref[subsys]++;
2763 ASSERT(z->zone_subsys_ref[subsys] != 0);
2764 list_insert_head(&z->zone_ref_list, ref);
2765 mutex_exit(&z->zone_lock);
2766 }
2767
2768 /*
2769 * Release the zone reference represented by the specified zone_ref_t.
2770 * The reference is invalid after it's released; however, the zone_ref_t
2771 * structure can be reused without having to invoke zone_init_ref().
2772 * subsys should be the same value that was passed to zone_hold_ref()
2773 * when the reference was acquired.
2774 */
2775 void
2776 zone_rele_ref(zone_ref_t *ref, zone_ref_subsys_t subsys)
2777 {
2778 zone_rele_common(ref->zref_zone, ref, subsys);
2779
2780 /*
2781 * Set the zone_ref_t's zref_zone field to NULL to generate panics
2782 * when consumers dereference the reference. This helps us catch
2783 * consumers who use released references. Furthermore, this lets
2784 * consumers reuse the zone_ref_t structure without having to
2785 * invoke zone_init_ref().
2786 */
2787 ref->zref_zone = NULL;
2788 }
2789
2790 void
2791 zone_cred_hold(zone_t *z)
2792 {
2793 mutex_enter(&z->zone_lock);
2794 z->zone_cred_ref++;
2795 ASSERT(z->zone_cred_ref != 0);
2796 mutex_exit(&z->zone_lock);
2797 }
2798
2799 void
2800 zone_cred_rele(zone_t *z)
2801 {
2802 boolean_t wakeup;
2803
2804 mutex_enter(&z->zone_lock);
2805 ASSERT(z->zone_cred_ref != 0);
2806 z->zone_cred_ref--;
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 /*
2814 * If zone_destroy is waiting for the cred references to drain
2815 * out, and they have, signal it.
2816 */
2817 wakeup = (zone_wait_for_cred && ZONE_IS_UNREF(z) &&
2818 zone_status_get(z) >= ZONE_IS_DEAD);
2819 mutex_exit(&z->zone_lock);
2820
2821 if (wakeup) {
2822 /*
2823 * Grabbing zonehash_lock here effectively synchronizes with
2824 * zone_destroy() to avoid missed signals.
2825 */
2826 mutex_enter(&zonehash_lock);
2827 cv_broadcast(&zone_destroy_cv);
2828 mutex_exit(&zonehash_lock);
2829 }
2830 }
2831
2832 void
2833 zone_task_hold(zone_t *z)
2834 {
2835 mutex_enter(&z->zone_lock);
2836 z->zone_ntasks++;
2837 ASSERT(z->zone_ntasks != 0);
2838 mutex_exit(&z->zone_lock);
2839 }
2840
2841 void
2842 zone_task_rele(zone_t *zone)
2843 {
2844 uint_t refcnt;
2845
2846 mutex_enter(&zone->zone_lock);
2847 ASSERT(zone->zone_ntasks != 0);
2848 refcnt = --zone->zone_ntasks;
2849 if (refcnt > 1) { /* Common case */
2850 mutex_exit(&zone->zone_lock);
2851 return;
2852 }
2853 zone_hold_locked(zone); /* so we can use the zone_t later */
2854 mutex_exit(&zone->zone_lock);
2855 if (refcnt == 1) {
2856 /*
2857 * See if the zone is shutting down.
2858 */
2859 mutex_enter(&zone_status_lock);
2860 if (zone_status_get(zone) != ZONE_IS_SHUTTING_DOWN) {
2861 goto out;
2862 }
2863
2864 /*
2865 * Make sure the ntasks didn't change since we
2866 * dropped zone_lock.
2867 */
2868 mutex_enter(&zone->zone_lock);
2869 if (refcnt != zone->zone_ntasks) {
2870 mutex_exit(&zone->zone_lock);
2871 goto out;
2872 }
2873 mutex_exit(&zone->zone_lock);
2874
2875 /*
2876 * No more user processes in the zone. The zone is empty.
2877 */
2878 zone_status_set(zone, ZONE_IS_EMPTY);
2879 goto out;
2880 }
2881
2882 ASSERT(refcnt == 0);
2883 /*
2884 * zsched has exited; the zone is dead.
2885 */
2886 zone->zone_zsched = NULL; /* paranoia */
2887 mutex_enter(&zone_status_lock);
2888 zone_status_set(zone, ZONE_IS_DEAD);
2889 out:
2890 mutex_exit(&zone_status_lock);
2891 zone_rele(zone);
2892 }
2893
2894 zoneid_t
2895 getzoneid(void)
2896 {
2897 return (curproc->p_zone->zone_id);
2898 }
2899
2900 /*
2901 * Internal versions of zone_find_by_*(). These don't zone_hold() or
2902 * check the validity of a zone's state.
2903 */
2904 static zone_t *
2905 zone_find_all_by_id(zoneid_t zoneid)
2906 {
2907 mod_hash_val_t hv;
2908 zone_t *zone = NULL;
2909
2910 ASSERT(MUTEX_HELD(&zonehash_lock));
2911
2912 if (mod_hash_find(zonehashbyid,
2913 (mod_hash_key_t)(uintptr_t)zoneid, &hv) == 0)
2914 zone = (zone_t *)hv;
2915 return (zone);
2916 }
2917
2918 static zone_t *
2919 zone_find_all_by_label(const ts_label_t *label)
2920 {
2921 mod_hash_val_t hv;
2922 zone_t *zone = NULL;
2923
2924 ASSERT(MUTEX_HELD(&zonehash_lock));
2925
2926 /*
2927 * zonehashbylabel is not maintained for unlabeled systems
2928 */
2929 if (!is_system_labeled())
2930 return (NULL);
2931 if (mod_hash_find(zonehashbylabel, (mod_hash_key_t)label, &hv) == 0)
2932 zone = (zone_t *)hv;
2933 return (zone);
2934 }
2935
2936 static zone_t *
2937 zone_find_all_by_name(char *name)
2938 {
2939 mod_hash_val_t hv;
2940 zone_t *zone = NULL;
2941
2942 ASSERT(MUTEX_HELD(&zonehash_lock));
2943
2944 if (mod_hash_find(zonehashbyname, (mod_hash_key_t)name, &hv) == 0)
2945 zone = (zone_t *)hv;
2946 return (zone);
2947 }
2948
2949 /*
2950 * Public interface for looking up a zone by zoneid. Only returns the zone if
2951 * it is fully initialized, and has not yet begun the zone_destroy() sequence.
2952 * Caller must call zone_rele() once it is done with the zone.
2953 *
2954 * The zone may begin the zone_destroy() sequence immediately after this
2955 * function returns, but may be safely used until zone_rele() is called.
2956 */
2957 zone_t *
2958 zone_find_by_id(zoneid_t zoneid)
2959 {
2960 zone_t *zone;
2961 zone_status_t status;
2962
2963 mutex_enter(&zonehash_lock);
2964 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
2965 mutex_exit(&zonehash_lock);
2966 return (NULL);
2967 }
2968 status = zone_status_get(zone);
2969 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
2970 /*
2971 * For all practical purposes the zone doesn't exist.
2972 */
2973 mutex_exit(&zonehash_lock);
2974 return (NULL);
2975 }
2976 zone_hold(zone);
2977 mutex_exit(&zonehash_lock);
2978 return (zone);
2979 }
2980
2981 /*
2982 * Similar to zone_find_by_id, but using zone label as the key.
2983 */
2984 zone_t *
2985 zone_find_by_label(const ts_label_t *label)
2986 {
2987 zone_t *zone;
2988 zone_status_t status;
2989
2990 mutex_enter(&zonehash_lock);
2991 if ((zone = zone_find_all_by_label(label)) == NULL) {
2992 mutex_exit(&zonehash_lock);
2993 return (NULL);
2994 }
2995
2996 status = zone_status_get(zone);
2997 if (status > ZONE_IS_DOWN) {
2998 /*
2999 * For all practical purposes the zone doesn't exist.
3000 */
3001 mutex_exit(&zonehash_lock);
3002 return (NULL);
3003 }
3004 zone_hold(zone);
3005 mutex_exit(&zonehash_lock);
3006 return (zone);
3007 }
3008
3009 /*
3010 * Similar to zone_find_by_id, but using zone name as the key.
3011 */
3012 zone_t *
3013 zone_find_by_name(char *name)
3014 {
3015 zone_t *zone;
3016 zone_status_t status;
3017
3018 mutex_enter(&zonehash_lock);
3019 if ((zone = zone_find_all_by_name(name)) == NULL) {
3020 mutex_exit(&zonehash_lock);
3021 return (NULL);
3022 }
3023 status = zone_status_get(zone);
3024 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3025 /*
3026 * For all practical purposes the zone doesn't exist.
3027 */
3028 mutex_exit(&zonehash_lock);
3029 return (NULL);
3030 }
3031 zone_hold(zone);
3032 mutex_exit(&zonehash_lock);
3033 return (zone);
3034 }
3035
3036 /*
3037 * Similar to zone_find_by_id(), using the path as a key. For instance,
3038 * if there is a zone "foo" rooted at /foo/root, and the path argument
3039 * is "/foo/root/proc", it will return the held zone_t corresponding to
3040 * zone "foo".
3041 *
3042 * zone_find_by_path() always returns a non-NULL value, since at the
3043 * very least every path will be contained in the global zone.
3044 *
3045 * As with the other zone_find_by_*() functions, the caller is
3046 * responsible for zone_rele()ing the return value of this function.
3047 */
3048 zone_t *
3049 zone_find_by_path(const char *path)
3050 {
3051 zone_t *zone;
3052 zone_t *zret = NULL;
3053 zone_status_t status;
3054
3055 if (path == NULL) {
3056 /*
3057 * Call from rootconf().
3058 */
3059 zone_hold(global_zone);
3060 return (global_zone);
3061 }
3062 ASSERT(*path == '/');
3063 mutex_enter(&zonehash_lock);
3064 for (zone = list_head(&zone_active); zone != NULL;
3065 zone = list_next(&zone_active, zone)) {
3066 if (ZONE_PATH_VISIBLE(path, zone))
3067 zret = zone;
3068 }
3069 ASSERT(zret != NULL);
3070 status = zone_status_get(zret);
3071 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3072 /*
3073 * Zone practically doesn't exist.
3074 */
3075 zret = global_zone;
3076 }
3077 zone_hold(zret);
3078 mutex_exit(&zonehash_lock);
3079 return (zret);
3080 }
3081
3082 /*
3083 * Public interface for updating per-zone load averages. Called once per
3084 * second.
3085 *
3086 * Based on loadavg_update(), genloadavg() and calcloadavg() from clock.c.
3087 */
3088 void
3089 zone_loadavg_update()
3090 {
3091 zone_t *zp;
3092 zone_status_t status;
3093 struct loadavg_s *lavg;
3094 hrtime_t zone_total;
3095 int i;
3096 hrtime_t hr_avg;
3097 int nrun;
3098 static int64_t f[3] = { 135, 27, 9 };
3099 int64_t q, r;
3100
3101 mutex_enter(&zonehash_lock);
3102 for (zp = list_head(&zone_active); zp != NULL;
3103 zp = list_next(&zone_active, zp)) {
3104 mutex_enter(&zp->zone_lock);
3105
3106 /* Skip zones that are on the way down or not yet up */
3107 status = zone_status_get(zp);
3108 if (status < ZONE_IS_READY || status >= ZONE_IS_DOWN) {
3109 /* For all practical purposes the zone doesn't exist. */
3110 mutex_exit(&zp->zone_lock);
3111 continue;
3112 }
3113
3114 /*
3115 * Update the 10 second moving average data in zone_loadavg.
3116 */
3117 lavg = &zp->zone_loadavg;
3118
3119 zone_total = zp->zone_utime + zp->zone_stime + zp->zone_wtime;
3120 scalehrtime(&zone_total);
3121
3122 /* The zone_total should always be increasing. */
3123 lavg->lg_loads[lavg->lg_cur] = (zone_total > lavg->lg_total) ?
3124 zone_total - lavg->lg_total : 0;
3125 lavg->lg_cur = (lavg->lg_cur + 1) % S_LOADAVG_SZ;
3126 /* lg_total holds the prev. 1 sec. total */
3127 lavg->lg_total = zone_total;
3128
3129 /*
3130 * To simplify the calculation, we don't calculate the load avg.
3131 * until the zone has been up for at least 10 seconds and our
3132 * moving average is thus full.
3133 */
3134 if ((lavg->lg_len + 1) < S_LOADAVG_SZ) {
3135 lavg->lg_len++;
3136 mutex_exit(&zp->zone_lock);
3137 continue;
3138 }
3139
3140 /* Now calculate the 1min, 5min, 15 min load avg. */
3141 hr_avg = 0;
3142 for (i = 0; i < S_LOADAVG_SZ; i++)
3143 hr_avg += lavg->lg_loads[i];
3144 hr_avg = hr_avg / S_LOADAVG_SZ;
3145 nrun = hr_avg / (NANOSEC / LGRP_LOADAVG_IN_THREAD_MAX);
3146
3147 /* Compute load avg. See comment in calcloadavg() */
3148 for (i = 0; i < 3; i++) {
3149 q = (zp->zone_hp_avenrun[i] >> 16) << 7;
3150 r = (zp->zone_hp_avenrun[i] & 0xffff) << 7;
3151 zp->zone_hp_avenrun[i] +=
3152 ((nrun - q) * f[i] - ((r * f[i]) >> 16)) >> 4;
3153
3154 /* avenrun[] can only hold 31 bits of load avg. */
3155 if (zp->zone_hp_avenrun[i] <
3156 ((uint64_t)1<<(31+16-FSHIFT)))
3157 zp->zone_avenrun[i] = (int32_t)
3158 (zp->zone_hp_avenrun[i] >> (16 - FSHIFT));
3159 else
3160 zp->zone_avenrun[i] = 0x7fffffff;
3161 }
3162
3163 mutex_exit(&zp->zone_lock);
3164 }
3165 mutex_exit(&zonehash_lock);
3166 }
3167
3168 /*
3169 * Get the number of cpus visible to this zone. The system-wide global
3170 * 'ncpus' is returned if pools are disabled, the caller is in the
3171 * global zone, or a NULL zone argument is passed in.
3172 */
3173 int
3174 zone_ncpus_get(zone_t *zone)
3175 {
3176 int myncpus = zone == NULL ? 0 : zone->zone_ncpus;
3177
3178 return (myncpus != 0 ? myncpus : ncpus);
3179 }
3180
3181 /*
3182 * Get the number of online cpus visible to this zone. The system-wide
3183 * global 'ncpus_online' is returned if pools are disabled, the caller
3184 * is in the global zone, or a NULL zone argument is passed in.
3185 */
3186 int
3187 zone_ncpus_online_get(zone_t *zone)
3188 {
3189 int myncpus_online = zone == NULL ? 0 : zone->zone_ncpus_online;
3190
3191 return (myncpus_online != 0 ? myncpus_online : ncpus_online);
3192 }
3193
3194 /*
3195 * Return the pool to which the zone is currently bound.
3196 */
3197 pool_t *
3198 zone_pool_get(zone_t *zone)
3199 {
3200 ASSERT(pool_lock_held());
3201
3202 return (zone->zone_pool);
3203 }
3204
3205 /*
3206 * Set the zone's pool pointer and update the zone's visibility to match
3207 * the resources in the new pool.
3208 */
3209 void
3210 zone_pool_set(zone_t *zone, pool_t *pool)
3211 {
3212 ASSERT(pool_lock_held());
3213 ASSERT(MUTEX_HELD(&cpu_lock));
3214
3215 zone->zone_pool = pool;
3216 zone_pset_set(zone, pool->pool_pset->pset_id);
3217 }
3218
3219 /*
3220 * Return the cached value of the id of the processor set to which the
3221 * zone is currently bound. The value will be ZONE_PS_INVAL if the pools
3222 * facility is disabled.
3223 */
3224 psetid_t
3225 zone_pset_get(zone_t *zone)
3226 {
3227 ASSERT(MUTEX_HELD(&cpu_lock));
3228
3229 return (zone->zone_psetid);
3230 }
3231
3232 /*
3233 * Set the cached value of the id of the processor set to which the zone
3234 * is currently bound. Also update the zone's visibility to match the
3235 * resources in the new processor set.
3236 */
3237 void
3238 zone_pset_set(zone_t *zone, psetid_t newpsetid)
3239 {
3240 psetid_t oldpsetid;
3241
3242 ASSERT(MUTEX_HELD(&cpu_lock));
3243 oldpsetid = zone_pset_get(zone);
3244
3245 if (oldpsetid == newpsetid)
3246 return;
3247 /*
3248 * Global zone sees all.
3249 */
3250 if (zone != global_zone) {
3251 zone->zone_psetid = newpsetid;
3252 if (newpsetid != ZONE_PS_INVAL)
3253 pool_pset_visibility_add(newpsetid, zone);
3254 if (oldpsetid != ZONE_PS_INVAL)
3255 pool_pset_visibility_remove(oldpsetid, zone);
3256 }
3257 /*
3258 * Disabling pools, so we should start using the global values
3259 * for ncpus and ncpus_online.
3260 */
3261 if (newpsetid == ZONE_PS_INVAL) {
3262 zone->zone_ncpus = 0;
3263 zone->zone_ncpus_online = 0;
3264 }
3265 }
3266
3267 /*
3268 * Walk the list of active zones and issue the provided callback for
3269 * each of them.
3270 *
3271 * Caller must not be holding any locks that may be acquired under
3272 * zonehash_lock. See comment at the beginning of the file for a list of
3273 * common locks and their interactions with zones.
3274 */
3275 int
3276 zone_walk(int (*cb)(zone_t *, void *), void *data)
3277 {
3278 zone_t *zone;
3279 int ret = 0;
3280 zone_status_t status;
3281
3282 mutex_enter(&zonehash_lock);
3283 for (zone = list_head(&zone_active); zone != NULL;
3284 zone = list_next(&zone_active, zone)) {
3285 /*
3286 * Skip zones that shouldn't be externally visible.
3287 */
3288 status = zone_status_get(zone);
3289 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN)
3290 continue;
3291 /*
3292 * Bail immediately if any callback invocation returns a
3293 * non-zero value.
3294 */
3295 ret = (*cb)(zone, data);
3296 if (ret != 0)
3297 break;
3298 }
3299 mutex_exit(&zonehash_lock);
3300 return (ret);
3301 }
3302
3303 static int
3304 zone_set_root(zone_t *zone, const char *upath)
3305 {
3306 vnode_t *vp;
3307 int trycount;
3308 int error = 0;
3309 char *path;
3310 struct pathname upn, pn;
3311 size_t pathlen;
3312
3313 if ((error = pn_get((char *)upath, UIO_USERSPACE, &upn)) != 0)
3314 return (error);
3315
3316 pn_alloc(&pn);
3317
3318 /* prevent infinite loop */
3319 trycount = 10;
3320 for (;;) {
3321 if (--trycount <= 0) {
3322 error = ESTALE;
3323 goto out;
3324 }
3325
3326 if ((error = lookuppn(&upn, &pn, FOLLOW, NULLVPP, &vp)) == 0) {
3327 /*
3328 * VOP_ACCESS() may cover 'vp' with a new
3329 * filesystem, if 'vp' is an autoFS vnode.
3330 * Get the new 'vp' if so.
3331 */
3332 if ((error =
3333 VOP_ACCESS(vp, VEXEC, 0, CRED(), NULL)) == 0 &&
3334 (!vn_ismntpt(vp) ||
3335 (error = traverse(&vp)) == 0)) {
3336 pathlen = pn.pn_pathlen + 2;
3337 path = kmem_alloc(pathlen, KM_SLEEP);
3338 (void) strncpy(path, pn.pn_path,
3339 pn.pn_pathlen + 1);
3340 path[pathlen - 2] = '/';
3341 path[pathlen - 1] = '\0';
3342 pn_free(&pn);
3343 pn_free(&upn);
3344
3345 /* Success! */
3346 break;
3347 }
3348 VN_RELE(vp);
3349 }
3350 if (error != ESTALE)
3351 goto out;
3352 }
3353
3354 ASSERT(error == 0);
3355 zone->zone_rootvp = vp; /* we hold a reference to vp */
3356 zone->zone_rootpath = path;
3357 zone->zone_rootpathlen = pathlen;
3358 if (pathlen > 5 && strcmp(path + pathlen - 5, "/lu/") == 0)
3359 zone->zone_flags |= ZF_IS_SCRATCH;
3360 return (0);
3361
3362 out:
3363 pn_free(&pn);
3364 pn_free(&upn);
3365 return (error);
3366 }
3367
3368 #define isalnum(c) (((c) >= '0' && (c) <= '9') || \
3369 ((c) >= 'a' && (c) <= 'z') || \
3370 ((c) >= 'A' && (c) <= 'Z'))
3371
3372 static int
3373 zone_set_name(zone_t *zone, const char *uname)
3374 {
3375 char *kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
3376 size_t len;
3377 int i, err;
3378
3379 if ((err = copyinstr(uname, kname, ZONENAME_MAX, &len)) != 0) {
3380 kmem_free(kname, ZONENAME_MAX);
3381 return (err); /* EFAULT or ENAMETOOLONG */
3382 }
3383
3384 /* must be less than ZONENAME_MAX */
3385 if (len == ZONENAME_MAX && kname[ZONENAME_MAX - 1] != '\0') {
3386 kmem_free(kname, ZONENAME_MAX);
3387 return (EINVAL);
3388 }
3389
3390 /*
3391 * Name must start with an alphanumeric and must contain only
3392 * alphanumerics, '-', '_' and '.'.
3393 */
3394 if (!isalnum(kname[0])) {
3395 kmem_free(kname, ZONENAME_MAX);
3396 return (EINVAL);
3397 }
3398 for (i = 1; i < len - 1; i++) {
3399 if (!isalnum(kname[i]) && kname[i] != '-' && kname[i] != '_' &&
3400 kname[i] != '.') {
3401 kmem_free(kname, ZONENAME_MAX);
3402 return (EINVAL);
3403 }
3404 }
3405
3406 zone->zone_name = kname;
3407 return (0);
3408 }
3409
3410 /*
3411 * Gets the 32-bit hostid of the specified zone as an unsigned int. If 'zonep'
3412 * is NULL or it points to a zone with no hostid emulation, then the machine's
3413 * hostid (i.e., the global zone's hostid) is returned. This function returns
3414 * zero if neither the zone nor the host machine (global zone) have hostids. It
3415 * returns HW_INVALID_HOSTID if the function attempts to return the machine's
3416 * hostid and the machine's hostid is invalid.
3417 */
3418 uint32_t
3419 zone_get_hostid(zone_t *zonep)
3420 {
3421 unsigned long machine_hostid;
3422
3423 if (zonep == NULL || zonep->zone_hostid == HW_INVALID_HOSTID) {
3424 if (ddi_strtoul(hw_serial, NULL, 10, &machine_hostid) != 0)
3425 return (HW_INVALID_HOSTID);
3426 return ((uint32_t)machine_hostid);
3427 }
3428 return (zonep->zone_hostid);
3429 }
3430
3431 /*
3432 * Similar to thread_create(), but makes sure the thread is in the appropriate
3433 * zone's zsched process (curproc->p_zone->zone_zsched) before returning.
3434 */
3435 /*ARGSUSED*/
3436 kthread_t *
3437 zthread_create(
3438 caddr_t stk,
3439 size_t stksize,
3440 void (*proc)(),
3441 void *arg,
3442 size_t len,
3443 pri_t pri)
3444 {
3445 kthread_t *t;
3446 zone_t *zone = curproc->p_zone;
3447 proc_t *pp = zone->zone_zsched;
3448
3449 zone_hold(zone); /* Reference to be dropped when thread exits */
3450
3451 /*
3452 * No-one should be trying to create threads if the zone is shutting
3453 * down and there aren't any kernel threads around. See comment
3454 * in zthread_exit().
3455 */
3456 ASSERT(!(zone->zone_kthreads == NULL &&
3457 zone_status_get(zone) >= ZONE_IS_EMPTY));
3458 /*
3459 * Create a thread, but don't let it run until we've finished setting
3460 * things up.
3461 */
3462 t = thread_create(stk, stksize, proc, arg, len, pp, TS_STOPPED, pri);
3463 ASSERT(t->t_forw == NULL);
3464 mutex_enter(&zone_status_lock);
3465 if (zone->zone_kthreads == NULL) {
3466 t->t_forw = t->t_back = t;
3467 } else {
3468 kthread_t *tx = zone->zone_kthreads;
3469
3470 t->t_forw = tx;
3471 t->t_back = tx->t_back;
3472 tx->t_back->t_forw = t;
3473 tx->t_back = t;
3474 }
3475 zone->zone_kthreads = t;
3476 mutex_exit(&zone_status_lock);
3477
3478 mutex_enter(&pp->p_lock);
3479 t->t_proc_flag |= TP_ZTHREAD;
3480 project_rele(t->t_proj);
3481 t->t_proj = project_hold(pp->p_task->tk_proj);
3482
3483 /*
3484 * Setup complete, let it run.
3485 */
3486 thread_lock(t);
3487 t->t_schedflag |= TS_ALLSTART;
3488 setrun_locked(t);
3489 thread_unlock(t);
3490
3491 mutex_exit(&pp->p_lock);
3492
3493 return (t);
3494 }
3495
3496 /*
3497 * Similar to thread_exit(). Must be called by threads created via
3498 * zthread_exit().
3499 */
3500 void
3501 zthread_exit(void)
3502 {
3503 kthread_t *t = curthread;
3504 proc_t *pp = curproc;
3505 zone_t *zone = pp->p_zone;
3506
3507 mutex_enter(&zone_status_lock);
3508
3509 /*
3510 * Reparent to p0
3511 */
3512 kpreempt_disable();
3513 mutex_enter(&pp->p_lock);
3514 t->t_proc_flag &= ~TP_ZTHREAD;
3515 t->t_procp = &p0;
3516 hat_thread_exit(t);
3517 mutex_exit(&pp->p_lock);
3518 kpreempt_enable();
3519
3520 if (t->t_back == t) {
3521 ASSERT(t->t_forw == t);
3522 /*
3523 * If the zone is empty, once the thread count
3524 * goes to zero no further kernel threads can be
3525 * created. This is because if the creator is a process
3526 * in the zone, then it must have exited before the zone
3527 * state could be set to ZONE_IS_EMPTY.
3528 * Otherwise, if the creator is a kernel thread in the
3529 * zone, the thread count is non-zero.
3530 *
3531 * This really means that non-zone kernel threads should
3532 * not create zone kernel threads.
3533 */
3534 zone->zone_kthreads = NULL;
3535 if (zone_status_get(zone) == ZONE_IS_EMPTY) {
3536 zone_status_set(zone, ZONE_IS_DOWN);
3537 /*
3538 * Remove any CPU caps on this zone.
3539 */
3540 cpucaps_zone_remove(zone);
3541 }
3542 } else {
3543 t->t_forw->t_back = t->t_back;
3544 t->t_back->t_forw = t->t_forw;
3545 if (zone->zone_kthreads == t)
3546 zone->zone_kthreads = t->t_forw;
3547 }
3548 mutex_exit(&zone_status_lock);
3549 zone_rele(zone);
3550 thread_exit();
3551 /* NOTREACHED */
3552 }
3553
3554 static void
3555 zone_chdir(vnode_t *vp, vnode_t **vpp, proc_t *pp)
3556 {
3557 vnode_t *oldvp;
3558
3559 /* we're going to hold a reference here to the directory */
3560 VN_HOLD(vp);
3561
3562 /* update abs cwd/root path see c2/audit.c */
3563 if (AU_AUDITING())
3564 audit_chdirec(vp, vpp);
3565
3566 mutex_enter(&pp->p_lock);
3567 oldvp = *vpp;
3568 *vpp = vp;
3569 mutex_exit(&pp->p_lock);
3570 if (oldvp != NULL)
3571 VN_RELE(oldvp);
3572 }
3573
3574 /*
3575 * Convert an rctl value represented by an nvlist_t into an rctl_val_t.
3576 */
3577 static int
3578 nvlist2rctlval(nvlist_t *nvl, rctl_val_t *rv)
3579 {
3580 nvpair_t *nvp = NULL;
3581 boolean_t priv_set = B_FALSE;
3582 boolean_t limit_set = B_FALSE;
3583 boolean_t action_set = B_FALSE;
3584
3585 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
3586 const char *name;
3587 uint64_t ui64;
3588
3589 name = nvpair_name(nvp);
3590 if (nvpair_type(nvp) != DATA_TYPE_UINT64)
3591 return (EINVAL);
3592 (void) nvpair_value_uint64(nvp, &ui64);
3593 if (strcmp(name, "privilege") == 0) {
3594 /*
3595 * Currently only privileged values are allowed, but
3596 * this may change in the future.
3597 */
3598 if (ui64 != RCPRIV_PRIVILEGED)
3599 return (EINVAL);
3600 rv->rcv_privilege = ui64;
3601 priv_set = B_TRUE;
3602 } else if (strcmp(name, "limit") == 0) {
3603 rv->rcv_value = ui64;
3604 limit_set = B_TRUE;
3605 } else if (strcmp(name, "action") == 0) {
3606 if (ui64 != RCTL_LOCAL_NOACTION &&
3607 ui64 != RCTL_LOCAL_DENY)
3608 return (EINVAL);
3609 rv->rcv_flagaction = ui64;
3610 action_set = B_TRUE;
3611 } else {
3612 return (EINVAL);
3613 }
3614 }
3615
3616 if (!(priv_set && limit_set && action_set))
3617 return (EINVAL);
3618 rv->rcv_action_signal = 0;
3619 rv->rcv_action_recipient = NULL;
3620 rv->rcv_action_recip_pid = -1;
3621 rv->rcv_firing_time = 0;
3622
3623 return (0);
3624 }
3625
3626 /*
3627 * Non-global zone version of start_init.
3628 */
3629 void
3630 zone_start_init(void)
3631 {
3632 proc_t *p = ttoproc(curthread);
3633 zone_t *z = p->p_zone;
3634
3635 ASSERT(!INGLOBALZONE(curproc));
3636
3637 /*
3638 * For all purposes (ZONE_ATTR_INITPID and restart_init),
3639 * storing just the pid of init is sufficient.
3640 */
3641 z->zone_proc_initpid = p->p_pid;
3642
3643 /*
3644 * We maintain zone_boot_err so that we can return the cause of the
3645 * failure back to the caller of the zone_boot syscall.
3646 */
3647 p->p_zone->zone_boot_err = start_init_common();
3648
3649 /*
3650 * We will prevent booting zones from becoming running zones if the
3651 * global zone is shutting down.
3652 */
3653 mutex_enter(&zone_status_lock);
3654 if (z->zone_boot_err != 0 || zone_status_get(global_zone) >=
3655 ZONE_IS_SHUTTING_DOWN) {
3656 /*
3657 * Make sure we are still in the booting state-- we could have
3658 * raced and already be shutting down, or even further along.
3659 */
3660 if (zone_status_get(z) == ZONE_IS_BOOTING) {
3661 zone_status_set(z, ZONE_IS_SHUTTING_DOWN);
3662 }
3663 mutex_exit(&zone_status_lock);
3664 /* It's gone bad, dispose of the process */
3665 if (proc_exit(CLD_EXITED, z->zone_boot_err) != 0) {
3666 mutex_enter(&p->p_lock);
3667 ASSERT(p->p_flag & SEXITLWPS);
3668 lwp_exit();
3669 }
3670 } else {
3671 if (zone_status_get(z) == ZONE_IS_BOOTING)
3672 zone_status_set(z, ZONE_IS_RUNNING);
3673 mutex_exit(&zone_status_lock);
3674 /* cause the process to return to userland. */
3675 lwp_rtt();
3676 }
3677 }
3678
3679 struct zsched_arg {
3680 zone_t *zone;
3681 nvlist_t *nvlist;
3682 };
3683
3684 /*
3685 * Per-zone "sched" workalike. The similarity to "sched" doesn't have
3686 * anything to do with scheduling, but rather with the fact that
3687 * per-zone kernel threads are parented to zsched, just like regular
3688 * kernel threads are parented to sched (p0).
3689 *
3690 * zsched is also responsible for launching init for the zone.
3691 */
3692 static void
3693 zsched(void *arg)
3694 {
3695 struct zsched_arg *za = arg;
3696 proc_t *pp = curproc;
3697 proc_t *initp = proc_init;
3698 zone_t *zone = za->zone;
3699 cred_t *cr, *oldcred;
3700 rctl_set_t *set;
3701 rctl_alloc_gp_t *gp;
3702 contract_t *ct = NULL;
3703 task_t *tk, *oldtk;
3704 rctl_entity_p_t e;
3705 kproject_t *pj;
3706
3707 nvlist_t *nvl = za->nvlist;
3708 nvpair_t *nvp = NULL;
3709
3710 bcopy("zsched", PTOU(pp)->u_psargs, sizeof ("zsched"));
3711 bcopy("zsched", PTOU(pp)->u_comm, sizeof ("zsched"));
3712 PTOU(pp)->u_argc = 0;
3713 PTOU(pp)->u_argv = NULL;
3714 PTOU(pp)->u_envp = NULL;
3715 closeall(P_FINFO(pp));
3716
3717 /*
3718 * We are this zone's "zsched" process. As the zone isn't generally
3719 * visible yet we don't need to grab any locks before initializing its
3720 * zone_proc pointer.
3721 */
3722 zone_hold(zone); /* this hold is released by zone_destroy() */
3723 zone->zone_zsched = pp;
3724 mutex_enter(&pp->p_lock);
3725 pp->p_zone = zone;
3726 mutex_exit(&pp->p_lock);
3727
3728 /*
3729 * Disassociate process from its 'parent'; parent ourselves to init
3730 * (pid 1) and change other values as needed.
3731 */
3732 sess_create();
3733
3734 mutex_enter(&pidlock);
3735 proc_detach(pp);
3736 pp->p_ppid = 1;
3737 pp->p_flag |= SZONETOP;
3738 pp->p_ancpid = 1;
3739 pp->p_parent = initp;
3740 pp->p_psibling = NULL;
3741 if (initp->p_child)
3742 initp->p_child->p_psibling = pp;
3743 pp->p_sibling = initp->p_child;
3744 initp->p_child = pp;
3745
3746 /* Decrement what newproc() incremented. */
3747 upcount_dec(crgetruid(CRED()), GLOBAL_ZONEID);
3748 /*
3749 * Our credentials are about to become kcred-like, so we don't care
3750 * about the caller's ruid.
3751 */
3752 upcount_inc(crgetruid(kcred), zone->zone_id);
3753 mutex_exit(&pidlock);
3754
3755 /*
3756 * getting out of global zone, so decrement lwp and process counts
3757 */
3758 pj = pp->p_task->tk_proj;
3759 mutex_enter(&global_zone->zone_nlwps_lock);
3760 pj->kpj_nlwps -= pp->p_lwpcnt;
3761 global_zone->zone_nlwps -= pp->p_lwpcnt;
3762 pj->kpj_nprocs--;
3763 global_zone->zone_nprocs--;
3764 mutex_exit(&global_zone->zone_nlwps_lock);
3765
3766 /*
3767 * Decrement locked memory counts on old zone and project.
3768 */
3769 mutex_enter(&global_zone->zone_mem_lock);
3770 global_zone->zone_locked_mem -= pp->p_locked_mem;
3771 pj->kpj_data.kpd_locked_mem -= pp->p_locked_mem;
3772 mutex_exit(&global_zone->zone_mem_lock);
3773
3774 /*
3775 * Create and join a new task in project '0' of this zone.
3776 *
3777 * We don't need to call holdlwps() since we know we're the only lwp in
3778 * this process.
3779 *
3780 * task_join() returns with p_lock held.
3781 */
3782 tk = task_create(0, zone);
3783 mutex_enter(&cpu_lock);
3784 oldtk = task_join(tk, 0);
3785
3786 pj = pp->p_task->tk_proj;
3787
3788 mutex_enter(&zone->zone_mem_lock);
3789 zone->zone_locked_mem += pp->p_locked_mem;
3790 pj->kpj_data.kpd_locked_mem += pp->p_locked_mem;
3791 mutex_exit(&zone->zone_mem_lock);
3792
3793 /*
3794 * add lwp and process counts to zsched's zone, and increment
3795 * project's task and process count due to the task created in
3796 * the above task_create.
3797 */
3798 mutex_enter(&zone->zone_nlwps_lock);
3799 pj->kpj_nlwps += pp->p_lwpcnt;
3800 pj->kpj_ntasks += 1;
3801 zone->zone_nlwps += pp->p_lwpcnt;
3802 pj->kpj_nprocs++;
3803 zone->zone_nprocs++;
3804 mutex_exit(&zone->zone_nlwps_lock);
3805
3806 mutex_exit(&curproc->p_lock);
3807 mutex_exit(&cpu_lock);
3808 task_rele(oldtk);
3809
3810 /*
3811 * The process was created by a process in the global zone, hence the
3812 * credentials are wrong. We might as well have kcred-ish credentials.
3813 */
3814 cr = zone->zone_kcred;
3815 crhold(cr);
3816 mutex_enter(&pp->p_crlock);
3817 oldcred = pp->p_cred;
3818 pp->p_cred = cr;
3819 mutex_exit(&pp->p_crlock);
3820 crfree(oldcred);
3821
3822 /*
3823 * Hold credentials again (for thread)
3824 */
3825 crhold(cr);
3826
3827 /*
3828 * p_lwpcnt can't change since this is a kernel process.
3829 */
3830 crset(pp, cr);
3831
3832 /*
3833 * Chroot
3834 */
3835 zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_cdir, pp);
3836 zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_rdir, pp);
3837
3838 /*
3839 * Initialize zone's rctl set.
3840 */
3841 set = rctl_set_create();
3842 gp = rctl_set_init_prealloc(RCENTITY_ZONE);
3843 mutex_enter(&pp->p_lock);
3844 e.rcep_p.zone = zone;
3845 e.rcep_t = RCENTITY_ZONE;
3846 zone->zone_rctls = rctl_set_init(RCENTITY_ZONE, pp, &e, set, gp);
3847 mutex_exit(&pp->p_lock);
3848 rctl_prealloc_destroy(gp);
3849
3850 /*
3851 * Apply the rctls passed in to zone_create(). This is basically a list
3852 * assignment: all of the old values are removed and the new ones
3853 * inserted. That is, if an empty list is passed in, all values are
3854 * removed.
3855 */
3856 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
3857 rctl_dict_entry_t *rde;
3858 rctl_hndl_t hndl;
3859 char *name;
3860 nvlist_t **nvlarray;
3861 uint_t i, nelem;
3862 int error; /* For ASSERT()s */
3863
3864 name = nvpair_name(nvp);
3865 hndl = rctl_hndl_lookup(name);
3866 ASSERT(hndl != -1);
3867 rde = rctl_dict_lookup_hndl(hndl);
3868 ASSERT(rde != NULL);
3869
3870 for (; /* ever */; ) {
3871 rctl_val_t oval;
3872
3873 mutex_enter(&pp->p_lock);
3874 error = rctl_local_get(hndl, NULL, &oval, pp);
3875 mutex_exit(&pp->p_lock);
3876 ASSERT(error == 0); /* Can't fail for RCTL_FIRST */
3877 ASSERT(oval.rcv_privilege != RCPRIV_BASIC);
3878 if (oval.rcv_privilege == RCPRIV_SYSTEM)
3879 break;
3880 mutex_enter(&pp->p_lock);
3881 error = rctl_local_delete(hndl, &oval, pp);
3882 mutex_exit(&pp->p_lock);
3883 ASSERT(error == 0);
3884 }
3885 error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
3886 ASSERT(error == 0);
3887 for (i = 0; i < nelem; i++) {
3888 rctl_val_t *nvalp;
3889
3890 nvalp = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
3891 error = nvlist2rctlval(nvlarray[i], nvalp);
3892 ASSERT(error == 0);
3893 /*
3894 * rctl_local_insert can fail if the value being
3895 * inserted is a duplicate; this is OK.
3896 */
3897 mutex_enter(&pp->p_lock);
3898 if (rctl_local_insert(hndl, nvalp, pp) != 0)
3899 kmem_cache_free(rctl_val_cache, nvalp);
3900 mutex_exit(&pp->p_lock);
3901 }
3902 }
3903 /*
3904 * Tell the world that we're done setting up.
3905 *
3906 * At this point we want to set the zone status to ZONE_IS_INITIALIZED
3907 * and atomically set the zone's processor set visibility. Once
3908 * we drop pool_lock() this zone will automatically get updated
3909 * to reflect any future changes to the pools configuration.
3910 *
3911 * Note that after we drop the locks below (zonehash_lock in
3912 * particular) other operations such as a zone_getattr call can
3913 * now proceed and observe the zone. That is the reason for doing a
3914 * state transition to the INITIALIZED state.
3915 */
3916 pool_lock();
3917 mutex_enter(&cpu_lock);
3918 mutex_enter(&zonehash_lock);
3919 zone_uniqid(zone);
3920 zone_zsd_configure(zone);
3921 if (pool_state == POOL_ENABLED)
3922 zone_pset_set(zone, pool_default->pool_pset->pset_id);
3923 mutex_enter(&zone_status_lock);
3924 ASSERT(zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
3925 zone_status_set(zone, ZONE_IS_INITIALIZED);
3926 mutex_exit(&zone_status_lock);
3927 mutex_exit(&zonehash_lock);
3928 mutex_exit(&cpu_lock);
3929 pool_unlock();
3930
3931 /* Now call the create callback for this key */
3932 zsd_apply_all_keys(zsd_apply_create, zone);
3933
3934 /* The callbacks are complete. Mark ZONE_IS_READY */
3935 mutex_enter(&zone_status_lock);
3936 ASSERT(zone_status_get(zone) == ZONE_IS_INITIALIZED);
3937 zone_status_set(zone, ZONE_IS_READY);
3938 mutex_exit(&zone_status_lock);
3939
3940 /*
3941 * Once we see the zone transition to the ZONE_IS_BOOTING state,
3942 * we launch init, and set the state to running.
3943 */
3944 zone_status_wait_cpr(zone, ZONE_IS_BOOTING, "zsched");
3945
3946 if (zone_status_get(zone) == ZONE_IS_BOOTING) {
3947 id_t cid;
3948
3949 /*
3950 * Ok, this is a little complicated. We need to grab the
3951 * zone's pool's scheduling class ID; note that by now, we
3952 * are already bound to a pool if we need to be (zoneadmd
3953 * will have done that to us while we're in the READY
3954 * state). *But* the scheduling class for the zone's 'init'
3955 * must be explicitly passed to newproc, which doesn't
3956 * respect pool bindings.
3957 *
3958 * We hold the pool_lock across the call to newproc() to
3959 * close the obvious race: the pool's scheduling class
3960 * could change before we manage to create the LWP with
3961 * classid 'cid'.
3962 */
3963 pool_lock();
3964 if (zone->zone_defaultcid > 0)
3965 cid = zone->zone_defaultcid;
3966 else
3967 cid = pool_get_class(zone->zone_pool);
3968 if (cid == -1)
3969 cid = defaultcid;
3970
3971 /*
3972 * If this fails, zone_boot will ultimately fail. The
3973 * state of the zone will be set to SHUTTING_DOWN-- userland
3974 * will have to tear down the zone, and fail, or try again.
3975 */
3976 if ((zone->zone_boot_err = newproc(zone_start_init, NULL, cid,
3977 minclsyspri - 1, &ct, 0)) != 0) {
3978 mutex_enter(&zone_status_lock);
3979 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
3980 mutex_exit(&zone_status_lock);
3981 } else {
3982 zone->zone_boot_time = gethrestime_sec();
3983 }
3984
3985 pool_unlock();
3986 }
3987
3988 /*
3989 * Wait for zone_destroy() to be called. This is what we spend
3990 * most of our life doing.
3991 */
3992 zone_status_wait_cpr(zone, ZONE_IS_DYING, "zsched");
3993
3994 if (ct)
3995 /*
3996 * At this point the process contract should be empty.
3997 * (Though if it isn't, it's not the end of the world.)
3998 */
3999 VERIFY(contract_abandon(ct, curproc, B_TRUE) == 0);
4000
4001 /*
4002 * Allow kcred to be freed when all referring processes
4003 * (including this one) go away. We can't just do this in
4004 * zone_free because we need to wait for the zone_cred_ref to
4005 * drop to 0 before calling zone_free, and the existence of
4006 * zone_kcred will prevent that. Thus, we call crfree here to
4007 * balance the crdup in zone_create. The crhold calls earlier
4008 * in zsched will be dropped when the thread and process exit.
4009 */
4010 crfree(zone->zone_kcred);
4011 zone->zone_kcred = NULL;
4012
4013 exit(CLD_EXITED, 0);
4014 }
4015
4016 /*
4017 * Helper function to determine if there are any submounts of the
4018 * provided path. Used to make sure the zone doesn't "inherit" any
4019 * mounts from before it is created.
4020 */
4021 static uint_t
4022 zone_mount_count(const char *rootpath)
4023 {
4024 vfs_t *vfsp;
4025 uint_t count = 0;
4026 size_t rootpathlen = strlen(rootpath);
4027
4028 /*
4029 * Holding zonehash_lock prevents race conditions with
4030 * vfs_list_add()/vfs_list_remove() since we serialize with
4031 * zone_find_by_path().
4032 */
4033 ASSERT(MUTEX_HELD(&zonehash_lock));
4034 /*
4035 * The rootpath must end with a '/'
4036 */
4037 ASSERT(rootpath[rootpathlen - 1] == '/');
4038
4039 /*
4040 * This intentionally does not count the rootpath itself if that
4041 * happens to be a mount point.
4042 */
4043 vfs_list_read_lock();
4044 vfsp = rootvfs;
4045 do {
4046 if (strncmp(rootpath, refstr_value(vfsp->vfs_mntpt),
4047 rootpathlen) == 0)
4048 count++;
4049 vfsp = vfsp->vfs_next;
4050 } while (vfsp != rootvfs);
4051 vfs_list_unlock();
4052 return (count);
4053 }
4054
4055 /*
4056 * Helper function to make sure that a zone created on 'rootpath'
4057 * wouldn't end up containing other zones' rootpaths.
4058 */
4059 static boolean_t
4060 zone_is_nested(const char *rootpath)
4061 {
4062 zone_t *zone;
4063 size_t rootpathlen = strlen(rootpath);
4064 size_t len;
4065
4066 ASSERT(MUTEX_HELD(&zonehash_lock));
4067
4068 /*
4069 * zone_set_root() appended '/' and '\0' at the end of rootpath
4070 */
4071 if ((rootpathlen <= 3) && (rootpath[0] == '/') &&
4072 (rootpath[1] == '/') && (rootpath[2] == '\0'))
4073 return (B_TRUE);
4074
4075 for (zone = list_head(&zone_active); zone != NULL;
4076 zone = list_next(&zone_active, zone)) {
4077 if (zone == global_zone)
4078 continue;
4079 len = strlen(zone->zone_rootpath);
4080 if (strncmp(rootpath, zone->zone_rootpath,
4081 MIN(rootpathlen, len)) == 0)
4082 return (B_TRUE);
4083 }
4084 return (B_FALSE);
4085 }
4086
4087 static int
4088 zone_set_privset(zone_t *zone, const priv_set_t *zone_privs,
4089 size_t zone_privssz)
4090 {
4091 priv_set_t *privs;
4092
4093 if (zone_privssz < sizeof (priv_set_t))
4094 return (ENOMEM);
4095
4096 privs = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
4097
4098 if (copyin(zone_privs, privs, sizeof (priv_set_t))) {
4099 kmem_free(privs, sizeof (priv_set_t));
4100 return (EFAULT);
4101 }
4102
4103 zone->zone_privset = privs;
4104 return (0);
4105 }
4106
4107 /*
4108 * We make creative use of nvlists to pass in rctls from userland. The list is
4109 * a list of the following structures:
4110 *
4111 * (name = rctl_name, value = nvpair_list_array)
4112 *
4113 * Where each element of the nvpair_list_array is of the form:
4114 *
4115 * [(name = "privilege", value = RCPRIV_PRIVILEGED),
4116 * (name = "limit", value = uint64_t),
4117 * (name = "action", value = (RCTL_LOCAL_NOACTION || RCTL_LOCAL_DENY))]
4118 */
4119 static int
4120 parse_rctls(caddr_t ubuf, size_t buflen, nvlist_t **nvlp)
4121 {
4122 nvpair_t *nvp = NULL;
4123 nvlist_t *nvl = NULL;
4124 char *kbuf;
4125 int error;
4126 rctl_val_t rv;
4127
4128 *nvlp = NULL;
4129
4130 if (buflen == 0)
4131 return (0);
4132
4133 if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL)
4134 return (ENOMEM);
4135 if (copyin(ubuf, kbuf, buflen)) {
4136 error = EFAULT;
4137 goto out;
4138 }
4139 if (nvlist_unpack(kbuf, buflen, &nvl, KM_SLEEP) != 0) {
4140 /*
4141 * nvl may have been allocated/free'd, but the value set to
4142 * non-NULL, so we reset it here.
4143 */
4144 nvl = NULL;
4145 error = EINVAL;
4146 goto out;
4147 }
4148 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
4149 rctl_dict_entry_t *rde;
4150 rctl_hndl_t hndl;
4151 nvlist_t **nvlarray;
4152 uint_t i, nelem;
4153 char *name;
4154
4155 error = EINVAL;
4156 name = nvpair_name(nvp);
4157 if (strncmp(nvpair_name(nvp), "zone.", sizeof ("zone.") - 1)
4158 != 0 || nvpair_type(nvp) != DATA_TYPE_NVLIST_ARRAY) {
4159 goto out;
4160 }
4161 if ((hndl = rctl_hndl_lookup(name)) == -1) {
4162 goto out;
4163 }
4164 rde = rctl_dict_lookup_hndl(hndl);
4165 error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
4166 ASSERT(error == 0);
4167 for (i = 0; i < nelem; i++) {
4168 if (error = nvlist2rctlval(nvlarray[i], &rv))
4169 goto out;
4170 }
4171 if (rctl_invalid_value(rde, &rv)) {
4172 error = EINVAL;
4173 goto out;
4174 }
4175 }
4176 error = 0;
4177 *nvlp = nvl;
4178 out:
4179 kmem_free(kbuf, buflen);
4180 if (error && nvl != NULL)
4181 nvlist_free(nvl);
4182 return (error);
4183 }
4184
4185 int
4186 zone_create_error(int er_error, int er_ext, int *er_out) {
4187 if (er_out != NULL) {
4188 if (copyout(&er_ext, er_out, sizeof (int))) {
4189 return (set_errno(EFAULT));
4190 }
4191 }
4192 return (set_errno(er_error));
4193 }
4194
4195 static int
4196 zone_set_label(zone_t *zone, const bslabel_t *lab, uint32_t doi)
4197 {
4198 ts_label_t *tsl;
4199 bslabel_t blab;
4200
4201 /* Get label from user */
4202 if (copyin(lab, &blab, sizeof (blab)) != 0)
4203 return (EFAULT);
4204 tsl = labelalloc(&blab, doi, KM_NOSLEEP);
4205 if (tsl == NULL)
4206 return (ENOMEM);
4207
4208 zone->zone_slabel = tsl;
4209 return (0);
4210 }
4211
4212 /*
4213 * Parses a comma-separated list of ZFS datasets into a per-zone dictionary.
4214 */
4215 static int
4216 parse_zfs(zone_t *zone, caddr_t ubuf, size_t buflen)
4217 {
4218 char *kbuf;
4219 char *dataset, *next;
4220 zone_dataset_t *zd;
4221 size_t len;
4222
4223 if (ubuf == NULL || buflen == 0)
4224 return (0);
4225
4226 if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL)
4227 return (ENOMEM);
4228
4229 if (copyin(ubuf, kbuf, buflen) != 0) {
4230 kmem_free(kbuf, buflen);
4231 return (EFAULT);
4232 }
4233
4234 dataset = next = kbuf;
4235 for (;;) {
4236 zd = kmem_alloc(sizeof (zone_dataset_t), KM_SLEEP);
4237
4238 next = strchr(dataset, ',');
4239
4240 if (next == NULL)
4241 len = strlen(dataset);
4242 else
4243 len = next - dataset;
4244
4245 zd->zd_dataset = kmem_alloc(len + 1, KM_SLEEP);
4246 bcopy(dataset, zd->zd_dataset, len);
4247 zd->zd_dataset[len] = '\0';
4248
4249 list_insert_head(&zone->zone_datasets, zd);
4250
4251 if (next == NULL)
4252 break;
4253
4254 dataset = next + 1;
4255 }
4256
4257 kmem_free(kbuf, buflen);
4258 return (0);
4259 }
4260
4261 /*
4262 * System call to create/initialize a new zone named 'zone_name', rooted
4263 * at 'zone_root', with a zone-wide privilege limit set of 'zone_privs',
4264 * and initialized with the zone-wide rctls described in 'rctlbuf', and
4265 * with labeling set by 'match', 'doi', and 'label'.
4266 *
4267 * If extended error is non-null, we may use it to return more detailed
4268 * error information.
4269 */
4270 static zoneid_t
4271 zone_create(const char *zone_name, const char *zone_root,
4272 const priv_set_t *zone_privs, size_t zone_privssz,
4273 caddr_t rctlbuf, size_t rctlbufsz,
4274 caddr_t zfsbuf, size_t zfsbufsz, int *extended_error,
4275 int match, uint32_t doi, const bslabel_t *label,
4276 int flags)
4277 {
4278 struct zsched_arg zarg;
4279 nvlist_t *rctls = NULL;
4280 proc_t *pp = curproc;
4281 zone_t *zone, *ztmp;
4282 zoneid_t zoneid;
4283 int error;
4284 int error2 = 0;
4285 char *str;
4286 cred_t *zkcr;
4287 boolean_t insert_label_hash;
4288
4289 if (secpolicy_zone_config(CRED()) != 0)
4290 return (set_errno(EPERM));
4291
4292 /* can't boot zone from within chroot environment */
4293 if (PTOU(pp)->u_rdir != NULL && PTOU(pp)->u_rdir != rootdir)
4294 return (zone_create_error(ENOTSUP, ZE_CHROOTED,
4295 extended_error));
4296
4297 zone = kmem_zalloc(sizeof (zone_t), KM_SLEEP);
4298 zoneid = zone->zone_id = id_alloc(zoneid_space);
4299 zone->zone_status = ZONE_IS_UNINITIALIZED;
4300 zone->zone_pool = pool_default;
4301 zone->zone_pool_mod = gethrtime();
4302 zone->zone_psetid = ZONE_PS_INVAL;
4303 zone->zone_ncpus = 0;
4304 zone->zone_ncpus_online = 0;
4305 zone->zone_restart_init = B_TRUE;
4306 zone->zone_brand = &native_brand;
4307 zone->zone_initname = NULL;
4308 mutex_init(&zone->zone_lock, NULL, MUTEX_DEFAULT, NULL);
4309 mutex_init(&zone->zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
4310 mutex_init(&zone->zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
4311 cv_init(&zone->zone_cv, NULL, CV_DEFAULT, NULL);
4312 list_create(&zone->zone_ref_list, sizeof (zone_ref_t),
4313 offsetof(zone_ref_t, zref_linkage));
4314 list_create(&zone->zone_zsd, sizeof (struct zsd_entry),
4315 offsetof(struct zsd_entry, zsd_linkage));
4316 list_create(&zone->zone_datasets, sizeof (zone_dataset_t),
4317 offsetof(zone_dataset_t, zd_linkage));
4318 list_create(&zone->zone_dl_list, sizeof (zone_dl_t),
4319 offsetof(zone_dl_t, zdl_linkage));
4320 rw_init(&zone->zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
4321 rw_init(&zone->zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL);
4322
4323 if (flags & ZCF_NET_EXCL) {
4324 zone->zone_flags |= ZF_NET_EXCL;
4325 }
4326
4327 if ((error = zone_set_name(zone, zone_name)) != 0) {
4328 zone_free(zone);
4329 return (zone_create_error(error, 0, extended_error));
4330 }
4331
4332 if ((error = zone_set_root(zone, zone_root)) != 0) {
4333 zone_free(zone);
4334 return (zone_create_error(error, 0, extended_error));
4335 }
4336 if ((error = zone_set_privset(zone, zone_privs, zone_privssz)) != 0) {
4337 zone_free(zone);
4338 return (zone_create_error(error, 0, extended_error));
4339 }
4340
4341 /* initialize node name to be the same as zone name */
4342 zone->zone_nodename = kmem_alloc(_SYS_NMLN, KM_SLEEP);
4343 (void) strncpy(zone->zone_nodename, zone->zone_name, _SYS_NMLN);
4344 zone->zone_nodename[_SYS_NMLN - 1] = '\0';
4345
4346 zone->zone_domain = kmem_alloc(_SYS_NMLN, KM_SLEEP);
4347 zone->zone_domain[0] = '\0';
4348 zone->zone_hostid = HW_INVALID_HOSTID;
4349 zone->zone_shares = 1;
4350 zone->zone_shmmax = 0;
4351 zone->zone_ipc.ipcq_shmmni = 0;
4352 zone->zone_ipc.ipcq_semmni = 0;
4353 zone->zone_ipc.ipcq_msgmni = 0;
4354 zone->zone_bootargs = NULL;
4355 zone->zone_fs_allowed = NULL;
4356 zone->zone_initname =
4357 kmem_alloc(strlen(zone_default_initname) + 1, KM_SLEEP);
4358 (void) strcpy(zone->zone_initname, zone_default_initname);
4359 zone->zone_nlwps = 0;
4360 zone->zone_nlwps_ctl = INT_MAX;
4361 zone->zone_nprocs = 0;
4362 zone->zone_nprocs_ctl = INT_MAX;
4363 zone->zone_locked_mem = 0;
4364 zone->zone_locked_mem_ctl = UINT64_MAX;
4365 zone->zone_max_swap = 0;
4366 zone->zone_max_swap_ctl = UINT64_MAX;
4367 zone->zone_max_lofi = 0;
4368 zone->zone_max_lofi_ctl = UINT64_MAX;
4369 zone0.zone_lockedmem_kstat = NULL;
4370 zone0.zone_swapresv_kstat = NULL;
4371
4372 /*
4373 * Zsched initializes the rctls.
4374 */
4375 zone->zone_rctls = NULL;
4376
4377 if ((error = parse_rctls(rctlbuf, rctlbufsz, &rctls)) != 0) {
4378 zone_free(zone);
4379 return (zone_create_error(error, 0, extended_error));
4380 }
4381
4382 if ((error = parse_zfs(zone, zfsbuf, zfsbufsz)) != 0) {
4383 zone_free(zone);
4384 return (set_errno(error));
4385 }
4386
4387 /*
4388 * Read in the trusted system parameters:
4389 * match flag and sensitivity label.
4390 */
4391 zone->zone_match = match;
4392 if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) {
4393 /* Fail if requested to set doi to anything but system's doi */
4394 if (doi != 0 && doi != default_doi) {
4395 zone_free(zone);
4396 return (set_errno(EINVAL));
4397 }
4398 /* Always apply system's doi to the zone */
4399 error = zone_set_label(zone, label, default_doi);
4400 if (error != 0) {
4401 zone_free(zone);
4402 return (set_errno(error));
4403 }
4404 insert_label_hash = B_TRUE;
4405 } else {
4406 /* all zones get an admin_low label if system is not labeled */
4407 zone->zone_slabel = l_admin_low;
4408 label_hold(l_admin_low);
4409 insert_label_hash = B_FALSE;
4410 }
4411
4412 /*
4413 * Stop all lwps since that's what normally happens as part of fork().
4414 * This needs to happen before we grab any locks to avoid deadlock
4415 * (another lwp in the process could be waiting for the held lock).
4416 */
4417 if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK)) {
4418 zone_free(zone);
4419 if (rctls)
4420 nvlist_free(rctls);
4421 return (zone_create_error(error, 0, extended_error));
4422 }
4423
4424 if (block_mounts(zone) == 0) {
4425 mutex_enter(&pp->p_lock);
4426 if (curthread != pp->p_agenttp)
4427 continuelwps(pp);
4428 mutex_exit(&pp->p_lock);
4429 zone_free(zone);
4430 if (rctls)
4431 nvlist_free(rctls);
4432 return (zone_create_error(error, 0, extended_error));
4433 }
4434
4435 /*
4436 * Set up credential for kernel access. After this, any errors
4437 * should go through the dance in errout rather than calling
4438 * zone_free directly.
4439 */
4440 zone->zone_kcred = crdup(kcred);
4441 crsetzone(zone->zone_kcred, zone);
4442 priv_intersect(zone->zone_privset, &CR_PPRIV(zone->zone_kcred));
4443 priv_intersect(zone->zone_privset, &CR_EPRIV(zone->zone_kcred));
4444 priv_intersect(zone->zone_privset, &CR_IPRIV(zone->zone_kcred));
4445 priv_intersect(zone->zone_privset, &CR_LPRIV(zone->zone_kcred));
4446
4447 mutex_enter(&zonehash_lock);
4448 /*
4449 * Make sure zone doesn't already exist.
4450 *
4451 * If the system and zone are labeled,
4452 * make sure no other zone exists that has the same label.
4453 */
4454 if ((ztmp = zone_find_all_by_name(zone->zone_name)) != NULL ||
4455 (insert_label_hash &&
4456 (ztmp = zone_find_all_by_label(zone->zone_slabel)) != NULL)) {
4457 zone_status_t status;
4458
4459 status = zone_status_get(ztmp);
4460 if (status == ZONE_IS_READY || status == ZONE_IS_RUNNING)
4461 error = EEXIST;
4462 else
4463 error = EBUSY;
4464
4465 if (insert_label_hash)
4466 error2 = ZE_LABELINUSE;
4467
4468 goto errout;
4469 }
4470
4471 /*
4472 * Don't allow zone creations which would cause one zone's rootpath to
4473 * be accessible from that of another (non-global) zone.
4474 */
4475 if (zone_is_nested(zone->zone_rootpath)) {
4476 error = EBUSY;
4477 goto errout;
4478 }
4479
4480 ASSERT(zonecount != 0); /* check for leaks */
4481 if (zonecount + 1 > maxzones) {
4482 error = ENOMEM;
4483 goto errout;
4484 }
4485
4486 if (zone_mount_count(zone->zone_rootpath) != 0) {
4487 error = EBUSY;
4488 error2 = ZE_AREMOUNTS;
4489 goto errout;
4490 }
4491
4492 /*
4493 * Zone is still incomplete, but we need to drop all locks while
4494 * zsched() initializes this zone's kernel process. We
4495 * optimistically add the zone to the hashtable and associated
4496 * lists so a parallel zone_create() doesn't try to create the
4497 * same zone.
4498 */
4499 zonecount++;
4500 (void) mod_hash_insert(zonehashbyid,
4501 (mod_hash_key_t)(uintptr_t)zone->zone_id,
4502 (mod_hash_val_t)(uintptr_t)zone);
4503 str = kmem_alloc(strlen(zone->zone_name) + 1, KM_SLEEP);
4504 (void) strcpy(str, zone->zone_name);
4505 (void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)str,
4506 (mod_hash_val_t)(uintptr_t)zone);
4507 if (insert_label_hash) {
4508 (void) mod_hash_insert(zonehashbylabel,
4509 (mod_hash_key_t)zone->zone_slabel, (mod_hash_val_t)zone);
4510 zone->zone_flags |= ZF_HASHED_LABEL;
4511 }
4512
4513 /*
4514 * Insert into active list. At this point there are no 'hold's
4515 * on the zone, but everyone else knows not to use it, so we can
4516 * continue to use it. zsched() will do a zone_hold() if the
4517 * newproc() is successful.
4518 */
4519 list_insert_tail(&zone_active, zone);
4520 mutex_exit(&zonehash_lock);
4521
4522 zarg.zone = zone;
4523 zarg.nvlist = rctls;
4524 /*
4525 * The process, task, and project rctls are probably wrong;
4526 * we need an interface to get the default values of all rctls,
4527 * and initialize zsched appropriately. I'm not sure that that
4528 * makes much of a difference, though.
4529 */
4530 error = newproc(zsched, (void *)&zarg, syscid, minclsyspri, NULL, 0);
4531 if (error != 0) {
4532 /*
4533 * We need to undo all globally visible state.
4534 */
4535 mutex_enter(&zonehash_lock);
4536 list_remove(&zone_active, zone);
4537 if (zone->zone_flags & ZF_HASHED_LABEL) {
4538 ASSERT(zone->zone_slabel != NULL);
4539 (void) mod_hash_destroy(zonehashbylabel,
4540 (mod_hash_key_t)zone->zone_slabel);
4541 }
4542 (void) mod_hash_destroy(zonehashbyname,
4543 (mod_hash_key_t)(uintptr_t)zone->zone_name);
4544 (void) mod_hash_destroy(zonehashbyid,
4545 (mod_hash_key_t)(uintptr_t)zone->zone_id);
4546 ASSERT(zonecount > 1);
4547 zonecount--;
4548 goto errout;
4549 }
4550
4551 /*
4552 * Zone creation can't fail from now on.
4553 */
4554
4555 /*
4556 * Create zone kstats
4557 */
4558 zone_kstat_create(zone);
4559
4560 /*
4561 * Let the other lwps continue.
4562 */
4563 mutex_enter(&pp->p_lock);
4564 if (curthread != pp->p_agenttp)
4565 continuelwps(pp);
4566 mutex_exit(&pp->p_lock);
4567
4568 /*
4569 * Wait for zsched to finish initializing the zone.
4570 */
4571 zone_status_wait(zone, ZONE_IS_READY);
4572 /*
4573 * The zone is fully visible, so we can let mounts progress.
4574 */
4575 resume_mounts(zone);
4576 if (rctls)
4577 nvlist_free(rctls);
4578
4579 return (zoneid);
4580
4581 errout:
4582 mutex_exit(&zonehash_lock);
4583 /*
4584 * Let the other lwps continue.
4585 */
4586 mutex_enter(&pp->p_lock);
4587 if (curthread != pp->p_agenttp)
4588 continuelwps(pp);
4589 mutex_exit(&pp->p_lock);
4590
4591 resume_mounts(zone);
4592 if (rctls)
4593 nvlist_free(rctls);
4594 /*
4595 * There is currently one reference to the zone, a cred_ref from
4596 * zone_kcred. To free the zone, we call crfree, which will call
4597 * zone_cred_rele, which will call zone_free.
4598 */
4599 ASSERT(zone->zone_cred_ref == 1);
4600 ASSERT(zone->zone_kcred->cr_ref == 1);
4601 ASSERT(zone->zone_ref == 0);
4602 zkcr = zone->zone_kcred;
4603 zone->zone_kcred = NULL;
4604 crfree(zkcr); /* triggers call to zone_free */
4605 return (zone_create_error(error, error2, extended_error));
4606 }
4607
4608 /*
4609 * Cause the zone to boot. This is pretty simple, since we let zoneadmd do
4610 * the heavy lifting. initname is the path to the program to launch
4611 * at the "top" of the zone; if this is NULL, we use the system default,
4612 * which is stored at zone_default_initname.
4613 */
4614 static int
4615 zone_boot(zoneid_t zoneid)
4616 {
4617 int err;
4618 zone_t *zone;
4619
4620 if (secpolicy_zone_config(CRED()) != 0)
4621 return (set_errno(EPERM));
4622 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4623 return (set_errno(EINVAL));
4624
4625 mutex_enter(&zonehash_lock);
4626 /*
4627 * Look for zone under hash lock to prevent races with calls to
4628 * zone_shutdown, zone_destroy, etc.
4629 */
4630 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4631 mutex_exit(&zonehash_lock);
4632 return (set_errno(EINVAL));
4633 }
4634
4635 mutex_enter(&zone_status_lock);
4636 if (zone_status_get(zone) != ZONE_IS_READY) {
4637 mutex_exit(&zone_status_lock);
4638 mutex_exit(&zonehash_lock);
4639 return (set_errno(EINVAL));
4640 }
4641 zone_status_set(zone, ZONE_IS_BOOTING);
4642 mutex_exit(&zone_status_lock);
4643
4644 zone_hold(zone); /* so we can use the zone_t later */
4645 mutex_exit(&zonehash_lock);
4646
4647 if (zone_status_wait_sig(zone, ZONE_IS_RUNNING) == 0) {
4648 zone_rele(zone);
4649 return (set_errno(EINTR));
4650 }
4651
4652 /*
4653 * Boot (starting init) might have failed, in which case the zone
4654 * will go to the SHUTTING_DOWN state; an appropriate errno will
4655 * be placed in zone->zone_boot_err, and so we return that.
4656 */
4657 err = zone->zone_boot_err;
4658 zone_rele(zone);
4659 return (err ? set_errno(err) : 0);
4660 }
4661
4662 /*
4663 * Kills all user processes in the zone, waiting for them all to exit
4664 * before returning.
4665 */
4666 static int
4667 zone_empty(zone_t *zone)
4668 {
4669 int waitstatus;
4670
4671 /*
4672 * We need to drop zonehash_lock before killing all
4673 * processes, otherwise we'll deadlock with zone_find_*
4674 * which can be called from the exit path.
4675 */
4676 ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
4677 while ((waitstatus = zone_status_timedwait_sig(zone,
4678 ddi_get_lbolt() + hz, ZONE_IS_EMPTY)) == -1) {
4679 killall(zone->zone_id);
4680 }
4681 /*
4682 * return EINTR if we were signaled
4683 */
4684 if (waitstatus == 0)
4685 return (EINTR);
4686 return (0);
4687 }
4688
4689 /*
4690 * This function implements the policy for zone visibility.
4691 *
4692 * In standard Solaris, a non-global zone can only see itself.
4693 *
4694 * In Trusted Extensions, a labeled zone can lookup any zone whose label
4695 * it dominates. For this test, the label of the global zone is treated as
4696 * admin_high so it is special-cased instead of being checked for dominance.
4697 *
4698 * Returns true if zone attributes are viewable, false otherwise.
4699 */
4700 static boolean_t
4701 zone_list_access(zone_t *zone)
4702 {
4703
4704 if (curproc->p_zone == global_zone ||
4705 curproc->p_zone == zone) {
4706 return (B_TRUE);
4707 } else if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) {
4708 bslabel_t *curproc_label;
4709 bslabel_t *zone_label;
4710
4711 curproc_label = label2bslabel(curproc->p_zone->zone_slabel);
4712 zone_label = label2bslabel(zone->zone_slabel);
4713
4714 if (zone->zone_id != GLOBAL_ZONEID &&
4715 bldominates(curproc_label, zone_label)) {
4716 return (B_TRUE);
4717 } else {
4718 return (B_FALSE);
4719 }
4720 } else {
4721 return (B_FALSE);
4722 }
4723 }
4724
4725 /*
4726 * Systemcall to start the zone's halt sequence. By the time this
4727 * function successfully returns, all user processes and kernel threads
4728 * executing in it will have exited, ZSD shutdown callbacks executed,
4729 * and the zone status set to ZONE_IS_DOWN.
4730 *
4731 * It is possible that the call will interrupt itself if the caller is the
4732 * parent of any process running in the zone, and doesn't have SIGCHLD blocked.
4733 */
4734 static int
4735 zone_shutdown(zoneid_t zoneid)
4736 {
4737 int error;
4738 zone_t *zone;
4739 zone_status_t status;
4740
4741 if (secpolicy_zone_config(CRED()) != 0)
4742 return (set_errno(EPERM));
4743 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4744 return (set_errno(EINVAL));
4745
4746 mutex_enter(&zonehash_lock);
4747 /*
4748 * Look for zone under hash lock to prevent races with other
4749 * calls to zone_shutdown and zone_destroy.
4750 */
4751 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4752 mutex_exit(&zonehash_lock);
4753 return (set_errno(EINVAL));
4754 }
4755
4756 /*
4757 * We have to drop zonehash_lock before calling block_mounts.
4758 * Hold the zone so we can continue to use the zone_t.
4759 */
4760 zone_hold(zone);
4761 mutex_exit(&zonehash_lock);
4762
4763 /*
4764 * Block mounts so that VFS_MOUNT() can get an accurate view of
4765 * the zone's status with regards to ZONE_IS_SHUTTING down.
4766 *
4767 * e.g. NFS can fail the mount if it determines that the zone
4768 * has already begun the shutdown sequence.
4769 *
4770 */
4771 if (block_mounts(zone) == 0) {
4772 zone_rele(zone);
4773 return (set_errno(EINTR));
4774 }
4775
4776 mutex_enter(&zonehash_lock);
4777 mutex_enter(&zone_status_lock);
4778 status = zone_status_get(zone);
4779 /*
4780 * Fail if the zone isn't fully initialized yet.
4781 */
4782 if (status < ZONE_IS_READY) {
4783 mutex_exit(&zone_status_lock);
4784 mutex_exit(&zonehash_lock);
4785 resume_mounts(zone);
4786 zone_rele(zone);
4787 return (set_errno(EINVAL));
4788 }
4789 /*
4790 * If conditions required for zone_shutdown() to return have been met,
4791 * return success.
4792 */
4793 if (status >= ZONE_IS_DOWN) {
4794 mutex_exit(&zone_status_lock);
4795 mutex_exit(&zonehash_lock);
4796 resume_mounts(zone);
4797 zone_rele(zone);
4798 return (0);
4799 }
4800 /*
4801 * If zone_shutdown() hasn't been called before, go through the motions.
4802 * If it has, there's nothing to do but wait for the kernel threads to
4803 * drain.
4804 */
4805 if (status < ZONE_IS_EMPTY) {
4806 uint_t ntasks;
4807
4808 mutex_enter(&zone->zone_lock);
4809 if ((ntasks = zone->zone_ntasks) != 1) {
4810 /*
4811 * There's still stuff running.
4812 */
4813 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
4814 }
4815 mutex_exit(&zone->zone_lock);
4816 if (ntasks == 1) {
4817 /*
4818 * The only way to create another task is through
4819 * zone_enter(), which will block until we drop
4820 * zonehash_lock. The zone is empty.
4821 */
4822 if (zone->zone_kthreads == NULL) {
4823 /*
4824 * Skip ahead to ZONE_IS_DOWN
4825 */
4826 zone_status_set(zone, ZONE_IS_DOWN);
4827 } else {
4828 zone_status_set(zone, ZONE_IS_EMPTY);
4829 }
4830 }
4831 }
4832 mutex_exit(&zone_status_lock);
4833 mutex_exit(&zonehash_lock);
4834 resume_mounts(zone);
4835
4836 if (error = zone_empty(zone)) {
4837 zone_rele(zone);
4838 return (set_errno(error));
4839 }
4840 /*
4841 * After the zone status goes to ZONE_IS_DOWN this zone will no
4842 * longer be notified of changes to the pools configuration, so
4843 * in order to not end up with a stale pool pointer, we point
4844 * ourselves at the default pool and remove all resource
4845 * visibility. This is especially important as the zone_t may
4846 * languish on the deathrow for a very long time waiting for
4847 * cred's to drain out.
4848 *
4849 * This rebinding of the zone can happen multiple times
4850 * (presumably due to interrupted or parallel systemcalls)
4851 * without any adverse effects.
4852 */
4853 if (pool_lock_intr() != 0) {
4854 zone_rele(zone);
4855 return (set_errno(EINTR));
4856 }
4857 if (pool_state == POOL_ENABLED) {
4858 mutex_enter(&cpu_lock);
4859 zone_pool_set(zone, pool_default);
4860 /*
4861 * The zone no longer needs to be able to see any cpus.
4862 */
4863 zone_pset_set(zone, ZONE_PS_INVAL);
4864 mutex_exit(&cpu_lock);
4865 }
4866 pool_unlock();
4867
4868 /*
4869 * ZSD shutdown callbacks can be executed multiple times, hence
4870 * it is safe to not be holding any locks across this call.
4871 */
4872 zone_zsd_callbacks(zone, ZSD_SHUTDOWN);
4873
4874 mutex_enter(&zone_status_lock);
4875 if (zone->zone_kthreads == NULL && zone_status_get(zone) < ZONE_IS_DOWN)
4876 zone_status_set(zone, ZONE_IS_DOWN);
4877 mutex_exit(&zone_status_lock);
4878
4879 /*
4880 * Wait for kernel threads to drain.
4881 */
4882 if (!zone_status_wait_sig(zone, ZONE_IS_DOWN)) {
4883 zone_rele(zone);
4884 return (set_errno(EINTR));
4885 }
4886
4887 /*
4888 * Zone can be become down/destroyable even if the above wait
4889 * returns EINTR, so any code added here may never execute.
4890 * (i.e. don't add code here)
4891 */
4892
4893 zone_rele(zone);
4894 return (0);
4895 }
4896
4897 /*
4898 * Log the specified zone's reference counts. The caller should not be
4899 * holding the zone's zone_lock.
4900 */
4901 static void
4902 zone_log_refcounts(zone_t *zone)
4903 {
4904 char *buffer;
4905 char *buffer_position;
4906 uint32_t buffer_size;
4907 uint32_t index;
4908 uint_t ref;
4909 uint_t cred_ref;
4910
4911 /*
4912 * Construct a string representing the subsystem-specific reference
4913 * counts. The counts are printed in ascending order by index into the
4914 * zone_t::zone_subsys_ref array. The list will be surrounded by
4915 * square brackets [] and will only contain nonzero reference counts.
4916 *
4917 * The buffer will hold two square bracket characters plus ten digits,
4918 * one colon, one space, one comma, and some characters for a
4919 * subsystem name per subsystem-specific reference count. (Unsigned 32-
4920 * bit integers have at most ten decimal digits.) The last
4921 * reference count's comma is replaced by the closing square
4922 * bracket and a NULL character to terminate the string.
4923 *
4924 * NOTE: We have to grab the zone's zone_lock to create a consistent
4925 * snapshot of the zone's reference counters.
4926 *
4927 * First, figure out how much space the string buffer will need.
4928 * The buffer's size is stored in buffer_size.
4929 */
4930 buffer_size = 2; /* for the square brackets */
4931 mutex_enter(&zone->zone_lock);
4932 zone->zone_flags |= ZF_REFCOUNTS_LOGGED;
4933 ref = zone->zone_ref;
4934 cred_ref = zone->zone_cred_ref;
4935 for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index)
4936 if (zone->zone_subsys_ref[index] != 0)
4937 buffer_size += strlen(zone_ref_subsys_names[index]) +
4938 13;
4939 if (buffer_size == 2) {
4940 /*
4941 * No subsystems had nonzero reference counts. Don't bother
4942 * with allocating a buffer; just log the general-purpose and
4943 * credential reference counts.
4944 */
4945 mutex_exit(&zone->zone_lock);
4946 (void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE,
4947 "Zone '%s' (ID: %d) is shutting down, but %u zone "
4948 "references and %u credential references are still extant",
4949 zone->zone_name, zone->zone_id, ref, cred_ref);
4950 return;
4951 }
4952
4953 /*
4954 * buffer_size contains the exact number of characters that the
4955 * buffer will need. Allocate the buffer and fill it with nonzero
4956 * subsystem-specific reference counts. Surround the results with
4957 * square brackets afterwards.
4958 */
4959 buffer = kmem_alloc(buffer_size, KM_SLEEP);
4960 buffer_position = &buffer[1];
4961 for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index) {
4962 /*
4963 * NOTE: The DDI's version of sprintf() returns a pointer to
4964 * the modified buffer rather than the number of bytes written
4965 * (as in snprintf(3C)). This is unfortunate and annoying.
4966 * Therefore, we'll use snprintf() with INT_MAX to get the
4967 * number of bytes written. Using INT_MAX is safe because
4968 * the buffer is perfectly sized for the data: we'll never
4969 * overrun the buffer.
4970 */
4971 if (zone->zone_subsys_ref[index] != 0)
4972 buffer_position += snprintf(buffer_position, INT_MAX,
4973 "%s: %u,", zone_ref_subsys_names[index],
4974 zone->zone_subsys_ref[index]);
4975 }
4976 mutex_exit(&zone->zone_lock);
4977 buffer[0] = '[';
4978 ASSERT((uintptr_t)(buffer_position - buffer) < buffer_size);
4979 ASSERT(buffer_position[0] == '\0' && buffer_position[-1] == ',');
4980 buffer_position[-1] = ']';
4981
4982 /*
4983 * Log the reference counts and free the message buffer.
4984 */
4985 (void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE,
4986 "Zone '%s' (ID: %d) is shutting down, but %u zone references and "
4987 "%u credential references are still extant %s", zone->zone_name,
4988 zone->zone_id, ref, cred_ref, buffer);
4989 kmem_free(buffer, buffer_size);
4990 }
4991
4992 /*
4993 * Systemcall entry point to finalize the zone halt process. The caller
4994 * must have already successfully called zone_shutdown().
4995 *
4996 * Upon successful completion, the zone will have been fully destroyed:
4997 * zsched will have exited, destructor callbacks executed, and the zone
4998 * removed from the list of active zones.
4999 */
5000 static int
5001 zone_destroy(zoneid_t zoneid)
5002 {
5003 uint64_t uniqid;
5004 zone_t *zone;
5005 zone_status_t status;
5006 clock_t wait_time;
5007 boolean_t log_refcounts;
5008
5009 if (secpolicy_zone_config(CRED()) != 0)
5010 return (set_errno(EPERM));
5011 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
5012 return (set_errno(EINVAL));
5013
5014 mutex_enter(&zonehash_lock);
5015 /*
5016 * Look for zone under hash lock to prevent races with other
5017 * calls to zone_destroy.
5018 */
5019 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5020 mutex_exit(&zonehash_lock);
5021 return (set_errno(EINVAL));
5022 }
5023
5024 if (zone_mount_count(zone->zone_rootpath) != 0) {
5025 mutex_exit(&zonehash_lock);
5026 return (set_errno(EBUSY));
5027 }
5028 mutex_enter(&zone_status_lock);
5029 status = zone_status_get(zone);
5030 if (status < ZONE_IS_DOWN) {
5031 mutex_exit(&zone_status_lock);
5032 mutex_exit(&zonehash_lock);
5033 return (set_errno(EBUSY));
5034 } else if (status == ZONE_IS_DOWN) {
5035 zone_status_set(zone, ZONE_IS_DYING); /* Tell zsched to exit */
5036 }
5037 mutex_exit(&zone_status_lock);
5038 zone_hold(zone);
5039 mutex_exit(&zonehash_lock);
5040
5041 /*
5042 * wait for zsched to exit
5043 */
5044 zone_status_wait(zone, ZONE_IS_DEAD);
5045 zone_zsd_callbacks(zone, ZSD_DESTROY);
5046 zone->zone_netstack = NULL;
5047 uniqid = zone->zone_uniqid;
5048 zone_rele(zone);
5049 zone = NULL; /* potentially free'd */
5050
5051 log_refcounts = B_FALSE;
5052 wait_time = SEC_TO_TICK(ZONE_DESTROY_TIMEOUT_SECS);
5053 mutex_enter(&zonehash_lock);
5054 for (; /* ever */; ) {
5055 boolean_t unref;
5056 boolean_t refs_have_been_logged;
5057
5058 if ((zone = zone_find_all_by_id(zoneid)) == NULL ||
5059 zone->zone_uniqid != uniqid) {
5060 /*
5061 * The zone has gone away. Necessary conditions
5062 * are met, so we return success.
5063 */
5064 mutex_exit(&zonehash_lock);
5065 return (0);
5066 }
5067 mutex_enter(&zone->zone_lock);
5068 unref = ZONE_IS_UNREF(zone);
5069 refs_have_been_logged = (zone->zone_flags &
5070 ZF_REFCOUNTS_LOGGED);
5071 mutex_exit(&zone->zone_lock);
5072 if (unref) {
5073 /*
5074 * There is only one reference to the zone -- that
5075 * added when the zone was added to the hashtables --
5076 * and things will remain this way until we drop
5077 * zonehash_lock... we can go ahead and cleanup the
5078 * zone.
5079 */
5080 break;
5081 }
5082
5083 /*
5084 * Wait for zone_rele_common() or zone_cred_rele() to signal
5085 * zone_destroy_cv. zone_destroy_cv is signaled only when
5086 * some zone's general-purpose reference count reaches one.
5087 * If ZONE_DESTROY_TIMEOUT_SECS seconds elapse while waiting
5088 * on zone_destroy_cv, then log the zone's reference counts and
5089 * continue to wait for zone_rele() and zone_cred_rele().
5090 */
5091 if (!refs_have_been_logged) {
5092 if (!log_refcounts) {
5093 /*
5094 * This thread hasn't timed out waiting on
5095 * zone_destroy_cv yet. Wait wait_time clock
5096 * ticks (initially ZONE_DESTROY_TIMEOUT_SECS
5097 * seconds) for the zone's references to clear.
5098 */
5099 ASSERT(wait_time > 0);
5100 wait_time = cv_reltimedwait_sig(
5101 &zone_destroy_cv, &zonehash_lock, wait_time,
5102 TR_SEC);
5103 if (wait_time > 0) {
5104 /*
5105 * A thread in zone_rele() or
5106 * zone_cred_rele() signaled
5107 * zone_destroy_cv before this thread's
5108 * wait timed out. The zone might have
5109 * only one reference left; find out!
5110 */
5111 continue;
5112 } else if (wait_time == 0) {
5113 /* The thread's process was signaled. */
5114 mutex_exit(&zonehash_lock);
5115 return (set_errno(EINTR));
5116 }
5117
5118 /*
5119 * The thread timed out while waiting on
5120 * zone_destroy_cv. Even though the thread
5121 * timed out, it has to check whether another
5122 * thread woke up from zone_destroy_cv and
5123 * destroyed the zone.
5124 *
5125 * If the zone still exists and has more than
5126 * one unreleased general-purpose reference,
5127 * then log the zone's reference counts.
5128 */
5129 log_refcounts = B_TRUE;
5130 continue;
5131 }
5132
5133 /*
5134 * The thread already timed out on zone_destroy_cv while
5135 * waiting for subsystems to release the zone's last
5136 * general-purpose references. Log the zone's reference
5137 * counts and wait indefinitely on zone_destroy_cv.
5138 */
5139 zone_log_refcounts(zone);
5140 }
5141 if (cv_wait_sig(&zone_destroy_cv, &zonehash_lock) == 0) {
5142 /* The thread's process was signaled. */
5143 mutex_exit(&zonehash_lock);
5144 return (set_errno(EINTR));
5145 }
5146 }
5147
5148 /*
5149 * Remove CPU cap for this zone now since we're not going to
5150 * fail below this point.
5151 */
5152 cpucaps_zone_remove(zone);
5153
5154 /* Get rid of the zone's kstats */
5155 zone_kstat_delete(zone);
5156
5157 /* remove the pfexecd doors */
5158 if (zone->zone_pfexecd != NULL) {
5159 klpd_freelist(&zone->zone_pfexecd);
5160 zone->zone_pfexecd = NULL;
5161 }
5162
5163 /* free brand specific data */
5164 if (ZONE_IS_BRANDED(zone))
5165 ZBROP(zone)->b_free_brand_data(zone);
5166
5167 /* Say goodbye to brand framework. */
5168 brand_unregister_zone(zone->zone_brand);
5169
5170 /*
5171 * It is now safe to let the zone be recreated; remove it from the
5172 * lists. The memory will not be freed until the last cred
5173 * reference goes away.
5174 */
5175 ASSERT(zonecount > 1); /* must be > 1; can't destroy global zone */
5176 zonecount--;
5177 /* remove from active list and hash tables */
5178 list_remove(&zone_active, zone);
5179 (void) mod_hash_destroy(zonehashbyname,
5180 (mod_hash_key_t)zone->zone_name);
5181 (void) mod_hash_destroy(zonehashbyid,
5182 (mod_hash_key_t)(uintptr_t)zone->zone_id);
5183 if (zone->zone_flags & ZF_HASHED_LABEL)
5184 (void) mod_hash_destroy(zonehashbylabel,
5185 (mod_hash_key_t)zone->zone_slabel);
5186 mutex_exit(&zonehash_lock);
5187
5188 /*
5189 * Release the root vnode; we're not using it anymore. Nor should any
5190 * other thread that might access it exist.
5191 */
5192 if (zone->zone_rootvp != NULL) {
5193 VN_RELE(zone->zone_rootvp);
5194 zone->zone_rootvp = NULL;
5195 }
5196
5197 /* add to deathrow list */
5198 mutex_enter(&zone_deathrow_lock);
5199 list_insert_tail(&zone_deathrow, zone);
5200 mutex_exit(&zone_deathrow_lock);
5201
5202 /*
5203 * Drop last reference (which was added by zsched()), this will
5204 * free the zone unless there are outstanding cred references.
5205 */
5206 zone_rele(zone);
5207 return (0);
5208 }
5209
5210 /*
5211 * Systemcall entry point for zone_getattr(2).
5212 */
5213 static ssize_t
5214 zone_getattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
5215 {
5216 size_t size;
5217 int error = 0, err;
5218 zone_t *zone;
5219 char *zonepath;
5220 char *outstr;
5221 zone_status_t zone_status;
5222 pid_t initpid;
5223 boolean_t global = (curzone == global_zone);
5224 boolean_t inzone = (curzone->zone_id == zoneid);
5225 ushort_t flags;
5226 zone_net_data_t *zbuf;
5227
5228 mutex_enter(&zonehash_lock);
5229 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5230 mutex_exit(&zonehash_lock);
5231 return (set_errno(EINVAL));
5232 }
5233 zone_status = zone_status_get(zone);
5234 if (zone_status < ZONE_IS_INITIALIZED) {
5235 mutex_exit(&zonehash_lock);
5236 return (set_errno(EINVAL));
5237 }
5238 zone_hold(zone);
5239 mutex_exit(&zonehash_lock);
5240
5241 /*
5242 * If not in the global zone, don't show information about other zones,
5243 * unless the system is labeled and the local zone's label dominates
5244 * the other zone.
5245 */
5246 if (!zone_list_access(zone)) {
5247 zone_rele(zone);
5248 return (set_errno(EINVAL));
5249 }
5250
5251 switch (attr) {
5252 case ZONE_ATTR_ROOT:
5253 if (global) {
5254 /*
5255 * Copy the path to trim the trailing "/" (except for
5256 * the global zone).
5257 */
5258 if (zone != global_zone)
5259 size = zone->zone_rootpathlen - 1;
5260 else
5261 size = zone->zone_rootpathlen;
5262 zonepath = kmem_alloc(size, KM_SLEEP);
5263 bcopy(zone->zone_rootpath, zonepath, size);
5264 zonepath[size - 1] = '\0';
5265 } else {
5266 if (inzone || !is_system_labeled()) {
5267 /*
5268 * Caller is not in the global zone.
5269 * if the query is on the current zone
5270 * or the system is not labeled,
5271 * just return faked-up path for current zone.
5272 */
5273 zonepath = "/";
5274 size = 2;
5275 } else {
5276 /*
5277 * Return related path for current zone.
5278 */
5279 int prefix_len = strlen(zone_prefix);
5280 int zname_len = strlen(zone->zone_name);
5281
5282 size = prefix_len + zname_len + 1;
5283 zonepath = kmem_alloc(size, KM_SLEEP);
5284 bcopy(zone_prefix, zonepath, prefix_len);
5285 bcopy(zone->zone_name, zonepath +
5286 prefix_len, zname_len);
5287 zonepath[size - 1] = '\0';
5288 }
5289 }
5290 if (bufsize > size)
5291 bufsize = size;
5292 if (buf != NULL) {
5293 err = copyoutstr(zonepath, buf, bufsize, NULL);
5294 if (err != 0 && err != ENAMETOOLONG)
5295 error = EFAULT;
5296 }
5297 if (global || (is_system_labeled() && !inzone))
5298 kmem_free(zonepath, size);
5299 break;
5300
5301 case ZONE_ATTR_NAME:
5302 size = strlen(zone->zone_name) + 1;
5303 if (bufsize > size)
5304 bufsize = size;
5305 if (buf != NULL) {
5306 err = copyoutstr(zone->zone_name, buf, bufsize, NULL);
5307 if (err != 0 && err != ENAMETOOLONG)
5308 error = EFAULT;
5309 }
5310 break;
5311
5312 case ZONE_ATTR_STATUS:
5313 /*
5314 * Since we're not holding zonehash_lock, the zone status
5315 * may be anything; leave it up to userland to sort it out.
5316 */
5317 size = sizeof (zone_status);
5318 if (bufsize > size)
5319 bufsize = size;
5320 zone_status = zone_status_get(zone);
5321 if (buf != NULL &&
5322 copyout(&zone_status, buf, bufsize) != 0)
5323 error = EFAULT;
5324 break;
5325 case ZONE_ATTR_FLAGS:
5326 size = sizeof (zone->zone_flags);
5327 if (bufsize > size)
5328 bufsize = size;
5329 flags = zone->zone_flags;
5330 if (buf != NULL &&
5331 copyout(&flags, buf, bufsize) != 0)
5332 error = EFAULT;
5333 break;
5334 case ZONE_ATTR_PRIVSET:
5335 size = sizeof (priv_set_t);
5336 if (bufsize > size)
5337 bufsize = size;
5338 if (buf != NULL &&
5339 copyout(zone->zone_privset, buf, bufsize) != 0)
5340 error = EFAULT;
5341 break;
5342 case ZONE_ATTR_UNIQID:
5343 size = sizeof (zone->zone_uniqid);
5344 if (bufsize > size)
5345 bufsize = size;
5346 if (buf != NULL &&
5347 copyout(&zone->zone_uniqid, buf, bufsize) != 0)
5348 error = EFAULT;
5349 break;
5350 case ZONE_ATTR_POOLID:
5351 {
5352 pool_t *pool;
5353 poolid_t poolid;
5354
5355 if (pool_lock_intr() != 0) {
5356 error = EINTR;
5357 break;
5358 }
5359 pool = zone_pool_get(zone);
5360 poolid = pool->pool_id;
5361 pool_unlock();
5362 size = sizeof (poolid);
5363 if (bufsize > size)
5364 bufsize = size;
5365 if (buf != NULL && copyout(&poolid, buf, size) != 0)
5366 error = EFAULT;
5367 }
5368 break;
5369 case ZONE_ATTR_SLBL:
5370 size = sizeof (bslabel_t);
5371 if (bufsize > size)
5372 bufsize = size;
5373 if (zone->zone_slabel == NULL)
5374 error = EINVAL;
5375 else if (buf != NULL &&
5376 copyout(label2bslabel(zone->zone_slabel), buf,
5377 bufsize) != 0)
5378 error = EFAULT;
5379 break;
5380 case ZONE_ATTR_INITPID:
5381 size = sizeof (initpid);
5382 if (bufsize > size)
5383 bufsize = size;
5384 initpid = zone->zone_proc_initpid;
5385 if (initpid == -1) {
5386 error = ESRCH;
5387 break;
5388 }
5389 if (buf != NULL &&
5390 copyout(&initpid, buf, bufsize) != 0)
5391 error = EFAULT;
5392 break;
5393 case ZONE_ATTR_BRAND:
5394 size = strlen(zone->zone_brand->b_name) + 1;
5395
5396 if (bufsize > size)
5397 bufsize = size;
5398 if (buf != NULL) {
5399 err = copyoutstr(zone->zone_brand->b_name, buf,
5400 bufsize, NULL);
5401 if (err != 0 && err != ENAMETOOLONG)
5402 error = EFAULT;
5403 }
5404 break;
5405 case ZONE_ATTR_INITNAME:
5406 size = strlen(zone->zone_initname) + 1;
5407 if (bufsize > size)
5408 bufsize = size;
5409 if (buf != NULL) {
5410 err = copyoutstr(zone->zone_initname, buf, bufsize,
5411 NULL);
5412 if (err != 0 && err != ENAMETOOLONG)
5413 error = EFAULT;
5414 }
5415 break;
5416 case ZONE_ATTR_BOOTARGS:
5417 if (zone->zone_bootargs == NULL)
5418 outstr = "";
5419 else
5420 outstr = zone->zone_bootargs;
5421 size = strlen(outstr) + 1;
5422 if (bufsize > size)
5423 bufsize = size;
5424 if (buf != NULL) {
5425 err = copyoutstr(outstr, buf, bufsize, NULL);
5426 if (err != 0 && err != ENAMETOOLONG)
5427 error = EFAULT;
5428 }
5429 break;
5430 case ZONE_ATTR_PHYS_MCAP:
5431 size = sizeof (zone->zone_phys_mcap);
5432 if (bufsize > size)
5433 bufsize = size;
5434 if (buf != NULL &&
5435 copyout(&zone->zone_phys_mcap, buf, bufsize) != 0)
5436 error = EFAULT;
5437 break;
5438 case ZONE_ATTR_SCHED_CLASS:
5439 mutex_enter(&class_lock);
5440
5441 if (zone->zone_defaultcid >= loaded_classes)
5442 outstr = "";
5443 else
5444 outstr = sclass[zone->zone_defaultcid].cl_name;
5445 size = strlen(outstr) + 1;
5446 if (bufsize > size)
5447 bufsize = size;
5448 if (buf != NULL) {
5449 err = copyoutstr(outstr, buf, bufsize, NULL);
5450 if (err != 0 && err != ENAMETOOLONG)
5451 error = EFAULT;
5452 }
5453
5454 mutex_exit(&class_lock);
5455 break;
5456 case ZONE_ATTR_HOSTID:
5457 if (zone->zone_hostid != HW_INVALID_HOSTID &&
5458 bufsize == sizeof (zone->zone_hostid)) {
5459 size = sizeof (zone->zone_hostid);
5460 if (buf != NULL && copyout(&zone->zone_hostid, buf,
5461 bufsize) != 0)
5462 error = EFAULT;
5463 } else {
5464 error = EINVAL;
5465 }
5466 break;
5467 case ZONE_ATTR_FS_ALLOWED:
5468 if (zone->zone_fs_allowed == NULL)
5469 outstr = "";
5470 else
5471 outstr = zone->zone_fs_allowed;
5472 size = strlen(outstr) + 1;
5473 if (bufsize > size)
5474 bufsize = size;
5475 if (buf != NULL) {
5476 err = copyoutstr(outstr, buf, bufsize, NULL);
5477 if (err != 0 && err != ENAMETOOLONG)
5478 error = EFAULT;
5479 }
5480 break;
5481 case ZONE_ATTR_NETWORK:
5482 zbuf = kmem_alloc(bufsize, KM_SLEEP);
5483 if (copyin(buf, zbuf, bufsize) != 0) {
5484 error = EFAULT;
5485 } else {
5486 error = zone_get_network(zoneid, zbuf);
5487 if (error == 0 && copyout(zbuf, buf, bufsize) != 0)
5488 error = EFAULT;
5489 }
5490 kmem_free(zbuf, bufsize);
5491 break;
5492 default:
5493 if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone)) {
5494 size = bufsize;
5495 error = ZBROP(zone)->b_getattr(zone, attr, buf, &size);
5496 } else {
5497 error = EINVAL;
5498 }
5499 }
5500 zone_rele(zone);
5501
5502 if (error)
5503 return (set_errno(error));
5504 return ((ssize_t)size);
5505 }
5506
5507 /*
5508 * Systemcall entry point for zone_setattr(2).
5509 */
5510 /*ARGSUSED*/
5511 static int
5512 zone_setattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
5513 {
5514 zone_t *zone;
5515 zone_status_t zone_status;
5516 int err = -1;
5517 zone_net_data_t *zbuf;
5518
5519 if (secpolicy_zone_config(CRED()) != 0)
5520 return (set_errno(EPERM));
5521
5522 /*
5523 * Only the ZONE_ATTR_PHYS_MCAP attribute can be set on the
5524 * global zone.
5525 */
5526 if (zoneid == GLOBAL_ZONEID && attr != ZONE_ATTR_PHYS_MCAP) {
5527 return (set_errno(EINVAL));
5528 }
5529
5530 mutex_enter(&zonehash_lock);
5531 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5532 mutex_exit(&zonehash_lock);
5533 return (set_errno(EINVAL));
5534 }
5535 zone_hold(zone);
5536 mutex_exit(&zonehash_lock);
5537
5538 /*
5539 * At present most attributes can only be set on non-running,
5540 * non-global zones.
5541 */
5542 zone_status = zone_status_get(zone);
5543 if (attr != ZONE_ATTR_PHYS_MCAP && zone_status > ZONE_IS_READY) {
5544 err = EINVAL;
5545 goto done;
5546 }
5547
5548 switch (attr) {
5549 case ZONE_ATTR_INITNAME:
5550 err = zone_set_initname(zone, (const char *)buf);
5551 break;
5552 case ZONE_ATTR_INITNORESTART:
5553 zone->zone_restart_init = B_FALSE;
5554 err = 0;
5555 break;
5556 case ZONE_ATTR_BOOTARGS:
5557 err = zone_set_bootargs(zone, (const char *)buf);
5558 break;
5559 case ZONE_ATTR_BRAND:
5560 err = zone_set_brand(zone, (const char *)buf);
5561 break;
5562 case ZONE_ATTR_FS_ALLOWED:
5563 err = zone_set_fs_allowed(zone, (const char *)buf);
5564 break;
5565 case ZONE_ATTR_PHYS_MCAP:
5566 err = zone_set_phys_mcap(zone, (const uint64_t *)buf);
5567 break;
5568 case ZONE_ATTR_SCHED_CLASS:
5569 err = zone_set_sched_class(zone, (const char *)buf);
5570 break;
5571 case ZONE_ATTR_HOSTID:
5572 if (bufsize == sizeof (zone->zone_hostid)) {
5573 if (copyin(buf, &zone->zone_hostid, bufsize) == 0)
5574 err = 0;
5575 else
5576 err = EFAULT;
5577 } else {
5578 err = EINVAL;
5579 }
5580 break;
5581 case ZONE_ATTR_NETWORK:
5582 if (bufsize > (PIPE_BUF + sizeof (zone_net_data_t))) {
5583 err = EINVAL;
5584 break;
5585 }
5586 zbuf = kmem_alloc(bufsize, KM_SLEEP);
5587 if (copyin(buf, zbuf, bufsize) != 0) {
5588 kmem_free(zbuf, bufsize);
5589 err = EFAULT;
5590 break;
5591 }
5592 err = zone_set_network(zoneid, zbuf);
5593 kmem_free(zbuf, bufsize);
5594 break;
5595 default:
5596 if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone))
5597 err = ZBROP(zone)->b_setattr(zone, attr, buf, bufsize);
5598 else
5599 err = EINVAL;
5600 }
5601
5602 done:
5603 zone_rele(zone);
5604 ASSERT(err != -1);
5605 return (err != 0 ? set_errno(err) : 0);
5606 }
5607
5608 /*
5609 * Return zero if the process has at least one vnode mapped in to its
5610 * address space which shouldn't be allowed to change zones.
5611 *
5612 * Also return zero if the process has any shared mappings which reserve
5613 * swap. This is because the counting for zone.max-swap does not allow swap
5614 * reservation to be shared between zones. zone swap reservation is counted
5615 * on zone->zone_max_swap.
5616 */
5617 static int
5618 as_can_change_zones(void)
5619 {
5620 proc_t *pp = curproc;
5621 struct seg *seg;
5622 struct as *as = pp->p_as;
5623 vnode_t *vp;
5624 int allow = 1;
5625
5626 ASSERT(pp->p_as != &kas);
5627 AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
5628 for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) {
5629
5630 /*
5631 * Cannot enter zone with shared anon memory which
5632 * reserves swap. See comment above.
5633 */
5634 if (seg_can_change_zones(seg) == B_FALSE) {
5635 allow = 0;
5636 break;
5637 }
5638 /*
5639 * if we can't get a backing vnode for this segment then skip
5640 * it.
5641 */
5642 vp = NULL;
5643 if (SEGOP_GETVP(seg, seg->s_base, &vp) != 0 || vp == NULL)
5644 continue;
5645 if (!vn_can_change_zones(vp)) { /* bail on first match */
5646 allow = 0;
5647 break;
5648 }
5649 }
5650 AS_LOCK_EXIT(as, &as->a_lock);
5651 return (allow);
5652 }
5653
5654 /*
5655 * Count swap reserved by curproc's address space
5656 */
5657 static size_t
5658 as_swresv(void)
5659 {
5660 proc_t *pp = curproc;
5661 struct seg *seg;
5662 struct as *as = pp->p_as;
5663 size_t swap = 0;
5664
5665 ASSERT(pp->p_as != &kas);
5666 ASSERT(AS_WRITE_HELD(as, &as->a_lock));
5667 for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg))
5668 swap += seg_swresv(seg);
5669
5670 return (swap);
5671 }
5672
5673 /*
5674 * Systemcall entry point for zone_enter().
5675 *
5676 * The current process is injected into said zone. In the process
5677 * it will change its project membership, privileges, rootdir/cwd,
5678 * zone-wide rctls, and pool association to match those of the zone.
5679 *
5680 * The first zone_enter() called while the zone is in the ZONE_IS_READY
5681 * state will transition it to ZONE_IS_RUNNING. Processes may only
5682 * enter a zone that is "ready" or "running".
5683 */
5684 static int
5685 zone_enter(zoneid_t zoneid)
5686 {
5687 zone_t *zone;
5688 vnode_t *vp;
5689 proc_t *pp = curproc;
5690 contract_t *ct;
5691 cont_process_t *ctp;
5692 task_t *tk, *oldtk;
5693 kproject_t *zone_proj0;
5694 cred_t *cr, *newcr;
5695 pool_t *oldpool, *newpool;
5696 sess_t *sp;
5697 uid_t uid;
5698 zone_status_t status;
5699 int err = 0;
5700 rctl_entity_p_t e;
5701 size_t swap;
5702 kthread_id_t t;
5703
5704 if (secpolicy_zone_config(CRED()) != 0)
5705 return (set_errno(EPERM));
5706 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
5707 return (set_errno(EINVAL));
5708
5709 /*
5710 * Stop all lwps so we don't need to hold a lock to look at
5711 * curproc->p_zone. This needs to happen before we grab any
5712 * locks to avoid deadlock (another lwp in the process could
5713 * be waiting for the held lock).
5714 */
5715 if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK))
5716 return (set_errno(EINTR));
5717
5718 /*
5719 * Make sure we're not changing zones with files open or mapped in
5720 * to our address space which shouldn't be changing zones.
5721 */
5722 if (!files_can_change_zones()) {
5723 err = EBADF;
5724 goto out;
5725 }
5726 if (!as_can_change_zones()) {
5727 err = EFAULT;
5728 goto out;
5729 }
5730
5731 mutex_enter(&zonehash_lock);
5732 if (pp->p_zone != global_zone) {
5733 mutex_exit(&zonehash_lock);
5734 err = EINVAL;
5735 goto out;
5736 }
5737
5738 zone = zone_find_all_by_id(zoneid);
5739 if (zone == NULL) {
5740 mutex_exit(&zonehash_lock);
5741 err = EINVAL;
5742 goto out;
5743 }
5744
5745 /*
5746 * To prevent processes in a zone from holding contracts on
5747 * extrazonal resources, and to avoid process contract
5748 * memberships which span zones, contract holders and processes
5749 * which aren't the sole members of their encapsulating process
5750 * contracts are not allowed to zone_enter.
5751 */
5752 ctp = pp->p_ct_process;
5753 ct = &ctp->conp_contract;
5754 mutex_enter(&ct->ct_lock);
5755 mutex_enter(&pp->p_lock);
5756 if ((avl_numnodes(&pp->p_ct_held) != 0) || (ctp->conp_nmembers != 1)) {
5757 mutex_exit(&pp->p_lock);
5758 mutex_exit(&ct->ct_lock);
5759 mutex_exit(&zonehash_lock);
5760 err = EINVAL;
5761 goto out;
5762 }
5763
5764 /*
5765 * Moreover, we don't allow processes whose encapsulating
5766 * process contracts have inherited extrazonal contracts.
5767 * While it would be easier to eliminate all process contracts
5768 * with inherited contracts, we need to be able to give a
5769 * restarted init (or other zone-penetrating process) its
5770 * predecessor's contracts.
5771 */
5772 if (ctp->conp_ninherited != 0) {
5773 contract_t *next;
5774 for (next = list_head(&ctp->conp_inherited); next;
5775 next = list_next(&ctp->conp_inherited, next)) {
5776 if (contract_getzuniqid(next) != zone->zone_uniqid) {
5777 mutex_exit(&pp->p_lock);
5778 mutex_exit(&ct->ct_lock);
5779 mutex_exit(&zonehash_lock);
5780 err = EINVAL;
5781 goto out;
5782 }
5783 }
5784 }
5785
5786 mutex_exit(&pp->p_lock);
5787 mutex_exit(&ct->ct_lock);
5788
5789 status = zone_status_get(zone);
5790 if (status < ZONE_IS_READY || status >= ZONE_IS_SHUTTING_DOWN) {
5791 /*
5792 * Can't join
5793 */
5794 mutex_exit(&zonehash_lock);
5795 err = EINVAL;
5796 goto out;
5797 }
5798
5799 /*
5800 * Make sure new priv set is within the permitted set for caller
5801 */
5802 if (!priv_issubset(zone->zone_privset, &CR_OPPRIV(CRED()))) {
5803 mutex_exit(&zonehash_lock);
5804 err = EPERM;
5805 goto out;
5806 }
5807 /*
5808 * We want to momentarily drop zonehash_lock while we optimistically
5809 * bind curproc to the pool it should be running in. This is safe
5810 * since the zone can't disappear (we have a hold on it).
5811 */
5812 zone_hold(zone);
5813 mutex_exit(&zonehash_lock);
5814
5815 /*
5816 * Grab pool_lock to keep the pools configuration from changing
5817 * and to stop ourselves from getting rebound to another pool
5818 * until we join the zone.
5819 */
5820 if (pool_lock_intr() != 0) {
5821 zone_rele(zone);
5822 err = EINTR;
5823 goto out;
5824 }
5825 ASSERT(secpolicy_pool(CRED()) == 0);
5826 /*
5827 * Bind ourselves to the pool currently associated with the zone.
5828 */
5829 oldpool = curproc->p_pool;
5830 newpool = zone_pool_get(zone);
5831 if (pool_state == POOL_ENABLED && newpool != oldpool &&
5832 (err = pool_do_bind(newpool, P_PID, P_MYID,
5833 POOL_BIND_ALL)) != 0) {
5834 pool_unlock();
5835 zone_rele(zone);
5836 goto out;
5837 }
5838
5839 /*
5840 * Grab cpu_lock now; we'll need it later when we call
5841 * task_join().
5842 */
5843 mutex_enter(&cpu_lock);
5844 mutex_enter(&zonehash_lock);
5845 /*
5846 * Make sure the zone hasn't moved on since we dropped zonehash_lock.
5847 */
5848 if (zone_status_get(zone) >= ZONE_IS_SHUTTING_DOWN) {
5849 /*
5850 * Can't join anymore.
5851 */
5852 mutex_exit(&zonehash_lock);
5853 mutex_exit(&cpu_lock);
5854 if (pool_state == POOL_ENABLED &&
5855 newpool != oldpool)
5856 (void) pool_do_bind(oldpool, P_PID, P_MYID,
5857 POOL_BIND_ALL);
5858 pool_unlock();
5859 zone_rele(zone);
5860 err = EINVAL;
5861 goto out;
5862 }
5863
5864 /*
5865 * a_lock must be held while transfering locked memory and swap
5866 * reservation from the global zone to the non global zone because
5867 * asynchronous faults on the processes' address space can lock
5868 * memory and reserve swap via MCL_FUTURE and MAP_NORESERVE
5869 * segments respectively.
5870 */
5871 AS_LOCK_ENTER(pp->as, &pp->p_as->a_lock, RW_WRITER);
5872 swap = as_swresv();
5873 mutex_enter(&pp->p_lock);
5874 zone_proj0 = zone->zone_zsched->p_task->tk_proj;
5875 /* verify that we do not exceed and task or lwp limits */
5876 mutex_enter(&zone->zone_nlwps_lock);
5877 /* add new lwps to zone and zone's proj0 */
5878 zone_proj0->kpj_nlwps += pp->p_lwpcnt;
5879 zone->zone_nlwps += pp->p_lwpcnt;
5880 /* add 1 task to zone's proj0 */
5881 zone_proj0->kpj_ntasks += 1;
5882
5883 zone_proj0->kpj_nprocs++;
5884 zone->zone_nprocs++;
5885 mutex_exit(&zone->zone_nlwps_lock);
5886
5887 mutex_enter(&zone->zone_mem_lock);
5888 zone->zone_locked_mem += pp->p_locked_mem;
5889 zone_proj0->kpj_data.kpd_locked_mem += pp->p_locked_mem;
5890 zone->zone_max_swap += swap;
5891 mutex_exit(&zone->zone_mem_lock);
5892
5893 mutex_enter(&(zone_proj0->kpj_data.kpd_crypto_lock));
5894 zone_proj0->kpj_data.kpd_crypto_mem += pp->p_crypto_mem;
5895 mutex_exit(&(zone_proj0->kpj_data.kpd_crypto_lock));
5896
5897 /* remove lwps and process from proc's old zone and old project */
5898 mutex_enter(&pp->p_zone->zone_nlwps_lock);
5899 pp->p_zone->zone_nlwps -= pp->p_lwpcnt;
5900 pp->p_task->tk_proj->kpj_nlwps -= pp->p_lwpcnt;
5901 pp->p_task->tk_proj->kpj_nprocs--;
5902 pp->p_zone->zone_nprocs--;
5903 mutex_exit(&pp->p_zone->zone_nlwps_lock);
5904
5905 mutex_enter(&pp->p_zone->zone_mem_lock);
5906 pp->p_zone->zone_locked_mem -= pp->p_locked_mem;
5907 pp->p_task->tk_proj->kpj_data.kpd_locked_mem -= pp->p_locked_mem;
5908 pp->p_zone->zone_max_swap -= swap;
5909 mutex_exit(&pp->p_zone->zone_mem_lock);
5910
5911 mutex_enter(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock));
5912 pp->p_task->tk_proj->kpj_data.kpd_crypto_mem -= pp->p_crypto_mem;
5913 mutex_exit(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock));
5914
5915 pp->p_flag |= SZONETOP;
5916 pp->p_zone = zone;
5917 mutex_exit(&pp->p_lock);
5918 AS_LOCK_EXIT(pp->p_as, &pp->p_as->a_lock);
5919
5920 /*
5921 * Joining the zone cannot fail from now on.
5922 *
5923 * This means that a lot of the following code can be commonized and
5924 * shared with zsched().
5925 */
5926
5927 /*
5928 * If the process contract fmri was inherited, we need to
5929 * flag this so that any contract status will not leak
5930 * extra zone information, svc_fmri in this case
5931 */
5932 if (ctp->conp_svc_ctid != ct->ct_id) {
5933 mutex_enter(&ct->ct_lock);
5934 ctp->conp_svc_zone_enter = ct->ct_id;
5935 mutex_exit(&ct->ct_lock);
5936 }
5937
5938 /*
5939 * Reset the encapsulating process contract's zone.
5940 */
5941 ASSERT(ct->ct_mzuniqid == GLOBAL_ZONEUNIQID);
5942 contract_setzuniqid(ct, zone->zone_uniqid);
5943
5944 /*
5945 * Create a new task and associate the process with the project keyed
5946 * by (projid,zoneid).
5947 *
5948 * We might as well be in project 0; the global zone's projid doesn't
5949 * make much sense in a zone anyhow.
5950 *
5951 * This also increments zone_ntasks, and returns with p_lock held.
5952 */
5953 tk = task_create(0, zone);
5954 oldtk = task_join(tk, 0);
5955 mutex_exit(&cpu_lock);
5956
5957 /*
5958 * call RCTLOP_SET functions on this proc
5959 */
5960 e.rcep_p.zone = zone;
5961 e.rcep_t = RCENTITY_ZONE;
5962 (void) rctl_set_dup(NULL, NULL, pp, &e, zone->zone_rctls, NULL,
5963 RCD_CALLBACK);
5964 mutex_exit(&pp->p_lock);
5965
5966 /*
5967 * We don't need to hold any of zsched's locks here; not only do we know
5968 * the process and zone aren't going away, we know its session isn't
5969 * changing either.
5970 *
5971 * By joining zsched's session here, we mimic the behavior in the
5972 * global zone of init's sid being the pid of sched. We extend this
5973 * to all zlogin-like zone_enter()'ing processes as well.
5974 */
5975 mutex_enter(&pidlock);
5976 sp = zone->zone_zsched->p_sessp;
5977 sess_hold(zone->zone_zsched);
5978 mutex_enter(&pp->p_lock);
5979 pgexit(pp);
5980 sess_rele(pp->p_sessp, B_TRUE);
5981 pp->p_sessp = sp;
5982 pgjoin(pp, zone->zone_zsched->p_pidp);
5983
5984 /*
5985 * If any threads are scheduled to be placed on zone wait queue they
5986 * should abandon the idea since the wait queue is changing.
5987 * We need to be holding pidlock & p_lock to do this.
5988 */
5989 if ((t = pp->p_tlist) != NULL) {
5990 do {
5991 thread_lock(t);
5992 /*
5993 * Kick this thread so that he doesn't sit
5994 * on a wrong wait queue.
5995 */
5996 if (ISWAITING(t))
5997 setrun_locked(t);
5998
5999 if (t->t_schedflag & TS_ANYWAITQ)
6000 t->t_schedflag &= ~ TS_ANYWAITQ;
6001
6002 thread_unlock(t);
6003 } while ((t = t->t_forw) != pp->p_tlist);
6004 }
6005
6006 /*
6007 * If there is a default scheduling class for the zone and it is not
6008 * the class we are currently in, change all of the threads in the
6009 * process to the new class. We need to be holding pidlock & p_lock
6010 * when we call parmsset so this is a good place to do it.
6011 */
6012 if (zone->zone_defaultcid > 0 &&
6013 zone->zone_defaultcid != curthread->t_cid) {
6014 pcparms_t pcparms;
6015
6016 pcparms.pc_cid = zone->zone_defaultcid;
6017 pcparms.pc_clparms[0] = 0;
6018
6019 /*
6020 * If setting the class fails, we still want to enter the zone.
6021 */
6022 if ((t = pp->p_tlist) != NULL) {
6023 do {
6024 (void) parmsset(&pcparms, t);
6025 } while ((t = t->t_forw) != pp->p_tlist);
6026 }
6027 }
6028
6029 mutex_exit(&pp->p_lock);
6030 mutex_exit(&pidlock);
6031
6032 mutex_exit(&zonehash_lock);
6033 /*
6034 * We're firmly in the zone; let pools progress.
6035 */
6036 pool_unlock();
6037 task_rele(oldtk);
6038 /*
6039 * We don't need to retain a hold on the zone since we already
6040 * incremented zone_ntasks, so the zone isn't going anywhere.
6041 */
6042 zone_rele(zone);
6043
6044 /*
6045 * Chroot
6046 */
6047 vp = zone->zone_rootvp;
6048 zone_chdir(vp, &PTOU(pp)->u_cdir, pp);
6049 zone_chdir(vp, &PTOU(pp)->u_rdir, pp);
6050
6051 /*
6052 * Change process credentials
6053 */
6054 newcr = cralloc();
6055 mutex_enter(&pp->p_crlock);
6056 cr = pp->p_cred;
6057 crcopy_to(cr, newcr);
6058 crsetzone(newcr, zone);
6059 pp->p_cred = newcr;
6060
6061 /*
6062 * Restrict all process privilege sets to zone limit
6063 */
6064 priv_intersect(zone->zone_privset, &CR_PPRIV(newcr));
6065 priv_intersect(zone->zone_privset, &CR_EPRIV(newcr));
6066 priv_intersect(zone->zone_privset, &CR_IPRIV(newcr));
6067 priv_intersect(zone->zone_privset, &CR_LPRIV(newcr));
6068 mutex_exit(&pp->p_crlock);
6069 crset(pp, newcr);
6070
6071 /*
6072 * Adjust upcount to reflect zone entry.
6073 */
6074 uid = crgetruid(newcr);
6075 mutex_enter(&pidlock);
6076 upcount_dec(uid, GLOBAL_ZONEID);
6077 upcount_inc(uid, zoneid);
6078 mutex_exit(&pidlock);
6079
6080 /*
6081 * Set up core file path and content.
6082 */
6083 set_core_defaults();
6084
6085 out:
6086 /*
6087 * Let the other lwps continue.
6088 */
6089 mutex_enter(&pp->p_lock);
6090 if (curthread != pp->p_agenttp)
6091 continuelwps(pp);
6092 mutex_exit(&pp->p_lock);
6093
6094 return (err != 0 ? set_errno(err) : 0);
6095 }
6096
6097 /*
6098 * Systemcall entry point for zone_list(2).
6099 *
6100 * Processes running in a (non-global) zone only see themselves.
6101 * On labeled systems, they see all zones whose label they dominate.
6102 */
6103 static int
6104 zone_list(zoneid_t *zoneidlist, uint_t *numzones)
6105 {
6106 zoneid_t *zoneids;
6107 zone_t *zone, *myzone;
6108 uint_t user_nzones, real_nzones;
6109 uint_t domi_nzones;
6110 int error;
6111
6112 if (copyin(numzones, &user_nzones, sizeof (uint_t)) != 0)
6113 return (set_errno(EFAULT));
6114
6115 myzone = curproc->p_zone;
6116 if (myzone != global_zone) {
6117 bslabel_t *mybslab;
6118
6119 if (!is_system_labeled()) {
6120 /* just return current zone */
6121 real_nzones = domi_nzones = 1;
6122 zoneids = kmem_alloc(sizeof (zoneid_t), KM_SLEEP);
6123 zoneids[0] = myzone->zone_id;
6124 } else {
6125 /* return all zones that are dominated */
6126 mutex_enter(&zonehash_lock);
6127 real_nzones = zonecount;
6128 domi_nzones = 0;
6129 if (real_nzones > 0) {
6130 zoneids = kmem_alloc(real_nzones *
6131 sizeof (zoneid_t), KM_SLEEP);
6132 mybslab = label2bslabel(myzone->zone_slabel);
6133 for (zone = list_head(&zone_active);
6134 zone != NULL;
6135 zone = list_next(&zone_active, zone)) {
6136 if (zone->zone_id == GLOBAL_ZONEID)
6137 continue;
6138 if (zone != myzone &&
6139 (zone->zone_flags & ZF_IS_SCRATCH))
6140 continue;
6141 /*
6142 * Note that a label always dominates
6143 * itself, so myzone is always included
6144 * in the list.
6145 */
6146 if (bldominates(mybslab,
6147 label2bslabel(zone->zone_slabel))) {
6148 zoneids[domi_nzones++] =
6149 zone->zone_id;
6150 }
6151 }
6152 }
6153 mutex_exit(&zonehash_lock);
6154 }
6155 } else {
6156 mutex_enter(&zonehash_lock);
6157 real_nzones = zonecount;
6158 domi_nzones = 0;
6159 if (real_nzones > 0) {
6160 zoneids = kmem_alloc(real_nzones * sizeof (zoneid_t),
6161 KM_SLEEP);
6162 for (zone = list_head(&zone_active); zone != NULL;
6163 zone = list_next(&zone_active, zone))
6164 zoneids[domi_nzones++] = zone->zone_id;
6165 ASSERT(domi_nzones == real_nzones);
6166 }
6167 mutex_exit(&zonehash_lock);
6168 }
6169
6170 /*
6171 * If user has allocated space for fewer entries than we found, then
6172 * return only up to his limit. Either way, tell him exactly how many
6173 * we found.
6174 */
6175 if (domi_nzones < user_nzones)
6176 user_nzones = domi_nzones;
6177 error = 0;
6178 if (copyout(&domi_nzones, numzones, sizeof (uint_t)) != 0) {
6179 error = EFAULT;
6180 } else if (zoneidlist != NULL && user_nzones != 0) {
6181 if (copyout(zoneids, zoneidlist,
6182 user_nzones * sizeof (zoneid_t)) != 0)
6183 error = EFAULT;
6184 }
6185
6186 if (real_nzones > 0)
6187 kmem_free(zoneids, real_nzones * sizeof (zoneid_t));
6188
6189 if (error != 0)
6190 return (set_errno(error));
6191 else
6192 return (0);
6193 }
6194
6195 /*
6196 * Systemcall entry point for zone_lookup(2).
6197 *
6198 * Non-global zones are only able to see themselves and (on labeled systems)
6199 * the zones they dominate.
6200 */
6201 static zoneid_t
6202 zone_lookup(const char *zone_name)
6203 {
6204 char *kname;
6205 zone_t *zone;
6206 zoneid_t zoneid;
6207 int err;
6208
6209 if (zone_name == NULL) {
6210 /* return caller's zone id */
6211 return (getzoneid());
6212 }
6213
6214 kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
6215 if ((err = copyinstr(zone_name, kname, ZONENAME_MAX, NULL)) != 0) {
6216 kmem_free(kname, ZONENAME_MAX);
6217 return (set_errno(err));
6218 }
6219
6220 mutex_enter(&zonehash_lock);
6221 zone = zone_find_all_by_name(kname);
6222 kmem_free(kname, ZONENAME_MAX);
6223 /*
6224 * In a non-global zone, can only lookup global and own name.
6225 * In Trusted Extensions zone label dominance rules apply.
6226 */
6227 if (zone == NULL ||
6228 zone_status_get(zone) < ZONE_IS_READY ||
6229 !zone_list_access(zone)) {
6230 mutex_exit(&zonehash_lock);
6231 return (set_errno(EINVAL));
6232 } else {
6233 zoneid = zone->zone_id;
6234 mutex_exit(&zonehash_lock);
6235 return (zoneid);
6236 }
6237 }
6238
6239 static int
6240 zone_version(int *version_arg)
6241 {
6242 int version = ZONE_SYSCALL_API_VERSION;
6243
6244 if (copyout(&version, version_arg, sizeof (int)) != 0)
6245 return (set_errno(EFAULT));
6246 return (0);
6247 }
6248
6249 /* ARGSUSED */
6250 long
6251 zone(int cmd, void *arg1, void *arg2, void *arg3, void *arg4)
6252 {
6253 zone_def zs;
6254 int err;
6255
6256 switch (cmd) {
6257 case ZONE_CREATE:
6258 if (get_udatamodel() == DATAMODEL_NATIVE) {
6259 if (copyin(arg1, &zs, sizeof (zone_def))) {
6260 return (set_errno(EFAULT));
6261 }
6262 } else {
6263 #ifdef _SYSCALL32_IMPL
6264 zone_def32 zs32;
6265
6266 if (copyin(arg1, &zs32, sizeof (zone_def32))) {
6267 return (set_errno(EFAULT));
6268 }
6269 zs.zone_name =
6270 (const char *)(unsigned long)zs32.zone_name;
6271 zs.zone_root =
6272 (const char *)(unsigned long)zs32.zone_root;
6273 zs.zone_privs =
6274 (const struct priv_set *)
6275 (unsigned long)zs32.zone_privs;
6276 zs.zone_privssz = zs32.zone_privssz;
6277 zs.rctlbuf = (caddr_t)(unsigned long)zs32.rctlbuf;
6278 zs.rctlbufsz = zs32.rctlbufsz;
6279 zs.zfsbuf = (caddr_t)(unsigned long)zs32.zfsbuf;
6280 zs.zfsbufsz = zs32.zfsbufsz;
6281 zs.extended_error =
6282 (int *)(unsigned long)zs32.extended_error;
6283 zs.match = zs32.match;
6284 zs.doi = zs32.doi;
6285 zs.label = (const bslabel_t *)(uintptr_t)zs32.label;
6286 zs.flags = zs32.flags;
6287 #else
6288 panic("get_udatamodel() returned bogus result\n");
6289 #endif
6290 }
6291
6292 return (zone_create(zs.zone_name, zs.zone_root,
6293 zs.zone_privs, zs.zone_privssz,
6294 (caddr_t)zs.rctlbuf, zs.rctlbufsz,
6295 (caddr_t)zs.zfsbuf, zs.zfsbufsz,
6296 zs.extended_error, zs.match, zs.doi,
6297 zs.label, zs.flags));
6298 case ZONE_BOOT:
6299 return (zone_boot((zoneid_t)(uintptr_t)arg1));
6300 case ZONE_DESTROY:
6301 return (zone_destroy((zoneid_t)(uintptr_t)arg1));
6302 case ZONE_GETATTR:
6303 return (zone_getattr((zoneid_t)(uintptr_t)arg1,
6304 (int)(uintptr_t)arg2, arg3, (size_t)arg4));
6305 case ZONE_SETATTR:
6306 return (zone_setattr((zoneid_t)(uintptr_t)arg1,
6307 (int)(uintptr_t)arg2, arg3, (size_t)arg4));
6308 case ZONE_ENTER:
6309 return (zone_enter((zoneid_t)(uintptr_t)arg1));
6310 case ZONE_LIST:
6311 return (zone_list((zoneid_t *)arg1, (uint_t *)arg2));
6312 case ZONE_SHUTDOWN:
6313 return (zone_shutdown((zoneid_t)(uintptr_t)arg1));
6314 case ZONE_LOOKUP:
6315 return (zone_lookup((const char *)arg1));
6316 case ZONE_VERSION:
6317 return (zone_version((int *)arg1));
6318 case ZONE_ADD_DATALINK:
6319 return (zone_add_datalink((zoneid_t)(uintptr_t)arg1,
6320 (datalink_id_t)(uintptr_t)arg2));
6321 case ZONE_DEL_DATALINK:
6322 return (zone_remove_datalink((zoneid_t)(uintptr_t)arg1,
6323 (datalink_id_t)(uintptr_t)arg2));
6324 case ZONE_CHECK_DATALINK: {
6325 zoneid_t zoneid;
6326 boolean_t need_copyout;
6327
6328 if (copyin(arg1, &zoneid, sizeof (zoneid)) != 0)
6329 return (EFAULT);
6330 need_copyout = (zoneid == ALL_ZONES);
6331 err = zone_check_datalink(&zoneid,
6332 (datalink_id_t)(uintptr_t)arg2);
6333 if (err == 0 && need_copyout) {
6334 if (copyout(&zoneid, arg1, sizeof (zoneid)) != 0)
6335 err = EFAULT;
6336 }
6337 return (err == 0 ? 0 : set_errno(err));
6338 }
6339 case ZONE_LIST_DATALINK:
6340 return (zone_list_datalink((zoneid_t)(uintptr_t)arg1,
6341 (int *)arg2, (datalink_id_t *)(uintptr_t)arg3));
6342 default:
6343 return (set_errno(EINVAL));
6344 }
6345 }
6346
6347 struct zarg {
6348 zone_t *zone;
6349 zone_cmd_arg_t arg;
6350 };
6351
6352 static int
6353 zone_lookup_door(const char *zone_name, door_handle_t *doorp)
6354 {
6355 char *buf;
6356 size_t buflen;
6357 int error;
6358
6359 buflen = sizeof (ZONE_DOOR_PATH) + strlen(zone_name);
6360 buf = kmem_alloc(buflen, KM_SLEEP);
6361 (void) snprintf(buf, buflen, ZONE_DOOR_PATH, zone_name);
6362 error = door_ki_open(buf, doorp);
6363 kmem_free(buf, buflen);
6364 return (error);
6365 }
6366
6367 static void
6368 zone_release_door(door_handle_t *doorp)
6369 {
6370 door_ki_rele(*doorp);
6371 *doorp = NULL;
6372 }
6373
6374 static void
6375 zone_ki_call_zoneadmd(struct zarg *zargp)
6376 {
6377 door_handle_t door = NULL;
6378 door_arg_t darg, save_arg;
6379 char *zone_name;
6380 size_t zone_namelen;
6381 zoneid_t zoneid;
6382 zone_t *zone;
6383 zone_cmd_arg_t arg;
6384 uint64_t uniqid;
6385 size_t size;
6386 int error;
6387 int retry;
6388
6389 zone = zargp->zone;
6390 arg = zargp->arg;
6391 kmem_free(zargp, sizeof (*zargp));
6392
6393 zone_namelen = strlen(zone->zone_name) + 1;
6394 zone_name = kmem_alloc(zone_namelen, KM_SLEEP);
6395 bcopy(zone->zone_name, zone_name, zone_namelen);
6396 zoneid = zone->zone_id;
6397 uniqid = zone->zone_uniqid;
6398 /*
6399 * zoneadmd may be down, but at least we can empty out the zone.
6400 * We can ignore the return value of zone_empty() since we're called
6401 * from a kernel thread and know we won't be delivered any signals.
6402 */
6403 ASSERT(curproc == &p0);
6404 (void) zone_empty(zone);
6405 ASSERT(zone_status_get(zone) >= ZONE_IS_EMPTY);
6406 zone_rele(zone);
6407
6408 size = sizeof (arg);
6409 darg.rbuf = (char *)&arg;
6410 darg.data_ptr = (char *)&arg;
6411 darg.rsize = size;
6412 darg.data_size = size;
6413 darg.desc_ptr = NULL;
6414 darg.desc_num = 0;
6415
6416 save_arg = darg;
6417 /*
6418 * Since we're not holding a reference to the zone, any number of
6419 * things can go wrong, including the zone disappearing before we get a
6420 * chance to talk to zoneadmd.
6421 */
6422 for (retry = 0; /* forever */; retry++) {
6423 if (door == NULL &&
6424 (error = zone_lookup_door(zone_name, &door)) != 0) {
6425 goto next;
6426 }
6427 ASSERT(door != NULL);
6428
6429 if ((error = door_ki_upcall_limited(door, &darg, NULL,
6430 SIZE_MAX, 0)) == 0) {
6431 break;
6432 }
6433 switch (error) {
6434 case EINTR:
6435 /* FALLTHROUGH */
6436 case EAGAIN: /* process may be forking */
6437 /*
6438 * Back off for a bit
6439 */
6440 break;
6441 case EBADF:
6442 zone_release_door(&door);
6443 if (zone_lookup_door(zone_name, &door) != 0) {
6444 /*
6445 * zoneadmd may be dead, but it may come back to
6446 * life later.
6447 */
6448 break;
6449 }
6450 break;
6451 default:
6452 cmn_err(CE_WARN,
6453 "zone_ki_call_zoneadmd: door_ki_upcall error %d\n",
6454 error);
6455 goto out;
6456 }
6457 next:
6458 /*
6459 * If this isn't the same zone_t that we originally had in mind,
6460 * then this is the same as if two kadmin requests come in at
6461 * the same time: the first one wins. This means we lose, so we
6462 * bail.
6463 */
6464 if ((zone = zone_find_by_id(zoneid)) == NULL) {
6465 /*
6466 * Problem is solved.
6467 */
6468 break;
6469 }
6470 if (zone->zone_uniqid != uniqid) {
6471 /*
6472 * zoneid recycled
6473 */
6474 zone_rele(zone);
6475 break;
6476 }
6477 /*
6478 * We could zone_status_timedwait(), but there doesn't seem to
6479 * be much point in doing that (plus, it would mean that
6480 * zone_free() isn't called until this thread exits).
6481 */
6482 zone_rele(zone);
6483 delay(hz);
6484 darg = save_arg;
6485 }
6486 out:
6487 if (door != NULL) {
6488 zone_release_door(&door);
6489 }
6490 kmem_free(zone_name, zone_namelen);
6491 thread_exit();
6492 }
6493
6494 /*
6495 * Entry point for uadmin() to tell the zone to go away or reboot. Analog to
6496 * kadmin(). The caller is a process in the zone.
6497 *
6498 * In order to shutdown the zone, we will hand off control to zoneadmd
6499 * (running in the global zone) via a door. We do a half-hearted job at
6500 * killing all processes in the zone, create a kernel thread to contact
6501 * zoneadmd, and make note of the "uniqid" of the zone. The uniqid is
6502 * a form of generation number used to let zoneadmd (as well as
6503 * zone_destroy()) know exactly which zone they're re talking about.
6504 */
6505 int
6506 zone_kadmin(int cmd, int fcn, const char *mdep, cred_t *credp)
6507 {
6508 struct zarg *zargp;
6509 zone_cmd_t zcmd;
6510 zone_t *zone;
6511
6512 zone = curproc->p_zone;
6513 ASSERT(getzoneid() != GLOBAL_ZONEID);
6514
6515 switch (cmd) {
6516 case A_SHUTDOWN:
6517 switch (fcn) {
6518 case AD_HALT:
6519 case AD_POWEROFF:
6520 zcmd = Z_HALT;
6521 break;
6522 case AD_BOOT:
6523 zcmd = Z_REBOOT;
6524 break;
6525 case AD_IBOOT:
6526 case AD_SBOOT:
6527 case AD_SIBOOT:
6528 case AD_NOSYNC:
6529 return (ENOTSUP);
6530 default:
6531 return (EINVAL);
6532 }
6533 break;
6534 case A_REBOOT:
6535 zcmd = Z_REBOOT;
6536 break;
6537 case A_FTRACE:
6538 case A_REMOUNT:
6539 case A_FREEZE:
6540 case A_DUMP:
6541 case A_CONFIG:
6542 return (ENOTSUP);
6543 default:
6544 ASSERT(cmd != A_SWAPCTL); /* handled by uadmin() */
6545 return (EINVAL);
6546 }
6547
6548 if (secpolicy_zone_admin(credp, B_FALSE))
6549 return (EPERM);
6550 mutex_enter(&zone_status_lock);
6551
6552 /*
6553 * zone_status can't be ZONE_IS_EMPTY or higher since curproc
6554 * is in the zone.
6555 */
6556 ASSERT(zone_status_get(zone) < ZONE_IS_EMPTY);
6557 if (zone_status_get(zone) > ZONE_IS_RUNNING) {
6558 /*
6559 * This zone is already on its way down.
6560 */
6561 mutex_exit(&zone_status_lock);
6562 return (0);
6563 }
6564 /*
6565 * Prevent future zone_enter()s
6566 */
6567 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
6568 mutex_exit(&zone_status_lock);
6569
6570 /*
6571 * Kill everyone now and call zoneadmd later.
6572 * zone_ki_call_zoneadmd() will do a more thorough job of this
6573 * later.
6574 */
6575 killall(zone->zone_id);
6576 /*
6577 * Now, create the thread to contact zoneadmd and do the rest of the
6578 * work. This thread can't be created in our zone otherwise
6579 * zone_destroy() would deadlock.
6580 */
6581 zargp = kmem_zalloc(sizeof (*zargp), KM_SLEEP);
6582 zargp->arg.cmd = zcmd;
6583 zargp->arg.uniqid = zone->zone_uniqid;
6584 zargp->zone = zone;
6585 (void) strcpy(zargp->arg.locale, "C");
6586 /* mdep was already copied in for us by uadmin */
6587 if (mdep != NULL)
6588 (void) strlcpy(zargp->arg.bootbuf, mdep,
6589 sizeof (zargp->arg.bootbuf));
6590 zone_hold(zone);
6591
6592 (void) thread_create(NULL, 0, zone_ki_call_zoneadmd, zargp, 0, &p0,
6593 TS_RUN, minclsyspri);
6594 exit(CLD_EXITED, 0);
6595
6596 return (EINVAL);
6597 }
6598
6599 /*
6600 * Entry point so kadmin(A_SHUTDOWN, ...) can set the global zone's
6601 * status to ZONE_IS_SHUTTING_DOWN.
6602 *
6603 * This function also shuts down all running zones to ensure that they won't
6604 * fork new processes.
6605 */
6606 void
6607 zone_shutdown_global(void)
6608 {
6609 zone_t *current_zonep;
6610
6611 ASSERT(INGLOBALZONE(curproc));
6612 mutex_enter(&zonehash_lock);
6613 mutex_enter(&zone_status_lock);
6614
6615 /* Modify the global zone's status first. */
6616 ASSERT(zone_status_get(global_zone) == ZONE_IS_RUNNING);
6617 zone_status_set(global_zone, ZONE_IS_SHUTTING_DOWN);
6618
6619 /*
6620 * Now change the states of all running zones to ZONE_IS_SHUTTING_DOWN.
6621 * We don't mark all zones with ZONE_IS_SHUTTING_DOWN because doing so
6622 * could cause assertions to fail (e.g., assertions about a zone's
6623 * state during initialization, readying, or booting) or produce races.
6624 * We'll let threads continue to initialize and ready new zones: they'll
6625 * fail to boot the new zones when they see that the global zone is
6626 * shutting down.
6627 */
6628 for (current_zonep = list_head(&zone_active); current_zonep != NULL;
6629 current_zonep = list_next(&zone_active, current_zonep)) {
6630 if (zone_status_get(current_zonep) == ZONE_IS_RUNNING)
6631 zone_status_set(current_zonep, ZONE_IS_SHUTTING_DOWN);
6632 }
6633 mutex_exit(&zone_status_lock);
6634 mutex_exit(&zonehash_lock);
6635 }
6636
6637 /*
6638 * Returns true if the named dataset is visible in the current zone.
6639 * The 'write' parameter is set to 1 if the dataset is also writable.
6640 */
6641 int
6642 zone_dataset_visible(const char *dataset, int *write)
6643 {
6644 static int zfstype = -1;
6645 zone_dataset_t *zd;
6646 size_t len;
6647 zone_t *zone = curproc->p_zone;
6648 const char *name = NULL;
6649 vfs_t *vfsp = NULL;
6650
6651 if (dataset[0] == '\0')
6652 return (0);
6653
6654 /*
6655 * Walk the list once, looking for datasets which match exactly, or
6656 * specify a dataset underneath an exported dataset. If found, return
6657 * true and note that it is writable.
6658 */
6659 for (zd = list_head(&zone->zone_datasets); zd != NULL;
6660 zd = list_next(&zone->zone_datasets, zd)) {
6661
6662 len = strlen(zd->zd_dataset);
6663 if (strlen(dataset) >= len &&
6664 bcmp(dataset, zd->zd_dataset, len) == 0 &&
6665 (dataset[len] == '\0' || dataset[len] == '/' ||
6666 dataset[len] == '@')) {
6667 if (write)
6668 *write = 1;
6669 return (1);
6670 }
6671 }
6672
6673 /*
6674 * Walk the list a second time, searching for datasets which are parents
6675 * of exported datasets. These should be visible, but read-only.
6676 *
6677 * Note that we also have to support forms such as 'pool/dataset/', with
6678 * a trailing slash.
6679 */
6680 for (zd = list_head(&zone->zone_datasets); zd != NULL;
6681 zd = list_next(&zone->zone_datasets, zd)) {
6682
6683 len = strlen(dataset);
6684 if (dataset[len - 1] == '/')
6685 len--; /* Ignore trailing slash */
6686 if (len < strlen(zd->zd_dataset) &&
6687 bcmp(dataset, zd->zd_dataset, len) == 0 &&
6688 zd->zd_dataset[len] == '/') {
6689 if (write)
6690 *write = 0;
6691 return (1);
6692 }
6693 }
6694
6695 /*
6696 * We reach here if the given dataset is not found in the zone_dataset
6697 * list. Check if this dataset was added as a filesystem (ie. "add fs")
6698 * instead of delegation. For this we search for the dataset in the
6699 * zone_vfslist of this zone. If found, return true and note that it is
6700 * not writable.
6701 */
6702
6703 /*
6704 * Initialize zfstype if it is not initialized yet.
6705 */
6706 if (zfstype == -1) {
6707 struct vfssw *vswp = vfs_getvfssw("zfs");
6708 zfstype = vswp - vfssw;
6709 vfs_unrefvfssw(vswp);
6710 }
6711
6712 vfs_list_read_lock();
6713 vfsp = zone->zone_vfslist;
6714 do {
6715 ASSERT(vfsp);
6716 if (vfsp->vfs_fstype == zfstype) {
6717 name = refstr_value(vfsp->vfs_resource);
6718
6719 /*
6720 * Check if we have an exact match.
6721 */
6722 if (strcmp(dataset, name) == 0) {
6723 vfs_list_unlock();
6724 if (write)
6725 *write = 0;
6726 return (1);
6727 }
6728 /*
6729 * We need to check if we are looking for parents of
6730 * a dataset. These should be visible, but read-only.
6731 */
6732 len = strlen(dataset);
6733 if (dataset[len - 1] == '/')
6734 len--;
6735
6736 if (len < strlen(name) &&
6737 bcmp(dataset, name, len) == 0 && name[len] == '/') {
6738 vfs_list_unlock();
6739 if (write)
6740 *write = 0;
6741 return (1);
6742 }
6743 }
6744 vfsp = vfsp->vfs_zone_next;
6745 } while (vfsp != zone->zone_vfslist);
6746
6747 vfs_list_unlock();
6748 return (0);
6749 }
6750
6751 /*
6752 * zone_find_by_any_path() -
6753 *
6754 * kernel-private routine similar to zone_find_by_path(), but which
6755 * effectively compares against zone paths rather than zonerootpath
6756 * (i.e., the last component of zonerootpaths, which should be "root/",
6757 * are not compared.) This is done in order to accurately identify all
6758 * paths, whether zone-visible or not, including those which are parallel
6759 * to /root/, such as /dev/, /home/, etc...
6760 *
6761 * If the specified path does not fall under any zone path then global
6762 * zone is returned.
6763 *
6764 * The treat_abs parameter indicates whether the path should be treated as
6765 * an absolute path although it does not begin with "/". (This supports
6766 * nfs mount syntax such as host:any/path.)
6767 *
6768 * The caller is responsible for zone_rele of the returned zone.
6769 */
6770 zone_t *
6771 zone_find_by_any_path(const char *path, boolean_t treat_abs)
6772 {
6773 zone_t *zone;
6774 int path_offset = 0;
6775
6776 if (path == NULL) {
6777 zone_hold(global_zone);
6778 return (global_zone);
6779 }
6780
6781 if (*path != '/') {
6782 ASSERT(treat_abs);
6783 path_offset = 1;
6784 }
6785
6786 mutex_enter(&zonehash_lock);
6787 for (zone = list_head(&zone_active); zone != NULL;
6788 zone = list_next(&zone_active, zone)) {
6789 char *c;
6790 size_t pathlen;
6791 char *rootpath_start;
6792
6793 if (zone == global_zone) /* skip global zone */
6794 continue;
6795
6796 /* scan backwards to find start of last component */
6797 c = zone->zone_rootpath + zone->zone_rootpathlen - 2;
6798 do {
6799 c--;
6800 } while (*c != '/');
6801
6802 pathlen = c - zone->zone_rootpath + 1 - path_offset;
6803 rootpath_start = (zone->zone_rootpath + path_offset);
6804 if (strncmp(path, rootpath_start, pathlen) == 0)
6805 break;
6806 }
6807 if (zone == NULL)
6808 zone = global_zone;
6809 zone_hold(zone);
6810 mutex_exit(&zonehash_lock);
6811 return (zone);
6812 }
6813
6814 /*
6815 * Finds a zone_dl_t with the given linkid in the given zone. Returns the
6816 * zone_dl_t pointer if found, and NULL otherwise.
6817 */
6818 static zone_dl_t *
6819 zone_find_dl(zone_t *zone, datalink_id_t linkid)
6820 {
6821 zone_dl_t *zdl;
6822
6823 ASSERT(mutex_owned(&zone->zone_lock));
6824 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
6825 zdl = list_next(&zone->zone_dl_list, zdl)) {
6826 if (zdl->zdl_id == linkid)
6827 break;
6828 }
6829 return (zdl);
6830 }
6831
6832 static boolean_t
6833 zone_dl_exists(zone_t *zone, datalink_id_t linkid)
6834 {
6835 boolean_t exists;
6836
6837 mutex_enter(&zone->zone_lock);
6838 exists = (zone_find_dl(zone, linkid) != NULL);
6839 mutex_exit(&zone->zone_lock);
6840 return (exists);
6841 }
6842
6843 /*
6844 * Add an data link name for the zone.
6845 */
6846 static int
6847 zone_add_datalink(zoneid_t zoneid, datalink_id_t linkid)
6848 {
6849 zone_dl_t *zdl;
6850 zone_t *zone;
6851 zone_t *thiszone;
6852
6853 if ((thiszone = zone_find_by_id(zoneid)) == NULL)
6854 return (set_errno(ENXIO));
6855
6856 /* Verify that the datalink ID doesn't already belong to a zone. */
6857 mutex_enter(&zonehash_lock);
6858 for (zone = list_head(&zone_active); zone != NULL;
6859 zone = list_next(&zone_active, zone)) {
6860 if (zone_dl_exists(zone, linkid)) {
6861 mutex_exit(&zonehash_lock);
6862 zone_rele(thiszone);
6863 return (set_errno((zone == thiszone) ? EEXIST : EPERM));
6864 }
6865 }
6866
6867 zdl = kmem_zalloc(sizeof (*zdl), KM_SLEEP);
6868 zdl->zdl_id = linkid;
6869 zdl->zdl_net = NULL;
6870 mutex_enter(&thiszone->zone_lock);
6871 list_insert_head(&thiszone->zone_dl_list, zdl);
6872 mutex_exit(&thiszone->zone_lock);
6873 mutex_exit(&zonehash_lock);
6874 zone_rele(thiszone);
6875 return (0);
6876 }
6877
6878 static int
6879 zone_remove_datalink(zoneid_t zoneid, datalink_id_t linkid)
6880 {
6881 zone_dl_t *zdl;
6882 zone_t *zone;
6883 int err = 0;
6884
6885 if ((zone = zone_find_by_id(zoneid)) == NULL)
6886 return (set_errno(EINVAL));
6887
6888 mutex_enter(&zone->zone_lock);
6889 if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
6890 err = ENXIO;
6891 } else {
6892 list_remove(&zone->zone_dl_list, zdl);
6893 if (zdl->zdl_net != NULL)
6894 nvlist_free(zdl->zdl_net);
6895 kmem_free(zdl, sizeof (zone_dl_t));
6896 }
6897 mutex_exit(&zone->zone_lock);
6898 zone_rele(zone);
6899 return (err == 0 ? 0 : set_errno(err));
6900 }
6901
6902 /*
6903 * Using the zoneidp as ALL_ZONES, we can lookup which zone has been assigned
6904 * the linkid. Otherwise we just check if the specified zoneidp has been
6905 * assigned the supplied linkid.
6906 */
6907 int
6908 zone_check_datalink(zoneid_t *zoneidp, datalink_id_t linkid)
6909 {
6910 zone_t *zone;
6911 int err = ENXIO;
6912
6913 if (*zoneidp != ALL_ZONES) {
6914 if ((zone = zone_find_by_id(*zoneidp)) != NULL) {
6915 if (zone_dl_exists(zone, linkid))
6916 err = 0;
6917 zone_rele(zone);
6918 }
6919 return (err);
6920 }
6921
6922 mutex_enter(&zonehash_lock);
6923 for (zone = list_head(&zone_active); zone != NULL;
6924 zone = list_next(&zone_active, zone)) {
6925 if (zone_dl_exists(zone, linkid)) {
6926 *zoneidp = zone->zone_id;
6927 err = 0;
6928 break;
6929 }
6930 }
6931 mutex_exit(&zonehash_lock);
6932 return (err);
6933 }
6934
6935 /*
6936 * Get the list of datalink IDs assigned to a zone.
6937 *
6938 * On input, *nump is the number of datalink IDs that can fit in the supplied
6939 * idarray. Upon return, *nump is either set to the number of datalink IDs
6940 * that were placed in the array if the array was large enough, or to the
6941 * number of datalink IDs that the function needs to place in the array if the
6942 * array is too small.
6943 */
6944 static int
6945 zone_list_datalink(zoneid_t zoneid, int *nump, datalink_id_t *idarray)
6946 {
6947 uint_t num, dlcount;
6948 zone_t *zone;
6949 zone_dl_t *zdl;
6950 datalink_id_t *idptr = idarray;
6951
6952 if (copyin(nump, &dlcount, sizeof (dlcount)) != 0)
6953 return (set_errno(EFAULT));
6954 if ((zone = zone_find_by_id(zoneid)) == NULL)
6955 return (set_errno(ENXIO));
6956
6957 num = 0;
6958 mutex_enter(&zone->zone_lock);
6959 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
6960 zdl = list_next(&zone->zone_dl_list, zdl)) {
6961 /*
6962 * If the list is bigger than what the caller supplied, just
6963 * count, don't do copyout.
6964 */
6965 if (++num > dlcount)
6966 continue;
6967 if (copyout(&zdl->zdl_id, idptr, sizeof (*idptr)) != 0) {
6968 mutex_exit(&zone->zone_lock);
6969 zone_rele(zone);
6970 return (set_errno(EFAULT));
6971 }
6972 idptr++;
6973 }
6974 mutex_exit(&zone->zone_lock);
6975 zone_rele(zone);
6976
6977 /* Increased or decreased, caller should be notified. */
6978 if (num != dlcount) {
6979 if (copyout(&num, nump, sizeof (num)) != 0)
6980 return (set_errno(EFAULT));
6981 }
6982 return (0);
6983 }
6984
6985 /*
6986 * Public interface for looking up a zone by zoneid. It's a customized version
6987 * for netstack_zone_create(). It can only be called from the zsd create
6988 * callbacks, since it doesn't have reference on the zone structure hence if
6989 * it is called elsewhere the zone could disappear after the zonehash_lock
6990 * is dropped.
6991 *
6992 * Furthermore it
6993 * 1. Doesn't check the status of the zone.
6994 * 2. It will be called even before zone_init is called, in that case the
6995 * address of zone0 is returned directly, and netstack_zone_create()
6996 * will only assign a value to zone0.zone_netstack, won't break anything.
6997 * 3. Returns without the zone being held.
6998 */
6999 zone_t *
7000 zone_find_by_id_nolock(zoneid_t zoneid)
7001 {
7002 zone_t *zone;
7003
7004 mutex_enter(&zonehash_lock);
7005 if (zonehashbyid == NULL)
7006 zone = &zone0;
7007 else
7008 zone = zone_find_all_by_id(zoneid);
7009 mutex_exit(&zonehash_lock);
7010 return (zone);
7011 }
7012
7013 /*
7014 * Walk the datalinks for a given zone
7015 */
7016 int
7017 zone_datalink_walk(zoneid_t zoneid, int (*cb)(datalink_id_t, void *),
7018 void *data)
7019 {
7020 zone_t *zone;
7021 zone_dl_t *zdl;
7022 datalink_id_t *idarray;
7023 uint_t idcount = 0;
7024 int i, ret = 0;
7025
7026 if ((zone = zone_find_by_id(zoneid)) == NULL)
7027 return (ENOENT);
7028
7029 /*
7030 * We first build an array of linkid's so that we can walk these and
7031 * execute the callback with the zone_lock dropped.
7032 */
7033 mutex_enter(&zone->zone_lock);
7034 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7035 zdl = list_next(&zone->zone_dl_list, zdl)) {
7036 idcount++;
7037 }
7038
7039 if (idcount == 0) {
7040 mutex_exit(&zone->zone_lock);
7041 zone_rele(zone);
7042 return (0);
7043 }
7044
7045 idarray = kmem_alloc(sizeof (datalink_id_t) * idcount, KM_NOSLEEP);
7046 if (idarray == NULL) {
7047 mutex_exit(&zone->zone_lock);
7048 zone_rele(zone);
7049 return (ENOMEM);
7050 }
7051
7052 for (i = 0, zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7053 i++, zdl = list_next(&zone->zone_dl_list, zdl)) {
7054 idarray[i] = zdl->zdl_id;
7055 }
7056
7057 mutex_exit(&zone->zone_lock);
7058
7059 for (i = 0; i < idcount && ret == 0; i++) {
7060 if ((ret = (*cb)(idarray[i], data)) != 0)
7061 break;
7062 }
7063
7064 zone_rele(zone);
7065 kmem_free(idarray, sizeof (datalink_id_t) * idcount);
7066 return (ret);
7067 }
7068
7069 static char *
7070 zone_net_type2name(int type)
7071 {
7072 switch (type) {
7073 case ZONE_NETWORK_ADDRESS:
7074 return (ZONE_NET_ADDRNAME);
7075 case ZONE_NETWORK_DEFROUTER:
7076 return (ZONE_NET_RTRNAME);
7077 default:
7078 return (NULL);
7079 }
7080 }
7081
7082 static int
7083 zone_set_network(zoneid_t zoneid, zone_net_data_t *znbuf)
7084 {
7085 zone_t *zone;
7086 zone_dl_t *zdl;
7087 nvlist_t *nvl;
7088 int err = 0;
7089 uint8_t *new = NULL;
7090 char *nvname;
7091 int bufsize;
7092 datalink_id_t linkid = znbuf->zn_linkid;
7093
7094 if (secpolicy_zone_config(CRED()) != 0)
7095 return (set_errno(EPERM));
7096
7097 if (zoneid == GLOBAL_ZONEID)
7098 return (set_errno(EINVAL));
7099
7100 nvname = zone_net_type2name(znbuf->zn_type);
7101 bufsize = znbuf->zn_len;
7102 new = znbuf->zn_val;
7103 if (nvname == NULL)
7104 return (set_errno(EINVAL));
7105
7106 if ((zone = zone_find_by_id(zoneid)) == NULL) {
7107 return (set_errno(EINVAL));
7108 }
7109
7110 mutex_enter(&zone->zone_lock);
7111 if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
7112 err = ENXIO;
7113 goto done;
7114 }
7115 if ((nvl = zdl->zdl_net) == NULL) {
7116 if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP)) {
7117 err = ENOMEM;
7118 goto done;
7119 } else {
7120 zdl->zdl_net = nvl;
7121 }
7122 }
7123 if (nvlist_exists(nvl, nvname)) {
7124 err = EINVAL;
7125 goto done;
7126 }
7127 err = nvlist_add_uint8_array(nvl, nvname, new, bufsize);
7128 ASSERT(err == 0);
7129 done:
7130 mutex_exit(&zone->zone_lock);
7131 zone_rele(zone);
7132 if (err != 0)
7133 return (set_errno(err));
7134 else
7135 return (0);
7136 }
7137
7138 static int
7139 zone_get_network(zoneid_t zoneid, zone_net_data_t *znbuf)
7140 {
7141 zone_t *zone;
7142 zone_dl_t *zdl;
7143 nvlist_t *nvl;
7144 uint8_t *ptr;
7145 uint_t psize;
7146 int err = 0;
7147 char *nvname;
7148 int bufsize;
7149 void *buf;
7150 datalink_id_t linkid = znbuf->zn_linkid;
7151
7152 if (zoneid == GLOBAL_ZONEID)
7153 return (set_errno(EINVAL));
7154
7155 nvname = zone_net_type2name(znbuf->zn_type);
7156 bufsize = znbuf->zn_len;
7157 buf = znbuf->zn_val;
7158
7159 if (nvname == NULL)
7160 return (set_errno(EINVAL));
7161 if ((zone = zone_find_by_id(zoneid)) == NULL)
7162 return (set_errno(EINVAL));
7163
7164 mutex_enter(&zone->zone_lock);
7165 if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
7166 err = ENXIO;
7167 goto done;
7168 }
7169 if ((nvl = zdl->zdl_net) == NULL || !nvlist_exists(nvl, nvname)) {
7170 err = ENOENT;
7171 goto done;
7172 }
7173 err = nvlist_lookup_uint8_array(nvl, nvname, &ptr, &psize);
7174 ASSERT(err == 0);
7175
7176 if (psize > bufsize) {
7177 err = ENOBUFS;
7178 goto done;
7179 }
7180 znbuf->zn_len = psize;
7181 bcopy(ptr, buf, psize);
7182 done:
7183 mutex_exit(&zone->zone_lock);
7184 zone_rele(zone);
7185 if (err != 0)
7186 return (set_errno(err));
7187 else
7188 return (0);
7189 }