Print this page
9936 atomic ops in syscall_mstate() induce significant overhead
9942 zone secflags are not initialized correctly
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/os/zone.c
+++ new/usr/src/uts/common/os/zone.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 * Copyright 2015, Joyent Inc. All rights reserved.
25 25 * Copyright (c) 2016 by Delphix. All rights reserved.
26 26 * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
27 27 */
28 28
29 29 /*
30 30 * Zones
31 31 *
32 32 * A zone is a named collection of processes, namespace constraints,
33 33 * and other system resources which comprise a secure and manageable
34 34 * application containment facility.
35 35 *
36 36 * Zones (represented by the reference counted zone_t) are tracked in
37 37 * the kernel in the zonehash. Elsewhere in the kernel, Zone IDs
38 38 * (zoneid_t) are used to track zone association. Zone IDs are
39 39 * dynamically generated when the zone is created; if a persistent
40 40 * identifier is needed (core files, accounting logs, audit trail,
41 41 * etc.), the zone name should be used.
42 42 *
43 43 *
44 44 * Global Zone:
45 45 *
46 46 * The global zone (zoneid 0) is automatically associated with all
47 47 * system resources that have not been bound to a user-created zone.
48 48 * This means that even systems where zones are not in active use
49 49 * have a global zone, and all processes, mounts, etc. are
50 50 * associated with that zone. The global zone is generally
51 51 * unconstrained in terms of privileges and access, though the usual
52 52 * credential and privilege based restrictions apply.
53 53 *
54 54 *
55 55 * Zone States:
56 56 *
57 57 * The states in which a zone may be in and the transitions are as
58 58 * follows:
59 59 *
60 60 * ZONE_IS_UNINITIALIZED: primordial state for a zone. The partially
61 61 * initialized zone is added to the list of active zones on the system but
62 62 * isn't accessible.
63 63 *
64 64 * ZONE_IS_INITIALIZED: Initialization complete except the ZSD callbacks are
65 65 * not yet completed. Not possible to enter the zone, but attributes can
66 66 * be retrieved.
67 67 *
68 68 * ZONE_IS_READY: zsched (the kernel dummy process for a zone) is
69 69 * ready. The zone is made visible after the ZSD constructor callbacks are
70 70 * executed. A zone remains in this state until it transitions into
71 71 * the ZONE_IS_BOOTING state as a result of a call to zone_boot().
72 72 *
73 73 * ZONE_IS_BOOTING: in this shortlived-state, zsched attempts to start
74 74 * init. Should that fail, the zone proceeds to the ZONE_IS_SHUTTING_DOWN
75 75 * state.
76 76 *
77 77 * ZONE_IS_RUNNING: The zone is open for business: zsched has
78 78 * successfully started init. A zone remains in this state until
79 79 * zone_shutdown() is called.
80 80 *
81 81 * ZONE_IS_SHUTTING_DOWN: zone_shutdown() has been called, the system is
82 82 * killing all processes running in the zone. The zone remains
83 83 * in this state until there are no more user processes running in the zone.
84 84 * zone_create(), zone_enter(), and zone_destroy() on this zone will fail.
85 85 * Since zone_shutdown() is restartable, it may be called successfully
86 86 * multiple times for the same zone_t. Setting of the zone's state to
87 87 * ZONE_IS_SHUTTING_DOWN is synchronized with mounts, so VOP_MOUNT() may check
88 88 * the zone's status without worrying about it being a moving target.
89 89 *
90 90 * ZONE_IS_EMPTY: zone_shutdown() has been called, and there
91 91 * are no more user processes in the zone. The zone remains in this
92 92 * state until there are no more kernel threads associated with the
93 93 * zone. zone_create(), zone_enter(), and zone_destroy() on this zone will
94 94 * fail.
95 95 *
96 96 * ZONE_IS_DOWN: All kernel threads doing work on behalf of the zone
97 97 * have exited. zone_shutdown() returns. Henceforth it is not possible to
98 98 * join the zone or create kernel threads therein.
99 99 *
100 100 * ZONE_IS_DYING: zone_destroy() has been called on the zone; zone
101 101 * remains in this state until zsched exits. Calls to zone_find_by_*()
102 102 * return NULL from now on.
103 103 *
104 104 * ZONE_IS_DEAD: zsched has exited (zone_ntasks == 0). There are no
105 105 * processes or threads doing work on behalf of the zone. The zone is
106 106 * removed from the list of active zones. zone_destroy() returns, and
107 107 * the zone can be recreated.
108 108 *
109 109 * ZONE_IS_FREE (internal state): zone_ref goes to 0, ZSD destructor
110 110 * callbacks are executed, and all memory associated with the zone is
111 111 * freed.
112 112 *
113 113 * Threads can wait for the zone to enter a requested state by using
114 114 * zone_status_wait() or zone_status_timedwait() with the desired
115 115 * state passed in as an argument. Zone state transitions are
116 116 * uni-directional; it is not possible to move back to an earlier state.
117 117 *
118 118 *
119 119 * Zone-Specific Data:
120 120 *
121 121 * Subsystems needing to maintain zone-specific data can store that
122 122 * data using the ZSD mechanism. This provides a zone-specific data
123 123 * store, similar to thread-specific data (see pthread_getspecific(3C)
124 124 * or the TSD code in uts/common/disp/thread.c. Also, ZSD can be used
125 125 * to register callbacks to be invoked when a zone is created, shut
126 126 * down, or destroyed. This can be used to initialize zone-specific
127 127 * data for new zones and to clean up when zones go away.
128 128 *
129 129 *
130 130 * Data Structures:
131 131 *
132 132 * The per-zone structure (zone_t) is reference counted, and freed
133 133 * when all references are released. zone_hold and zone_rele can be
134 134 * used to adjust the reference count. In addition, reference counts
135 135 * associated with the cred_t structure are tracked separately using
136 136 * zone_cred_hold and zone_cred_rele.
137 137 *
138 138 * Pointers to active zone_t's are stored in two hash tables; one
139 139 * for searching by id, the other for searching by name. Lookups
140 140 * can be performed on either basis, using zone_find_by_id and
141 141 * zone_find_by_name. Both return zone_t pointers with the zone
142 142 * held, so zone_rele should be called when the pointer is no longer
143 143 * needed. Zones can also be searched by path; zone_find_by_path
144 144 * returns the zone with which a path name is associated (global
145 145 * zone if the path is not within some other zone's file system
146 146 * hierarchy). This currently requires iterating through each zone,
147 147 * so it is slower than an id or name search via a hash table.
148 148 *
149 149 *
150 150 * Locking:
151 151 *
152 152 * zonehash_lock: This is a top-level global lock used to protect the
153 153 * zone hash tables and lists. Zones cannot be created or destroyed
154 154 * while this lock is held.
155 155 * zone_status_lock: This is a global lock protecting zone state.
156 156 * Zones cannot change state while this lock is held. It also
157 157 * protects the list of kernel threads associated with a zone.
158 158 * zone_lock: This is a per-zone lock used to protect several fields of
159 159 * the zone_t (see <sys/zone.h> for details). In addition, holding
160 160 * this lock means that the zone cannot go away.
161 161 * zone_nlwps_lock: This is a per-zone lock used to protect the fields
162 162 * related to the zone.max-lwps rctl.
|
↓ open down ↓ |
162 lines elided |
↑ open up ↑ |
163 163 * zone_mem_lock: This is a per-zone lock used to protect the fields
164 164 * related to the zone.max-locked-memory and zone.max-swap rctls.
165 165 * zone_rctl_lock: This is a per-zone lock used to protect other rctls,
166 166 * currently just max_lofi
167 167 * zsd_key_lock: This is a global lock protecting the key state for ZSD.
168 168 * zone_deathrow_lock: This is a global lock protecting the "deathrow"
169 169 * list (a list of zones in the ZONE_IS_DEAD state).
170 170 *
171 171 * Ordering requirements:
172 172 * pool_lock --> cpu_lock --> zonehash_lock --> zone_status_lock -->
173 - * zone_lock --> zsd_key_lock --> pidlock --> p_lock
173 + * zone_lock --> zsd_key_lock --> pidlock --> p_lock
174 174 *
175 175 * When taking zone_mem_lock or zone_nlwps_lock, the lock ordering is:
176 176 * zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_mem_lock
177 177 * zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_nlwps_lock
178 178 *
179 179 * Blocking memory allocations are permitted while holding any of the
180 180 * zone locks.
181 181 *
182 182 *
183 183 * System Call Interface:
184 184 *
185 185 * The zone subsystem can be managed and queried from user level with
186 186 * the following system calls (all subcodes of the primary "zone"
187 187 * system call):
188 188 * - zone_create: creates a zone with selected attributes (name,
189 189 * root path, privileges, resource controls, ZFS datasets)
190 190 * - zone_enter: allows the current process to enter a zone
191 191 * - zone_getattr: reports attributes of a zone
192 192 * - zone_setattr: set attributes of a zone
193 193 * - zone_boot: set 'init' running for the zone
194 194 * - zone_list: lists all zones active in the system
195 195 * - zone_lookup: looks up zone id based on name
196 196 * - zone_shutdown: initiates shutdown process (see states above)
197 197 * - zone_destroy: completes shutdown process (see states above)
198 198 *
199 199 */
200 200
201 201 #include <sys/priv_impl.h>
202 202 #include <sys/cred.h>
203 203 #include <c2/audit.h>
204 204 #include <sys/debug.h>
205 205 #include <sys/file.h>
206 206 #include <sys/kmem.h>
207 207 #include <sys/kstat.h>
208 208 #include <sys/mutex.h>
209 209 #include <sys/note.h>
210 210 #include <sys/pathname.h>
211 211 #include <sys/proc.h>
212 212 #include <sys/project.h>
213 213 #include <sys/sysevent.h>
214 214 #include <sys/task.h>
215 215 #include <sys/systm.h>
216 216 #include <sys/types.h>
217 217 #include <sys/utsname.h>
218 218 #include <sys/vnode.h>
219 219 #include <sys/vfs.h>
220 220 #include <sys/systeminfo.h>
221 221 #include <sys/policy.h>
222 222 #include <sys/cred_impl.h>
223 223 #include <sys/contract_impl.h>
224 224 #include <sys/contract/process_impl.h>
225 225 #include <sys/class.h>
226 226 #include <sys/pool.h>
227 227 #include <sys/pool_pset.h>
228 228 #include <sys/pset.h>
229 229 #include <sys/strlog.h>
230 230 #include <sys/sysmacros.h>
231 231 #include <sys/callb.h>
232 232 #include <sys/vmparam.h>
233 233 #include <sys/corectl.h>
234 234 #include <sys/ipc_impl.h>
235 235 #include <sys/klpd.h>
236 236
237 237 #include <sys/door.h>
238 238 #include <sys/cpuvar.h>
239 239 #include <sys/sdt.h>
240 240
241 241 #include <sys/uadmin.h>
242 242 #include <sys/session.h>
243 243 #include <sys/cmn_err.h>
244 244 #include <sys/modhash.h>
245 245 #include <sys/sunddi.h>
246 246 #include <sys/nvpair.h>
247 247 #include <sys/rctl.h>
248 248 #include <sys/fss.h>
249 249 #include <sys/brand.h>
250 250 #include <sys/zone.h>
251 251 #include <net/if.h>
252 252 #include <sys/cpucaps.h>
253 253 #include <vm/seg.h>
254 254 #include <sys/mac.h>
255 255
256 256 /*
257 257 * This constant specifies the number of seconds that threads waiting for
258 258 * subsystems to release a zone's general-purpose references will wait before
259 259 * they log the zone's reference counts. The constant's value shouldn't
260 260 * be so small that reference counts are unnecessarily reported for zones
261 261 * whose references are slowly released. On the other hand, it shouldn't be so
262 262 * large that users reboot their systems out of frustration over hung zones
263 263 * before the system logs the zones' reference counts.
264 264 */
265 265 #define ZONE_DESTROY_TIMEOUT_SECS 60
266 266
267 267 /* List of data link IDs which are accessible from the zone */
268 268 typedef struct zone_dl {
269 269 datalink_id_t zdl_id;
270 270 nvlist_t *zdl_net;
271 271 list_node_t zdl_linkage;
272 272 } zone_dl_t;
273 273
274 274 /*
275 275 * cv used to signal that all references to the zone have been released. This
276 276 * needs to be global since there may be multiple waiters, and the first to
277 277 * wake up will free the zone_t, hence we cannot use zone->zone_cv.
278 278 */
279 279 static kcondvar_t zone_destroy_cv;
280 280 /*
281 281 * Lock used to serialize access to zone_cv. This could have been per-zone,
282 282 * but then we'd need another lock for zone_destroy_cv, and why bother?
283 283 */
284 284 static kmutex_t zone_status_lock;
285 285
286 286 /*
287 287 * ZSD-related global variables.
288 288 */
289 289 static kmutex_t zsd_key_lock; /* protects the following two */
290 290 /*
291 291 * The next caller of zone_key_create() will be assigned a key of ++zsd_keyval.
292 292 */
293 293 static zone_key_t zsd_keyval = 0;
294 294 /*
295 295 * Global list of registered keys. We use this when a new zone is created.
296 296 */
297 297 static list_t zsd_registered_keys;
298 298
299 299 int zone_hash_size = 256;
300 300 static mod_hash_t *zonehashbyname, *zonehashbyid, *zonehashbylabel;
301 301 static kmutex_t zonehash_lock;
302 302 static uint_t zonecount;
303 303 static id_space_t *zoneid_space;
304 304
305 305 /*
306 306 * The global zone (aka zone0) is the all-seeing, all-knowing zone in which the
307 307 * kernel proper runs, and which manages all other zones.
308 308 *
309 309 * Although not declared as static, the variable "zone0" should not be used
310 310 * except for by code that needs to reference the global zone early on in boot,
311 311 * before it is fully initialized. All other consumers should use
312 312 * 'global_zone'.
313 313 */
314 314 zone_t zone0;
315 315 zone_t *global_zone = NULL; /* Set when the global zone is initialized */
316 316
317 317 /*
318 318 * List of active zones, protected by zonehash_lock.
319 319 */
320 320 static list_t zone_active;
321 321
322 322 /*
323 323 * List of destroyed zones that still have outstanding cred references.
324 324 * Used for debugging. Uses a separate lock to avoid lock ordering
325 325 * problems in zone_free.
326 326 */
327 327 static list_t zone_deathrow;
328 328 static kmutex_t zone_deathrow_lock;
329 329
330 330 /* number of zones is limited by virtual interface limit in IP */
331 331 uint_t maxzones = 8192;
332 332
333 333 /* Event channel to sent zone state change notifications */
334 334 evchan_t *zone_event_chan;
335 335
336 336 /*
337 337 * This table holds the mapping from kernel zone states to
338 338 * states visible in the state notification API.
339 339 * The idea is that we only expose "obvious" states and
340 340 * do not expose states which are just implementation details.
341 341 */
342 342 const char *zone_status_table[] = {
343 343 ZONE_EVENT_UNINITIALIZED, /* uninitialized */
344 344 ZONE_EVENT_INITIALIZED, /* initialized */
345 345 ZONE_EVENT_READY, /* ready */
346 346 ZONE_EVENT_READY, /* booting */
347 347 ZONE_EVENT_RUNNING, /* running */
348 348 ZONE_EVENT_SHUTTING_DOWN, /* shutting_down */
349 349 ZONE_EVENT_SHUTTING_DOWN, /* empty */
350 350 ZONE_EVENT_SHUTTING_DOWN, /* down */
351 351 ZONE_EVENT_SHUTTING_DOWN, /* dying */
352 352 ZONE_EVENT_UNINITIALIZED, /* dead */
353 353 };
354 354
355 355 /*
356 356 * This array contains the names of the subsystems listed in zone_ref_subsys_t
357 357 * (see sys/zone.h).
358 358 */
359 359 static char *zone_ref_subsys_names[] = {
360 360 "NFS", /* ZONE_REF_NFS */
361 361 "NFSv4", /* ZONE_REF_NFSV4 */
362 362 "SMBFS", /* ZONE_REF_SMBFS */
363 363 "MNTFS", /* ZONE_REF_MNTFS */
364 364 "LOFI", /* ZONE_REF_LOFI */
365 365 "VFS", /* ZONE_REF_VFS */
366 366 "IPC" /* ZONE_REF_IPC */
367 367 };
368 368
369 369 /*
370 370 * This isn't static so lint doesn't complain.
371 371 */
372 372 rctl_hndl_t rc_zone_cpu_shares;
373 373 rctl_hndl_t rc_zone_locked_mem;
374 374 rctl_hndl_t rc_zone_max_swap;
375 375 rctl_hndl_t rc_zone_max_lofi;
376 376 rctl_hndl_t rc_zone_cpu_cap;
377 377 rctl_hndl_t rc_zone_nlwps;
378 378 rctl_hndl_t rc_zone_nprocs;
379 379 rctl_hndl_t rc_zone_shmmax;
380 380 rctl_hndl_t rc_zone_shmmni;
381 381 rctl_hndl_t rc_zone_semmni;
382 382 rctl_hndl_t rc_zone_msgmni;
383 383
384 384 const char * const zone_default_initname = "/sbin/init";
385 385 static char * const zone_prefix = "/zone/";
386 386 static int zone_shutdown(zoneid_t zoneid);
387 387 static int zone_add_datalink(zoneid_t, datalink_id_t);
388 388 static int zone_remove_datalink(zoneid_t, datalink_id_t);
389 389 static int zone_list_datalink(zoneid_t, int *, datalink_id_t *);
390 390 static int zone_set_network(zoneid_t, zone_net_data_t *);
391 391 static int zone_get_network(zoneid_t, zone_net_data_t *);
392 392
393 393 typedef boolean_t zsd_applyfn_t(kmutex_t *, boolean_t, zone_t *, zone_key_t);
394 394
395 395 static void zsd_apply_all_zones(zsd_applyfn_t *, zone_key_t);
396 396 static void zsd_apply_all_keys(zsd_applyfn_t *, zone_t *);
397 397 static boolean_t zsd_apply_create(kmutex_t *, boolean_t, zone_t *, zone_key_t);
398 398 static boolean_t zsd_apply_shutdown(kmutex_t *, boolean_t, zone_t *,
399 399 zone_key_t);
400 400 static boolean_t zsd_apply_destroy(kmutex_t *, boolean_t, zone_t *, zone_key_t);
401 401 static boolean_t zsd_wait_for_creator(zone_t *, struct zsd_entry *,
402 402 kmutex_t *);
403 403 static boolean_t zsd_wait_for_inprogress(zone_t *, struct zsd_entry *,
404 404 kmutex_t *);
405 405
406 406 /*
407 407 * Bump this number when you alter the zone syscall interfaces; this is
408 408 * because we need to have support for previous API versions in libc
409 409 * to support patching; libc calls into the kernel to determine this number.
410 410 *
411 411 * Version 1 of the API is the version originally shipped with Solaris 10
412 412 * Version 2 alters the zone_create system call in order to support more
413 413 * arguments by moving the args into a structure; and to do better
414 414 * error reporting when zone_create() fails.
415 415 * Version 3 alters the zone_create system call in order to support the
416 416 * import of ZFS datasets to zones.
417 417 * Version 4 alters the zone_create system call in order to support
418 418 * Trusted Extensions.
419 419 * Version 5 alters the zone_boot system call, and converts its old
420 420 * bootargs parameter to be set by the zone_setattr API instead.
421 421 * Version 6 adds the flag argument to zone_create.
422 422 */
423 423 static const int ZONE_SYSCALL_API_VERSION = 6;
424 424
425 425 /*
426 426 * Certain filesystems (such as NFS and autofs) need to know which zone
427 427 * the mount is being placed in. Because of this, we need to be able to
428 428 * ensure that a zone isn't in the process of being created/destroyed such
429 429 * that nfs_mount() thinks it is in the global/NGZ zone, while by the time
430 430 * it gets added the list of mounted zones, it ends up on the wrong zone's
431 431 * mount list. Since a zone can't reside on an NFS file system, we don't
432 432 * have to worry about the zonepath itself.
433 433 *
434 434 * The following functions: block_mounts()/resume_mounts() and
435 435 * mount_in_progress()/mount_completed() are used by zones and the VFS
436 436 * layer (respectively) to synchronize zone state transitions and new
437 437 * mounts within a zone. This syncronization is on a per-zone basis, so
438 438 * activity for one zone will not interfere with activity for another zone.
439 439 *
440 440 * The semantics are like a reader-reader lock such that there may
441 441 * either be multiple mounts (or zone state transitions, if that weren't
442 442 * serialized by zonehash_lock) in progress at the same time, but not
443 443 * both.
444 444 *
445 445 * We use cv's so the user can ctrl-C out of the operation if it's
446 446 * taking too long.
447 447 *
448 448 * The semantics are such that there is unfair bias towards the
449 449 * "current" operation. This means that zone halt may starve if
450 450 * there is a rapid succession of new mounts coming in to the zone.
451 451 */
452 452 /*
453 453 * Prevent new mounts from progressing to the point of calling
454 454 * VFS_MOUNT(). If there are already mounts in this "region", wait for
455 455 * them to complete.
456 456 */
457 457 static int
458 458 block_mounts(zone_t *zp)
459 459 {
460 460 int retval = 0;
461 461
462 462 /*
463 463 * Since it may block for a long time, block_mounts() shouldn't be
464 464 * called with zonehash_lock held.
465 465 */
466 466 ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
467 467 mutex_enter(&zp->zone_mount_lock);
468 468 while (zp->zone_mounts_in_progress > 0) {
469 469 if (cv_wait_sig(&zp->zone_mount_cv, &zp->zone_mount_lock) == 0)
470 470 goto signaled;
471 471 }
472 472 /*
473 473 * A negative value of mounts_in_progress indicates that mounts
474 474 * have been blocked by (-mounts_in_progress) different callers
475 475 * (remotely possible if two threads enter zone_shutdown at the same
476 476 * time).
477 477 */
478 478 zp->zone_mounts_in_progress--;
479 479 retval = 1;
480 480 signaled:
481 481 mutex_exit(&zp->zone_mount_lock);
482 482 return (retval);
483 483 }
484 484
485 485 /*
486 486 * The VFS layer may progress with new mounts as far as we're concerned.
487 487 * Allow them to progress if we were the last obstacle.
488 488 */
489 489 static void
490 490 resume_mounts(zone_t *zp)
491 491 {
492 492 mutex_enter(&zp->zone_mount_lock);
493 493 if (++zp->zone_mounts_in_progress == 0)
494 494 cv_broadcast(&zp->zone_mount_cv);
495 495 mutex_exit(&zp->zone_mount_lock);
496 496 }
497 497
498 498 /*
499 499 * The VFS layer is busy with a mount; this zone should wait until all
500 500 * of its mounts are completed to progress.
501 501 */
502 502 void
503 503 mount_in_progress(zone_t *zp)
504 504 {
505 505 mutex_enter(&zp->zone_mount_lock);
506 506 while (zp->zone_mounts_in_progress < 0)
507 507 cv_wait(&zp->zone_mount_cv, &zp->zone_mount_lock);
508 508 zp->zone_mounts_in_progress++;
509 509 mutex_exit(&zp->zone_mount_lock);
510 510 }
511 511
512 512 /*
513 513 * VFS is done with one mount; wake up any waiting block_mounts()
514 514 * callers if this is the last mount.
515 515 */
516 516 void
517 517 mount_completed(zone_t *zp)
518 518 {
519 519 mutex_enter(&zp->zone_mount_lock);
520 520 if (--zp->zone_mounts_in_progress == 0)
521 521 cv_broadcast(&zp->zone_mount_cv);
522 522 mutex_exit(&zp->zone_mount_lock);
523 523 }
524 524
525 525 /*
526 526 * ZSD routines.
527 527 *
528 528 * Zone Specific Data (ZSD) is modeled after Thread Specific Data as
529 529 * defined by the pthread_key_create() and related interfaces.
530 530 *
531 531 * Kernel subsystems may register one or more data items and/or
532 532 * callbacks to be executed when a zone is created, shutdown, or
533 533 * destroyed.
534 534 *
535 535 * Unlike the thread counterpart, destructor callbacks will be executed
536 536 * even if the data pointer is NULL and/or there are no constructor
537 537 * callbacks, so it is the responsibility of such callbacks to check for
538 538 * NULL data values if necessary.
539 539 *
540 540 * The locking strategy and overall picture is as follows:
541 541 *
542 542 * When someone calls zone_key_create(), a template ZSD entry is added to the
543 543 * global list "zsd_registered_keys", protected by zsd_key_lock. While
544 544 * holding that lock all the existing zones are marked as
545 545 * ZSD_CREATE_NEEDED and a copy of the ZSD entry added to the per-zone
546 546 * zone_zsd list (protected by zone_lock). The global list is updated first
547 547 * (under zone_key_lock) to make sure that newly created zones use the
548 548 * most recent list of keys. Then under zonehash_lock we walk the zones
549 549 * and mark them. Similar locking is used in zone_key_delete().
550 550 *
551 551 * The actual create, shutdown, and destroy callbacks are done without
552 552 * holding any lock. And zsd_flags are used to ensure that the operations
553 553 * completed so that when zone_key_create (and zone_create) is done, as well as
554 554 * zone_key_delete (and zone_destroy) is done, all the necessary callbacks
555 555 * are completed.
556 556 *
557 557 * When new zones are created constructor callbacks for all registered ZSD
558 558 * entries will be called. That also uses the above two phases of marking
559 559 * what needs to be done, and then running the callbacks without holding
560 560 * any locks.
561 561 *
562 562 * The framework does not provide any locking around zone_getspecific() and
563 563 * zone_setspecific() apart from that needed for internal consistency, so
564 564 * callers interested in atomic "test-and-set" semantics will need to provide
565 565 * their own locking.
566 566 */
567 567
568 568 /*
569 569 * Helper function to find the zsd_entry associated with the key in the
570 570 * given list.
571 571 */
572 572 static struct zsd_entry *
573 573 zsd_find(list_t *l, zone_key_t key)
574 574 {
575 575 struct zsd_entry *zsd;
576 576
577 577 for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
578 578 if (zsd->zsd_key == key) {
579 579 return (zsd);
580 580 }
581 581 }
582 582 return (NULL);
583 583 }
584 584
585 585 /*
586 586 * Helper function to find the zsd_entry associated with the key in the
587 587 * given list. Move it to the front of the list.
588 588 */
589 589 static struct zsd_entry *
590 590 zsd_find_mru(list_t *l, zone_key_t key)
591 591 {
592 592 struct zsd_entry *zsd;
593 593
594 594 for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) {
595 595 if (zsd->zsd_key == key) {
596 596 /*
597 597 * Move to head of list to keep list in MRU order.
598 598 */
599 599 if (zsd != list_head(l)) {
600 600 list_remove(l, zsd);
601 601 list_insert_head(l, zsd);
602 602 }
603 603 return (zsd);
604 604 }
605 605 }
606 606 return (NULL);
607 607 }
608 608
609 609 void
610 610 zone_key_create(zone_key_t *keyp, void *(*create)(zoneid_t),
611 611 void (*shutdown)(zoneid_t, void *), void (*destroy)(zoneid_t, void *))
612 612 {
613 613 struct zsd_entry *zsdp;
614 614 struct zsd_entry *t;
615 615 struct zone *zone;
616 616 zone_key_t key;
617 617
618 618 zsdp = kmem_zalloc(sizeof (*zsdp), KM_SLEEP);
619 619 zsdp->zsd_data = NULL;
620 620 zsdp->zsd_create = create;
621 621 zsdp->zsd_shutdown = shutdown;
622 622 zsdp->zsd_destroy = destroy;
623 623
624 624 /*
625 625 * Insert in global list of callbacks. Makes future zone creations
626 626 * see it.
627 627 */
628 628 mutex_enter(&zsd_key_lock);
629 629 key = zsdp->zsd_key = ++zsd_keyval;
630 630 ASSERT(zsd_keyval != 0);
631 631 list_insert_tail(&zsd_registered_keys, zsdp);
632 632 mutex_exit(&zsd_key_lock);
633 633
634 634 /*
635 635 * Insert for all existing zones and mark them as needing
636 636 * a create callback.
637 637 */
638 638 mutex_enter(&zonehash_lock); /* stop the world */
639 639 for (zone = list_head(&zone_active); zone != NULL;
640 640 zone = list_next(&zone_active, zone)) {
641 641 zone_status_t status;
642 642
643 643 mutex_enter(&zone->zone_lock);
644 644
645 645 /* Skip zones that are on the way down or not yet up */
646 646 status = zone_status_get(zone);
647 647 if (status >= ZONE_IS_DOWN ||
648 648 status == ZONE_IS_UNINITIALIZED) {
649 649 mutex_exit(&zone->zone_lock);
650 650 continue;
651 651 }
652 652
653 653 t = zsd_find_mru(&zone->zone_zsd, key);
654 654 if (t != NULL) {
655 655 /*
656 656 * A zsd_configure already inserted it after
657 657 * we dropped zsd_key_lock above.
658 658 */
659 659 mutex_exit(&zone->zone_lock);
660 660 continue;
661 661 }
662 662 t = kmem_zalloc(sizeof (*t), KM_SLEEP);
663 663 t->zsd_key = key;
664 664 t->zsd_create = create;
665 665 t->zsd_shutdown = shutdown;
666 666 t->zsd_destroy = destroy;
667 667 if (create != NULL) {
668 668 t->zsd_flags = ZSD_CREATE_NEEDED;
669 669 DTRACE_PROBE2(zsd__create__needed,
670 670 zone_t *, zone, zone_key_t, key);
671 671 }
672 672 list_insert_tail(&zone->zone_zsd, t);
673 673 mutex_exit(&zone->zone_lock);
674 674 }
675 675 mutex_exit(&zonehash_lock);
676 676
677 677 if (create != NULL) {
678 678 /* Now call the create callback for this key */
679 679 zsd_apply_all_zones(zsd_apply_create, key);
680 680 }
681 681 /*
682 682 * It is safe for consumers to use the key now, make it
683 683 * globally visible. Specifically zone_getspecific() will
684 684 * always successfully return the zone specific data associated
685 685 * with the key.
686 686 */
687 687 *keyp = key;
688 688
689 689 }
690 690
691 691 /*
692 692 * Function called when a module is being unloaded, or otherwise wishes
693 693 * to unregister its ZSD key and callbacks.
694 694 *
695 695 * Remove from the global list and determine the functions that need to
696 696 * be called under a global lock. Then call the functions without
697 697 * holding any locks. Finally free up the zone_zsd entries. (The apply
698 698 * functions need to access the zone_zsd entries to find zsd_data etc.)
699 699 */
700 700 int
701 701 zone_key_delete(zone_key_t key)
702 702 {
703 703 struct zsd_entry *zsdp = NULL;
704 704 zone_t *zone;
705 705
706 706 mutex_enter(&zsd_key_lock);
707 707 zsdp = zsd_find_mru(&zsd_registered_keys, key);
708 708 if (zsdp == NULL) {
709 709 mutex_exit(&zsd_key_lock);
710 710 return (-1);
711 711 }
712 712 list_remove(&zsd_registered_keys, zsdp);
713 713 mutex_exit(&zsd_key_lock);
714 714
715 715 mutex_enter(&zonehash_lock);
716 716 for (zone = list_head(&zone_active); zone != NULL;
717 717 zone = list_next(&zone_active, zone)) {
718 718 struct zsd_entry *del;
719 719
720 720 mutex_enter(&zone->zone_lock);
721 721 del = zsd_find_mru(&zone->zone_zsd, key);
722 722 if (del == NULL) {
723 723 /*
724 724 * Somebody else got here first e.g the zone going
725 725 * away.
726 726 */
727 727 mutex_exit(&zone->zone_lock);
728 728 continue;
729 729 }
730 730 ASSERT(del->zsd_shutdown == zsdp->zsd_shutdown);
731 731 ASSERT(del->zsd_destroy == zsdp->zsd_destroy);
732 732 if (del->zsd_shutdown != NULL &&
733 733 (del->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
734 734 del->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
735 735 DTRACE_PROBE2(zsd__shutdown__needed,
736 736 zone_t *, zone, zone_key_t, key);
737 737 }
738 738 if (del->zsd_destroy != NULL &&
739 739 (del->zsd_flags & ZSD_DESTROY_ALL) == 0) {
740 740 del->zsd_flags |= ZSD_DESTROY_NEEDED;
741 741 DTRACE_PROBE2(zsd__destroy__needed,
742 742 zone_t *, zone, zone_key_t, key);
743 743 }
744 744 mutex_exit(&zone->zone_lock);
745 745 }
746 746 mutex_exit(&zonehash_lock);
747 747 kmem_free(zsdp, sizeof (*zsdp));
748 748
749 749 /* Now call the shutdown and destroy callback for this key */
750 750 zsd_apply_all_zones(zsd_apply_shutdown, key);
751 751 zsd_apply_all_zones(zsd_apply_destroy, key);
752 752
753 753 /* Now we can free up the zsdp structures in each zone */
754 754 mutex_enter(&zonehash_lock);
755 755 for (zone = list_head(&zone_active); zone != NULL;
756 756 zone = list_next(&zone_active, zone)) {
757 757 struct zsd_entry *del;
758 758
759 759 mutex_enter(&zone->zone_lock);
760 760 del = zsd_find(&zone->zone_zsd, key);
761 761 if (del != NULL) {
762 762 list_remove(&zone->zone_zsd, del);
763 763 ASSERT(!(del->zsd_flags & ZSD_ALL_INPROGRESS));
764 764 kmem_free(del, sizeof (*del));
765 765 }
766 766 mutex_exit(&zone->zone_lock);
767 767 }
768 768 mutex_exit(&zonehash_lock);
769 769
770 770 return (0);
771 771 }
772 772
773 773 /*
774 774 * ZSD counterpart of pthread_setspecific().
775 775 *
776 776 * Since all zsd callbacks, including those with no create function,
777 777 * have an entry in zone_zsd, if the key is registered it is part of
778 778 * the zone_zsd list.
779 779 * Return an error if the key wasn't registerd.
780 780 */
781 781 int
782 782 zone_setspecific(zone_key_t key, zone_t *zone, const void *data)
783 783 {
784 784 struct zsd_entry *t;
785 785
786 786 mutex_enter(&zone->zone_lock);
787 787 t = zsd_find_mru(&zone->zone_zsd, key);
788 788 if (t != NULL) {
789 789 /*
790 790 * Replace old value with new
791 791 */
792 792 t->zsd_data = (void *)data;
793 793 mutex_exit(&zone->zone_lock);
794 794 return (0);
795 795 }
796 796 mutex_exit(&zone->zone_lock);
797 797 return (-1);
798 798 }
799 799
800 800 /*
801 801 * ZSD counterpart of pthread_getspecific().
802 802 */
803 803 void *
804 804 zone_getspecific(zone_key_t key, zone_t *zone)
805 805 {
806 806 struct zsd_entry *t;
807 807 void *data;
808 808
809 809 mutex_enter(&zone->zone_lock);
810 810 t = zsd_find_mru(&zone->zone_zsd, key);
811 811 data = (t == NULL ? NULL : t->zsd_data);
812 812 mutex_exit(&zone->zone_lock);
813 813 return (data);
814 814 }
815 815
816 816 /*
817 817 * Function used to initialize a zone's list of ZSD callbacks and data
818 818 * when the zone is being created. The callbacks are initialized from
819 819 * the template list (zsd_registered_keys). The constructor callback is
820 820 * executed later (once the zone exists and with locks dropped).
821 821 */
822 822 static void
823 823 zone_zsd_configure(zone_t *zone)
824 824 {
825 825 struct zsd_entry *zsdp;
826 826 struct zsd_entry *t;
827 827
828 828 ASSERT(MUTEX_HELD(&zonehash_lock));
829 829 ASSERT(list_head(&zone->zone_zsd) == NULL);
830 830 mutex_enter(&zone->zone_lock);
831 831 mutex_enter(&zsd_key_lock);
832 832 for (zsdp = list_head(&zsd_registered_keys); zsdp != NULL;
833 833 zsdp = list_next(&zsd_registered_keys, zsdp)) {
834 834 /*
835 835 * Since this zone is ZONE_IS_UNCONFIGURED, zone_key_create
836 836 * should not have added anything to it.
837 837 */
838 838 ASSERT(zsd_find(&zone->zone_zsd, zsdp->zsd_key) == NULL);
839 839
840 840 t = kmem_zalloc(sizeof (*t), KM_SLEEP);
841 841 t->zsd_key = zsdp->zsd_key;
842 842 t->zsd_create = zsdp->zsd_create;
843 843 t->zsd_shutdown = zsdp->zsd_shutdown;
844 844 t->zsd_destroy = zsdp->zsd_destroy;
845 845 if (zsdp->zsd_create != NULL) {
846 846 t->zsd_flags = ZSD_CREATE_NEEDED;
847 847 DTRACE_PROBE2(zsd__create__needed,
848 848 zone_t *, zone, zone_key_t, zsdp->zsd_key);
849 849 }
850 850 list_insert_tail(&zone->zone_zsd, t);
851 851 }
852 852 mutex_exit(&zsd_key_lock);
853 853 mutex_exit(&zone->zone_lock);
854 854 }
855 855
856 856 enum zsd_callback_type { ZSD_CREATE, ZSD_SHUTDOWN, ZSD_DESTROY };
857 857
858 858 /*
859 859 * Helper function to execute shutdown or destructor callbacks.
860 860 */
861 861 static void
862 862 zone_zsd_callbacks(zone_t *zone, enum zsd_callback_type ct)
863 863 {
864 864 struct zsd_entry *t;
865 865
866 866 ASSERT(ct == ZSD_SHUTDOWN || ct == ZSD_DESTROY);
867 867 ASSERT(ct != ZSD_SHUTDOWN || zone_status_get(zone) >= ZONE_IS_EMPTY);
868 868 ASSERT(ct != ZSD_DESTROY || zone_status_get(zone) >= ZONE_IS_DOWN);
869 869
870 870 /*
871 871 * Run the callback solely based on what is registered for the zone
872 872 * in zone_zsd. The global list can change independently of this
873 873 * as keys are registered and unregistered and we don't register new
874 874 * callbacks for a zone that is in the process of going away.
875 875 */
876 876 mutex_enter(&zone->zone_lock);
877 877 for (t = list_head(&zone->zone_zsd); t != NULL;
878 878 t = list_next(&zone->zone_zsd, t)) {
879 879 zone_key_t key = t->zsd_key;
880 880
881 881 /* Skip if no callbacks registered */
882 882
883 883 if (ct == ZSD_SHUTDOWN) {
884 884 if (t->zsd_shutdown != NULL &&
885 885 (t->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) {
886 886 t->zsd_flags |= ZSD_SHUTDOWN_NEEDED;
887 887 DTRACE_PROBE2(zsd__shutdown__needed,
888 888 zone_t *, zone, zone_key_t, key);
889 889 }
890 890 } else {
891 891 if (t->zsd_destroy != NULL &&
892 892 (t->zsd_flags & ZSD_DESTROY_ALL) == 0) {
893 893 t->zsd_flags |= ZSD_DESTROY_NEEDED;
894 894 DTRACE_PROBE2(zsd__destroy__needed,
895 895 zone_t *, zone, zone_key_t, key);
896 896 }
897 897 }
898 898 }
899 899 mutex_exit(&zone->zone_lock);
900 900
901 901 /* Now call the shutdown and destroy callback for this key */
902 902 zsd_apply_all_keys(zsd_apply_shutdown, zone);
903 903 zsd_apply_all_keys(zsd_apply_destroy, zone);
904 904
905 905 }
906 906
907 907 /*
908 908 * Called when the zone is going away; free ZSD-related memory, and
909 909 * destroy the zone_zsd list.
910 910 */
911 911 static void
912 912 zone_free_zsd(zone_t *zone)
913 913 {
914 914 struct zsd_entry *t, *next;
915 915
916 916 /*
917 917 * Free all the zsd_entry's we had on this zone.
918 918 */
919 919 mutex_enter(&zone->zone_lock);
920 920 for (t = list_head(&zone->zone_zsd); t != NULL; t = next) {
921 921 next = list_next(&zone->zone_zsd, t);
922 922 list_remove(&zone->zone_zsd, t);
923 923 ASSERT(!(t->zsd_flags & ZSD_ALL_INPROGRESS));
924 924 kmem_free(t, sizeof (*t));
925 925 }
926 926 list_destroy(&zone->zone_zsd);
927 927 mutex_exit(&zone->zone_lock);
928 928
929 929 }
930 930
931 931 /*
932 932 * Apply a function to all zones for particular key value.
933 933 *
934 934 * The applyfn has to drop zonehash_lock if it does some work, and
935 935 * then reacquire it before it returns.
936 936 * When the lock is dropped we don't follow list_next even
937 937 * if it is possible to do so without any hazards. This is
938 938 * because we want the design to allow for the list of zones
939 939 * to change in any arbitrary way during the time the
940 940 * lock was dropped.
941 941 *
942 942 * It is safe to restart the loop at list_head since the applyfn
943 943 * changes the zsd_flags as it does work, so a subsequent
944 944 * pass through will have no effect in applyfn, hence the loop will terminate
945 945 * in at worst O(N^2).
946 946 */
947 947 static void
948 948 zsd_apply_all_zones(zsd_applyfn_t *applyfn, zone_key_t key)
949 949 {
950 950 zone_t *zone;
951 951
952 952 mutex_enter(&zonehash_lock);
953 953 zone = list_head(&zone_active);
954 954 while (zone != NULL) {
955 955 if ((applyfn)(&zonehash_lock, B_FALSE, zone, key)) {
956 956 /* Lock dropped - restart at head */
957 957 zone = list_head(&zone_active);
958 958 } else {
959 959 zone = list_next(&zone_active, zone);
960 960 }
961 961 }
962 962 mutex_exit(&zonehash_lock);
963 963 }
964 964
965 965 /*
966 966 * Apply a function to all keys for a particular zone.
967 967 *
968 968 * The applyfn has to drop zonehash_lock if it does some work, and
969 969 * then reacquire it before it returns.
970 970 * When the lock is dropped we don't follow list_next even
971 971 * if it is possible to do so without any hazards. This is
972 972 * because we want the design to allow for the list of zsd callbacks
973 973 * to change in any arbitrary way during the time the
974 974 * lock was dropped.
975 975 *
976 976 * It is safe to restart the loop at list_head since the applyfn
977 977 * changes the zsd_flags as it does work, so a subsequent
978 978 * pass through will have no effect in applyfn, hence the loop will terminate
979 979 * in at worst O(N^2).
980 980 */
981 981 static void
982 982 zsd_apply_all_keys(zsd_applyfn_t *applyfn, zone_t *zone)
983 983 {
984 984 struct zsd_entry *t;
985 985
986 986 mutex_enter(&zone->zone_lock);
987 987 t = list_head(&zone->zone_zsd);
988 988 while (t != NULL) {
989 989 if ((applyfn)(NULL, B_TRUE, zone, t->zsd_key)) {
990 990 /* Lock dropped - restart at head */
991 991 t = list_head(&zone->zone_zsd);
992 992 } else {
993 993 t = list_next(&zone->zone_zsd, t);
994 994 }
995 995 }
996 996 mutex_exit(&zone->zone_lock);
997 997 }
998 998
999 999 /*
1000 1000 * Call the create function for the zone and key if CREATE_NEEDED
1001 1001 * is set.
1002 1002 * If some other thread gets here first and sets CREATE_INPROGRESS, then
1003 1003 * we wait for that thread to complete so that we can ensure that
1004 1004 * all the callbacks are done when we've looped over all zones/keys.
1005 1005 *
1006 1006 * When we call the create function, we drop the global held by the
1007 1007 * caller, and return true to tell the caller it needs to re-evalute the
1008 1008 * state.
1009 1009 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1010 1010 * remains held on exit.
1011 1011 */
1012 1012 static boolean_t
1013 1013 zsd_apply_create(kmutex_t *lockp, boolean_t zone_lock_held,
1014 1014 zone_t *zone, zone_key_t key)
1015 1015 {
1016 1016 void *result;
1017 1017 struct zsd_entry *t;
1018 1018 boolean_t dropped;
1019 1019
1020 1020 if (lockp != NULL) {
1021 1021 ASSERT(MUTEX_HELD(lockp));
1022 1022 }
1023 1023 if (zone_lock_held) {
1024 1024 ASSERT(MUTEX_HELD(&zone->zone_lock));
1025 1025 } else {
1026 1026 mutex_enter(&zone->zone_lock);
1027 1027 }
1028 1028
1029 1029 t = zsd_find(&zone->zone_zsd, key);
1030 1030 if (t == NULL) {
1031 1031 /*
1032 1032 * Somebody else got here first e.g the zone going
1033 1033 * away.
1034 1034 */
1035 1035 if (!zone_lock_held)
1036 1036 mutex_exit(&zone->zone_lock);
1037 1037 return (B_FALSE);
1038 1038 }
1039 1039 dropped = B_FALSE;
1040 1040 if (zsd_wait_for_inprogress(zone, t, lockp))
1041 1041 dropped = B_TRUE;
1042 1042
1043 1043 if (t->zsd_flags & ZSD_CREATE_NEEDED) {
1044 1044 t->zsd_flags &= ~ZSD_CREATE_NEEDED;
1045 1045 t->zsd_flags |= ZSD_CREATE_INPROGRESS;
1046 1046 DTRACE_PROBE2(zsd__create__inprogress,
1047 1047 zone_t *, zone, zone_key_t, key);
1048 1048 mutex_exit(&zone->zone_lock);
1049 1049 if (lockp != NULL)
1050 1050 mutex_exit(lockp);
1051 1051
1052 1052 dropped = B_TRUE;
1053 1053 ASSERT(t->zsd_create != NULL);
1054 1054 DTRACE_PROBE2(zsd__create__start,
1055 1055 zone_t *, zone, zone_key_t, key);
1056 1056
1057 1057 result = (*t->zsd_create)(zone->zone_id);
1058 1058
1059 1059 DTRACE_PROBE2(zsd__create__end,
1060 1060 zone_t *, zone, voidn *, result);
1061 1061
1062 1062 ASSERT(result != NULL);
1063 1063 if (lockp != NULL)
1064 1064 mutex_enter(lockp);
1065 1065 mutex_enter(&zone->zone_lock);
1066 1066 t->zsd_data = result;
1067 1067 t->zsd_flags &= ~ZSD_CREATE_INPROGRESS;
1068 1068 t->zsd_flags |= ZSD_CREATE_COMPLETED;
1069 1069 cv_broadcast(&t->zsd_cv);
1070 1070 DTRACE_PROBE2(zsd__create__completed,
1071 1071 zone_t *, zone, zone_key_t, key);
1072 1072 }
1073 1073 if (!zone_lock_held)
1074 1074 mutex_exit(&zone->zone_lock);
1075 1075 return (dropped);
1076 1076 }
1077 1077
1078 1078 /*
1079 1079 * Call the shutdown function for the zone and key if SHUTDOWN_NEEDED
1080 1080 * is set.
1081 1081 * If some other thread gets here first and sets *_INPROGRESS, then
1082 1082 * we wait for that thread to complete so that we can ensure that
1083 1083 * all the callbacks are done when we've looped over all zones/keys.
1084 1084 *
1085 1085 * When we call the shutdown function, we drop the global held by the
1086 1086 * caller, and return true to tell the caller it needs to re-evalute the
1087 1087 * state.
1088 1088 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1089 1089 * remains held on exit.
1090 1090 */
1091 1091 static boolean_t
1092 1092 zsd_apply_shutdown(kmutex_t *lockp, boolean_t zone_lock_held,
1093 1093 zone_t *zone, zone_key_t key)
1094 1094 {
1095 1095 struct zsd_entry *t;
1096 1096 void *data;
1097 1097 boolean_t dropped;
1098 1098
1099 1099 if (lockp != NULL) {
1100 1100 ASSERT(MUTEX_HELD(lockp));
1101 1101 }
1102 1102 if (zone_lock_held) {
1103 1103 ASSERT(MUTEX_HELD(&zone->zone_lock));
1104 1104 } else {
1105 1105 mutex_enter(&zone->zone_lock);
1106 1106 }
1107 1107
1108 1108 t = zsd_find(&zone->zone_zsd, key);
1109 1109 if (t == NULL) {
1110 1110 /*
1111 1111 * Somebody else got here first e.g the zone going
1112 1112 * away.
1113 1113 */
1114 1114 if (!zone_lock_held)
1115 1115 mutex_exit(&zone->zone_lock);
1116 1116 return (B_FALSE);
1117 1117 }
1118 1118 dropped = B_FALSE;
1119 1119 if (zsd_wait_for_creator(zone, t, lockp))
1120 1120 dropped = B_TRUE;
1121 1121
1122 1122 if (zsd_wait_for_inprogress(zone, t, lockp))
1123 1123 dropped = B_TRUE;
1124 1124
1125 1125 if (t->zsd_flags & ZSD_SHUTDOWN_NEEDED) {
1126 1126 t->zsd_flags &= ~ZSD_SHUTDOWN_NEEDED;
1127 1127 t->zsd_flags |= ZSD_SHUTDOWN_INPROGRESS;
1128 1128 DTRACE_PROBE2(zsd__shutdown__inprogress,
1129 1129 zone_t *, zone, zone_key_t, key);
1130 1130 mutex_exit(&zone->zone_lock);
1131 1131 if (lockp != NULL)
1132 1132 mutex_exit(lockp);
1133 1133 dropped = B_TRUE;
1134 1134
1135 1135 ASSERT(t->zsd_shutdown != NULL);
1136 1136 data = t->zsd_data;
1137 1137
1138 1138 DTRACE_PROBE2(zsd__shutdown__start,
1139 1139 zone_t *, zone, zone_key_t, key);
1140 1140
1141 1141 (t->zsd_shutdown)(zone->zone_id, data);
1142 1142 DTRACE_PROBE2(zsd__shutdown__end,
1143 1143 zone_t *, zone, zone_key_t, key);
1144 1144
1145 1145 if (lockp != NULL)
1146 1146 mutex_enter(lockp);
1147 1147 mutex_enter(&zone->zone_lock);
1148 1148 t->zsd_flags &= ~ZSD_SHUTDOWN_INPROGRESS;
1149 1149 t->zsd_flags |= ZSD_SHUTDOWN_COMPLETED;
1150 1150 cv_broadcast(&t->zsd_cv);
1151 1151 DTRACE_PROBE2(zsd__shutdown__completed,
1152 1152 zone_t *, zone, zone_key_t, key);
1153 1153 }
1154 1154 if (!zone_lock_held)
1155 1155 mutex_exit(&zone->zone_lock);
1156 1156 return (dropped);
1157 1157 }
1158 1158
1159 1159 /*
1160 1160 * Call the destroy function for the zone and key if DESTROY_NEEDED
1161 1161 * is set.
1162 1162 * If some other thread gets here first and sets *_INPROGRESS, then
1163 1163 * we wait for that thread to complete so that we can ensure that
1164 1164 * all the callbacks are done when we've looped over all zones/keys.
1165 1165 *
1166 1166 * When we call the destroy function, we drop the global held by the
1167 1167 * caller, and return true to tell the caller it needs to re-evalute the
1168 1168 * state.
1169 1169 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock
1170 1170 * remains held on exit.
1171 1171 */
1172 1172 static boolean_t
1173 1173 zsd_apply_destroy(kmutex_t *lockp, boolean_t zone_lock_held,
1174 1174 zone_t *zone, zone_key_t key)
1175 1175 {
1176 1176 struct zsd_entry *t;
1177 1177 void *data;
1178 1178 boolean_t dropped;
1179 1179
1180 1180 if (lockp != NULL) {
1181 1181 ASSERT(MUTEX_HELD(lockp));
1182 1182 }
1183 1183 if (zone_lock_held) {
1184 1184 ASSERT(MUTEX_HELD(&zone->zone_lock));
1185 1185 } else {
1186 1186 mutex_enter(&zone->zone_lock);
1187 1187 }
1188 1188
1189 1189 t = zsd_find(&zone->zone_zsd, key);
1190 1190 if (t == NULL) {
1191 1191 /*
1192 1192 * Somebody else got here first e.g the zone going
1193 1193 * away.
1194 1194 */
1195 1195 if (!zone_lock_held)
1196 1196 mutex_exit(&zone->zone_lock);
1197 1197 return (B_FALSE);
1198 1198 }
1199 1199 dropped = B_FALSE;
1200 1200 if (zsd_wait_for_creator(zone, t, lockp))
1201 1201 dropped = B_TRUE;
1202 1202
1203 1203 if (zsd_wait_for_inprogress(zone, t, lockp))
1204 1204 dropped = B_TRUE;
1205 1205
1206 1206 if (t->zsd_flags & ZSD_DESTROY_NEEDED) {
1207 1207 t->zsd_flags &= ~ZSD_DESTROY_NEEDED;
1208 1208 t->zsd_flags |= ZSD_DESTROY_INPROGRESS;
1209 1209 DTRACE_PROBE2(zsd__destroy__inprogress,
1210 1210 zone_t *, zone, zone_key_t, key);
1211 1211 mutex_exit(&zone->zone_lock);
1212 1212 if (lockp != NULL)
1213 1213 mutex_exit(lockp);
1214 1214 dropped = B_TRUE;
1215 1215
1216 1216 ASSERT(t->zsd_destroy != NULL);
1217 1217 data = t->zsd_data;
1218 1218 DTRACE_PROBE2(zsd__destroy__start,
1219 1219 zone_t *, zone, zone_key_t, key);
1220 1220
1221 1221 (t->zsd_destroy)(zone->zone_id, data);
1222 1222 DTRACE_PROBE2(zsd__destroy__end,
1223 1223 zone_t *, zone, zone_key_t, key);
1224 1224
1225 1225 if (lockp != NULL)
1226 1226 mutex_enter(lockp);
1227 1227 mutex_enter(&zone->zone_lock);
1228 1228 t->zsd_data = NULL;
1229 1229 t->zsd_flags &= ~ZSD_DESTROY_INPROGRESS;
1230 1230 t->zsd_flags |= ZSD_DESTROY_COMPLETED;
1231 1231 cv_broadcast(&t->zsd_cv);
1232 1232 DTRACE_PROBE2(zsd__destroy__completed,
1233 1233 zone_t *, zone, zone_key_t, key);
1234 1234 }
1235 1235 if (!zone_lock_held)
1236 1236 mutex_exit(&zone->zone_lock);
1237 1237 return (dropped);
1238 1238 }
1239 1239
1240 1240 /*
1241 1241 * Wait for any CREATE_NEEDED flag to be cleared.
1242 1242 * Returns true if lockp was temporarily dropped while waiting.
1243 1243 */
1244 1244 static boolean_t
1245 1245 zsd_wait_for_creator(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
1246 1246 {
1247 1247 boolean_t dropped = B_FALSE;
1248 1248
1249 1249 while (t->zsd_flags & ZSD_CREATE_NEEDED) {
1250 1250 DTRACE_PROBE2(zsd__wait__for__creator,
1251 1251 zone_t *, zone, struct zsd_entry *, t);
1252 1252 if (lockp != NULL) {
1253 1253 dropped = B_TRUE;
1254 1254 mutex_exit(lockp);
1255 1255 }
1256 1256 cv_wait(&t->zsd_cv, &zone->zone_lock);
1257 1257 if (lockp != NULL) {
1258 1258 /* First drop zone_lock to preserve order */
1259 1259 mutex_exit(&zone->zone_lock);
1260 1260 mutex_enter(lockp);
1261 1261 mutex_enter(&zone->zone_lock);
1262 1262 }
1263 1263 }
1264 1264 return (dropped);
1265 1265 }
1266 1266
1267 1267 /*
1268 1268 * Wait for any INPROGRESS flag to be cleared.
1269 1269 * Returns true if lockp was temporarily dropped while waiting.
1270 1270 */
1271 1271 static boolean_t
1272 1272 zsd_wait_for_inprogress(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp)
1273 1273 {
1274 1274 boolean_t dropped = B_FALSE;
1275 1275
1276 1276 while (t->zsd_flags & ZSD_ALL_INPROGRESS) {
1277 1277 DTRACE_PROBE2(zsd__wait__for__inprogress,
1278 1278 zone_t *, zone, struct zsd_entry *, t);
1279 1279 if (lockp != NULL) {
1280 1280 dropped = B_TRUE;
1281 1281 mutex_exit(lockp);
1282 1282 }
1283 1283 cv_wait(&t->zsd_cv, &zone->zone_lock);
1284 1284 if (lockp != NULL) {
1285 1285 /* First drop zone_lock to preserve order */
1286 1286 mutex_exit(&zone->zone_lock);
1287 1287 mutex_enter(lockp);
1288 1288 mutex_enter(&zone->zone_lock);
1289 1289 }
1290 1290 }
1291 1291 return (dropped);
1292 1292 }
1293 1293
1294 1294 /*
1295 1295 * Frees memory associated with the zone dataset list.
1296 1296 */
1297 1297 static void
1298 1298 zone_free_datasets(zone_t *zone)
1299 1299 {
1300 1300 zone_dataset_t *t, *next;
1301 1301
1302 1302 for (t = list_head(&zone->zone_datasets); t != NULL; t = next) {
1303 1303 next = list_next(&zone->zone_datasets, t);
1304 1304 list_remove(&zone->zone_datasets, t);
1305 1305 kmem_free(t->zd_dataset, strlen(t->zd_dataset) + 1);
1306 1306 kmem_free(t, sizeof (*t));
1307 1307 }
1308 1308 list_destroy(&zone->zone_datasets);
1309 1309 }
1310 1310
1311 1311 /*
1312 1312 * zone.cpu-shares resource control support.
1313 1313 */
1314 1314 /*ARGSUSED*/
1315 1315 static rctl_qty_t
1316 1316 zone_cpu_shares_usage(rctl_t *rctl, struct proc *p)
1317 1317 {
1318 1318 ASSERT(MUTEX_HELD(&p->p_lock));
1319 1319 return (p->p_zone->zone_shares);
1320 1320 }
1321 1321
1322 1322 /*ARGSUSED*/
1323 1323 static int
1324 1324 zone_cpu_shares_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1325 1325 rctl_qty_t nv)
1326 1326 {
1327 1327 ASSERT(MUTEX_HELD(&p->p_lock));
1328 1328 ASSERT(e->rcep_t == RCENTITY_ZONE);
1329 1329 if (e->rcep_p.zone == NULL)
1330 1330 return (0);
1331 1331
1332 1332 e->rcep_p.zone->zone_shares = nv;
1333 1333 return (0);
1334 1334 }
1335 1335
1336 1336 static rctl_ops_t zone_cpu_shares_ops = {
1337 1337 rcop_no_action,
1338 1338 zone_cpu_shares_usage,
1339 1339 zone_cpu_shares_set,
1340 1340 rcop_no_test
1341 1341 };
1342 1342
1343 1343 /*
1344 1344 * zone.cpu-cap resource control support.
1345 1345 */
1346 1346 /*ARGSUSED*/
1347 1347 static rctl_qty_t
1348 1348 zone_cpu_cap_get(rctl_t *rctl, struct proc *p)
1349 1349 {
1350 1350 ASSERT(MUTEX_HELD(&p->p_lock));
1351 1351 return (cpucaps_zone_get(p->p_zone));
1352 1352 }
1353 1353
1354 1354 /*ARGSUSED*/
1355 1355 static int
1356 1356 zone_cpu_cap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1357 1357 rctl_qty_t nv)
1358 1358 {
1359 1359 zone_t *zone = e->rcep_p.zone;
1360 1360
1361 1361 ASSERT(MUTEX_HELD(&p->p_lock));
1362 1362 ASSERT(e->rcep_t == RCENTITY_ZONE);
1363 1363
1364 1364 if (zone == NULL)
1365 1365 return (0);
1366 1366
1367 1367 /*
1368 1368 * set cap to the new value.
1369 1369 */
1370 1370 return (cpucaps_zone_set(zone, nv));
1371 1371 }
1372 1372
1373 1373 static rctl_ops_t zone_cpu_cap_ops = {
1374 1374 rcop_no_action,
1375 1375 zone_cpu_cap_get,
1376 1376 zone_cpu_cap_set,
1377 1377 rcop_no_test
1378 1378 };
1379 1379
1380 1380 /*ARGSUSED*/
1381 1381 static rctl_qty_t
1382 1382 zone_lwps_usage(rctl_t *r, proc_t *p)
1383 1383 {
1384 1384 rctl_qty_t nlwps;
1385 1385 zone_t *zone = p->p_zone;
1386 1386
1387 1387 ASSERT(MUTEX_HELD(&p->p_lock));
1388 1388
1389 1389 mutex_enter(&zone->zone_nlwps_lock);
1390 1390 nlwps = zone->zone_nlwps;
1391 1391 mutex_exit(&zone->zone_nlwps_lock);
1392 1392
1393 1393 return (nlwps);
1394 1394 }
1395 1395
1396 1396 /*ARGSUSED*/
1397 1397 static int
1398 1398 zone_lwps_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
1399 1399 rctl_qty_t incr, uint_t flags)
1400 1400 {
1401 1401 rctl_qty_t nlwps;
1402 1402
1403 1403 ASSERT(MUTEX_HELD(&p->p_lock));
1404 1404 ASSERT(e->rcep_t == RCENTITY_ZONE);
1405 1405 if (e->rcep_p.zone == NULL)
1406 1406 return (0);
1407 1407 ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
1408 1408 nlwps = e->rcep_p.zone->zone_nlwps;
1409 1409
1410 1410 if (nlwps + incr > rcntl->rcv_value)
1411 1411 return (1);
1412 1412
1413 1413 return (0);
1414 1414 }
1415 1415
1416 1416 /*ARGSUSED*/
1417 1417 static int
1418 1418 zone_lwps_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv)
1419 1419 {
1420 1420 ASSERT(MUTEX_HELD(&p->p_lock));
1421 1421 ASSERT(e->rcep_t == RCENTITY_ZONE);
1422 1422 if (e->rcep_p.zone == NULL)
1423 1423 return (0);
1424 1424 e->rcep_p.zone->zone_nlwps_ctl = nv;
1425 1425 return (0);
1426 1426 }
1427 1427
1428 1428 static rctl_ops_t zone_lwps_ops = {
1429 1429 rcop_no_action,
1430 1430 zone_lwps_usage,
1431 1431 zone_lwps_set,
1432 1432 zone_lwps_test,
1433 1433 };
1434 1434
1435 1435 /*ARGSUSED*/
1436 1436 static rctl_qty_t
1437 1437 zone_procs_usage(rctl_t *r, proc_t *p)
1438 1438 {
1439 1439 rctl_qty_t nprocs;
1440 1440 zone_t *zone = p->p_zone;
1441 1441
1442 1442 ASSERT(MUTEX_HELD(&p->p_lock));
1443 1443
1444 1444 mutex_enter(&zone->zone_nlwps_lock);
1445 1445 nprocs = zone->zone_nprocs;
1446 1446 mutex_exit(&zone->zone_nlwps_lock);
1447 1447
1448 1448 return (nprocs);
1449 1449 }
1450 1450
1451 1451 /*ARGSUSED*/
1452 1452 static int
1453 1453 zone_procs_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl,
1454 1454 rctl_qty_t incr, uint_t flags)
1455 1455 {
1456 1456 rctl_qty_t nprocs;
1457 1457
1458 1458 ASSERT(MUTEX_HELD(&p->p_lock));
1459 1459 ASSERT(e->rcep_t == RCENTITY_ZONE);
1460 1460 if (e->rcep_p.zone == NULL)
1461 1461 return (0);
1462 1462 ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock)));
1463 1463 nprocs = e->rcep_p.zone->zone_nprocs;
1464 1464
1465 1465 if (nprocs + incr > rcntl->rcv_value)
1466 1466 return (1);
1467 1467
1468 1468 return (0);
1469 1469 }
1470 1470
1471 1471 /*ARGSUSED*/
1472 1472 static int
1473 1473 zone_procs_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv)
1474 1474 {
1475 1475 ASSERT(MUTEX_HELD(&p->p_lock));
1476 1476 ASSERT(e->rcep_t == RCENTITY_ZONE);
1477 1477 if (e->rcep_p.zone == NULL)
1478 1478 return (0);
1479 1479 e->rcep_p.zone->zone_nprocs_ctl = nv;
1480 1480 return (0);
1481 1481 }
1482 1482
1483 1483 static rctl_ops_t zone_procs_ops = {
1484 1484 rcop_no_action,
1485 1485 zone_procs_usage,
1486 1486 zone_procs_set,
1487 1487 zone_procs_test,
1488 1488 };
1489 1489
1490 1490 /*ARGSUSED*/
1491 1491 static rctl_qty_t
1492 1492 zone_shmmax_usage(rctl_t *rctl, struct proc *p)
1493 1493 {
1494 1494 ASSERT(MUTEX_HELD(&p->p_lock));
1495 1495 return (p->p_zone->zone_shmmax);
1496 1496 }
1497 1497
1498 1498 /*ARGSUSED*/
1499 1499 static int
1500 1500 zone_shmmax_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1501 1501 rctl_qty_t incr, uint_t flags)
1502 1502 {
1503 1503 rctl_qty_t v;
1504 1504 ASSERT(MUTEX_HELD(&p->p_lock));
1505 1505 ASSERT(e->rcep_t == RCENTITY_ZONE);
1506 1506 v = e->rcep_p.zone->zone_shmmax + incr;
1507 1507 if (v > rval->rcv_value)
1508 1508 return (1);
1509 1509 return (0);
1510 1510 }
1511 1511
1512 1512 static rctl_ops_t zone_shmmax_ops = {
1513 1513 rcop_no_action,
1514 1514 zone_shmmax_usage,
1515 1515 rcop_no_set,
1516 1516 zone_shmmax_test
1517 1517 };
1518 1518
1519 1519 /*ARGSUSED*/
1520 1520 static rctl_qty_t
1521 1521 zone_shmmni_usage(rctl_t *rctl, struct proc *p)
1522 1522 {
1523 1523 ASSERT(MUTEX_HELD(&p->p_lock));
1524 1524 return (p->p_zone->zone_ipc.ipcq_shmmni);
1525 1525 }
1526 1526
1527 1527 /*ARGSUSED*/
1528 1528 static int
1529 1529 zone_shmmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1530 1530 rctl_qty_t incr, uint_t flags)
1531 1531 {
1532 1532 rctl_qty_t v;
1533 1533 ASSERT(MUTEX_HELD(&p->p_lock));
1534 1534 ASSERT(e->rcep_t == RCENTITY_ZONE);
1535 1535 v = e->rcep_p.zone->zone_ipc.ipcq_shmmni + incr;
1536 1536 if (v > rval->rcv_value)
1537 1537 return (1);
1538 1538 return (0);
1539 1539 }
1540 1540
1541 1541 static rctl_ops_t zone_shmmni_ops = {
1542 1542 rcop_no_action,
1543 1543 zone_shmmni_usage,
1544 1544 rcop_no_set,
1545 1545 zone_shmmni_test
1546 1546 };
1547 1547
1548 1548 /*ARGSUSED*/
1549 1549 static rctl_qty_t
1550 1550 zone_semmni_usage(rctl_t *rctl, struct proc *p)
1551 1551 {
1552 1552 ASSERT(MUTEX_HELD(&p->p_lock));
1553 1553 return (p->p_zone->zone_ipc.ipcq_semmni);
1554 1554 }
1555 1555
1556 1556 /*ARGSUSED*/
1557 1557 static int
1558 1558 zone_semmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1559 1559 rctl_qty_t incr, uint_t flags)
1560 1560 {
1561 1561 rctl_qty_t v;
1562 1562 ASSERT(MUTEX_HELD(&p->p_lock));
1563 1563 ASSERT(e->rcep_t == RCENTITY_ZONE);
1564 1564 v = e->rcep_p.zone->zone_ipc.ipcq_semmni + incr;
1565 1565 if (v > rval->rcv_value)
1566 1566 return (1);
1567 1567 return (0);
1568 1568 }
1569 1569
1570 1570 static rctl_ops_t zone_semmni_ops = {
1571 1571 rcop_no_action,
1572 1572 zone_semmni_usage,
1573 1573 rcop_no_set,
1574 1574 zone_semmni_test
1575 1575 };
1576 1576
1577 1577 /*ARGSUSED*/
1578 1578 static rctl_qty_t
1579 1579 zone_msgmni_usage(rctl_t *rctl, struct proc *p)
1580 1580 {
1581 1581 ASSERT(MUTEX_HELD(&p->p_lock));
1582 1582 return (p->p_zone->zone_ipc.ipcq_msgmni);
1583 1583 }
1584 1584
1585 1585 /*ARGSUSED*/
1586 1586 static int
1587 1587 zone_msgmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval,
1588 1588 rctl_qty_t incr, uint_t flags)
1589 1589 {
1590 1590 rctl_qty_t v;
1591 1591 ASSERT(MUTEX_HELD(&p->p_lock));
1592 1592 ASSERT(e->rcep_t == RCENTITY_ZONE);
1593 1593 v = e->rcep_p.zone->zone_ipc.ipcq_msgmni + incr;
1594 1594 if (v > rval->rcv_value)
1595 1595 return (1);
1596 1596 return (0);
1597 1597 }
1598 1598
1599 1599 static rctl_ops_t zone_msgmni_ops = {
1600 1600 rcop_no_action,
1601 1601 zone_msgmni_usage,
1602 1602 rcop_no_set,
1603 1603 zone_msgmni_test
1604 1604 };
1605 1605
1606 1606 /*ARGSUSED*/
1607 1607 static rctl_qty_t
1608 1608 zone_locked_mem_usage(rctl_t *rctl, struct proc *p)
1609 1609 {
1610 1610 rctl_qty_t q;
1611 1611 ASSERT(MUTEX_HELD(&p->p_lock));
1612 1612 mutex_enter(&p->p_zone->zone_mem_lock);
1613 1613 q = p->p_zone->zone_locked_mem;
1614 1614 mutex_exit(&p->p_zone->zone_mem_lock);
1615 1615 return (q);
1616 1616 }
1617 1617
1618 1618 /*ARGSUSED*/
1619 1619 static int
1620 1620 zone_locked_mem_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1621 1621 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1622 1622 {
1623 1623 rctl_qty_t q;
1624 1624 zone_t *z;
1625 1625
1626 1626 z = e->rcep_p.zone;
1627 1627 ASSERT(MUTEX_HELD(&p->p_lock));
1628 1628 ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1629 1629 q = z->zone_locked_mem;
1630 1630 if (q + incr > rcntl->rcv_value)
1631 1631 return (1);
1632 1632 return (0);
1633 1633 }
1634 1634
1635 1635 /*ARGSUSED*/
1636 1636 static int
1637 1637 zone_locked_mem_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1638 1638 rctl_qty_t nv)
1639 1639 {
1640 1640 ASSERT(MUTEX_HELD(&p->p_lock));
1641 1641 ASSERT(e->rcep_t == RCENTITY_ZONE);
1642 1642 if (e->rcep_p.zone == NULL)
1643 1643 return (0);
1644 1644 e->rcep_p.zone->zone_locked_mem_ctl = nv;
1645 1645 return (0);
1646 1646 }
1647 1647
1648 1648 static rctl_ops_t zone_locked_mem_ops = {
1649 1649 rcop_no_action,
1650 1650 zone_locked_mem_usage,
1651 1651 zone_locked_mem_set,
1652 1652 zone_locked_mem_test
1653 1653 };
1654 1654
1655 1655 /*ARGSUSED*/
1656 1656 static rctl_qty_t
1657 1657 zone_max_swap_usage(rctl_t *rctl, struct proc *p)
1658 1658 {
1659 1659 rctl_qty_t q;
1660 1660 zone_t *z = p->p_zone;
1661 1661
1662 1662 ASSERT(MUTEX_HELD(&p->p_lock));
1663 1663 mutex_enter(&z->zone_mem_lock);
1664 1664 q = z->zone_max_swap;
1665 1665 mutex_exit(&z->zone_mem_lock);
1666 1666 return (q);
1667 1667 }
1668 1668
1669 1669 /*ARGSUSED*/
1670 1670 static int
1671 1671 zone_max_swap_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1672 1672 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1673 1673 {
1674 1674 rctl_qty_t q;
1675 1675 zone_t *z;
1676 1676
1677 1677 z = e->rcep_p.zone;
1678 1678 ASSERT(MUTEX_HELD(&p->p_lock));
1679 1679 ASSERT(MUTEX_HELD(&z->zone_mem_lock));
1680 1680 q = z->zone_max_swap;
1681 1681 if (q + incr > rcntl->rcv_value)
1682 1682 return (1);
1683 1683 return (0);
1684 1684 }
1685 1685
1686 1686 /*ARGSUSED*/
1687 1687 static int
1688 1688 zone_max_swap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1689 1689 rctl_qty_t nv)
1690 1690 {
1691 1691 ASSERT(MUTEX_HELD(&p->p_lock));
1692 1692 ASSERT(e->rcep_t == RCENTITY_ZONE);
1693 1693 if (e->rcep_p.zone == NULL)
1694 1694 return (0);
1695 1695 e->rcep_p.zone->zone_max_swap_ctl = nv;
1696 1696 return (0);
1697 1697 }
1698 1698
1699 1699 static rctl_ops_t zone_max_swap_ops = {
1700 1700 rcop_no_action,
1701 1701 zone_max_swap_usage,
1702 1702 zone_max_swap_set,
1703 1703 zone_max_swap_test
1704 1704 };
1705 1705
1706 1706 /*ARGSUSED*/
1707 1707 static rctl_qty_t
1708 1708 zone_max_lofi_usage(rctl_t *rctl, struct proc *p)
1709 1709 {
1710 1710 rctl_qty_t q;
1711 1711 zone_t *z = p->p_zone;
1712 1712
1713 1713 ASSERT(MUTEX_HELD(&p->p_lock));
1714 1714 mutex_enter(&z->zone_rctl_lock);
1715 1715 q = z->zone_max_lofi;
1716 1716 mutex_exit(&z->zone_rctl_lock);
1717 1717 return (q);
1718 1718 }
1719 1719
1720 1720 /*ARGSUSED*/
1721 1721 static int
1722 1722 zone_max_lofi_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e,
1723 1723 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags)
1724 1724 {
1725 1725 rctl_qty_t q;
1726 1726 zone_t *z;
1727 1727
1728 1728 z = e->rcep_p.zone;
1729 1729 ASSERT(MUTEX_HELD(&p->p_lock));
1730 1730 ASSERT(MUTEX_HELD(&z->zone_rctl_lock));
1731 1731 q = z->zone_max_lofi;
1732 1732 if (q + incr > rcntl->rcv_value)
1733 1733 return (1);
1734 1734 return (0);
1735 1735 }
1736 1736
1737 1737 /*ARGSUSED*/
1738 1738 static int
1739 1739 zone_max_lofi_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e,
1740 1740 rctl_qty_t nv)
1741 1741 {
1742 1742 ASSERT(MUTEX_HELD(&p->p_lock));
1743 1743 ASSERT(e->rcep_t == RCENTITY_ZONE);
1744 1744 if (e->rcep_p.zone == NULL)
1745 1745 return (0);
1746 1746 e->rcep_p.zone->zone_max_lofi_ctl = nv;
1747 1747 return (0);
1748 1748 }
1749 1749
1750 1750 static rctl_ops_t zone_max_lofi_ops = {
1751 1751 rcop_no_action,
1752 1752 zone_max_lofi_usage,
1753 1753 zone_max_lofi_set,
1754 1754 zone_max_lofi_test
1755 1755 };
1756 1756
1757 1757 /*
1758 1758 * Helper function to brand the zone with a unique ID.
1759 1759 */
1760 1760 static void
1761 1761 zone_uniqid(zone_t *zone)
1762 1762 {
1763 1763 static uint64_t uniqid = 0;
1764 1764
1765 1765 ASSERT(MUTEX_HELD(&zonehash_lock));
1766 1766 zone->zone_uniqid = uniqid++;
1767 1767 }
1768 1768
1769 1769 /*
1770 1770 * Returns a held pointer to the "kcred" for the specified zone.
1771 1771 */
1772 1772 struct cred *
1773 1773 zone_get_kcred(zoneid_t zoneid)
1774 1774 {
1775 1775 zone_t *zone;
1776 1776 cred_t *cr;
1777 1777
1778 1778 if ((zone = zone_find_by_id(zoneid)) == NULL)
1779 1779 return (NULL);
1780 1780 cr = zone->zone_kcred;
1781 1781 crhold(cr);
1782 1782 zone_rele(zone);
1783 1783 return (cr);
1784 1784 }
1785 1785
1786 1786 static int
1787 1787 zone_lockedmem_kstat_update(kstat_t *ksp, int rw)
1788 1788 {
1789 1789 zone_t *zone = ksp->ks_private;
1790 1790 zone_kstat_t *zk = ksp->ks_data;
1791 1791
1792 1792 if (rw == KSTAT_WRITE)
1793 1793 return (EACCES);
1794 1794
1795 1795 zk->zk_usage.value.ui64 = zone->zone_locked_mem;
1796 1796 zk->zk_value.value.ui64 = zone->zone_locked_mem_ctl;
1797 1797 return (0);
1798 1798 }
1799 1799
1800 1800 static int
1801 1801 zone_nprocs_kstat_update(kstat_t *ksp, int rw)
1802 1802 {
1803 1803 zone_t *zone = ksp->ks_private;
1804 1804 zone_kstat_t *zk = ksp->ks_data;
1805 1805
1806 1806 if (rw == KSTAT_WRITE)
1807 1807 return (EACCES);
1808 1808
1809 1809 zk->zk_usage.value.ui64 = zone->zone_nprocs;
1810 1810 zk->zk_value.value.ui64 = zone->zone_nprocs_ctl;
1811 1811 return (0);
1812 1812 }
1813 1813
1814 1814 static int
1815 1815 zone_swapresv_kstat_update(kstat_t *ksp, int rw)
1816 1816 {
1817 1817 zone_t *zone = ksp->ks_private;
1818 1818 zone_kstat_t *zk = ksp->ks_data;
1819 1819
1820 1820 if (rw == KSTAT_WRITE)
1821 1821 return (EACCES);
1822 1822
1823 1823 zk->zk_usage.value.ui64 = zone->zone_max_swap;
1824 1824 zk->zk_value.value.ui64 = zone->zone_max_swap_ctl;
1825 1825 return (0);
1826 1826 }
1827 1827
1828 1828 static kstat_t *
1829 1829 zone_kstat_create_common(zone_t *zone, char *name,
1830 1830 int (*updatefunc) (kstat_t *, int))
1831 1831 {
1832 1832 kstat_t *ksp;
1833 1833 zone_kstat_t *zk;
1834 1834
1835 1835 ksp = rctl_kstat_create_zone(zone, name, KSTAT_TYPE_NAMED,
1836 1836 sizeof (zone_kstat_t) / sizeof (kstat_named_t),
1837 1837 KSTAT_FLAG_VIRTUAL);
1838 1838
1839 1839 if (ksp == NULL)
1840 1840 return (NULL);
1841 1841
1842 1842 zk = ksp->ks_data = kmem_alloc(sizeof (zone_kstat_t), KM_SLEEP);
1843 1843 ksp->ks_data_size += strlen(zone->zone_name) + 1;
1844 1844 kstat_named_init(&zk->zk_zonename, "zonename", KSTAT_DATA_STRING);
1845 1845 kstat_named_setstr(&zk->zk_zonename, zone->zone_name);
1846 1846 kstat_named_init(&zk->zk_usage, "usage", KSTAT_DATA_UINT64);
1847 1847 kstat_named_init(&zk->zk_value, "value", KSTAT_DATA_UINT64);
1848 1848 ksp->ks_update = updatefunc;
1849 1849 ksp->ks_private = zone;
1850 1850 kstat_install(ksp);
1851 1851 return (ksp);
1852 1852 }
1853 1853
1854 1854
1855 1855 static int
1856 1856 zone_mcap_kstat_update(kstat_t *ksp, int rw)
1857 1857 {
1858 1858 zone_t *zone = ksp->ks_private;
1859 1859 zone_mcap_kstat_t *zmp = ksp->ks_data;
1860 1860
1861 1861 if (rw == KSTAT_WRITE)
1862 1862 return (EACCES);
1863 1863
1864 1864 zmp->zm_pgpgin.value.ui64 = zone->zone_pgpgin;
1865 1865 zmp->zm_anonpgin.value.ui64 = zone->zone_anonpgin;
1866 1866 zmp->zm_execpgin.value.ui64 = zone->zone_execpgin;
1867 1867 zmp->zm_fspgin.value.ui64 = zone->zone_fspgin;
1868 1868 zmp->zm_anon_alloc_fail.value.ui64 = zone->zone_anon_alloc_fail;
1869 1869
1870 1870 return (0);
1871 1871 }
1872 1872
1873 1873 static kstat_t *
1874 1874 zone_mcap_kstat_create(zone_t *zone)
1875 1875 {
1876 1876 kstat_t *ksp;
1877 1877 zone_mcap_kstat_t *zmp;
1878 1878
1879 1879 if ((ksp = kstat_create_zone("memory_cap", zone->zone_id,
1880 1880 zone->zone_name, "zone_memory_cap", KSTAT_TYPE_NAMED,
1881 1881 sizeof (zone_mcap_kstat_t) / sizeof (kstat_named_t),
1882 1882 KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
1883 1883 return (NULL);
1884 1884
1885 1885 if (zone->zone_id != GLOBAL_ZONEID)
1886 1886 kstat_zone_add(ksp, GLOBAL_ZONEID);
1887 1887
1888 1888 zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_mcap_kstat_t), KM_SLEEP);
1889 1889 ksp->ks_data_size += strlen(zone->zone_name) + 1;
1890 1890 ksp->ks_lock = &zone->zone_mcap_lock;
1891 1891 zone->zone_mcap_stats = zmp;
1892 1892
1893 1893 /* The kstat "name" field is not large enough for a full zonename */
1894 1894 kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
1895 1895 kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
1896 1896 kstat_named_init(&zmp->zm_pgpgin, "pgpgin", KSTAT_DATA_UINT64);
1897 1897 kstat_named_init(&zmp->zm_anonpgin, "anonpgin", KSTAT_DATA_UINT64);
1898 1898 kstat_named_init(&zmp->zm_execpgin, "execpgin", KSTAT_DATA_UINT64);
1899 1899 kstat_named_init(&zmp->zm_fspgin, "fspgin", KSTAT_DATA_UINT64);
1900 1900 kstat_named_init(&zmp->zm_anon_alloc_fail, "anon_alloc_fail",
1901 1901 KSTAT_DATA_UINT64);
1902 1902
1903 1903 ksp->ks_update = zone_mcap_kstat_update;
1904 1904 ksp->ks_private = zone;
|
↓ open down ↓ |
1721 lines elided |
↑ open up ↑ |
1905 1905
1906 1906 kstat_install(ksp);
1907 1907 return (ksp);
1908 1908 }
1909 1909
1910 1910 static int
1911 1911 zone_misc_kstat_update(kstat_t *ksp, int rw)
1912 1912 {
1913 1913 zone_t *zone = ksp->ks_private;
1914 1914 zone_misc_kstat_t *zmp = ksp->ks_data;
1915 - hrtime_t tmp;
1915 + hrtime_t hrtime;
1916 + uint64_t tmp;
1916 1917
1917 1918 if (rw == KSTAT_WRITE)
1918 1919 return (EACCES);
1919 1920
1920 - tmp = zone->zone_utime;
1921 - scalehrtime(&tmp);
1922 - zmp->zm_utime.value.ui64 = tmp;
1923 - tmp = zone->zone_stime;
1924 - scalehrtime(&tmp);
1925 - zmp->zm_stime.value.ui64 = tmp;
1926 - tmp = zone->zone_wtime;
1927 - scalehrtime(&tmp);
1928 - zmp->zm_wtime.value.ui64 = tmp;
1921 + tmp = cpu_uarray_sum(zone->zone_ustate, ZONE_USTATE_STIME);
1922 + hrtime = UINT64_OVERFLOW_TO_INT64(tmp);
1923 + scalehrtime(&hrtime);
1924 + zmp->zm_stime.value.ui64 = hrtime;
1929 1925
1926 + tmp = cpu_uarray_sum(zone->zone_ustate, ZONE_USTATE_UTIME);
1927 + hrtime = UINT64_OVERFLOW_TO_INT64(tmp);
1928 + scalehrtime(&hrtime);
1929 + zmp->zm_utime.value.ui64 = hrtime;
1930 +
1931 + tmp = cpu_uarray_sum(zone->zone_ustate, ZONE_USTATE_WTIME);
1932 + hrtime = UINT64_OVERFLOW_TO_INT64(tmp);
1933 + scalehrtime(&hrtime);
1934 + zmp->zm_wtime.value.ui64 = hrtime;
1935 +
1930 1936 zmp->zm_avenrun1.value.ui32 = zone->zone_avenrun[0];
1931 1937 zmp->zm_avenrun5.value.ui32 = zone->zone_avenrun[1];
1932 1938 zmp->zm_avenrun15.value.ui32 = zone->zone_avenrun[2];
1933 1939
1934 1940 zmp->zm_ffcap.value.ui32 = zone->zone_ffcap;
1935 1941 zmp->zm_ffnoproc.value.ui32 = zone->zone_ffnoproc;
1936 1942 zmp->zm_ffnomem.value.ui32 = zone->zone_ffnomem;
1937 1943 zmp->zm_ffmisc.value.ui32 = zone->zone_ffmisc;
1938 1944
1939 1945 zmp->zm_nested_intp.value.ui32 = zone->zone_nested_intp;
1940 1946
1941 1947 zmp->zm_init_pid.value.ui32 = zone->zone_proc_initpid;
1942 1948 zmp->zm_boot_time.value.ui64 = (uint64_t)zone->zone_boot_time;
1943 1949
1944 1950 return (0);
1945 1951 }
1946 1952
1947 1953 static kstat_t *
1948 1954 zone_misc_kstat_create(zone_t *zone)
1949 1955 {
1950 1956 kstat_t *ksp;
1951 1957 zone_misc_kstat_t *zmp;
1952 1958
1953 1959 if ((ksp = kstat_create_zone("zones", zone->zone_id,
1954 1960 zone->zone_name, "zone_misc", KSTAT_TYPE_NAMED,
1955 1961 sizeof (zone_misc_kstat_t) / sizeof (kstat_named_t),
1956 1962 KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL)
1957 1963 return (NULL);
1958 1964
1959 1965 if (zone->zone_id != GLOBAL_ZONEID)
1960 1966 kstat_zone_add(ksp, GLOBAL_ZONEID);
1961 1967
1962 1968 zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_misc_kstat_t), KM_SLEEP);
1963 1969 ksp->ks_data_size += strlen(zone->zone_name) + 1;
1964 1970 ksp->ks_lock = &zone->zone_misc_lock;
1965 1971 zone->zone_misc_stats = zmp;
1966 1972
1967 1973 /* The kstat "name" field is not large enough for a full zonename */
1968 1974 kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING);
1969 1975 kstat_named_setstr(&zmp->zm_zonename, zone->zone_name);
1970 1976 kstat_named_init(&zmp->zm_utime, "nsec_user", KSTAT_DATA_UINT64);
1971 1977 kstat_named_init(&zmp->zm_stime, "nsec_sys", KSTAT_DATA_UINT64);
1972 1978 kstat_named_init(&zmp->zm_wtime, "nsec_waitrq", KSTAT_DATA_UINT64);
1973 1979 kstat_named_init(&zmp->zm_avenrun1, "avenrun_1min", KSTAT_DATA_UINT32);
1974 1980 kstat_named_init(&zmp->zm_avenrun5, "avenrun_5min", KSTAT_DATA_UINT32);
1975 1981 kstat_named_init(&zmp->zm_avenrun15, "avenrun_15min",
1976 1982 KSTAT_DATA_UINT32);
1977 1983 kstat_named_init(&zmp->zm_ffcap, "forkfail_cap", KSTAT_DATA_UINT32);
1978 1984 kstat_named_init(&zmp->zm_ffnoproc, "forkfail_noproc",
1979 1985 KSTAT_DATA_UINT32);
1980 1986 kstat_named_init(&zmp->zm_ffnomem, "forkfail_nomem", KSTAT_DATA_UINT32);
1981 1987 kstat_named_init(&zmp->zm_ffmisc, "forkfail_misc", KSTAT_DATA_UINT32);
1982 1988 kstat_named_init(&zmp->zm_nested_intp, "nested_interp",
1983 1989 KSTAT_DATA_UINT32);
1984 1990 kstat_named_init(&zmp->zm_init_pid, "init_pid", KSTAT_DATA_UINT32);
1985 1991 kstat_named_init(&zmp->zm_boot_time, "boot_time", KSTAT_DATA_UINT64);
1986 1992
1987 1993 ksp->ks_update = zone_misc_kstat_update;
1988 1994 ksp->ks_private = zone;
1989 1995
1990 1996 kstat_install(ksp);
1991 1997 return (ksp);
1992 1998 }
1993 1999
1994 2000 static void
1995 2001 zone_kstat_create(zone_t *zone)
1996 2002 {
1997 2003 zone->zone_lockedmem_kstat = zone_kstat_create_common(zone,
1998 2004 "lockedmem", zone_lockedmem_kstat_update);
1999 2005 zone->zone_swapresv_kstat = zone_kstat_create_common(zone,
2000 2006 "swapresv", zone_swapresv_kstat_update);
2001 2007 zone->zone_nprocs_kstat = zone_kstat_create_common(zone,
2002 2008 "nprocs", zone_nprocs_kstat_update);
2003 2009
2004 2010 if ((zone->zone_mcap_ksp = zone_mcap_kstat_create(zone)) == NULL) {
2005 2011 zone->zone_mcap_stats = kmem_zalloc(
2006 2012 sizeof (zone_mcap_kstat_t), KM_SLEEP);
2007 2013 }
2008 2014
2009 2015 if ((zone->zone_misc_ksp = zone_misc_kstat_create(zone)) == NULL) {
2010 2016 zone->zone_misc_stats = kmem_zalloc(
2011 2017 sizeof (zone_misc_kstat_t), KM_SLEEP);
2012 2018 }
2013 2019 }
2014 2020
2015 2021 static void
2016 2022 zone_kstat_delete_common(kstat_t **pkstat, size_t datasz)
2017 2023 {
2018 2024 void *data;
2019 2025
2020 2026 if (*pkstat != NULL) {
2021 2027 data = (*pkstat)->ks_data;
2022 2028 kstat_delete(*pkstat);
2023 2029 kmem_free(data, datasz);
2024 2030 *pkstat = NULL;
2025 2031 }
2026 2032 }
2027 2033
2028 2034 static void
2029 2035 zone_kstat_delete(zone_t *zone)
2030 2036 {
2031 2037 zone_kstat_delete_common(&zone->zone_lockedmem_kstat,
2032 2038 sizeof (zone_kstat_t));
2033 2039 zone_kstat_delete_common(&zone->zone_swapresv_kstat,
2034 2040 sizeof (zone_kstat_t));
2035 2041 zone_kstat_delete_common(&zone->zone_nprocs_kstat,
2036 2042 sizeof (zone_kstat_t));
2037 2043 zone_kstat_delete_common(&zone->zone_mcap_ksp,
2038 2044 sizeof (zone_mcap_kstat_t));
2039 2045 zone_kstat_delete_common(&zone->zone_misc_ksp,
2040 2046 sizeof (zone_misc_kstat_t));
2041 2047 }
2042 2048
2043 2049 /*
2044 2050 * Called very early on in boot to initialize the ZSD list so that
2045 2051 * zone_key_create() can be called before zone_init(). It also initializes
2046 2052 * portions of zone0 which may be used before zone_init() is called. The
2047 2053 * variable "global_zone" will be set when zone0 is fully initialized by
2048 2054 * zone_init().
2049 2055 */
2050 2056 void
2051 2057 zone_zsd_init(void)
2052 2058 {
2053 2059 mutex_init(&zonehash_lock, NULL, MUTEX_DEFAULT, NULL);
2054 2060 mutex_init(&zsd_key_lock, NULL, MUTEX_DEFAULT, NULL);
2055 2061 list_create(&zsd_registered_keys, sizeof (struct zsd_entry),
2056 2062 offsetof(struct zsd_entry, zsd_linkage));
2057 2063 list_create(&zone_active, sizeof (zone_t),
2058 2064 offsetof(zone_t, zone_linkage));
2059 2065 list_create(&zone_deathrow, sizeof (zone_t),
2060 2066 offsetof(zone_t, zone_linkage));
2061 2067
2062 2068 mutex_init(&zone0.zone_lock, NULL, MUTEX_DEFAULT, NULL);
2063 2069 mutex_init(&zone0.zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
2064 2070 mutex_init(&zone0.zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
2065 2071 zone0.zone_shares = 1;
2066 2072 zone0.zone_nlwps = 0;
2067 2073 zone0.zone_nlwps_ctl = INT_MAX;
2068 2074 zone0.zone_nprocs = 0;
2069 2075 zone0.zone_nprocs_ctl = INT_MAX;
2070 2076 zone0.zone_locked_mem = 0;
2071 2077 zone0.zone_locked_mem_ctl = UINT64_MAX;
2072 2078 ASSERT(zone0.zone_max_swap == 0);
2073 2079 zone0.zone_max_swap_ctl = UINT64_MAX;
2074 2080 zone0.zone_max_lofi = 0;
2075 2081 zone0.zone_max_lofi_ctl = UINT64_MAX;
2076 2082 zone0.zone_shmmax = 0;
2077 2083 zone0.zone_ipc.ipcq_shmmni = 0;
2078 2084 zone0.zone_ipc.ipcq_semmni = 0;
2079 2085 zone0.zone_ipc.ipcq_msgmni = 0;
2080 2086 zone0.zone_name = GLOBAL_ZONENAME;
2081 2087 zone0.zone_nodename = utsname.nodename;
2082 2088 zone0.zone_domain = srpc_domain;
2083 2089 zone0.zone_hostid = HW_INVALID_HOSTID;
2084 2090 zone0.zone_fs_allowed = NULL;
2085 2091 psecflags_default(&zone0.zone_secflags);
2086 2092 zone0.zone_ref = 1;
2087 2093 zone0.zone_id = GLOBAL_ZONEID;
2088 2094 zone0.zone_status = ZONE_IS_RUNNING;
2089 2095 zone0.zone_rootpath = "/";
|
↓ open down ↓ |
150 lines elided |
↑ open up ↑ |
2090 2096 zone0.zone_rootpathlen = 2;
2091 2097 zone0.zone_psetid = ZONE_PS_INVAL;
2092 2098 zone0.zone_ncpus = 0;
2093 2099 zone0.zone_ncpus_online = 0;
2094 2100 zone0.zone_proc_initpid = 1;
2095 2101 zone0.zone_initname = initname;
2096 2102 zone0.zone_lockedmem_kstat = NULL;
2097 2103 zone0.zone_swapresv_kstat = NULL;
2098 2104 zone0.zone_nprocs_kstat = NULL;
2099 2105
2100 - zone0.zone_stime = 0;
2101 - zone0.zone_utime = 0;
2102 - zone0.zone_wtime = 0;
2103 -
2104 2106 list_create(&zone0.zone_ref_list, sizeof (zone_ref_t),
2105 2107 offsetof(zone_ref_t, zref_linkage));
2106 2108 list_create(&zone0.zone_zsd, sizeof (struct zsd_entry),
2107 2109 offsetof(struct zsd_entry, zsd_linkage));
2108 2110 list_insert_head(&zone_active, &zone0);
2109 2111
2110 2112 /*
2111 2113 * The root filesystem is not mounted yet, so zone_rootvp cannot be set
2112 2114 * to anything meaningful. It is assigned to be 'rootdir' in
2113 2115 * vfs_mountroot().
2114 2116 */
2115 2117 zone0.zone_rootvp = NULL;
2116 2118 zone0.zone_vfslist = NULL;
2117 2119 zone0.zone_bootargs = initargs;
2118 2120 zone0.zone_privset = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
2119 2121 /*
2120 2122 * The global zone has all privileges
2121 2123 */
2122 2124 priv_fillset(zone0.zone_privset);
2123 2125 /*
2124 2126 * Add p0 to the global zone
2125 2127 */
2126 2128 zone0.zone_zsched = &p0;
2127 2129 p0.p_zone = &zone0;
2128 2130 }
2129 2131
2130 2132 /*
2131 2133 * Compute a hash value based on the contents of the label and the DOI. The
2132 2134 * hash algorithm is somewhat arbitrary, but is based on the observation that
2133 2135 * humans will likely pick labels that differ by amounts that work out to be
2134 2136 * multiples of the number of hash chains, and thus stirring in some primes
2135 2137 * should help.
2136 2138 */
2137 2139 static uint_t
2138 2140 hash_bylabel(void *hdata, mod_hash_key_t key)
2139 2141 {
2140 2142 const ts_label_t *lab = (ts_label_t *)key;
2141 2143 const uint32_t *up, *ue;
2142 2144 uint_t hash;
2143 2145 int i;
2144 2146
2145 2147 _NOTE(ARGUNUSED(hdata));
2146 2148
2147 2149 hash = lab->tsl_doi + (lab->tsl_doi << 1);
2148 2150 /* we depend on alignment of label, but not representation */
2149 2151 up = (const uint32_t *)&lab->tsl_label;
2150 2152 ue = up + sizeof (lab->tsl_label) / sizeof (*up);
2151 2153 i = 1;
2152 2154 while (up < ue) {
2153 2155 /* using 2^n + 1, 1 <= n <= 16 as source of many primes */
2154 2156 hash += *up + (*up << ((i % 16) + 1));
2155 2157 up++;
2156 2158 i++;
2157 2159 }
2158 2160 return (hash);
2159 2161 }
2160 2162
2161 2163 /*
2162 2164 * All that mod_hash cares about here is zero (equal) versus non-zero (not
2163 2165 * equal). This may need to be changed if less than / greater than is ever
2164 2166 * needed.
2165 2167 */
2166 2168 static int
2167 2169 hash_labelkey_cmp(mod_hash_key_t key1, mod_hash_key_t key2)
2168 2170 {
2169 2171 ts_label_t *lab1 = (ts_label_t *)key1;
2170 2172 ts_label_t *lab2 = (ts_label_t *)key2;
2171 2173
2172 2174 return (label_equal(lab1, lab2) ? 0 : 1);
2173 2175 }
2174 2176
2175 2177 /*
2176 2178 * Called by main() to initialize the zones framework.
2177 2179 */
2178 2180 void
2179 2181 zone_init(void)
2180 2182 {
2181 2183 rctl_dict_entry_t *rde;
2182 2184 rctl_val_t *dval;
2183 2185 rctl_set_t *set;
2184 2186 rctl_alloc_gp_t *gp;
2185 2187 rctl_entity_p_t e;
2186 2188 int res;
2187 2189
2188 2190 ASSERT(curproc == &p0);
2189 2191
2190 2192 /*
2191 2193 * Create ID space for zone IDs. ID 0 is reserved for the
2192 2194 * global zone.
2193 2195 */
2194 2196 zoneid_space = id_space_create("zoneid_space", 1, MAX_ZONEID);
2195 2197
2196 2198 /*
2197 2199 * Initialize generic zone resource controls, if any.
2198 2200 */
2199 2201 rc_zone_cpu_shares = rctl_register("zone.cpu-shares",
2200 2202 RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_NEVER |
2201 2203 RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT | RCTL_GLOBAL_SYSLOG_NEVER,
2202 2204 FSS_MAXSHARES, FSS_MAXSHARES, &zone_cpu_shares_ops);
2203 2205
2204 2206 rc_zone_cpu_cap = rctl_register("zone.cpu-cap",
2205 2207 RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_ALWAYS |
2206 2208 RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |RCTL_GLOBAL_SYSLOG_NEVER |
2207 2209 RCTL_GLOBAL_INFINITE,
2208 2210 MAXCAP, MAXCAP, &zone_cpu_cap_ops);
2209 2211
2210 2212 rc_zone_nlwps = rctl_register("zone.max-lwps", RCENTITY_ZONE,
2211 2213 RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
2212 2214 INT_MAX, INT_MAX, &zone_lwps_ops);
2213 2215
2214 2216 rc_zone_nprocs = rctl_register("zone.max-processes", RCENTITY_ZONE,
2215 2217 RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT,
2216 2218 INT_MAX, INT_MAX, &zone_procs_ops);
2217 2219
2218 2220 /*
2219 2221 * System V IPC resource controls
2220 2222 */
2221 2223 rc_zone_msgmni = rctl_register("zone.max-msg-ids",
2222 2224 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2223 2225 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_msgmni_ops);
2224 2226
2225 2227 rc_zone_semmni = rctl_register("zone.max-sem-ids",
2226 2228 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2227 2229 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_semmni_ops);
2228 2230
2229 2231 rc_zone_shmmni = rctl_register("zone.max-shm-ids",
2230 2232 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2231 2233 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_shmmni_ops);
2232 2234
2233 2235 rc_zone_shmmax = rctl_register("zone.max-shm-memory",
2234 2236 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC |
2235 2237 RCTL_GLOBAL_BYTES, UINT64_MAX, UINT64_MAX, &zone_shmmax_ops);
2236 2238
2237 2239 /*
2238 2240 * Create a rctl_val with PRIVILEGED, NOACTION, value = 1. Then attach
2239 2241 * this at the head of the rctl_dict_entry for ``zone.cpu-shares''.
2240 2242 */
2241 2243 dval = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
2242 2244 bzero(dval, sizeof (rctl_val_t));
2243 2245 dval->rcv_value = 1;
2244 2246 dval->rcv_privilege = RCPRIV_PRIVILEGED;
2245 2247 dval->rcv_flagaction = RCTL_LOCAL_NOACTION;
2246 2248 dval->rcv_action_recip_pid = -1;
2247 2249
2248 2250 rde = rctl_dict_lookup("zone.cpu-shares");
2249 2251 (void) rctl_val_list_insert(&rde->rcd_default_value, dval);
2250 2252
2251 2253 rc_zone_locked_mem = rctl_register("zone.max-locked-memory",
2252 2254 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
2253 2255 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2254 2256 &zone_locked_mem_ops);
2255 2257
2256 2258 rc_zone_max_swap = rctl_register("zone.max-swap",
2257 2259 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES |
2258 2260 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2259 2261 &zone_max_swap_ops);
2260 2262
2261 2263 rc_zone_max_lofi = rctl_register("zone.max-lofi",
2262 2264 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |
2263 2265 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX,
2264 2266 &zone_max_lofi_ops);
2265 2267
2266 2268 /*
2267 2269 * Initialize the ``global zone''.
2268 2270 */
2269 2271 set = rctl_set_create();
2270 2272 gp = rctl_set_init_prealloc(RCENTITY_ZONE);
2271 2273 mutex_enter(&p0.p_lock);
2272 2274 e.rcep_p.zone = &zone0;
2273 2275 e.rcep_t = RCENTITY_ZONE;
2274 2276 zone0.zone_rctls = rctl_set_init(RCENTITY_ZONE, &p0, &e, set,
2275 2277 gp);
2276 2278
2277 2279 zone0.zone_nlwps = p0.p_lwpcnt;
2278 2280 zone0.zone_nprocs = 1;
2279 2281 zone0.zone_ntasks = 1;
2280 2282 mutex_exit(&p0.p_lock);
2281 2283 zone0.zone_restart_init = B_TRUE;
2282 2284 zone0.zone_brand = &native_brand;
2283 2285 rctl_prealloc_destroy(gp);
2284 2286 /*
2285 2287 * pool_default hasn't been initialized yet, so we let pool_init()
2286 2288 * take care of making sure the global zone is in the default pool.
2287 2289 */
2288 2290
2289 2291 /*
2290 2292 * Initialize global zone kstats
2291 2293 */
2292 2294 zone_kstat_create(&zone0);
2293 2295
2294 2296 /*
2295 2297 * Initialize zone label.
2296 2298 * mlp are initialized when tnzonecfg is loaded.
|
↓ open down ↓ |
183 lines elided |
↑ open up ↑ |
2297 2299 */
2298 2300 zone0.zone_slabel = l_admin_low;
2299 2301 rw_init(&zone0.zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
2300 2302 label_hold(l_admin_low);
2301 2303
2302 2304 /*
2303 2305 * Initialise the lock for the database structure used by mntfs.
2304 2306 */
2305 2307 rw_init(&zone0.zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL);
2306 2308
2309 + zone0.zone_ustate = cpu_uarray_zalloc(ZONE_USTATE_MAX, KM_SLEEP);
2310 +
2307 2311 mutex_enter(&zonehash_lock);
2308 2312 zone_uniqid(&zone0);
2309 2313 ASSERT(zone0.zone_uniqid == GLOBAL_ZONEUNIQID);
2310 2314
2311 2315 zonehashbyid = mod_hash_create_idhash("zone_by_id", zone_hash_size,
2312 2316 mod_hash_null_valdtor);
2313 2317 zonehashbyname = mod_hash_create_strhash("zone_by_name",
2314 2318 zone_hash_size, mod_hash_null_valdtor);
2315 2319 /*
2316 2320 * maintain zonehashbylabel only for labeled systems
2317 2321 */
2318 2322 if (is_system_labeled())
2319 2323 zonehashbylabel = mod_hash_create_extended("zone_by_label",
2320 2324 zone_hash_size, mod_hash_null_keydtor,
2321 2325 mod_hash_null_valdtor, hash_bylabel, NULL,
2322 2326 hash_labelkey_cmp, KM_SLEEP);
2323 2327 zonecount = 1;
2324 2328
2325 2329 (void) mod_hash_insert(zonehashbyid, (mod_hash_key_t)GLOBAL_ZONEID,
2326 2330 (mod_hash_val_t)&zone0);
2327 2331 (void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)zone0.zone_name,
2328 2332 (mod_hash_val_t)&zone0);
2329 2333 if (is_system_labeled()) {
2330 2334 zone0.zone_flags |= ZF_HASHED_LABEL;
2331 2335 (void) mod_hash_insert(zonehashbylabel,
2332 2336 (mod_hash_key_t)zone0.zone_slabel, (mod_hash_val_t)&zone0);
2333 2337 }
2334 2338 mutex_exit(&zonehash_lock);
2335 2339
2336 2340 /*
2337 2341 * We avoid setting zone_kcred until now, since kcred is initialized
2338 2342 * sometime after zone_zsd_init() and before zone_init().
2339 2343 */
2340 2344 zone0.zone_kcred = kcred;
2341 2345 /*
2342 2346 * The global zone is fully initialized (except for zone_rootvp which
2343 2347 * will be set when the root filesystem is mounted).
2344 2348 */
2345 2349 global_zone = &zone0;
2346 2350
2347 2351 /*
2348 2352 * Setup an event channel to send zone status change notifications on
2349 2353 */
2350 2354 res = sysevent_evc_bind(ZONE_EVENT_CHANNEL, &zone_event_chan,
2351 2355 EVCH_CREAT);
2352 2356
2353 2357 if (res)
2354 2358 panic("Sysevent_evc_bind failed during zone setup.\n");
2355 2359
2356 2360 }
2357 2361
2358 2362 static void
2359 2363 zone_free(zone_t *zone)
2360 2364 {
2361 2365 ASSERT(zone != global_zone);
2362 2366 ASSERT(zone->zone_ntasks == 0);
2363 2367 ASSERT(zone->zone_nlwps == 0);
2364 2368 ASSERT(zone->zone_nprocs == 0);
2365 2369 ASSERT(zone->zone_cred_ref == 0);
2366 2370 ASSERT(zone->zone_kcred == NULL);
2367 2371 ASSERT(zone_status_get(zone) == ZONE_IS_DEAD ||
2368 2372 zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
2369 2373 ASSERT(list_is_empty(&zone->zone_ref_list));
2370 2374
2371 2375 /*
2372 2376 * Remove any zone caps.
2373 2377 */
2374 2378 cpucaps_zone_remove(zone);
2375 2379
2376 2380 ASSERT(zone->zone_cpucap == NULL);
2377 2381
2378 2382 /* remove from deathrow list */
2379 2383 if (zone_status_get(zone) == ZONE_IS_DEAD) {
2380 2384 ASSERT(zone->zone_ref == 0);
|
↓ open down ↓ |
64 lines elided |
↑ open up ↑ |
2381 2385 mutex_enter(&zone_deathrow_lock);
2382 2386 list_remove(&zone_deathrow, zone);
2383 2387 mutex_exit(&zone_deathrow_lock);
2384 2388 }
2385 2389
2386 2390 list_destroy(&zone->zone_ref_list);
2387 2391 zone_free_zsd(zone);
2388 2392 zone_free_datasets(zone);
2389 2393 list_destroy(&zone->zone_dl_list);
2390 2394
2395 + cpu_uarray_free(zone->zone_ustate);
2396 +
2391 2397 if (zone->zone_rootvp != NULL)
2392 2398 VN_RELE(zone->zone_rootvp);
2393 2399 if (zone->zone_rootpath)
2394 2400 kmem_free(zone->zone_rootpath, zone->zone_rootpathlen);
2395 2401 if (zone->zone_name != NULL)
2396 2402 kmem_free(zone->zone_name, ZONENAME_MAX);
2397 2403 if (zone->zone_slabel != NULL)
2398 2404 label_rele(zone->zone_slabel);
2399 2405 if (zone->zone_nodename != NULL)
2400 2406 kmem_free(zone->zone_nodename, _SYS_NMLN);
2401 2407 if (zone->zone_domain != NULL)
2402 2408 kmem_free(zone->zone_domain, _SYS_NMLN);
2403 2409 if (zone->zone_privset != NULL)
2404 2410 kmem_free(zone->zone_privset, sizeof (priv_set_t));
2405 2411 if (zone->zone_rctls != NULL)
2406 2412 rctl_set_free(zone->zone_rctls);
2407 2413 if (zone->zone_bootargs != NULL)
2408 2414 strfree(zone->zone_bootargs);
2409 2415 if (zone->zone_initname != NULL)
2410 2416 strfree(zone->zone_initname);
2411 2417 if (zone->zone_fs_allowed != NULL)
2412 2418 strfree(zone->zone_fs_allowed);
2413 2419 if (zone->zone_pfexecd != NULL)
2414 2420 klpd_freelist(&zone->zone_pfexecd);
2415 2421 id_free(zoneid_space, zone->zone_id);
2416 2422 mutex_destroy(&zone->zone_lock);
2417 2423 cv_destroy(&zone->zone_cv);
2418 2424 rw_destroy(&zone->zone_mlps.mlpl_rwlock);
2419 2425 rw_destroy(&zone->zone_mntfs_db_lock);
2420 2426 kmem_free(zone, sizeof (zone_t));
2421 2427 }
2422 2428
2423 2429 /*
2424 2430 * See block comment at the top of this file for information about zone
2425 2431 * status values.
2426 2432 */
2427 2433 /*
2428 2434 * Convenience function for setting zone status.
2429 2435 */
2430 2436 static void
2431 2437 zone_status_set(zone_t *zone, zone_status_t status)
2432 2438 {
2433 2439
2434 2440 nvlist_t *nvl = NULL;
2435 2441 ASSERT(MUTEX_HELD(&zone_status_lock));
2436 2442 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE &&
2437 2443 status >= zone_status_get(zone));
2438 2444
2439 2445 if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) ||
2440 2446 nvlist_add_string(nvl, ZONE_CB_NAME, zone->zone_name) ||
2441 2447 nvlist_add_string(nvl, ZONE_CB_NEWSTATE,
2442 2448 zone_status_table[status]) ||
2443 2449 nvlist_add_string(nvl, ZONE_CB_OLDSTATE,
2444 2450 zone_status_table[zone->zone_status]) ||
2445 2451 nvlist_add_int32(nvl, ZONE_CB_ZONEID, zone->zone_id) ||
2446 2452 nvlist_add_uint64(nvl, ZONE_CB_TIMESTAMP, (uint64_t)gethrtime()) ||
2447 2453 sysevent_evc_publish(zone_event_chan, ZONE_EVENT_STATUS_CLASS,
2448 2454 ZONE_EVENT_STATUS_SUBCLASS, "sun.com", "kernel", nvl, EVCH_SLEEP)) {
2449 2455 #ifdef DEBUG
2450 2456 (void) printf(
2451 2457 "Failed to allocate and send zone state change event.\n");
2452 2458 #endif
2453 2459 }
2454 2460 nvlist_free(nvl);
2455 2461
2456 2462 zone->zone_status = status;
2457 2463
2458 2464 cv_broadcast(&zone->zone_cv);
2459 2465 }
2460 2466
2461 2467 /*
2462 2468 * Public function to retrieve the zone status. The zone status may
2463 2469 * change after it is retrieved.
2464 2470 */
2465 2471 zone_status_t
2466 2472 zone_status_get(zone_t *zone)
2467 2473 {
2468 2474 return (zone->zone_status);
2469 2475 }
2470 2476
2471 2477 static int
2472 2478 zone_set_bootargs(zone_t *zone, const char *zone_bootargs)
2473 2479 {
2474 2480 char *buf = kmem_zalloc(BOOTARGS_MAX, KM_SLEEP);
2475 2481 int err = 0;
2476 2482
2477 2483 ASSERT(zone != global_zone);
2478 2484 if ((err = copyinstr(zone_bootargs, buf, BOOTARGS_MAX, NULL)) != 0)
2479 2485 goto done; /* EFAULT or ENAMETOOLONG */
2480 2486
2481 2487 if (zone->zone_bootargs != NULL)
2482 2488 strfree(zone->zone_bootargs);
2483 2489
2484 2490 zone->zone_bootargs = strdup(buf);
2485 2491
2486 2492 done:
2487 2493 kmem_free(buf, BOOTARGS_MAX);
2488 2494 return (err);
2489 2495 }
2490 2496
2491 2497 static int
2492 2498 zone_set_brand(zone_t *zone, const char *brand)
2493 2499 {
2494 2500 struct brand_attr *attrp;
2495 2501 brand_t *bp;
2496 2502
2497 2503 attrp = kmem_alloc(sizeof (struct brand_attr), KM_SLEEP);
2498 2504 if (copyin(brand, attrp, sizeof (struct brand_attr)) != 0) {
2499 2505 kmem_free(attrp, sizeof (struct brand_attr));
2500 2506 return (EFAULT);
2501 2507 }
2502 2508
2503 2509 bp = brand_register_zone(attrp);
2504 2510 kmem_free(attrp, sizeof (struct brand_attr));
2505 2511 if (bp == NULL)
2506 2512 return (EINVAL);
2507 2513
2508 2514 /*
2509 2515 * This is the only place where a zone can change it's brand.
2510 2516 * We already need to hold zone_status_lock to check the zone
2511 2517 * status, so we'll just use that lock to serialize zone
2512 2518 * branding requests as well.
2513 2519 */
2514 2520 mutex_enter(&zone_status_lock);
2515 2521
2516 2522 /* Re-Branding is not allowed and the zone can't be booted yet */
2517 2523 if ((ZONE_IS_BRANDED(zone)) ||
2518 2524 (zone_status_get(zone) >= ZONE_IS_BOOTING)) {
2519 2525 mutex_exit(&zone_status_lock);
2520 2526 brand_unregister_zone(bp);
2521 2527 return (EINVAL);
2522 2528 }
2523 2529
2524 2530 /* set up the brand specific data */
2525 2531 zone->zone_brand = bp;
2526 2532 ZBROP(zone)->b_init_brand_data(zone);
2527 2533
2528 2534 mutex_exit(&zone_status_lock);
2529 2535 return (0);
2530 2536 }
2531 2537
2532 2538 static int
2533 2539 zone_set_secflags(zone_t *zone, const psecflags_t *zone_secflags)
2534 2540 {
2535 2541 int err = 0;
2536 2542 psecflags_t psf;
2537 2543
2538 2544 ASSERT(zone != global_zone);
2539 2545
2540 2546 if ((err = copyin(zone_secflags, &psf, sizeof (psf))) != 0)
2541 2547 return (err);
2542 2548
2543 2549 if (zone_status_get(zone) > ZONE_IS_READY)
2544 2550 return (EINVAL);
2545 2551
2546 2552 if (!psecflags_validate(&psf))
2547 2553 return (EINVAL);
2548 2554
2549 2555 (void) memcpy(&zone->zone_secflags, &psf, sizeof (psf));
2550 2556
2551 2557 /* Set security flags on the zone's zsched */
2552 2558 (void) memcpy(&zone->zone_zsched->p_secflags, &zone->zone_secflags,
2553 2559 sizeof (zone->zone_zsched->p_secflags));
2554 2560
2555 2561 return (0);
2556 2562 }
2557 2563
2558 2564 static int
2559 2565 zone_set_fs_allowed(zone_t *zone, const char *zone_fs_allowed)
2560 2566 {
2561 2567 char *buf = kmem_zalloc(ZONE_FS_ALLOWED_MAX, KM_SLEEP);
2562 2568 int err = 0;
2563 2569
2564 2570 ASSERT(zone != global_zone);
2565 2571 if ((err = copyinstr(zone_fs_allowed, buf,
2566 2572 ZONE_FS_ALLOWED_MAX, NULL)) != 0)
2567 2573 goto done;
2568 2574
2569 2575 if (zone->zone_fs_allowed != NULL)
2570 2576 strfree(zone->zone_fs_allowed);
2571 2577
2572 2578 zone->zone_fs_allowed = strdup(buf);
2573 2579
2574 2580 done:
2575 2581 kmem_free(buf, ZONE_FS_ALLOWED_MAX);
2576 2582 return (err);
2577 2583 }
2578 2584
2579 2585 static int
2580 2586 zone_set_initname(zone_t *zone, const char *zone_initname)
2581 2587 {
2582 2588 char initname[INITNAME_SZ];
2583 2589 size_t len;
2584 2590 int err = 0;
2585 2591
2586 2592 ASSERT(zone != global_zone);
2587 2593 if ((err = copyinstr(zone_initname, initname, INITNAME_SZ, &len)) != 0)
2588 2594 return (err); /* EFAULT or ENAMETOOLONG */
2589 2595
2590 2596 if (zone->zone_initname != NULL)
2591 2597 strfree(zone->zone_initname);
2592 2598
2593 2599 zone->zone_initname = kmem_alloc(strlen(initname) + 1, KM_SLEEP);
2594 2600 (void) strcpy(zone->zone_initname, initname);
2595 2601 return (0);
2596 2602 }
2597 2603
2598 2604 static int
2599 2605 zone_set_phys_mcap(zone_t *zone, const uint64_t *zone_mcap)
2600 2606 {
2601 2607 uint64_t mcap;
2602 2608 int err = 0;
2603 2609
2604 2610 if ((err = copyin(zone_mcap, &mcap, sizeof (uint64_t))) == 0)
2605 2611 zone->zone_phys_mcap = mcap;
2606 2612
2607 2613 return (err);
2608 2614 }
2609 2615
2610 2616 static int
2611 2617 zone_set_sched_class(zone_t *zone, const char *new_class)
2612 2618 {
2613 2619 char sched_class[PC_CLNMSZ];
2614 2620 id_t classid;
2615 2621 int err;
2616 2622
2617 2623 ASSERT(zone != global_zone);
2618 2624 if ((err = copyinstr(new_class, sched_class, PC_CLNMSZ, NULL)) != 0)
2619 2625 return (err); /* EFAULT or ENAMETOOLONG */
2620 2626
2621 2627 if (getcid(sched_class, &classid) != 0 || CLASS_KERNEL(classid))
2622 2628 return (set_errno(EINVAL));
2623 2629 zone->zone_defaultcid = classid;
2624 2630 ASSERT(zone->zone_defaultcid > 0 &&
2625 2631 zone->zone_defaultcid < loaded_classes);
2626 2632
2627 2633 return (0);
2628 2634 }
2629 2635
2630 2636 /*
2631 2637 * Block indefinitely waiting for (zone_status >= status)
2632 2638 */
2633 2639 void
2634 2640 zone_status_wait(zone_t *zone, zone_status_t status)
2635 2641 {
2636 2642 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2637 2643
2638 2644 mutex_enter(&zone_status_lock);
2639 2645 while (zone->zone_status < status) {
2640 2646 cv_wait(&zone->zone_cv, &zone_status_lock);
2641 2647 }
2642 2648 mutex_exit(&zone_status_lock);
2643 2649 }
2644 2650
2645 2651 /*
2646 2652 * Private CPR-safe version of zone_status_wait().
2647 2653 */
2648 2654 static void
2649 2655 zone_status_wait_cpr(zone_t *zone, zone_status_t status, char *str)
2650 2656 {
2651 2657 callb_cpr_t cprinfo;
2652 2658
2653 2659 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2654 2660
2655 2661 CALLB_CPR_INIT(&cprinfo, &zone_status_lock, callb_generic_cpr,
2656 2662 str);
2657 2663 mutex_enter(&zone_status_lock);
2658 2664 while (zone->zone_status < status) {
2659 2665 CALLB_CPR_SAFE_BEGIN(&cprinfo);
2660 2666 cv_wait(&zone->zone_cv, &zone_status_lock);
2661 2667 CALLB_CPR_SAFE_END(&cprinfo, &zone_status_lock);
2662 2668 }
2663 2669 /*
2664 2670 * zone_status_lock is implicitly released by the following.
2665 2671 */
2666 2672 CALLB_CPR_EXIT(&cprinfo);
2667 2673 }
2668 2674
2669 2675 /*
2670 2676 * Block until zone enters requested state or signal is received. Return (0)
2671 2677 * if signaled, non-zero otherwise.
2672 2678 */
2673 2679 int
2674 2680 zone_status_wait_sig(zone_t *zone, zone_status_t status)
2675 2681 {
2676 2682 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2677 2683
2678 2684 mutex_enter(&zone_status_lock);
2679 2685 while (zone->zone_status < status) {
2680 2686 if (!cv_wait_sig(&zone->zone_cv, &zone_status_lock)) {
2681 2687 mutex_exit(&zone_status_lock);
2682 2688 return (0);
2683 2689 }
2684 2690 }
2685 2691 mutex_exit(&zone_status_lock);
2686 2692 return (1);
2687 2693 }
2688 2694
2689 2695 /*
2690 2696 * Block until the zone enters the requested state or the timeout expires,
2691 2697 * whichever happens first. Return (-1) if operation timed out, time remaining
2692 2698 * otherwise.
2693 2699 */
2694 2700 clock_t
2695 2701 zone_status_timedwait(zone_t *zone, clock_t tim, zone_status_t status)
2696 2702 {
2697 2703 clock_t timeleft = 0;
2698 2704
2699 2705 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2700 2706
2701 2707 mutex_enter(&zone_status_lock);
2702 2708 while (zone->zone_status < status && timeleft != -1) {
2703 2709 timeleft = cv_timedwait(&zone->zone_cv, &zone_status_lock, tim);
2704 2710 }
2705 2711 mutex_exit(&zone_status_lock);
2706 2712 return (timeleft);
2707 2713 }
2708 2714
2709 2715 /*
2710 2716 * Block until the zone enters the requested state, the current process is
2711 2717 * signaled, or the timeout expires, whichever happens first. Return (-1) if
2712 2718 * operation timed out, 0 if signaled, time remaining otherwise.
2713 2719 */
2714 2720 clock_t
2715 2721 zone_status_timedwait_sig(zone_t *zone, clock_t tim, zone_status_t status)
2716 2722 {
2717 2723 clock_t timeleft = tim - ddi_get_lbolt();
2718 2724
2719 2725 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE);
2720 2726
2721 2727 mutex_enter(&zone_status_lock);
2722 2728 while (zone->zone_status < status) {
2723 2729 timeleft = cv_timedwait_sig(&zone->zone_cv, &zone_status_lock,
2724 2730 tim);
2725 2731 if (timeleft <= 0)
2726 2732 break;
2727 2733 }
2728 2734 mutex_exit(&zone_status_lock);
2729 2735 return (timeleft);
2730 2736 }
2731 2737
2732 2738 /*
2733 2739 * Zones have two reference counts: one for references from credential
2734 2740 * structures (zone_cred_ref), and one (zone_ref) for everything else.
2735 2741 * This is so we can allow a zone to be rebooted while there are still
2736 2742 * outstanding cred references, since certain drivers cache dblks (which
2737 2743 * implicitly results in cached creds). We wait for zone_ref to drop to
2738 2744 * 0 (actually 1), but not zone_cred_ref. The zone structure itself is
2739 2745 * later freed when the zone_cred_ref drops to 0, though nothing other
2740 2746 * than the zone id and privilege set should be accessed once the zone
2741 2747 * is "dead".
2742 2748 *
2743 2749 * A debugging flag, zone_wait_for_cred, can be set to a non-zero value
2744 2750 * to force halt/reboot to block waiting for the zone_cred_ref to drop
2745 2751 * to 0. This can be useful to flush out other sources of cached creds
2746 2752 * that may be less innocuous than the driver case.
2747 2753 *
2748 2754 * Zones also provide a tracked reference counting mechanism in which zone
2749 2755 * references are represented by "crumbs" (zone_ref structures). Crumbs help
2750 2756 * debuggers determine the sources of leaked zone references. See
2751 2757 * zone_hold_ref() and zone_rele_ref() below for more information.
2752 2758 */
2753 2759
2754 2760 int zone_wait_for_cred = 0;
2755 2761
2756 2762 static void
2757 2763 zone_hold_locked(zone_t *z)
2758 2764 {
2759 2765 ASSERT(MUTEX_HELD(&z->zone_lock));
2760 2766 z->zone_ref++;
2761 2767 ASSERT(z->zone_ref != 0);
2762 2768 }
2763 2769
2764 2770 /*
2765 2771 * Increment the specified zone's reference count. The zone's zone_t structure
2766 2772 * will not be freed as long as the zone's reference count is nonzero.
2767 2773 * Decrement the zone's reference count via zone_rele().
2768 2774 *
2769 2775 * NOTE: This function should only be used to hold zones for short periods of
2770 2776 * time. Use zone_hold_ref() if the zone must be held for a long time.
2771 2777 */
2772 2778 void
2773 2779 zone_hold(zone_t *z)
2774 2780 {
2775 2781 mutex_enter(&z->zone_lock);
2776 2782 zone_hold_locked(z);
2777 2783 mutex_exit(&z->zone_lock);
2778 2784 }
2779 2785
2780 2786 /*
2781 2787 * If the non-cred ref count drops to 1 and either the cred ref count
2782 2788 * is 0 or we aren't waiting for cred references, the zone is ready to
2783 2789 * be destroyed.
2784 2790 */
2785 2791 #define ZONE_IS_UNREF(zone) ((zone)->zone_ref == 1 && \
2786 2792 (!zone_wait_for_cred || (zone)->zone_cred_ref == 0))
2787 2793
2788 2794 /*
2789 2795 * Common zone reference release function invoked by zone_rele() and
2790 2796 * zone_rele_ref(). If subsys is ZONE_REF_NUM_SUBSYS, then the specified
2791 2797 * zone's subsystem-specific reference counters are not affected by the
2792 2798 * release. If ref is not NULL, then the zone_ref_t to which it refers is
2793 2799 * removed from the specified zone's reference list. ref must be non-NULL iff
2794 2800 * subsys is not ZONE_REF_NUM_SUBSYS.
2795 2801 */
2796 2802 static void
2797 2803 zone_rele_common(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
2798 2804 {
2799 2805 boolean_t wakeup;
2800 2806
2801 2807 mutex_enter(&z->zone_lock);
2802 2808 ASSERT(z->zone_ref != 0);
2803 2809 z->zone_ref--;
2804 2810 if (subsys != ZONE_REF_NUM_SUBSYS) {
2805 2811 ASSERT(z->zone_subsys_ref[subsys] != 0);
2806 2812 z->zone_subsys_ref[subsys]--;
2807 2813 list_remove(&z->zone_ref_list, ref);
2808 2814 }
2809 2815 if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
2810 2816 /* no more refs, free the structure */
2811 2817 mutex_exit(&z->zone_lock);
2812 2818 zone_free(z);
2813 2819 return;
2814 2820 }
2815 2821 /* signal zone_destroy so the zone can finish halting */
2816 2822 wakeup = (ZONE_IS_UNREF(z) && zone_status_get(z) >= ZONE_IS_DEAD);
2817 2823 mutex_exit(&z->zone_lock);
2818 2824
2819 2825 if (wakeup) {
2820 2826 /*
2821 2827 * Grabbing zonehash_lock here effectively synchronizes with
2822 2828 * zone_destroy() to avoid missed signals.
2823 2829 */
2824 2830 mutex_enter(&zonehash_lock);
2825 2831 cv_broadcast(&zone_destroy_cv);
2826 2832 mutex_exit(&zonehash_lock);
2827 2833 }
2828 2834 }
2829 2835
2830 2836 /*
2831 2837 * Decrement the specified zone's reference count. The specified zone will
2832 2838 * cease to exist after this function returns if the reference count drops to
2833 2839 * zero. This function should be paired with zone_hold().
2834 2840 */
2835 2841 void
2836 2842 zone_rele(zone_t *z)
2837 2843 {
2838 2844 zone_rele_common(z, NULL, ZONE_REF_NUM_SUBSYS);
2839 2845 }
2840 2846
2841 2847 /*
2842 2848 * Initialize a zone reference structure. This function must be invoked for
2843 2849 * a reference structure before the structure is passed to zone_hold_ref().
2844 2850 */
2845 2851 void
2846 2852 zone_init_ref(zone_ref_t *ref)
2847 2853 {
2848 2854 ref->zref_zone = NULL;
2849 2855 list_link_init(&ref->zref_linkage);
2850 2856 }
2851 2857
2852 2858 /*
2853 2859 * Acquire a reference to zone z. The caller must specify the
2854 2860 * zone_ref_subsys_t constant associated with its subsystem. The specified
2855 2861 * zone_ref_t structure will represent a reference to the specified zone. Use
2856 2862 * zone_rele_ref() to release the reference.
2857 2863 *
2858 2864 * The referenced zone_t structure will not be freed as long as the zone_t's
2859 2865 * zone_status field is not ZONE_IS_DEAD and the zone has outstanding
2860 2866 * references.
2861 2867 *
2862 2868 * NOTE: The zone_ref_t structure must be initialized before it is used.
2863 2869 * See zone_init_ref() above.
2864 2870 */
2865 2871 void
2866 2872 zone_hold_ref(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys)
2867 2873 {
2868 2874 ASSERT(subsys >= 0 && subsys < ZONE_REF_NUM_SUBSYS);
2869 2875
2870 2876 /*
2871 2877 * Prevent consumers from reusing a reference structure before
2872 2878 * releasing it.
2873 2879 */
2874 2880 VERIFY(ref->zref_zone == NULL);
2875 2881
2876 2882 ref->zref_zone = z;
2877 2883 mutex_enter(&z->zone_lock);
2878 2884 zone_hold_locked(z);
2879 2885 z->zone_subsys_ref[subsys]++;
2880 2886 ASSERT(z->zone_subsys_ref[subsys] != 0);
2881 2887 list_insert_head(&z->zone_ref_list, ref);
2882 2888 mutex_exit(&z->zone_lock);
2883 2889 }
2884 2890
2885 2891 /*
2886 2892 * Release the zone reference represented by the specified zone_ref_t.
2887 2893 * The reference is invalid after it's released; however, the zone_ref_t
2888 2894 * structure can be reused without having to invoke zone_init_ref().
2889 2895 * subsys should be the same value that was passed to zone_hold_ref()
2890 2896 * when the reference was acquired.
2891 2897 */
2892 2898 void
2893 2899 zone_rele_ref(zone_ref_t *ref, zone_ref_subsys_t subsys)
2894 2900 {
2895 2901 zone_rele_common(ref->zref_zone, ref, subsys);
2896 2902
2897 2903 /*
2898 2904 * Set the zone_ref_t's zref_zone field to NULL to generate panics
2899 2905 * when consumers dereference the reference. This helps us catch
2900 2906 * consumers who use released references. Furthermore, this lets
2901 2907 * consumers reuse the zone_ref_t structure without having to
2902 2908 * invoke zone_init_ref().
2903 2909 */
2904 2910 ref->zref_zone = NULL;
2905 2911 }
2906 2912
2907 2913 void
2908 2914 zone_cred_hold(zone_t *z)
2909 2915 {
2910 2916 mutex_enter(&z->zone_lock);
2911 2917 z->zone_cred_ref++;
2912 2918 ASSERT(z->zone_cred_ref != 0);
2913 2919 mutex_exit(&z->zone_lock);
2914 2920 }
2915 2921
2916 2922 void
2917 2923 zone_cred_rele(zone_t *z)
2918 2924 {
2919 2925 boolean_t wakeup;
2920 2926
2921 2927 mutex_enter(&z->zone_lock);
2922 2928 ASSERT(z->zone_cred_ref != 0);
2923 2929 z->zone_cred_ref--;
2924 2930 if (z->zone_ref == 0 && z->zone_cred_ref == 0) {
2925 2931 /* no more refs, free the structure */
2926 2932 mutex_exit(&z->zone_lock);
2927 2933 zone_free(z);
2928 2934 return;
2929 2935 }
2930 2936 /*
2931 2937 * If zone_destroy is waiting for the cred references to drain
2932 2938 * out, and they have, signal it.
2933 2939 */
2934 2940 wakeup = (zone_wait_for_cred && ZONE_IS_UNREF(z) &&
2935 2941 zone_status_get(z) >= ZONE_IS_DEAD);
2936 2942 mutex_exit(&z->zone_lock);
2937 2943
2938 2944 if (wakeup) {
2939 2945 /*
2940 2946 * Grabbing zonehash_lock here effectively synchronizes with
2941 2947 * zone_destroy() to avoid missed signals.
2942 2948 */
2943 2949 mutex_enter(&zonehash_lock);
2944 2950 cv_broadcast(&zone_destroy_cv);
2945 2951 mutex_exit(&zonehash_lock);
2946 2952 }
2947 2953 }
2948 2954
2949 2955 void
2950 2956 zone_task_hold(zone_t *z)
2951 2957 {
2952 2958 mutex_enter(&z->zone_lock);
2953 2959 z->zone_ntasks++;
2954 2960 ASSERT(z->zone_ntasks != 0);
2955 2961 mutex_exit(&z->zone_lock);
2956 2962 }
2957 2963
2958 2964 void
2959 2965 zone_task_rele(zone_t *zone)
2960 2966 {
2961 2967 uint_t refcnt;
2962 2968
2963 2969 mutex_enter(&zone->zone_lock);
2964 2970 ASSERT(zone->zone_ntasks != 0);
2965 2971 refcnt = --zone->zone_ntasks;
2966 2972 if (refcnt > 1) { /* Common case */
2967 2973 mutex_exit(&zone->zone_lock);
2968 2974 return;
2969 2975 }
2970 2976 zone_hold_locked(zone); /* so we can use the zone_t later */
2971 2977 mutex_exit(&zone->zone_lock);
2972 2978 if (refcnt == 1) {
2973 2979 /*
2974 2980 * See if the zone is shutting down.
2975 2981 */
2976 2982 mutex_enter(&zone_status_lock);
2977 2983 if (zone_status_get(zone) != ZONE_IS_SHUTTING_DOWN) {
2978 2984 goto out;
2979 2985 }
2980 2986
2981 2987 /*
2982 2988 * Make sure the ntasks didn't change since we
2983 2989 * dropped zone_lock.
2984 2990 */
2985 2991 mutex_enter(&zone->zone_lock);
2986 2992 if (refcnt != zone->zone_ntasks) {
2987 2993 mutex_exit(&zone->zone_lock);
2988 2994 goto out;
2989 2995 }
2990 2996 mutex_exit(&zone->zone_lock);
2991 2997
2992 2998 /*
2993 2999 * No more user processes in the zone. The zone is empty.
2994 3000 */
2995 3001 zone_status_set(zone, ZONE_IS_EMPTY);
2996 3002 goto out;
2997 3003 }
2998 3004
2999 3005 ASSERT(refcnt == 0);
3000 3006 /*
3001 3007 * zsched has exited; the zone is dead.
3002 3008 */
3003 3009 zone->zone_zsched = NULL; /* paranoia */
3004 3010 mutex_enter(&zone_status_lock);
3005 3011 zone_status_set(zone, ZONE_IS_DEAD);
3006 3012 out:
3007 3013 mutex_exit(&zone_status_lock);
3008 3014 zone_rele(zone);
3009 3015 }
3010 3016
3011 3017 zoneid_t
3012 3018 getzoneid(void)
3013 3019 {
3014 3020 return (curproc->p_zone->zone_id);
3015 3021 }
3016 3022
3017 3023 /*
3018 3024 * Internal versions of zone_find_by_*(). These don't zone_hold() or
3019 3025 * check the validity of a zone's state.
3020 3026 */
3021 3027 static zone_t *
3022 3028 zone_find_all_by_id(zoneid_t zoneid)
3023 3029 {
3024 3030 mod_hash_val_t hv;
3025 3031 zone_t *zone = NULL;
3026 3032
3027 3033 ASSERT(MUTEX_HELD(&zonehash_lock));
3028 3034
3029 3035 if (mod_hash_find(zonehashbyid,
3030 3036 (mod_hash_key_t)(uintptr_t)zoneid, &hv) == 0)
3031 3037 zone = (zone_t *)hv;
3032 3038 return (zone);
3033 3039 }
3034 3040
3035 3041 static zone_t *
3036 3042 zone_find_all_by_label(const ts_label_t *label)
3037 3043 {
3038 3044 mod_hash_val_t hv;
3039 3045 zone_t *zone = NULL;
3040 3046
3041 3047 ASSERT(MUTEX_HELD(&zonehash_lock));
3042 3048
3043 3049 /*
3044 3050 * zonehashbylabel is not maintained for unlabeled systems
3045 3051 */
3046 3052 if (!is_system_labeled())
3047 3053 return (NULL);
3048 3054 if (mod_hash_find(zonehashbylabel, (mod_hash_key_t)label, &hv) == 0)
3049 3055 zone = (zone_t *)hv;
3050 3056 return (zone);
3051 3057 }
3052 3058
3053 3059 static zone_t *
3054 3060 zone_find_all_by_name(char *name)
3055 3061 {
3056 3062 mod_hash_val_t hv;
3057 3063 zone_t *zone = NULL;
3058 3064
3059 3065 ASSERT(MUTEX_HELD(&zonehash_lock));
3060 3066
3061 3067 if (mod_hash_find(zonehashbyname, (mod_hash_key_t)name, &hv) == 0)
3062 3068 zone = (zone_t *)hv;
3063 3069 return (zone);
3064 3070 }
3065 3071
3066 3072 /*
3067 3073 * Public interface for looking up a zone by zoneid. Only returns the zone if
3068 3074 * it is fully initialized, and has not yet begun the zone_destroy() sequence.
3069 3075 * Caller must call zone_rele() once it is done with the zone.
3070 3076 *
3071 3077 * The zone may begin the zone_destroy() sequence immediately after this
3072 3078 * function returns, but may be safely used until zone_rele() is called.
3073 3079 */
3074 3080 zone_t *
3075 3081 zone_find_by_id(zoneid_t zoneid)
3076 3082 {
3077 3083 zone_t *zone;
3078 3084 zone_status_t status;
3079 3085
3080 3086 mutex_enter(&zonehash_lock);
3081 3087 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
3082 3088 mutex_exit(&zonehash_lock);
3083 3089 return (NULL);
3084 3090 }
3085 3091 status = zone_status_get(zone);
3086 3092 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3087 3093 /*
3088 3094 * For all practical purposes the zone doesn't exist.
3089 3095 */
3090 3096 mutex_exit(&zonehash_lock);
3091 3097 return (NULL);
3092 3098 }
3093 3099 zone_hold(zone);
3094 3100 mutex_exit(&zonehash_lock);
3095 3101 return (zone);
3096 3102 }
3097 3103
3098 3104 /*
3099 3105 * Similar to zone_find_by_id, but using zone label as the key.
3100 3106 */
3101 3107 zone_t *
3102 3108 zone_find_by_label(const ts_label_t *label)
3103 3109 {
3104 3110 zone_t *zone;
3105 3111 zone_status_t status;
3106 3112
3107 3113 mutex_enter(&zonehash_lock);
3108 3114 if ((zone = zone_find_all_by_label(label)) == NULL) {
3109 3115 mutex_exit(&zonehash_lock);
3110 3116 return (NULL);
3111 3117 }
3112 3118
3113 3119 status = zone_status_get(zone);
3114 3120 if (status > ZONE_IS_DOWN) {
3115 3121 /*
3116 3122 * For all practical purposes the zone doesn't exist.
3117 3123 */
3118 3124 mutex_exit(&zonehash_lock);
3119 3125 return (NULL);
3120 3126 }
3121 3127 zone_hold(zone);
3122 3128 mutex_exit(&zonehash_lock);
3123 3129 return (zone);
3124 3130 }
3125 3131
3126 3132 /*
3127 3133 * Similar to zone_find_by_id, but using zone name as the key.
3128 3134 */
3129 3135 zone_t *
3130 3136 zone_find_by_name(char *name)
3131 3137 {
3132 3138 zone_t *zone;
3133 3139 zone_status_t status;
3134 3140
3135 3141 mutex_enter(&zonehash_lock);
3136 3142 if ((zone = zone_find_all_by_name(name)) == NULL) {
3137 3143 mutex_exit(&zonehash_lock);
3138 3144 return (NULL);
3139 3145 }
3140 3146 status = zone_status_get(zone);
3141 3147 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3142 3148 /*
3143 3149 * For all practical purposes the zone doesn't exist.
3144 3150 */
3145 3151 mutex_exit(&zonehash_lock);
3146 3152 return (NULL);
3147 3153 }
3148 3154 zone_hold(zone);
3149 3155 mutex_exit(&zonehash_lock);
3150 3156 return (zone);
3151 3157 }
3152 3158
3153 3159 /*
3154 3160 * Similar to zone_find_by_id(), using the path as a key. For instance,
3155 3161 * if there is a zone "foo" rooted at /foo/root, and the path argument
3156 3162 * is "/foo/root/proc", it will return the held zone_t corresponding to
3157 3163 * zone "foo".
3158 3164 *
3159 3165 * zone_find_by_path() always returns a non-NULL value, since at the
3160 3166 * very least every path will be contained in the global zone.
3161 3167 *
3162 3168 * As with the other zone_find_by_*() functions, the caller is
3163 3169 * responsible for zone_rele()ing the return value of this function.
3164 3170 */
3165 3171 zone_t *
3166 3172 zone_find_by_path(const char *path)
3167 3173 {
3168 3174 zone_t *zone;
3169 3175 zone_t *zret = NULL;
3170 3176 zone_status_t status;
3171 3177
3172 3178 if (path == NULL) {
3173 3179 /*
3174 3180 * Call from rootconf().
3175 3181 */
3176 3182 zone_hold(global_zone);
3177 3183 return (global_zone);
3178 3184 }
3179 3185 ASSERT(*path == '/');
3180 3186 mutex_enter(&zonehash_lock);
3181 3187 for (zone = list_head(&zone_active); zone != NULL;
3182 3188 zone = list_next(&zone_active, zone)) {
3183 3189 if (ZONE_PATH_VISIBLE(path, zone))
3184 3190 zret = zone;
3185 3191 }
3186 3192 ASSERT(zret != NULL);
3187 3193 status = zone_status_get(zret);
3188 3194 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) {
3189 3195 /*
3190 3196 * Zone practically doesn't exist.
3191 3197 */
3192 3198 zret = global_zone;
3193 3199 }
3194 3200 zone_hold(zret);
3195 3201 mutex_exit(&zonehash_lock);
|
↓ open down ↓ |
795 lines elided |
↑ open up ↑ |
3196 3202 return (zret);
3197 3203 }
3198 3204
3199 3205 /*
3200 3206 * Public interface for updating per-zone load averages. Called once per
3201 3207 * second.
3202 3208 *
3203 3209 * Based on loadavg_update(), genloadavg() and calcloadavg() from clock.c.
3204 3210 */
3205 3211 void
3206 -zone_loadavg_update()
3212 +zone_loadavg_update(void)
3207 3213 {
3208 3214 zone_t *zp;
3209 3215 zone_status_t status;
3210 3216 struct loadavg_s *lavg;
3211 3217 hrtime_t zone_total;
3218 + uint64_t tmp;
3212 3219 int i;
3213 3220 hrtime_t hr_avg;
3214 3221 int nrun;
3215 3222 static int64_t f[3] = { 135, 27, 9 };
3216 3223 int64_t q, r;
3217 3224
3218 3225 mutex_enter(&zonehash_lock);
3219 3226 for (zp = list_head(&zone_active); zp != NULL;
3220 3227 zp = list_next(&zone_active, zp)) {
3221 3228 mutex_enter(&zp->zone_lock);
3222 3229
3223 3230 /* Skip zones that are on the way down or not yet up */
3224 3231 status = zone_status_get(zp);
3225 3232 if (status < ZONE_IS_READY || status >= ZONE_IS_DOWN) {
|
↓ open down ↓ |
4 lines elided |
↑ open up ↑ |
3226 3233 /* For all practical purposes the zone doesn't exist. */
3227 3234 mutex_exit(&zp->zone_lock);
3228 3235 continue;
3229 3236 }
3230 3237
3231 3238 /*
3232 3239 * Update the 10 second moving average data in zone_loadavg.
3233 3240 */
3234 3241 lavg = &zp->zone_loadavg;
3235 3242
3236 - zone_total = zp->zone_utime + zp->zone_stime + zp->zone_wtime;
3243 + tmp = cpu_uarray_sum_all(zp->zone_ustate);
3244 + zone_total = UINT64_OVERFLOW_TO_INT64(tmp);
3245 +
3237 3246 scalehrtime(&zone_total);
3238 3247
3239 3248 /* The zone_total should always be increasing. */
3240 3249 lavg->lg_loads[lavg->lg_cur] = (zone_total > lavg->lg_total) ?
3241 3250 zone_total - lavg->lg_total : 0;
3242 3251 lavg->lg_cur = (lavg->lg_cur + 1) % S_LOADAVG_SZ;
3243 3252 /* lg_total holds the prev. 1 sec. total */
3244 3253 lavg->lg_total = zone_total;
3245 3254
3246 3255 /*
3247 3256 * To simplify the calculation, we don't calculate the load avg.
3248 3257 * until the zone has been up for at least 10 seconds and our
3249 3258 * moving average is thus full.
3250 3259 */
3251 3260 if ((lavg->lg_len + 1) < S_LOADAVG_SZ) {
3252 3261 lavg->lg_len++;
3253 3262 mutex_exit(&zp->zone_lock);
3254 3263 continue;
3255 3264 }
3256 3265
3257 3266 /* Now calculate the 1min, 5min, 15 min load avg. */
3258 3267 hr_avg = 0;
3259 3268 for (i = 0; i < S_LOADAVG_SZ; i++)
3260 3269 hr_avg += lavg->lg_loads[i];
3261 3270 hr_avg = hr_avg / S_LOADAVG_SZ;
3262 3271 nrun = hr_avg / (NANOSEC / LGRP_LOADAVG_IN_THREAD_MAX);
3263 3272
3264 3273 /* Compute load avg. See comment in calcloadavg() */
3265 3274 for (i = 0; i < 3; i++) {
3266 3275 q = (zp->zone_hp_avenrun[i] >> 16) << 7;
3267 3276 r = (zp->zone_hp_avenrun[i] & 0xffff) << 7;
3268 3277 zp->zone_hp_avenrun[i] +=
3269 3278 ((nrun - q) * f[i] - ((r * f[i]) >> 16)) >> 4;
3270 3279
3271 3280 /* avenrun[] can only hold 31 bits of load avg. */
3272 3281 if (zp->zone_hp_avenrun[i] <
3273 3282 ((uint64_t)1<<(31+16-FSHIFT)))
3274 3283 zp->zone_avenrun[i] = (int32_t)
3275 3284 (zp->zone_hp_avenrun[i] >> (16 - FSHIFT));
3276 3285 else
3277 3286 zp->zone_avenrun[i] = 0x7fffffff;
3278 3287 }
3279 3288
3280 3289 mutex_exit(&zp->zone_lock);
3281 3290 }
3282 3291 mutex_exit(&zonehash_lock);
3283 3292 }
3284 3293
3285 3294 /*
3286 3295 * Get the number of cpus visible to this zone. The system-wide global
3287 3296 * 'ncpus' is returned if pools are disabled, the caller is in the
3288 3297 * global zone, or a NULL zone argument is passed in.
3289 3298 */
3290 3299 int
3291 3300 zone_ncpus_get(zone_t *zone)
3292 3301 {
3293 3302 int myncpus = zone == NULL ? 0 : zone->zone_ncpus;
3294 3303
3295 3304 return (myncpus != 0 ? myncpus : ncpus);
3296 3305 }
3297 3306
3298 3307 /*
3299 3308 * Get the number of online cpus visible to this zone. The system-wide
3300 3309 * global 'ncpus_online' is returned if pools are disabled, the caller
3301 3310 * is in the global zone, or a NULL zone argument is passed in.
3302 3311 */
3303 3312 int
3304 3313 zone_ncpus_online_get(zone_t *zone)
3305 3314 {
3306 3315 int myncpus_online = zone == NULL ? 0 : zone->zone_ncpus_online;
3307 3316
3308 3317 return (myncpus_online != 0 ? myncpus_online : ncpus_online);
3309 3318 }
3310 3319
3311 3320 /*
3312 3321 * Return the pool to which the zone is currently bound.
3313 3322 */
3314 3323 pool_t *
3315 3324 zone_pool_get(zone_t *zone)
3316 3325 {
3317 3326 ASSERT(pool_lock_held());
3318 3327
3319 3328 return (zone->zone_pool);
3320 3329 }
3321 3330
3322 3331 /*
3323 3332 * Set the zone's pool pointer and update the zone's visibility to match
3324 3333 * the resources in the new pool.
3325 3334 */
3326 3335 void
3327 3336 zone_pool_set(zone_t *zone, pool_t *pool)
3328 3337 {
3329 3338 ASSERT(pool_lock_held());
3330 3339 ASSERT(MUTEX_HELD(&cpu_lock));
3331 3340
3332 3341 zone->zone_pool = pool;
3333 3342 zone_pset_set(zone, pool->pool_pset->pset_id);
3334 3343 }
3335 3344
3336 3345 /*
3337 3346 * Return the cached value of the id of the processor set to which the
3338 3347 * zone is currently bound. The value will be ZONE_PS_INVAL if the pools
3339 3348 * facility is disabled.
3340 3349 */
3341 3350 psetid_t
3342 3351 zone_pset_get(zone_t *zone)
3343 3352 {
3344 3353 ASSERT(MUTEX_HELD(&cpu_lock));
3345 3354
3346 3355 return (zone->zone_psetid);
3347 3356 }
3348 3357
3349 3358 /*
3350 3359 * Set the cached value of the id of the processor set to which the zone
3351 3360 * is currently bound. Also update the zone's visibility to match the
3352 3361 * resources in the new processor set.
3353 3362 */
3354 3363 void
3355 3364 zone_pset_set(zone_t *zone, psetid_t newpsetid)
3356 3365 {
3357 3366 psetid_t oldpsetid;
3358 3367
3359 3368 ASSERT(MUTEX_HELD(&cpu_lock));
3360 3369 oldpsetid = zone_pset_get(zone);
3361 3370
3362 3371 if (oldpsetid == newpsetid)
3363 3372 return;
3364 3373 /*
3365 3374 * Global zone sees all.
3366 3375 */
3367 3376 if (zone != global_zone) {
3368 3377 zone->zone_psetid = newpsetid;
3369 3378 if (newpsetid != ZONE_PS_INVAL)
3370 3379 pool_pset_visibility_add(newpsetid, zone);
3371 3380 if (oldpsetid != ZONE_PS_INVAL)
3372 3381 pool_pset_visibility_remove(oldpsetid, zone);
3373 3382 }
3374 3383 /*
3375 3384 * Disabling pools, so we should start using the global values
3376 3385 * for ncpus and ncpus_online.
3377 3386 */
3378 3387 if (newpsetid == ZONE_PS_INVAL) {
3379 3388 zone->zone_ncpus = 0;
3380 3389 zone->zone_ncpus_online = 0;
3381 3390 }
3382 3391 }
3383 3392
3384 3393 /*
3385 3394 * Walk the list of active zones and issue the provided callback for
3386 3395 * each of them.
3387 3396 *
3388 3397 * Caller must not be holding any locks that may be acquired under
3389 3398 * zonehash_lock. See comment at the beginning of the file for a list of
3390 3399 * common locks and their interactions with zones.
3391 3400 */
3392 3401 int
3393 3402 zone_walk(int (*cb)(zone_t *, void *), void *data)
3394 3403 {
3395 3404 zone_t *zone;
3396 3405 int ret = 0;
3397 3406 zone_status_t status;
3398 3407
3399 3408 mutex_enter(&zonehash_lock);
3400 3409 for (zone = list_head(&zone_active); zone != NULL;
3401 3410 zone = list_next(&zone_active, zone)) {
3402 3411 /*
3403 3412 * Skip zones that shouldn't be externally visible.
3404 3413 */
3405 3414 status = zone_status_get(zone);
3406 3415 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN)
3407 3416 continue;
3408 3417 /*
3409 3418 * Bail immediately if any callback invocation returns a
3410 3419 * non-zero value.
3411 3420 */
3412 3421 ret = (*cb)(zone, data);
3413 3422 if (ret != 0)
3414 3423 break;
3415 3424 }
3416 3425 mutex_exit(&zonehash_lock);
3417 3426 return (ret);
3418 3427 }
3419 3428
3420 3429 static int
3421 3430 zone_set_root(zone_t *zone, const char *upath)
3422 3431 {
3423 3432 vnode_t *vp;
3424 3433 int trycount;
3425 3434 int error = 0;
3426 3435 char *path;
3427 3436 struct pathname upn, pn;
3428 3437 size_t pathlen;
3429 3438
3430 3439 if ((error = pn_get((char *)upath, UIO_USERSPACE, &upn)) != 0)
3431 3440 return (error);
3432 3441
3433 3442 pn_alloc(&pn);
3434 3443
3435 3444 /* prevent infinite loop */
3436 3445 trycount = 10;
3437 3446 for (;;) {
3438 3447 if (--trycount <= 0) {
3439 3448 error = ESTALE;
3440 3449 goto out;
3441 3450 }
3442 3451
3443 3452 if ((error = lookuppn(&upn, &pn, FOLLOW, NULLVPP, &vp)) == 0) {
3444 3453 /*
3445 3454 * VOP_ACCESS() may cover 'vp' with a new
3446 3455 * filesystem, if 'vp' is an autoFS vnode.
3447 3456 * Get the new 'vp' if so.
3448 3457 */
3449 3458 if ((error =
3450 3459 VOP_ACCESS(vp, VEXEC, 0, CRED(), NULL)) == 0 &&
3451 3460 (!vn_ismntpt(vp) ||
3452 3461 (error = traverse(&vp)) == 0)) {
3453 3462 pathlen = pn.pn_pathlen + 2;
3454 3463 path = kmem_alloc(pathlen, KM_SLEEP);
3455 3464 (void) strncpy(path, pn.pn_path,
3456 3465 pn.pn_pathlen + 1);
3457 3466 path[pathlen - 2] = '/';
3458 3467 path[pathlen - 1] = '\0';
3459 3468 pn_free(&pn);
3460 3469 pn_free(&upn);
3461 3470
3462 3471 /* Success! */
3463 3472 break;
3464 3473 }
3465 3474 VN_RELE(vp);
3466 3475 }
3467 3476 if (error != ESTALE)
3468 3477 goto out;
3469 3478 }
3470 3479
3471 3480 ASSERT(error == 0);
3472 3481 zone->zone_rootvp = vp; /* we hold a reference to vp */
3473 3482 zone->zone_rootpath = path;
3474 3483 zone->zone_rootpathlen = pathlen;
3475 3484 if (pathlen > 5 && strcmp(path + pathlen - 5, "/lu/") == 0)
3476 3485 zone->zone_flags |= ZF_IS_SCRATCH;
3477 3486 return (0);
3478 3487
3479 3488 out:
3480 3489 pn_free(&pn);
3481 3490 pn_free(&upn);
3482 3491 return (error);
3483 3492 }
3484 3493
3485 3494 #define isalnum(c) (((c) >= '0' && (c) <= '9') || \
3486 3495 ((c) >= 'a' && (c) <= 'z') || \
3487 3496 ((c) >= 'A' && (c) <= 'Z'))
3488 3497
3489 3498 static int
3490 3499 zone_set_name(zone_t *zone, const char *uname)
3491 3500 {
3492 3501 char *kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
3493 3502 size_t len;
3494 3503 int i, err;
3495 3504
3496 3505 if ((err = copyinstr(uname, kname, ZONENAME_MAX, &len)) != 0) {
3497 3506 kmem_free(kname, ZONENAME_MAX);
3498 3507 return (err); /* EFAULT or ENAMETOOLONG */
3499 3508 }
3500 3509
3501 3510 /* must be less than ZONENAME_MAX */
3502 3511 if (len == ZONENAME_MAX && kname[ZONENAME_MAX - 1] != '\0') {
3503 3512 kmem_free(kname, ZONENAME_MAX);
3504 3513 return (EINVAL);
3505 3514 }
3506 3515
3507 3516 /*
3508 3517 * Name must start with an alphanumeric and must contain only
3509 3518 * alphanumerics, '-', '_' and '.'.
3510 3519 */
3511 3520 if (!isalnum(kname[0])) {
3512 3521 kmem_free(kname, ZONENAME_MAX);
3513 3522 return (EINVAL);
3514 3523 }
3515 3524 for (i = 1; i < len - 1; i++) {
3516 3525 if (!isalnum(kname[i]) && kname[i] != '-' && kname[i] != '_' &&
3517 3526 kname[i] != '.') {
3518 3527 kmem_free(kname, ZONENAME_MAX);
3519 3528 return (EINVAL);
3520 3529 }
3521 3530 }
3522 3531
3523 3532 zone->zone_name = kname;
3524 3533 return (0);
3525 3534 }
3526 3535
3527 3536 /*
3528 3537 * Gets the 32-bit hostid of the specified zone as an unsigned int. If 'zonep'
3529 3538 * is NULL or it points to a zone with no hostid emulation, then the machine's
3530 3539 * hostid (i.e., the global zone's hostid) is returned. This function returns
3531 3540 * zero if neither the zone nor the host machine (global zone) have hostids. It
3532 3541 * returns HW_INVALID_HOSTID if the function attempts to return the machine's
3533 3542 * hostid and the machine's hostid is invalid.
3534 3543 */
3535 3544 uint32_t
3536 3545 zone_get_hostid(zone_t *zonep)
3537 3546 {
3538 3547 unsigned long machine_hostid;
3539 3548
3540 3549 if (zonep == NULL || zonep->zone_hostid == HW_INVALID_HOSTID) {
3541 3550 if (ddi_strtoul(hw_serial, NULL, 10, &machine_hostid) != 0)
3542 3551 return (HW_INVALID_HOSTID);
3543 3552 return ((uint32_t)machine_hostid);
3544 3553 }
3545 3554 return (zonep->zone_hostid);
3546 3555 }
3547 3556
3548 3557 /*
3549 3558 * Similar to thread_create(), but makes sure the thread is in the appropriate
3550 3559 * zone's zsched process (curproc->p_zone->zone_zsched) before returning.
3551 3560 */
3552 3561 /*ARGSUSED*/
3553 3562 kthread_t *
3554 3563 zthread_create(
3555 3564 caddr_t stk,
3556 3565 size_t stksize,
3557 3566 void (*proc)(),
3558 3567 void *arg,
3559 3568 size_t len,
3560 3569 pri_t pri)
3561 3570 {
3562 3571 kthread_t *t;
3563 3572 zone_t *zone = curproc->p_zone;
3564 3573 proc_t *pp = zone->zone_zsched;
3565 3574
3566 3575 zone_hold(zone); /* Reference to be dropped when thread exits */
3567 3576
3568 3577 /*
3569 3578 * No-one should be trying to create threads if the zone is shutting
3570 3579 * down and there aren't any kernel threads around. See comment
3571 3580 * in zthread_exit().
3572 3581 */
3573 3582 ASSERT(!(zone->zone_kthreads == NULL &&
3574 3583 zone_status_get(zone) >= ZONE_IS_EMPTY));
3575 3584 /*
3576 3585 * Create a thread, but don't let it run until we've finished setting
3577 3586 * things up.
3578 3587 */
3579 3588 t = thread_create(stk, stksize, proc, arg, len, pp, TS_STOPPED, pri);
3580 3589 ASSERT(t->t_forw == NULL);
3581 3590 mutex_enter(&zone_status_lock);
3582 3591 if (zone->zone_kthreads == NULL) {
3583 3592 t->t_forw = t->t_back = t;
3584 3593 } else {
3585 3594 kthread_t *tx = zone->zone_kthreads;
3586 3595
3587 3596 t->t_forw = tx;
3588 3597 t->t_back = tx->t_back;
3589 3598 tx->t_back->t_forw = t;
3590 3599 tx->t_back = t;
3591 3600 }
3592 3601 zone->zone_kthreads = t;
3593 3602 mutex_exit(&zone_status_lock);
3594 3603
3595 3604 mutex_enter(&pp->p_lock);
3596 3605 t->t_proc_flag |= TP_ZTHREAD;
3597 3606 project_rele(t->t_proj);
3598 3607 t->t_proj = project_hold(pp->p_task->tk_proj);
3599 3608
3600 3609 /*
3601 3610 * Setup complete, let it run.
3602 3611 */
3603 3612 thread_lock(t);
3604 3613 t->t_schedflag |= TS_ALLSTART;
3605 3614 setrun_locked(t);
3606 3615 thread_unlock(t);
3607 3616
3608 3617 mutex_exit(&pp->p_lock);
3609 3618
3610 3619 return (t);
3611 3620 }
3612 3621
3613 3622 /*
3614 3623 * Similar to thread_exit(). Must be called by threads created via
3615 3624 * zthread_exit().
3616 3625 */
3617 3626 void
3618 3627 zthread_exit(void)
3619 3628 {
3620 3629 kthread_t *t = curthread;
3621 3630 proc_t *pp = curproc;
3622 3631 zone_t *zone = pp->p_zone;
3623 3632
3624 3633 mutex_enter(&zone_status_lock);
3625 3634
3626 3635 /*
3627 3636 * Reparent to p0
3628 3637 */
3629 3638 kpreempt_disable();
3630 3639 mutex_enter(&pp->p_lock);
3631 3640 t->t_proc_flag &= ~TP_ZTHREAD;
3632 3641 t->t_procp = &p0;
3633 3642 hat_thread_exit(t);
3634 3643 mutex_exit(&pp->p_lock);
3635 3644 kpreempt_enable();
3636 3645
3637 3646 if (t->t_back == t) {
3638 3647 ASSERT(t->t_forw == t);
3639 3648 /*
3640 3649 * If the zone is empty, once the thread count
3641 3650 * goes to zero no further kernel threads can be
3642 3651 * created. This is because if the creator is a process
3643 3652 * in the zone, then it must have exited before the zone
3644 3653 * state could be set to ZONE_IS_EMPTY.
3645 3654 * Otherwise, if the creator is a kernel thread in the
3646 3655 * zone, the thread count is non-zero.
3647 3656 *
3648 3657 * This really means that non-zone kernel threads should
3649 3658 * not create zone kernel threads.
3650 3659 */
3651 3660 zone->zone_kthreads = NULL;
3652 3661 if (zone_status_get(zone) == ZONE_IS_EMPTY) {
3653 3662 zone_status_set(zone, ZONE_IS_DOWN);
3654 3663 /*
3655 3664 * Remove any CPU caps on this zone.
3656 3665 */
3657 3666 cpucaps_zone_remove(zone);
3658 3667 }
3659 3668 } else {
3660 3669 t->t_forw->t_back = t->t_back;
3661 3670 t->t_back->t_forw = t->t_forw;
3662 3671 if (zone->zone_kthreads == t)
3663 3672 zone->zone_kthreads = t->t_forw;
3664 3673 }
3665 3674 mutex_exit(&zone_status_lock);
3666 3675 zone_rele(zone);
3667 3676 thread_exit();
3668 3677 /* NOTREACHED */
3669 3678 }
3670 3679
3671 3680 static void
3672 3681 zone_chdir(vnode_t *vp, vnode_t **vpp, proc_t *pp)
3673 3682 {
3674 3683 vnode_t *oldvp;
3675 3684
3676 3685 /* we're going to hold a reference here to the directory */
3677 3686 VN_HOLD(vp);
3678 3687
3679 3688 /* update abs cwd/root path see c2/audit.c */
3680 3689 if (AU_AUDITING())
3681 3690 audit_chdirec(vp, vpp);
3682 3691
3683 3692 mutex_enter(&pp->p_lock);
3684 3693 oldvp = *vpp;
3685 3694 *vpp = vp;
3686 3695 mutex_exit(&pp->p_lock);
3687 3696 if (oldvp != NULL)
3688 3697 VN_RELE(oldvp);
3689 3698 }
3690 3699
3691 3700 /*
3692 3701 * Convert an rctl value represented by an nvlist_t into an rctl_val_t.
3693 3702 */
3694 3703 static int
3695 3704 nvlist2rctlval(nvlist_t *nvl, rctl_val_t *rv)
3696 3705 {
3697 3706 nvpair_t *nvp = NULL;
3698 3707 boolean_t priv_set = B_FALSE;
3699 3708 boolean_t limit_set = B_FALSE;
3700 3709 boolean_t action_set = B_FALSE;
3701 3710
3702 3711 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
3703 3712 const char *name;
3704 3713 uint64_t ui64;
3705 3714
3706 3715 name = nvpair_name(nvp);
3707 3716 if (nvpair_type(nvp) != DATA_TYPE_UINT64)
3708 3717 return (EINVAL);
3709 3718 (void) nvpair_value_uint64(nvp, &ui64);
3710 3719 if (strcmp(name, "privilege") == 0) {
3711 3720 /*
3712 3721 * Currently only privileged values are allowed, but
3713 3722 * this may change in the future.
3714 3723 */
3715 3724 if (ui64 != RCPRIV_PRIVILEGED)
3716 3725 return (EINVAL);
3717 3726 rv->rcv_privilege = ui64;
3718 3727 priv_set = B_TRUE;
3719 3728 } else if (strcmp(name, "limit") == 0) {
3720 3729 rv->rcv_value = ui64;
3721 3730 limit_set = B_TRUE;
3722 3731 } else if (strcmp(name, "action") == 0) {
3723 3732 if (ui64 != RCTL_LOCAL_NOACTION &&
3724 3733 ui64 != RCTL_LOCAL_DENY)
3725 3734 return (EINVAL);
3726 3735 rv->rcv_flagaction = ui64;
3727 3736 action_set = B_TRUE;
3728 3737 } else {
3729 3738 return (EINVAL);
3730 3739 }
3731 3740 }
3732 3741
3733 3742 if (!(priv_set && limit_set && action_set))
3734 3743 return (EINVAL);
3735 3744 rv->rcv_action_signal = 0;
3736 3745 rv->rcv_action_recipient = NULL;
3737 3746 rv->rcv_action_recip_pid = -1;
3738 3747 rv->rcv_firing_time = 0;
3739 3748
3740 3749 return (0);
3741 3750 }
3742 3751
3743 3752 /*
3744 3753 * Non-global zone version of start_init.
3745 3754 */
3746 3755 void
3747 3756 zone_start_init(void)
3748 3757 {
3749 3758 proc_t *p = ttoproc(curthread);
3750 3759 zone_t *z = p->p_zone;
3751 3760
3752 3761 ASSERT(!INGLOBALZONE(curproc));
3753 3762
3754 3763 /*
3755 3764 * For all purposes (ZONE_ATTR_INITPID and restart_init),
3756 3765 * storing just the pid of init is sufficient.
3757 3766 */
3758 3767 z->zone_proc_initpid = p->p_pid;
3759 3768
3760 3769 /*
3761 3770 * We maintain zone_boot_err so that we can return the cause of the
3762 3771 * failure back to the caller of the zone_boot syscall.
3763 3772 */
3764 3773 p->p_zone->zone_boot_err = start_init_common();
3765 3774
3766 3775 /*
3767 3776 * We will prevent booting zones from becoming running zones if the
3768 3777 * global zone is shutting down.
3769 3778 */
3770 3779 mutex_enter(&zone_status_lock);
3771 3780 if (z->zone_boot_err != 0 || zone_status_get(global_zone) >=
3772 3781 ZONE_IS_SHUTTING_DOWN) {
3773 3782 /*
3774 3783 * Make sure we are still in the booting state-- we could have
3775 3784 * raced and already be shutting down, or even further along.
3776 3785 */
3777 3786 if (zone_status_get(z) == ZONE_IS_BOOTING) {
3778 3787 zone_status_set(z, ZONE_IS_SHUTTING_DOWN);
3779 3788 }
3780 3789 mutex_exit(&zone_status_lock);
3781 3790 /* It's gone bad, dispose of the process */
3782 3791 if (proc_exit(CLD_EXITED, z->zone_boot_err) != 0) {
3783 3792 mutex_enter(&p->p_lock);
3784 3793 ASSERT(p->p_flag & SEXITLWPS);
3785 3794 lwp_exit();
3786 3795 }
3787 3796 } else {
3788 3797 if (zone_status_get(z) == ZONE_IS_BOOTING)
3789 3798 zone_status_set(z, ZONE_IS_RUNNING);
3790 3799 mutex_exit(&zone_status_lock);
3791 3800 /* cause the process to return to userland. */
3792 3801 lwp_rtt();
3793 3802 }
3794 3803 }
3795 3804
3796 3805 struct zsched_arg {
3797 3806 zone_t *zone;
3798 3807 nvlist_t *nvlist;
3799 3808 };
3800 3809
3801 3810 /*
3802 3811 * Per-zone "sched" workalike. The similarity to "sched" doesn't have
3803 3812 * anything to do with scheduling, but rather with the fact that
3804 3813 * per-zone kernel threads are parented to zsched, just like regular
3805 3814 * kernel threads are parented to sched (p0).
3806 3815 *
3807 3816 * zsched is also responsible for launching init for the zone.
3808 3817 */
3809 3818 static void
3810 3819 zsched(void *arg)
3811 3820 {
3812 3821 struct zsched_arg *za = arg;
3813 3822 proc_t *pp = curproc;
3814 3823 proc_t *initp = proc_init;
3815 3824 zone_t *zone = za->zone;
3816 3825 cred_t *cr, *oldcred;
3817 3826 rctl_set_t *set;
3818 3827 rctl_alloc_gp_t *gp;
3819 3828 contract_t *ct = NULL;
3820 3829 task_t *tk, *oldtk;
3821 3830 rctl_entity_p_t e;
3822 3831 kproject_t *pj;
3823 3832
3824 3833 nvlist_t *nvl = za->nvlist;
3825 3834 nvpair_t *nvp = NULL;
3826 3835
3827 3836 bcopy("zsched", PTOU(pp)->u_psargs, sizeof ("zsched"));
3828 3837 bcopy("zsched", PTOU(pp)->u_comm, sizeof ("zsched"));
3829 3838 PTOU(pp)->u_argc = 0;
3830 3839 PTOU(pp)->u_argv = NULL;
3831 3840 PTOU(pp)->u_envp = NULL;
3832 3841 PTOU(pp)->u_commpagep = NULL;
3833 3842 closeall(P_FINFO(pp));
3834 3843
3835 3844 /*
3836 3845 * We are this zone's "zsched" process. As the zone isn't generally
3837 3846 * visible yet we don't need to grab any locks before initializing its
3838 3847 * zone_proc pointer.
3839 3848 */
3840 3849 zone_hold(zone); /* this hold is released by zone_destroy() */
3841 3850 zone->zone_zsched = pp;
3842 3851 mutex_enter(&pp->p_lock);
3843 3852 pp->p_zone = zone;
3844 3853 mutex_exit(&pp->p_lock);
3845 3854
3846 3855 /*
3847 3856 * Disassociate process from its 'parent'; parent ourselves to init
3848 3857 * (pid 1) and change other values as needed.
3849 3858 */
3850 3859 sess_create();
3851 3860
3852 3861 mutex_enter(&pidlock);
3853 3862 proc_detach(pp);
3854 3863 pp->p_ppid = 1;
3855 3864 pp->p_flag |= SZONETOP;
3856 3865 pp->p_ancpid = 1;
3857 3866 pp->p_parent = initp;
3858 3867 pp->p_psibling = NULL;
3859 3868 if (initp->p_child)
3860 3869 initp->p_child->p_psibling = pp;
3861 3870 pp->p_sibling = initp->p_child;
3862 3871 initp->p_child = pp;
3863 3872
3864 3873 /* Decrement what newproc() incremented. */
3865 3874 upcount_dec(crgetruid(CRED()), GLOBAL_ZONEID);
3866 3875 /*
3867 3876 * Our credentials are about to become kcred-like, so we don't care
3868 3877 * about the caller's ruid.
3869 3878 */
3870 3879 upcount_inc(crgetruid(kcred), zone->zone_id);
3871 3880 mutex_exit(&pidlock);
3872 3881
3873 3882 /*
3874 3883 * getting out of global zone, so decrement lwp and process counts
3875 3884 */
3876 3885 pj = pp->p_task->tk_proj;
3877 3886 mutex_enter(&global_zone->zone_nlwps_lock);
3878 3887 pj->kpj_nlwps -= pp->p_lwpcnt;
3879 3888 global_zone->zone_nlwps -= pp->p_lwpcnt;
3880 3889 pj->kpj_nprocs--;
3881 3890 global_zone->zone_nprocs--;
3882 3891 mutex_exit(&global_zone->zone_nlwps_lock);
3883 3892
3884 3893 /*
3885 3894 * Decrement locked memory counts on old zone and project.
3886 3895 */
3887 3896 mutex_enter(&global_zone->zone_mem_lock);
3888 3897 global_zone->zone_locked_mem -= pp->p_locked_mem;
3889 3898 pj->kpj_data.kpd_locked_mem -= pp->p_locked_mem;
3890 3899 mutex_exit(&global_zone->zone_mem_lock);
3891 3900
3892 3901 /*
3893 3902 * Create and join a new task in project '0' of this zone.
3894 3903 *
3895 3904 * We don't need to call holdlwps() since we know we're the only lwp in
3896 3905 * this process.
3897 3906 *
3898 3907 * task_join() returns with p_lock held.
3899 3908 */
3900 3909 tk = task_create(0, zone);
3901 3910 mutex_enter(&cpu_lock);
3902 3911 oldtk = task_join(tk, 0);
3903 3912
3904 3913 pj = pp->p_task->tk_proj;
3905 3914
3906 3915 mutex_enter(&zone->zone_mem_lock);
3907 3916 zone->zone_locked_mem += pp->p_locked_mem;
3908 3917 pj->kpj_data.kpd_locked_mem += pp->p_locked_mem;
3909 3918 mutex_exit(&zone->zone_mem_lock);
3910 3919
3911 3920 /*
3912 3921 * add lwp and process counts to zsched's zone, and increment
3913 3922 * project's task and process count due to the task created in
3914 3923 * the above task_create.
3915 3924 */
3916 3925 mutex_enter(&zone->zone_nlwps_lock);
3917 3926 pj->kpj_nlwps += pp->p_lwpcnt;
3918 3927 pj->kpj_ntasks += 1;
3919 3928 zone->zone_nlwps += pp->p_lwpcnt;
3920 3929 pj->kpj_nprocs++;
3921 3930 zone->zone_nprocs++;
3922 3931 mutex_exit(&zone->zone_nlwps_lock);
3923 3932
3924 3933 mutex_exit(&curproc->p_lock);
3925 3934 mutex_exit(&cpu_lock);
3926 3935 task_rele(oldtk);
3927 3936
3928 3937 /*
3929 3938 * The process was created by a process in the global zone, hence the
3930 3939 * credentials are wrong. We might as well have kcred-ish credentials.
3931 3940 */
3932 3941 cr = zone->zone_kcred;
3933 3942 crhold(cr);
3934 3943 mutex_enter(&pp->p_crlock);
3935 3944 oldcred = pp->p_cred;
3936 3945 pp->p_cred = cr;
3937 3946 mutex_exit(&pp->p_crlock);
3938 3947 crfree(oldcred);
3939 3948
3940 3949 /*
3941 3950 * Hold credentials again (for thread)
3942 3951 */
3943 3952 crhold(cr);
3944 3953
3945 3954 /*
3946 3955 * p_lwpcnt can't change since this is a kernel process.
3947 3956 */
3948 3957 crset(pp, cr);
3949 3958
3950 3959 /*
3951 3960 * Chroot
3952 3961 */
3953 3962 zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_cdir, pp);
3954 3963 zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_rdir, pp);
3955 3964
3956 3965 /*
3957 3966 * Initialize zone's rctl set.
3958 3967 */
3959 3968 set = rctl_set_create();
3960 3969 gp = rctl_set_init_prealloc(RCENTITY_ZONE);
3961 3970 mutex_enter(&pp->p_lock);
3962 3971 e.rcep_p.zone = zone;
3963 3972 e.rcep_t = RCENTITY_ZONE;
3964 3973 zone->zone_rctls = rctl_set_init(RCENTITY_ZONE, pp, &e, set, gp);
3965 3974 mutex_exit(&pp->p_lock);
3966 3975 rctl_prealloc_destroy(gp);
3967 3976
3968 3977 /*
3969 3978 * Apply the rctls passed in to zone_create(). This is basically a list
3970 3979 * assignment: all of the old values are removed and the new ones
3971 3980 * inserted. That is, if an empty list is passed in, all values are
3972 3981 * removed.
3973 3982 */
3974 3983 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
3975 3984 rctl_dict_entry_t *rde;
3976 3985 rctl_hndl_t hndl;
3977 3986 char *name;
3978 3987 nvlist_t **nvlarray;
3979 3988 uint_t i, nelem;
3980 3989 int error; /* For ASSERT()s */
3981 3990
3982 3991 name = nvpair_name(nvp);
3983 3992 hndl = rctl_hndl_lookup(name);
3984 3993 ASSERT(hndl != -1);
3985 3994 rde = rctl_dict_lookup_hndl(hndl);
3986 3995 ASSERT(rde != NULL);
3987 3996
3988 3997 for (; /* ever */; ) {
3989 3998 rctl_val_t oval;
3990 3999
3991 4000 mutex_enter(&pp->p_lock);
3992 4001 error = rctl_local_get(hndl, NULL, &oval, pp);
3993 4002 mutex_exit(&pp->p_lock);
3994 4003 ASSERT(error == 0); /* Can't fail for RCTL_FIRST */
3995 4004 ASSERT(oval.rcv_privilege != RCPRIV_BASIC);
3996 4005 if (oval.rcv_privilege == RCPRIV_SYSTEM)
3997 4006 break;
3998 4007 mutex_enter(&pp->p_lock);
3999 4008 error = rctl_local_delete(hndl, &oval, pp);
4000 4009 mutex_exit(&pp->p_lock);
4001 4010 ASSERT(error == 0);
4002 4011 }
4003 4012 error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
4004 4013 ASSERT(error == 0);
4005 4014 for (i = 0; i < nelem; i++) {
4006 4015 rctl_val_t *nvalp;
4007 4016
4008 4017 nvalp = kmem_cache_alloc(rctl_val_cache, KM_SLEEP);
4009 4018 error = nvlist2rctlval(nvlarray[i], nvalp);
4010 4019 ASSERT(error == 0);
4011 4020 /*
4012 4021 * rctl_local_insert can fail if the value being
4013 4022 * inserted is a duplicate; this is OK.
4014 4023 */
4015 4024 mutex_enter(&pp->p_lock);
4016 4025 if (rctl_local_insert(hndl, nvalp, pp) != 0)
4017 4026 kmem_cache_free(rctl_val_cache, nvalp);
4018 4027 mutex_exit(&pp->p_lock);
4019 4028 }
4020 4029 }
4021 4030
4022 4031 /*
4023 4032 * Tell the world that we're done setting up.
4024 4033 *
4025 4034 * At this point we want to set the zone status to ZONE_IS_INITIALIZED
4026 4035 * and atomically set the zone's processor set visibility. Once
4027 4036 * we drop pool_lock() this zone will automatically get updated
4028 4037 * to reflect any future changes to the pools configuration.
4029 4038 *
4030 4039 * Note that after we drop the locks below (zonehash_lock in
4031 4040 * particular) other operations such as a zone_getattr call can
4032 4041 * now proceed and observe the zone. That is the reason for doing a
4033 4042 * state transition to the INITIALIZED state.
4034 4043 */
4035 4044 pool_lock();
4036 4045 mutex_enter(&cpu_lock);
4037 4046 mutex_enter(&zonehash_lock);
4038 4047 zone_uniqid(zone);
4039 4048 zone_zsd_configure(zone);
4040 4049 if (pool_state == POOL_ENABLED)
4041 4050 zone_pset_set(zone, pool_default->pool_pset->pset_id);
4042 4051 mutex_enter(&zone_status_lock);
4043 4052 ASSERT(zone_status_get(zone) == ZONE_IS_UNINITIALIZED);
4044 4053 zone_status_set(zone, ZONE_IS_INITIALIZED);
4045 4054 mutex_exit(&zone_status_lock);
4046 4055 mutex_exit(&zonehash_lock);
4047 4056 mutex_exit(&cpu_lock);
4048 4057 pool_unlock();
4049 4058
4050 4059 /* Now call the create callback for this key */
4051 4060 zsd_apply_all_keys(zsd_apply_create, zone);
4052 4061
4053 4062 /* The callbacks are complete. Mark ZONE_IS_READY */
4054 4063 mutex_enter(&zone_status_lock);
4055 4064 ASSERT(zone_status_get(zone) == ZONE_IS_INITIALIZED);
4056 4065 zone_status_set(zone, ZONE_IS_READY);
4057 4066 mutex_exit(&zone_status_lock);
4058 4067
4059 4068 /*
4060 4069 * Once we see the zone transition to the ZONE_IS_BOOTING state,
4061 4070 * we launch init, and set the state to running.
4062 4071 */
4063 4072 zone_status_wait_cpr(zone, ZONE_IS_BOOTING, "zsched");
4064 4073
4065 4074 if (zone_status_get(zone) == ZONE_IS_BOOTING) {
4066 4075 id_t cid;
4067 4076
4068 4077 /*
4069 4078 * Ok, this is a little complicated. We need to grab the
4070 4079 * zone's pool's scheduling class ID; note that by now, we
4071 4080 * are already bound to a pool if we need to be (zoneadmd
4072 4081 * will have done that to us while we're in the READY
4073 4082 * state). *But* the scheduling class for the zone's 'init'
4074 4083 * must be explicitly passed to newproc, which doesn't
4075 4084 * respect pool bindings.
4076 4085 *
4077 4086 * We hold the pool_lock across the call to newproc() to
4078 4087 * close the obvious race: the pool's scheduling class
4079 4088 * could change before we manage to create the LWP with
4080 4089 * classid 'cid'.
4081 4090 */
4082 4091 pool_lock();
4083 4092 if (zone->zone_defaultcid > 0)
4084 4093 cid = zone->zone_defaultcid;
4085 4094 else
4086 4095 cid = pool_get_class(zone->zone_pool);
4087 4096 if (cid == -1)
4088 4097 cid = defaultcid;
4089 4098
4090 4099 /*
4091 4100 * If this fails, zone_boot will ultimately fail. The
4092 4101 * state of the zone will be set to SHUTTING_DOWN-- userland
4093 4102 * will have to tear down the zone, and fail, or try again.
4094 4103 */
4095 4104 if ((zone->zone_boot_err = newproc(zone_start_init, NULL, cid,
4096 4105 minclsyspri - 1, &ct, 0)) != 0) {
4097 4106 mutex_enter(&zone_status_lock);
4098 4107 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
4099 4108 mutex_exit(&zone_status_lock);
4100 4109 } else {
4101 4110 zone->zone_boot_time = gethrestime_sec();
4102 4111 }
4103 4112
4104 4113 pool_unlock();
4105 4114 }
4106 4115
4107 4116 /*
4108 4117 * Wait for zone_destroy() to be called. This is what we spend
4109 4118 * most of our life doing.
4110 4119 */
4111 4120 zone_status_wait_cpr(zone, ZONE_IS_DYING, "zsched");
4112 4121
4113 4122 if (ct)
4114 4123 /*
4115 4124 * At this point the process contract should be empty.
4116 4125 * (Though if it isn't, it's not the end of the world.)
4117 4126 */
4118 4127 VERIFY(contract_abandon(ct, curproc, B_TRUE) == 0);
4119 4128
4120 4129 /*
4121 4130 * Allow kcred to be freed when all referring processes
4122 4131 * (including this one) go away. We can't just do this in
4123 4132 * zone_free because we need to wait for the zone_cred_ref to
4124 4133 * drop to 0 before calling zone_free, and the existence of
4125 4134 * zone_kcred will prevent that. Thus, we call crfree here to
4126 4135 * balance the crdup in zone_create. The crhold calls earlier
4127 4136 * in zsched will be dropped when the thread and process exit.
4128 4137 */
4129 4138 crfree(zone->zone_kcred);
4130 4139 zone->zone_kcred = NULL;
4131 4140
4132 4141 exit(CLD_EXITED, 0);
4133 4142 }
4134 4143
4135 4144 /*
4136 4145 * Helper function to determine if there are any submounts of the
4137 4146 * provided path. Used to make sure the zone doesn't "inherit" any
4138 4147 * mounts from before it is created.
4139 4148 */
4140 4149 static uint_t
4141 4150 zone_mount_count(const char *rootpath)
4142 4151 {
4143 4152 vfs_t *vfsp;
4144 4153 uint_t count = 0;
4145 4154 size_t rootpathlen = strlen(rootpath);
4146 4155
4147 4156 /*
4148 4157 * Holding zonehash_lock prevents race conditions with
4149 4158 * vfs_list_add()/vfs_list_remove() since we serialize with
4150 4159 * zone_find_by_path().
4151 4160 */
4152 4161 ASSERT(MUTEX_HELD(&zonehash_lock));
4153 4162 /*
4154 4163 * The rootpath must end with a '/'
4155 4164 */
4156 4165 ASSERT(rootpath[rootpathlen - 1] == '/');
4157 4166
4158 4167 /*
4159 4168 * This intentionally does not count the rootpath itself if that
4160 4169 * happens to be a mount point.
4161 4170 */
4162 4171 vfs_list_read_lock();
4163 4172 vfsp = rootvfs;
4164 4173 do {
4165 4174 if (strncmp(rootpath, refstr_value(vfsp->vfs_mntpt),
4166 4175 rootpathlen) == 0)
4167 4176 count++;
4168 4177 vfsp = vfsp->vfs_next;
4169 4178 } while (vfsp != rootvfs);
4170 4179 vfs_list_unlock();
4171 4180 return (count);
4172 4181 }
4173 4182
4174 4183 /*
4175 4184 * Helper function to make sure that a zone created on 'rootpath'
4176 4185 * wouldn't end up containing other zones' rootpaths.
4177 4186 */
4178 4187 static boolean_t
4179 4188 zone_is_nested(const char *rootpath)
4180 4189 {
4181 4190 zone_t *zone;
4182 4191 size_t rootpathlen = strlen(rootpath);
4183 4192 size_t len;
4184 4193
4185 4194 ASSERT(MUTEX_HELD(&zonehash_lock));
4186 4195
4187 4196 /*
4188 4197 * zone_set_root() appended '/' and '\0' at the end of rootpath
4189 4198 */
4190 4199 if ((rootpathlen <= 3) && (rootpath[0] == '/') &&
4191 4200 (rootpath[1] == '/') && (rootpath[2] == '\0'))
4192 4201 return (B_TRUE);
4193 4202
4194 4203 for (zone = list_head(&zone_active); zone != NULL;
4195 4204 zone = list_next(&zone_active, zone)) {
4196 4205 if (zone == global_zone)
4197 4206 continue;
4198 4207 len = strlen(zone->zone_rootpath);
4199 4208 if (strncmp(rootpath, zone->zone_rootpath,
4200 4209 MIN(rootpathlen, len)) == 0)
4201 4210 return (B_TRUE);
4202 4211 }
4203 4212 return (B_FALSE);
4204 4213 }
4205 4214
4206 4215 static int
4207 4216 zone_set_privset(zone_t *zone, const priv_set_t *zone_privs,
4208 4217 size_t zone_privssz)
4209 4218 {
4210 4219 priv_set_t *privs;
4211 4220
4212 4221 if (zone_privssz < sizeof (priv_set_t))
4213 4222 return (ENOMEM);
4214 4223
4215 4224 privs = kmem_alloc(sizeof (priv_set_t), KM_SLEEP);
4216 4225
4217 4226 if (copyin(zone_privs, privs, sizeof (priv_set_t))) {
4218 4227 kmem_free(privs, sizeof (priv_set_t));
4219 4228 return (EFAULT);
4220 4229 }
4221 4230
4222 4231 zone->zone_privset = privs;
4223 4232 return (0);
4224 4233 }
|
↓ open down ↓ |
978 lines elided |
↑ open up ↑ |
4225 4234
4226 4235 /*
4227 4236 * We make creative use of nvlists to pass in rctls from userland. The list is
4228 4237 * a list of the following structures:
4229 4238 *
4230 4239 * (name = rctl_name, value = nvpair_list_array)
4231 4240 *
4232 4241 * Where each element of the nvpair_list_array is of the form:
4233 4242 *
4234 4243 * [(name = "privilege", value = RCPRIV_PRIVILEGED),
4235 - * (name = "limit", value = uint64_t),
4236 - * (name = "action", value = (RCTL_LOCAL_NOACTION || RCTL_LOCAL_DENY))]
4244 + * (name = "limit", value = uint64_t),
4245 + * (name = "action", value = (RCTL_LOCAL_NOACTION || RCTL_LOCAL_DENY))]
4237 4246 */
4238 4247 static int
4239 4248 parse_rctls(caddr_t ubuf, size_t buflen, nvlist_t **nvlp)
4240 4249 {
4241 4250 nvpair_t *nvp = NULL;
4242 4251 nvlist_t *nvl = NULL;
4243 4252 char *kbuf;
4244 4253 int error;
4245 4254 rctl_val_t rv;
4246 4255
4247 4256 *nvlp = NULL;
4248 4257
4249 4258 if (buflen == 0)
4250 4259 return (0);
4251 4260
4252 4261 if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL)
4253 4262 return (ENOMEM);
4254 4263 if (copyin(ubuf, kbuf, buflen)) {
4255 4264 error = EFAULT;
4256 4265 goto out;
4257 4266 }
4258 4267 if (nvlist_unpack(kbuf, buflen, &nvl, KM_SLEEP) != 0) {
4259 4268 /*
4260 4269 * nvl may have been allocated/free'd, but the value set to
4261 4270 * non-NULL, so we reset it here.
4262 4271 */
4263 4272 nvl = NULL;
4264 4273 error = EINVAL;
4265 4274 goto out;
4266 4275 }
4267 4276 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
4268 4277 rctl_dict_entry_t *rde;
4269 4278 rctl_hndl_t hndl;
4270 4279 nvlist_t **nvlarray;
4271 4280 uint_t i, nelem;
4272 4281 char *name;
4273 4282
4274 4283 error = EINVAL;
4275 4284 name = nvpair_name(nvp);
4276 4285 if (strncmp(nvpair_name(nvp), "zone.", sizeof ("zone.") - 1)
4277 4286 != 0 || nvpair_type(nvp) != DATA_TYPE_NVLIST_ARRAY) {
4278 4287 goto out;
4279 4288 }
4280 4289 if ((hndl = rctl_hndl_lookup(name)) == -1) {
4281 4290 goto out;
4282 4291 }
4283 4292 rde = rctl_dict_lookup_hndl(hndl);
4284 4293 error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem);
4285 4294 ASSERT(error == 0);
4286 4295 for (i = 0; i < nelem; i++) {
4287 4296 if (error = nvlist2rctlval(nvlarray[i], &rv))
4288 4297 goto out;
4289 4298 }
4290 4299 if (rctl_invalid_value(rde, &rv)) {
4291 4300 error = EINVAL;
4292 4301 goto out;
4293 4302 }
4294 4303 }
4295 4304 error = 0;
4296 4305 *nvlp = nvl;
4297 4306 out:
4298 4307 kmem_free(kbuf, buflen);
4299 4308 if (error && nvl != NULL)
4300 4309 nvlist_free(nvl);
4301 4310 return (error);
4302 4311 }
4303 4312
4304 4313 int
4305 4314 zone_create_error(int er_error, int er_ext, int *er_out)
4306 4315 {
4307 4316 if (er_out != NULL) {
4308 4317 if (copyout(&er_ext, er_out, sizeof (int))) {
4309 4318 return (set_errno(EFAULT));
4310 4319 }
4311 4320 }
4312 4321 return (set_errno(er_error));
4313 4322 }
4314 4323
4315 4324 static int
4316 4325 zone_set_label(zone_t *zone, const bslabel_t *lab, uint32_t doi)
4317 4326 {
4318 4327 ts_label_t *tsl;
4319 4328 bslabel_t blab;
4320 4329
4321 4330 /* Get label from user */
4322 4331 if (copyin(lab, &blab, sizeof (blab)) != 0)
4323 4332 return (EFAULT);
4324 4333 tsl = labelalloc(&blab, doi, KM_NOSLEEP);
4325 4334 if (tsl == NULL)
4326 4335 return (ENOMEM);
4327 4336
4328 4337 zone->zone_slabel = tsl;
4329 4338 return (0);
4330 4339 }
4331 4340
4332 4341 /*
4333 4342 * Parses a comma-separated list of ZFS datasets into a per-zone dictionary.
4334 4343 */
4335 4344 static int
4336 4345 parse_zfs(zone_t *zone, caddr_t ubuf, size_t buflen)
4337 4346 {
4338 4347 char *kbuf;
4339 4348 char *dataset, *next;
4340 4349 zone_dataset_t *zd;
4341 4350 size_t len;
4342 4351
4343 4352 if (ubuf == NULL || buflen == 0)
4344 4353 return (0);
4345 4354
4346 4355 if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL)
4347 4356 return (ENOMEM);
4348 4357
4349 4358 if (copyin(ubuf, kbuf, buflen) != 0) {
4350 4359 kmem_free(kbuf, buflen);
4351 4360 return (EFAULT);
4352 4361 }
4353 4362
4354 4363 dataset = next = kbuf;
4355 4364 for (;;) {
4356 4365 zd = kmem_alloc(sizeof (zone_dataset_t), KM_SLEEP);
4357 4366
4358 4367 next = strchr(dataset, ',');
4359 4368
4360 4369 if (next == NULL)
4361 4370 len = strlen(dataset);
4362 4371 else
4363 4372 len = next - dataset;
4364 4373
4365 4374 zd->zd_dataset = kmem_alloc(len + 1, KM_SLEEP);
4366 4375 bcopy(dataset, zd->zd_dataset, len);
4367 4376 zd->zd_dataset[len] = '\0';
4368 4377
4369 4378 list_insert_head(&zone->zone_datasets, zd);
4370 4379
4371 4380 if (next == NULL)
4372 4381 break;
4373 4382
4374 4383 dataset = next + 1;
4375 4384 }
4376 4385
4377 4386 kmem_free(kbuf, buflen);
4378 4387 return (0);
4379 4388 }
4380 4389
4381 4390 /*
4382 4391 * System call to create/initialize a new zone named 'zone_name', rooted
4383 4392 * at 'zone_root', with a zone-wide privilege limit set of 'zone_privs',
4384 4393 * and initialized with the zone-wide rctls described in 'rctlbuf', and
4385 4394 * with labeling set by 'match', 'doi', and 'label'.
4386 4395 *
4387 4396 * If extended error is non-null, we may use it to return more detailed
4388 4397 * error information.
4389 4398 */
4390 4399 static zoneid_t
4391 4400 zone_create(const char *zone_name, const char *zone_root,
4392 4401 const priv_set_t *zone_privs, size_t zone_privssz,
4393 4402 caddr_t rctlbuf, size_t rctlbufsz,
4394 4403 caddr_t zfsbuf, size_t zfsbufsz, int *extended_error,
4395 4404 int match, uint32_t doi, const bslabel_t *label,
4396 4405 int flags)
4397 4406 {
4398 4407 struct zsched_arg zarg;
4399 4408 nvlist_t *rctls = NULL;
4400 4409 proc_t *pp = curproc;
4401 4410 zone_t *zone, *ztmp;
4402 4411 zoneid_t zoneid, start = GLOBAL_ZONEID;
4403 4412 int error;
4404 4413 int error2 = 0;
4405 4414 char *str;
4406 4415 cred_t *zkcr;
4407 4416 boolean_t insert_label_hash;
4408 4417
4409 4418 if (secpolicy_zone_config(CRED()) != 0)
4410 4419 return (set_errno(EPERM));
4411 4420
4412 4421 /* can't boot zone from within chroot environment */
4413 4422 if (PTOU(pp)->u_rdir != NULL && PTOU(pp)->u_rdir != rootdir)
4414 4423 return (zone_create_error(ENOTSUP, ZE_CHROOTED,
4415 4424 extended_error));
4416 4425 /*
4417 4426 * As the first step of zone creation, we want to allocate a zoneid.
4418 4427 * This allocation is complicated by the fact that netstacks use the
4419 4428 * zoneid to determine their stackid, but netstacks themselves are
4420 4429 * freed asynchronously with respect to zone destruction. This means
4421 4430 * that a netstack reference leak (or in principle, an extraordinarily
4422 4431 * long netstack reference hold) could result in a zoneid being
4423 4432 * allocated that in fact corresponds to a stackid from an active
4424 4433 * (referenced) netstack -- unleashing all sorts of havoc when that
4425 4434 * netstack is actually (re)used. (In the abstract, we might wish a
4426 4435 * zoneid to not be deallocated until its last referencing netstack
4427 4436 * has been released, but netstacks lack a backpointer into their
4428 4437 * referencing zone -- and changing them to have such a pointer would
4429 4438 * be substantial, to put it euphemistically.) To avoid this, we
4430 4439 * detect this condition on allocation: if we have allocated a zoneid
4431 4440 * that corresponds to a netstack that's still in use, we warn about
4432 4441 * it (as it is much more likely to be a reference leak than an actual
4433 4442 * netstack reference), free it, and allocate another. That these
4434 4443 * identifers are allocated out of an ID space assures that we won't
4435 4444 * see the identifier we just allocated.
4436 4445 */
4437 4446 for (;;) {
4438 4447 zoneid = id_alloc(zoneid_space);
4439 4448
4440 4449 if (!netstack_inuse_by_stackid(zoneid_to_netstackid(zoneid)))
4441 4450 break;
4442 4451
4443 4452 id_free(zoneid_space, zoneid);
4444 4453
4445 4454 if (start == GLOBAL_ZONEID) {
4446 4455 start = zoneid;
4447 4456 } else if (zoneid == start) {
4448 4457 /*
4449 4458 * We have managed to iterate over the entire available
4450 4459 * zoneid space -- there are no identifiers available,
4451 4460 * presumably due to some number of leaked netstack
4452 4461 * references. While it's in principle possible for us
4453 4462 * to continue to try, it seems wiser to give up at
4454 4463 * this point to warn and fail explicitly with a
4455 4464 * distinctive error.
4456 4465 */
4457 4466 cmn_err(CE_WARN, "zone_create() failed: all available "
4458 4467 "zone IDs have netstacks still in use");
4459 4468 return (set_errno(ENFILE));
4460 4469 }
4461 4470
4462 4471 cmn_err(CE_WARN, "unable to reuse zone ID %d; "
4463 4472 "netstack still in use", zoneid);
4464 4473 }
4465 4474
4466 4475 zone = kmem_zalloc(sizeof (zone_t), KM_SLEEP);
4467 4476 zone->zone_id = zoneid;
4468 4477 zone->zone_status = ZONE_IS_UNINITIALIZED;
4469 4478 zone->zone_pool = pool_default;
4470 4479 zone->zone_pool_mod = gethrtime();
4471 4480 zone->zone_psetid = ZONE_PS_INVAL;
4472 4481 zone->zone_ncpus = 0;
4473 4482 zone->zone_ncpus_online = 0;
4474 4483 zone->zone_restart_init = B_TRUE;
4475 4484 zone->zone_brand = &native_brand;
4476 4485 zone->zone_initname = NULL;
4477 4486 mutex_init(&zone->zone_lock, NULL, MUTEX_DEFAULT, NULL);
4478 4487 mutex_init(&zone->zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL);
4479 4488 mutex_init(&zone->zone_mem_lock, NULL, MUTEX_DEFAULT, NULL);
4480 4489 cv_init(&zone->zone_cv, NULL, CV_DEFAULT, NULL);
4481 4490 list_create(&zone->zone_ref_list, sizeof (zone_ref_t),
4482 4491 offsetof(zone_ref_t, zref_linkage));
4483 4492 list_create(&zone->zone_zsd, sizeof (struct zsd_entry),
4484 4493 offsetof(struct zsd_entry, zsd_linkage));
4485 4494 list_create(&zone->zone_datasets, sizeof (zone_dataset_t),
4486 4495 offsetof(zone_dataset_t, zd_linkage));
4487 4496 list_create(&zone->zone_dl_list, sizeof (zone_dl_t),
4488 4497 offsetof(zone_dl_t, zdl_linkage));
4489 4498 rw_init(&zone->zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL);
4490 4499 rw_init(&zone->zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL);
4491 4500
4492 4501 if (flags & ZCF_NET_EXCL) {
4493 4502 zone->zone_flags |= ZF_NET_EXCL;
4494 4503 }
4495 4504
4496 4505 if ((error = zone_set_name(zone, zone_name)) != 0) {
4497 4506 zone_free(zone);
4498 4507 return (zone_create_error(error, 0, extended_error));
4499 4508 }
4500 4509
4501 4510 if ((error = zone_set_root(zone, zone_root)) != 0) {
4502 4511 zone_free(zone);
4503 4512 return (zone_create_error(error, 0, extended_error));
4504 4513 }
4505 4514 if ((error = zone_set_privset(zone, zone_privs, zone_privssz)) != 0) {
4506 4515 zone_free(zone);
4507 4516 return (zone_create_error(error, 0, extended_error));
4508 4517 }
4509 4518
4510 4519 /* initialize node name to be the same as zone name */
4511 4520 zone->zone_nodename = kmem_alloc(_SYS_NMLN, KM_SLEEP);
4512 4521 (void) strncpy(zone->zone_nodename, zone->zone_name, _SYS_NMLN);
4513 4522 zone->zone_nodename[_SYS_NMLN - 1] = '\0';
4514 4523
4515 4524 zone->zone_domain = kmem_alloc(_SYS_NMLN, KM_SLEEP);
|
↓ open down ↓ |
269 lines elided |
↑ open up ↑ |
4516 4525 zone->zone_domain[0] = '\0';
4517 4526 zone->zone_hostid = HW_INVALID_HOSTID;
4518 4527 zone->zone_shares = 1;
4519 4528 zone->zone_shmmax = 0;
4520 4529 zone->zone_ipc.ipcq_shmmni = 0;
4521 4530 zone->zone_ipc.ipcq_semmni = 0;
4522 4531 zone->zone_ipc.ipcq_msgmni = 0;
4523 4532 zone->zone_bootargs = NULL;
4524 4533 zone->zone_fs_allowed = NULL;
4525 4534
4526 - secflags_zero(&zone0.zone_secflags.psf_lower);
4527 - secflags_zero(&zone0.zone_secflags.psf_effective);
4528 - secflags_zero(&zone0.zone_secflags.psf_inherit);
4529 - secflags_fullset(&zone0.zone_secflags.psf_upper);
4535 + psecflags_default(&zone->zone_secflags);
4530 4536
4531 4537 zone->zone_initname =
4532 4538 kmem_alloc(strlen(zone_default_initname) + 1, KM_SLEEP);
4533 4539 (void) strcpy(zone->zone_initname, zone_default_initname);
4534 4540 zone->zone_nlwps = 0;
4535 4541 zone->zone_nlwps_ctl = INT_MAX;
4536 4542 zone->zone_nprocs = 0;
4537 4543 zone->zone_nprocs_ctl = INT_MAX;
4538 4544 zone->zone_locked_mem = 0;
4539 4545 zone->zone_locked_mem_ctl = UINT64_MAX;
4540 4546 zone->zone_max_swap = 0;
4541 4547 zone->zone_max_swap_ctl = UINT64_MAX;
4542 4548 zone->zone_max_lofi = 0;
4543 4549 zone->zone_max_lofi_ctl = UINT64_MAX;
4544 4550 zone0.zone_lockedmem_kstat = NULL;
4545 4551 zone0.zone_swapresv_kstat = NULL;
4546 4552
4553 + zone->zone_ustate = cpu_uarray_zalloc(ZONE_USTATE_MAX, KM_SLEEP);
4554 +
4547 4555 /*
4548 4556 * Zsched initializes the rctls.
4549 4557 */
4550 4558 zone->zone_rctls = NULL;
4551 4559
4552 4560 if ((error = parse_rctls(rctlbuf, rctlbufsz, &rctls)) != 0) {
4553 4561 zone_free(zone);
4554 4562 return (zone_create_error(error, 0, extended_error));
4555 4563 }
4556 4564
4557 4565 if ((error = parse_zfs(zone, zfsbuf, zfsbufsz)) != 0) {
4558 4566 zone_free(zone);
4559 4567 return (set_errno(error));
4560 4568 }
4561 4569
4562 4570 /*
4563 4571 * Read in the trusted system parameters:
4564 4572 * match flag and sensitivity label.
4565 4573 */
4566 4574 zone->zone_match = match;
4567 4575 if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) {
4568 4576 /* Fail if requested to set doi to anything but system's doi */
4569 4577 if (doi != 0 && doi != default_doi) {
4570 4578 zone_free(zone);
4571 4579 return (set_errno(EINVAL));
4572 4580 }
4573 4581 /* Always apply system's doi to the zone */
4574 4582 error = zone_set_label(zone, label, default_doi);
4575 4583 if (error != 0) {
4576 4584 zone_free(zone);
4577 4585 return (set_errno(error));
4578 4586 }
4579 4587 insert_label_hash = B_TRUE;
4580 4588 } else {
4581 4589 /* all zones get an admin_low label if system is not labeled */
4582 4590 zone->zone_slabel = l_admin_low;
4583 4591 label_hold(l_admin_low);
4584 4592 insert_label_hash = B_FALSE;
4585 4593 }
4586 4594
4587 4595 /*
4588 4596 * Stop all lwps since that's what normally happens as part of fork().
4589 4597 * This needs to happen before we grab any locks to avoid deadlock
4590 4598 * (another lwp in the process could be waiting for the held lock).
4591 4599 */
4592 4600 if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK)) {
4593 4601 zone_free(zone);
4594 4602 nvlist_free(rctls);
4595 4603 return (zone_create_error(error, 0, extended_error));
4596 4604 }
4597 4605
4598 4606 if (block_mounts(zone) == 0) {
4599 4607 mutex_enter(&pp->p_lock);
4600 4608 if (curthread != pp->p_agenttp)
4601 4609 continuelwps(pp);
4602 4610 mutex_exit(&pp->p_lock);
4603 4611 zone_free(zone);
4604 4612 nvlist_free(rctls);
4605 4613 return (zone_create_error(error, 0, extended_error));
4606 4614 }
4607 4615
4608 4616 /*
4609 4617 * Set up credential for kernel access. After this, any errors
4610 4618 * should go through the dance in errout rather than calling
4611 4619 * zone_free directly.
4612 4620 */
4613 4621 zone->zone_kcred = crdup(kcred);
4614 4622 crsetzone(zone->zone_kcred, zone);
4615 4623 priv_intersect(zone->zone_privset, &CR_PPRIV(zone->zone_kcred));
4616 4624 priv_intersect(zone->zone_privset, &CR_EPRIV(zone->zone_kcred));
4617 4625 priv_intersect(zone->zone_privset, &CR_IPRIV(zone->zone_kcred));
4618 4626 priv_intersect(zone->zone_privset, &CR_LPRIV(zone->zone_kcred));
4619 4627
4620 4628 mutex_enter(&zonehash_lock);
4621 4629 /*
4622 4630 * Make sure zone doesn't already exist.
4623 4631 *
4624 4632 * If the system and zone are labeled,
4625 4633 * make sure no other zone exists that has the same label.
4626 4634 */
4627 4635 if ((ztmp = zone_find_all_by_name(zone->zone_name)) != NULL ||
4628 4636 (insert_label_hash &&
4629 4637 (ztmp = zone_find_all_by_label(zone->zone_slabel)) != NULL)) {
4630 4638 zone_status_t status;
4631 4639
4632 4640 status = zone_status_get(ztmp);
4633 4641 if (status == ZONE_IS_READY || status == ZONE_IS_RUNNING)
4634 4642 error = EEXIST;
4635 4643 else
4636 4644 error = EBUSY;
4637 4645
4638 4646 if (insert_label_hash)
4639 4647 error2 = ZE_LABELINUSE;
4640 4648
4641 4649 goto errout;
4642 4650 }
4643 4651
4644 4652 /*
4645 4653 * Don't allow zone creations which would cause one zone's rootpath to
4646 4654 * be accessible from that of another (non-global) zone.
4647 4655 */
4648 4656 if (zone_is_nested(zone->zone_rootpath)) {
4649 4657 error = EBUSY;
4650 4658 goto errout;
4651 4659 }
4652 4660
4653 4661 ASSERT(zonecount != 0); /* check for leaks */
4654 4662 if (zonecount + 1 > maxzones) {
4655 4663 error = ENOMEM;
4656 4664 goto errout;
4657 4665 }
4658 4666
4659 4667 if (zone_mount_count(zone->zone_rootpath) != 0) {
4660 4668 error = EBUSY;
4661 4669 error2 = ZE_AREMOUNTS;
4662 4670 goto errout;
4663 4671 }
4664 4672
4665 4673 /*
4666 4674 * Zone is still incomplete, but we need to drop all locks while
4667 4675 * zsched() initializes this zone's kernel process. We
4668 4676 * optimistically add the zone to the hashtable and associated
4669 4677 * lists so a parallel zone_create() doesn't try to create the
4670 4678 * same zone.
4671 4679 */
4672 4680 zonecount++;
4673 4681 (void) mod_hash_insert(zonehashbyid,
4674 4682 (mod_hash_key_t)(uintptr_t)zone->zone_id,
4675 4683 (mod_hash_val_t)(uintptr_t)zone);
4676 4684 str = kmem_alloc(strlen(zone->zone_name) + 1, KM_SLEEP);
4677 4685 (void) strcpy(str, zone->zone_name);
4678 4686 (void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)str,
4679 4687 (mod_hash_val_t)(uintptr_t)zone);
4680 4688 if (insert_label_hash) {
4681 4689 (void) mod_hash_insert(zonehashbylabel,
4682 4690 (mod_hash_key_t)zone->zone_slabel, (mod_hash_val_t)zone);
4683 4691 zone->zone_flags |= ZF_HASHED_LABEL;
4684 4692 }
4685 4693
4686 4694 /*
4687 4695 * Insert into active list. At this point there are no 'hold's
4688 4696 * on the zone, but everyone else knows not to use it, so we can
4689 4697 * continue to use it. zsched() will do a zone_hold() if the
4690 4698 * newproc() is successful.
4691 4699 */
4692 4700 list_insert_tail(&zone_active, zone);
4693 4701 mutex_exit(&zonehash_lock);
4694 4702
4695 4703 zarg.zone = zone;
4696 4704 zarg.nvlist = rctls;
4697 4705 /*
4698 4706 * The process, task, and project rctls are probably wrong;
4699 4707 * we need an interface to get the default values of all rctls,
4700 4708 * and initialize zsched appropriately. I'm not sure that that
4701 4709 * makes much of a difference, though.
4702 4710 */
4703 4711 error = newproc(zsched, (void *)&zarg, syscid, minclsyspri, NULL, 0);
4704 4712 if (error != 0) {
4705 4713 /*
4706 4714 * We need to undo all globally visible state.
4707 4715 */
4708 4716 mutex_enter(&zonehash_lock);
4709 4717 list_remove(&zone_active, zone);
4710 4718 if (zone->zone_flags & ZF_HASHED_LABEL) {
4711 4719 ASSERT(zone->zone_slabel != NULL);
4712 4720 (void) mod_hash_destroy(zonehashbylabel,
4713 4721 (mod_hash_key_t)zone->zone_slabel);
4714 4722 }
4715 4723 (void) mod_hash_destroy(zonehashbyname,
4716 4724 (mod_hash_key_t)(uintptr_t)zone->zone_name);
4717 4725 (void) mod_hash_destroy(zonehashbyid,
4718 4726 (mod_hash_key_t)(uintptr_t)zone->zone_id);
4719 4727 ASSERT(zonecount > 1);
4720 4728 zonecount--;
4721 4729 goto errout;
4722 4730 }
4723 4731
4724 4732 /*
4725 4733 * Zone creation can't fail from now on.
4726 4734 */
4727 4735
4728 4736 /*
4729 4737 * Create zone kstats
4730 4738 */
4731 4739 zone_kstat_create(zone);
4732 4740
4733 4741 /*
4734 4742 * Let the other lwps continue.
4735 4743 */
4736 4744 mutex_enter(&pp->p_lock);
4737 4745 if (curthread != pp->p_agenttp)
4738 4746 continuelwps(pp);
4739 4747 mutex_exit(&pp->p_lock);
4740 4748
4741 4749 /*
4742 4750 * Wait for zsched to finish initializing the zone.
4743 4751 */
4744 4752 zone_status_wait(zone, ZONE_IS_READY);
4745 4753 /*
4746 4754 * The zone is fully visible, so we can let mounts progress.
4747 4755 */
4748 4756 resume_mounts(zone);
4749 4757 nvlist_free(rctls);
4750 4758
4751 4759 return (zoneid);
4752 4760
4753 4761 errout:
4754 4762 mutex_exit(&zonehash_lock);
4755 4763 /*
4756 4764 * Let the other lwps continue.
4757 4765 */
4758 4766 mutex_enter(&pp->p_lock);
4759 4767 if (curthread != pp->p_agenttp)
4760 4768 continuelwps(pp);
4761 4769 mutex_exit(&pp->p_lock);
4762 4770
4763 4771 resume_mounts(zone);
4764 4772 nvlist_free(rctls);
4765 4773 /*
4766 4774 * There is currently one reference to the zone, a cred_ref from
4767 4775 * zone_kcred. To free the zone, we call crfree, which will call
4768 4776 * zone_cred_rele, which will call zone_free.
4769 4777 */
4770 4778 ASSERT(zone->zone_cred_ref == 1);
4771 4779 ASSERT(zone->zone_kcred->cr_ref == 1);
4772 4780 ASSERT(zone->zone_ref == 0);
4773 4781 zkcr = zone->zone_kcred;
4774 4782 zone->zone_kcred = NULL;
4775 4783 crfree(zkcr); /* triggers call to zone_free */
4776 4784 return (zone_create_error(error, error2, extended_error));
4777 4785 }
4778 4786
4779 4787 /*
4780 4788 * Cause the zone to boot. This is pretty simple, since we let zoneadmd do
4781 4789 * the heavy lifting. initname is the path to the program to launch
4782 4790 * at the "top" of the zone; if this is NULL, we use the system default,
4783 4791 * which is stored at zone_default_initname.
4784 4792 */
4785 4793 static int
4786 4794 zone_boot(zoneid_t zoneid)
4787 4795 {
4788 4796 int err;
4789 4797 zone_t *zone;
4790 4798
4791 4799 if (secpolicy_zone_config(CRED()) != 0)
4792 4800 return (set_errno(EPERM));
4793 4801 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4794 4802 return (set_errno(EINVAL));
4795 4803
4796 4804 mutex_enter(&zonehash_lock);
4797 4805 /*
4798 4806 * Look for zone under hash lock to prevent races with calls to
4799 4807 * zone_shutdown, zone_destroy, etc.
4800 4808 */
4801 4809 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4802 4810 mutex_exit(&zonehash_lock);
4803 4811 return (set_errno(EINVAL));
4804 4812 }
4805 4813
4806 4814 mutex_enter(&zone_status_lock);
4807 4815 if (zone_status_get(zone) != ZONE_IS_READY) {
4808 4816 mutex_exit(&zone_status_lock);
4809 4817 mutex_exit(&zonehash_lock);
4810 4818 return (set_errno(EINVAL));
4811 4819 }
4812 4820 zone_status_set(zone, ZONE_IS_BOOTING);
4813 4821 mutex_exit(&zone_status_lock);
4814 4822
4815 4823 zone_hold(zone); /* so we can use the zone_t later */
4816 4824 mutex_exit(&zonehash_lock);
4817 4825
4818 4826 if (zone_status_wait_sig(zone, ZONE_IS_RUNNING) == 0) {
4819 4827 zone_rele(zone);
4820 4828 return (set_errno(EINTR));
4821 4829 }
4822 4830
4823 4831 /*
4824 4832 * Boot (starting init) might have failed, in which case the zone
4825 4833 * will go to the SHUTTING_DOWN state; an appropriate errno will
4826 4834 * be placed in zone->zone_boot_err, and so we return that.
4827 4835 */
4828 4836 err = zone->zone_boot_err;
4829 4837 zone_rele(zone);
4830 4838 return (err ? set_errno(err) : 0);
4831 4839 }
4832 4840
4833 4841 /*
4834 4842 * Kills all user processes in the zone, waiting for them all to exit
4835 4843 * before returning.
4836 4844 */
4837 4845 static int
4838 4846 zone_empty(zone_t *zone)
4839 4847 {
4840 4848 int waitstatus;
4841 4849
4842 4850 /*
4843 4851 * We need to drop zonehash_lock before killing all
4844 4852 * processes, otherwise we'll deadlock with zone_find_*
4845 4853 * which can be called from the exit path.
4846 4854 */
4847 4855 ASSERT(MUTEX_NOT_HELD(&zonehash_lock));
4848 4856 while ((waitstatus = zone_status_timedwait_sig(zone,
4849 4857 ddi_get_lbolt() + hz, ZONE_IS_EMPTY)) == -1) {
4850 4858 killall(zone->zone_id);
4851 4859 }
4852 4860 /*
4853 4861 * return EINTR if we were signaled
4854 4862 */
4855 4863 if (waitstatus == 0)
4856 4864 return (EINTR);
4857 4865 return (0);
4858 4866 }
4859 4867
4860 4868 /*
4861 4869 * This function implements the policy for zone visibility.
4862 4870 *
4863 4871 * In standard Solaris, a non-global zone can only see itself.
4864 4872 *
4865 4873 * In Trusted Extensions, a labeled zone can lookup any zone whose label
4866 4874 * it dominates. For this test, the label of the global zone is treated as
4867 4875 * admin_high so it is special-cased instead of being checked for dominance.
4868 4876 *
4869 4877 * Returns true if zone attributes are viewable, false otherwise.
4870 4878 */
4871 4879 static boolean_t
4872 4880 zone_list_access(zone_t *zone)
4873 4881 {
4874 4882
4875 4883 if (curproc->p_zone == global_zone ||
4876 4884 curproc->p_zone == zone) {
4877 4885 return (B_TRUE);
4878 4886 } else if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) {
4879 4887 bslabel_t *curproc_label;
4880 4888 bslabel_t *zone_label;
4881 4889
4882 4890 curproc_label = label2bslabel(curproc->p_zone->zone_slabel);
4883 4891 zone_label = label2bslabel(zone->zone_slabel);
4884 4892
4885 4893 if (zone->zone_id != GLOBAL_ZONEID &&
4886 4894 bldominates(curproc_label, zone_label)) {
4887 4895 return (B_TRUE);
4888 4896 } else {
4889 4897 return (B_FALSE);
4890 4898 }
4891 4899 } else {
4892 4900 return (B_FALSE);
4893 4901 }
4894 4902 }
4895 4903
4896 4904 /*
4897 4905 * Systemcall to start the zone's halt sequence. By the time this
4898 4906 * function successfully returns, all user processes and kernel threads
4899 4907 * executing in it will have exited, ZSD shutdown callbacks executed,
4900 4908 * and the zone status set to ZONE_IS_DOWN.
4901 4909 *
4902 4910 * It is possible that the call will interrupt itself if the caller is the
4903 4911 * parent of any process running in the zone, and doesn't have SIGCHLD blocked.
4904 4912 */
4905 4913 static int
4906 4914 zone_shutdown(zoneid_t zoneid)
4907 4915 {
4908 4916 int error;
4909 4917 zone_t *zone;
4910 4918 zone_status_t status;
4911 4919
4912 4920 if (secpolicy_zone_config(CRED()) != 0)
4913 4921 return (set_errno(EPERM));
4914 4922 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
4915 4923 return (set_errno(EINVAL));
4916 4924
4917 4925 mutex_enter(&zonehash_lock);
4918 4926 /*
4919 4927 * Look for zone under hash lock to prevent races with other
4920 4928 * calls to zone_shutdown and zone_destroy.
4921 4929 */
4922 4930 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
4923 4931 mutex_exit(&zonehash_lock);
4924 4932 return (set_errno(EINVAL));
4925 4933 }
4926 4934
4927 4935 /*
4928 4936 * We have to drop zonehash_lock before calling block_mounts.
4929 4937 * Hold the zone so we can continue to use the zone_t.
4930 4938 */
4931 4939 zone_hold(zone);
4932 4940 mutex_exit(&zonehash_lock);
4933 4941
4934 4942 /*
4935 4943 * Block mounts so that VFS_MOUNT() can get an accurate view of
4936 4944 * the zone's status with regards to ZONE_IS_SHUTTING down.
4937 4945 *
4938 4946 * e.g. NFS can fail the mount if it determines that the zone
4939 4947 * has already begun the shutdown sequence.
4940 4948 *
4941 4949 */
4942 4950 if (block_mounts(zone) == 0) {
4943 4951 zone_rele(zone);
4944 4952 return (set_errno(EINTR));
4945 4953 }
4946 4954
4947 4955 mutex_enter(&zonehash_lock);
4948 4956 mutex_enter(&zone_status_lock);
4949 4957 status = zone_status_get(zone);
4950 4958 /*
4951 4959 * Fail if the zone isn't fully initialized yet.
4952 4960 */
4953 4961 if (status < ZONE_IS_READY) {
4954 4962 mutex_exit(&zone_status_lock);
4955 4963 mutex_exit(&zonehash_lock);
4956 4964 resume_mounts(zone);
4957 4965 zone_rele(zone);
4958 4966 return (set_errno(EINVAL));
4959 4967 }
4960 4968 /*
4961 4969 * If conditions required for zone_shutdown() to return have been met,
4962 4970 * return success.
4963 4971 */
4964 4972 if (status >= ZONE_IS_DOWN) {
4965 4973 mutex_exit(&zone_status_lock);
4966 4974 mutex_exit(&zonehash_lock);
4967 4975 resume_mounts(zone);
4968 4976 zone_rele(zone);
4969 4977 return (0);
4970 4978 }
4971 4979 /*
4972 4980 * If zone_shutdown() hasn't been called before, go through the motions.
4973 4981 * If it has, there's nothing to do but wait for the kernel threads to
4974 4982 * drain.
4975 4983 */
4976 4984 if (status < ZONE_IS_EMPTY) {
4977 4985 uint_t ntasks;
4978 4986
4979 4987 mutex_enter(&zone->zone_lock);
4980 4988 if ((ntasks = zone->zone_ntasks) != 1) {
4981 4989 /*
4982 4990 * There's still stuff running.
4983 4991 */
4984 4992 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
4985 4993 }
4986 4994 mutex_exit(&zone->zone_lock);
4987 4995 if (ntasks == 1) {
4988 4996 /*
4989 4997 * The only way to create another task is through
4990 4998 * zone_enter(), which will block until we drop
4991 4999 * zonehash_lock. The zone is empty.
4992 5000 */
4993 5001 if (zone->zone_kthreads == NULL) {
4994 5002 /*
4995 5003 * Skip ahead to ZONE_IS_DOWN
4996 5004 */
4997 5005 zone_status_set(zone, ZONE_IS_DOWN);
4998 5006 } else {
4999 5007 zone_status_set(zone, ZONE_IS_EMPTY);
5000 5008 }
5001 5009 }
5002 5010 }
5003 5011 mutex_exit(&zone_status_lock);
5004 5012 mutex_exit(&zonehash_lock);
5005 5013 resume_mounts(zone);
5006 5014
5007 5015 if (error = zone_empty(zone)) {
5008 5016 zone_rele(zone);
5009 5017 return (set_errno(error));
5010 5018 }
5011 5019 /*
5012 5020 * After the zone status goes to ZONE_IS_DOWN this zone will no
5013 5021 * longer be notified of changes to the pools configuration, so
5014 5022 * in order to not end up with a stale pool pointer, we point
5015 5023 * ourselves at the default pool and remove all resource
5016 5024 * visibility. This is especially important as the zone_t may
5017 5025 * languish on the deathrow for a very long time waiting for
5018 5026 * cred's to drain out.
5019 5027 *
5020 5028 * This rebinding of the zone can happen multiple times
5021 5029 * (presumably due to interrupted or parallel systemcalls)
5022 5030 * without any adverse effects.
5023 5031 */
5024 5032 if (pool_lock_intr() != 0) {
5025 5033 zone_rele(zone);
5026 5034 return (set_errno(EINTR));
5027 5035 }
5028 5036 if (pool_state == POOL_ENABLED) {
5029 5037 mutex_enter(&cpu_lock);
5030 5038 zone_pool_set(zone, pool_default);
5031 5039 /*
5032 5040 * The zone no longer needs to be able to see any cpus.
5033 5041 */
5034 5042 zone_pset_set(zone, ZONE_PS_INVAL);
5035 5043 mutex_exit(&cpu_lock);
5036 5044 }
5037 5045 pool_unlock();
5038 5046
5039 5047 /*
5040 5048 * ZSD shutdown callbacks can be executed multiple times, hence
5041 5049 * it is safe to not be holding any locks across this call.
5042 5050 */
5043 5051 zone_zsd_callbacks(zone, ZSD_SHUTDOWN);
5044 5052
5045 5053 mutex_enter(&zone_status_lock);
5046 5054 if (zone->zone_kthreads == NULL && zone_status_get(zone) < ZONE_IS_DOWN)
5047 5055 zone_status_set(zone, ZONE_IS_DOWN);
5048 5056 mutex_exit(&zone_status_lock);
5049 5057
5050 5058 /*
5051 5059 * Wait for kernel threads to drain.
5052 5060 */
5053 5061 if (!zone_status_wait_sig(zone, ZONE_IS_DOWN)) {
5054 5062 zone_rele(zone);
5055 5063 return (set_errno(EINTR));
5056 5064 }
5057 5065
5058 5066 /*
5059 5067 * Zone can be become down/destroyable even if the above wait
5060 5068 * returns EINTR, so any code added here may never execute.
5061 5069 * (i.e. don't add code here)
5062 5070 */
5063 5071
5064 5072 zone_rele(zone);
5065 5073 return (0);
5066 5074 }
5067 5075
5068 5076 /*
5069 5077 * Log the specified zone's reference counts. The caller should not be
5070 5078 * holding the zone's zone_lock.
5071 5079 */
5072 5080 static void
5073 5081 zone_log_refcounts(zone_t *zone)
5074 5082 {
5075 5083 char *buffer;
5076 5084 char *buffer_position;
5077 5085 uint32_t buffer_size;
5078 5086 uint32_t index;
5079 5087 uint_t ref;
5080 5088 uint_t cred_ref;
5081 5089
5082 5090 /*
5083 5091 * Construct a string representing the subsystem-specific reference
5084 5092 * counts. The counts are printed in ascending order by index into the
5085 5093 * zone_t::zone_subsys_ref array. The list will be surrounded by
5086 5094 * square brackets [] and will only contain nonzero reference counts.
5087 5095 *
5088 5096 * The buffer will hold two square bracket characters plus ten digits,
5089 5097 * one colon, one space, one comma, and some characters for a
5090 5098 * subsystem name per subsystem-specific reference count. (Unsigned 32-
5091 5099 * bit integers have at most ten decimal digits.) The last
5092 5100 * reference count's comma is replaced by the closing square
5093 5101 * bracket and a NULL character to terminate the string.
5094 5102 *
5095 5103 * NOTE: We have to grab the zone's zone_lock to create a consistent
5096 5104 * snapshot of the zone's reference counters.
5097 5105 *
5098 5106 * First, figure out how much space the string buffer will need.
5099 5107 * The buffer's size is stored in buffer_size.
5100 5108 */
5101 5109 buffer_size = 2; /* for the square brackets */
5102 5110 mutex_enter(&zone->zone_lock);
5103 5111 zone->zone_flags |= ZF_REFCOUNTS_LOGGED;
5104 5112 ref = zone->zone_ref;
5105 5113 cred_ref = zone->zone_cred_ref;
5106 5114 for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index)
5107 5115 if (zone->zone_subsys_ref[index] != 0)
5108 5116 buffer_size += strlen(zone_ref_subsys_names[index]) +
5109 5117 13;
5110 5118 if (buffer_size == 2) {
5111 5119 /*
5112 5120 * No subsystems had nonzero reference counts. Don't bother
5113 5121 * with allocating a buffer; just log the general-purpose and
5114 5122 * credential reference counts.
5115 5123 */
5116 5124 mutex_exit(&zone->zone_lock);
5117 5125 (void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE,
5118 5126 "Zone '%s' (ID: %d) is shutting down, but %u zone "
5119 5127 "references and %u credential references are still extant",
5120 5128 zone->zone_name, zone->zone_id, ref, cred_ref);
5121 5129 return;
5122 5130 }
5123 5131
5124 5132 /*
5125 5133 * buffer_size contains the exact number of characters that the
5126 5134 * buffer will need. Allocate the buffer and fill it with nonzero
5127 5135 * subsystem-specific reference counts. Surround the results with
5128 5136 * square brackets afterwards.
5129 5137 */
5130 5138 buffer = kmem_alloc(buffer_size, KM_SLEEP);
5131 5139 buffer_position = &buffer[1];
5132 5140 for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index) {
5133 5141 /*
5134 5142 * NOTE: The DDI's version of sprintf() returns a pointer to
5135 5143 * the modified buffer rather than the number of bytes written
5136 5144 * (as in snprintf(3C)). This is unfortunate and annoying.
5137 5145 * Therefore, we'll use snprintf() with INT_MAX to get the
5138 5146 * number of bytes written. Using INT_MAX is safe because
5139 5147 * the buffer is perfectly sized for the data: we'll never
5140 5148 * overrun the buffer.
5141 5149 */
5142 5150 if (zone->zone_subsys_ref[index] != 0)
5143 5151 buffer_position += snprintf(buffer_position, INT_MAX,
5144 5152 "%s: %u,", zone_ref_subsys_names[index],
5145 5153 zone->zone_subsys_ref[index]);
5146 5154 }
5147 5155 mutex_exit(&zone->zone_lock);
5148 5156 buffer[0] = '[';
5149 5157 ASSERT((uintptr_t)(buffer_position - buffer) < buffer_size);
5150 5158 ASSERT(buffer_position[0] == '\0' && buffer_position[-1] == ',');
5151 5159 buffer_position[-1] = ']';
5152 5160
5153 5161 /*
5154 5162 * Log the reference counts and free the message buffer.
5155 5163 */
5156 5164 (void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE,
5157 5165 "Zone '%s' (ID: %d) is shutting down, but %u zone references and "
5158 5166 "%u credential references are still extant %s", zone->zone_name,
5159 5167 zone->zone_id, ref, cred_ref, buffer);
5160 5168 kmem_free(buffer, buffer_size);
5161 5169 }
5162 5170
5163 5171 /*
5164 5172 * Systemcall entry point to finalize the zone halt process. The caller
5165 5173 * must have already successfully called zone_shutdown().
5166 5174 *
5167 5175 * Upon successful completion, the zone will have been fully destroyed:
5168 5176 * zsched will have exited, destructor callbacks executed, and the zone
5169 5177 * removed from the list of active zones.
5170 5178 */
5171 5179 static int
5172 5180 zone_destroy(zoneid_t zoneid)
5173 5181 {
5174 5182 uint64_t uniqid;
5175 5183 zone_t *zone;
5176 5184 zone_status_t status;
5177 5185 clock_t wait_time;
5178 5186 boolean_t log_refcounts;
5179 5187
5180 5188 if (secpolicy_zone_config(CRED()) != 0)
5181 5189 return (set_errno(EPERM));
5182 5190 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
5183 5191 return (set_errno(EINVAL));
5184 5192
5185 5193 mutex_enter(&zonehash_lock);
5186 5194 /*
5187 5195 * Look for zone under hash lock to prevent races with other
5188 5196 * calls to zone_destroy.
5189 5197 */
5190 5198 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5191 5199 mutex_exit(&zonehash_lock);
5192 5200 return (set_errno(EINVAL));
5193 5201 }
5194 5202
5195 5203 if (zone_mount_count(zone->zone_rootpath) != 0) {
5196 5204 mutex_exit(&zonehash_lock);
5197 5205 return (set_errno(EBUSY));
5198 5206 }
5199 5207 mutex_enter(&zone_status_lock);
5200 5208 status = zone_status_get(zone);
5201 5209 if (status < ZONE_IS_DOWN) {
5202 5210 mutex_exit(&zone_status_lock);
5203 5211 mutex_exit(&zonehash_lock);
5204 5212 return (set_errno(EBUSY));
5205 5213 } else if (status == ZONE_IS_DOWN) {
5206 5214 zone_status_set(zone, ZONE_IS_DYING); /* Tell zsched to exit */
5207 5215 }
5208 5216 mutex_exit(&zone_status_lock);
5209 5217 zone_hold(zone);
5210 5218 mutex_exit(&zonehash_lock);
5211 5219
5212 5220 /*
5213 5221 * wait for zsched to exit
5214 5222 */
5215 5223 zone_status_wait(zone, ZONE_IS_DEAD);
5216 5224 zone_zsd_callbacks(zone, ZSD_DESTROY);
5217 5225 zone->zone_netstack = NULL;
5218 5226 uniqid = zone->zone_uniqid;
5219 5227 zone_rele(zone);
5220 5228 zone = NULL; /* potentially free'd */
5221 5229
5222 5230 log_refcounts = B_FALSE;
5223 5231 wait_time = SEC_TO_TICK(ZONE_DESTROY_TIMEOUT_SECS);
5224 5232 mutex_enter(&zonehash_lock);
5225 5233 for (; /* ever */; ) {
5226 5234 boolean_t unref;
5227 5235 boolean_t refs_have_been_logged;
5228 5236
5229 5237 if ((zone = zone_find_all_by_id(zoneid)) == NULL ||
5230 5238 zone->zone_uniqid != uniqid) {
5231 5239 /*
5232 5240 * The zone has gone away. Necessary conditions
5233 5241 * are met, so we return success.
5234 5242 */
5235 5243 mutex_exit(&zonehash_lock);
5236 5244 return (0);
5237 5245 }
5238 5246 mutex_enter(&zone->zone_lock);
5239 5247 unref = ZONE_IS_UNREF(zone);
5240 5248 refs_have_been_logged = (zone->zone_flags &
5241 5249 ZF_REFCOUNTS_LOGGED);
5242 5250 mutex_exit(&zone->zone_lock);
5243 5251 if (unref) {
5244 5252 /*
5245 5253 * There is only one reference to the zone -- that
5246 5254 * added when the zone was added to the hashtables --
5247 5255 * and things will remain this way until we drop
5248 5256 * zonehash_lock... we can go ahead and cleanup the
5249 5257 * zone.
5250 5258 */
5251 5259 break;
5252 5260 }
5253 5261
5254 5262 /*
5255 5263 * Wait for zone_rele_common() or zone_cred_rele() to signal
5256 5264 * zone_destroy_cv. zone_destroy_cv is signaled only when
5257 5265 * some zone's general-purpose reference count reaches one.
5258 5266 * If ZONE_DESTROY_TIMEOUT_SECS seconds elapse while waiting
5259 5267 * on zone_destroy_cv, then log the zone's reference counts and
5260 5268 * continue to wait for zone_rele() and zone_cred_rele().
5261 5269 */
5262 5270 if (!refs_have_been_logged) {
5263 5271 if (!log_refcounts) {
5264 5272 /*
5265 5273 * This thread hasn't timed out waiting on
5266 5274 * zone_destroy_cv yet. Wait wait_time clock
5267 5275 * ticks (initially ZONE_DESTROY_TIMEOUT_SECS
5268 5276 * seconds) for the zone's references to clear.
5269 5277 */
5270 5278 ASSERT(wait_time > 0);
5271 5279 wait_time = cv_reltimedwait_sig(
5272 5280 &zone_destroy_cv, &zonehash_lock, wait_time,
5273 5281 TR_SEC);
5274 5282 if (wait_time > 0) {
5275 5283 /*
5276 5284 * A thread in zone_rele() or
5277 5285 * zone_cred_rele() signaled
5278 5286 * zone_destroy_cv before this thread's
5279 5287 * wait timed out. The zone might have
5280 5288 * only one reference left; find out!
5281 5289 */
5282 5290 continue;
5283 5291 } else if (wait_time == 0) {
5284 5292 /* The thread's process was signaled. */
5285 5293 mutex_exit(&zonehash_lock);
5286 5294 return (set_errno(EINTR));
5287 5295 }
5288 5296
5289 5297 /*
5290 5298 * The thread timed out while waiting on
5291 5299 * zone_destroy_cv. Even though the thread
5292 5300 * timed out, it has to check whether another
5293 5301 * thread woke up from zone_destroy_cv and
5294 5302 * destroyed the zone.
5295 5303 *
5296 5304 * If the zone still exists and has more than
5297 5305 * one unreleased general-purpose reference,
5298 5306 * then log the zone's reference counts.
5299 5307 */
5300 5308 log_refcounts = B_TRUE;
5301 5309 continue;
5302 5310 }
5303 5311
5304 5312 /*
5305 5313 * The thread already timed out on zone_destroy_cv while
5306 5314 * waiting for subsystems to release the zone's last
5307 5315 * general-purpose references. Log the zone's reference
5308 5316 * counts and wait indefinitely on zone_destroy_cv.
5309 5317 */
5310 5318 zone_log_refcounts(zone);
5311 5319 }
5312 5320 if (cv_wait_sig(&zone_destroy_cv, &zonehash_lock) == 0) {
5313 5321 /* The thread's process was signaled. */
5314 5322 mutex_exit(&zonehash_lock);
5315 5323 return (set_errno(EINTR));
5316 5324 }
5317 5325 }
5318 5326
5319 5327 /*
5320 5328 * Remove CPU cap for this zone now since we're not going to
5321 5329 * fail below this point.
5322 5330 */
5323 5331 cpucaps_zone_remove(zone);
5324 5332
5325 5333 /* Get rid of the zone's kstats */
5326 5334 zone_kstat_delete(zone);
5327 5335
5328 5336 /* remove the pfexecd doors */
5329 5337 if (zone->zone_pfexecd != NULL) {
5330 5338 klpd_freelist(&zone->zone_pfexecd);
5331 5339 zone->zone_pfexecd = NULL;
5332 5340 }
5333 5341
5334 5342 /* free brand specific data */
5335 5343 if (ZONE_IS_BRANDED(zone))
5336 5344 ZBROP(zone)->b_free_brand_data(zone);
5337 5345
5338 5346 /* Say goodbye to brand framework. */
5339 5347 brand_unregister_zone(zone->zone_brand);
5340 5348
5341 5349 /*
5342 5350 * It is now safe to let the zone be recreated; remove it from the
5343 5351 * lists. The memory will not be freed until the last cred
5344 5352 * reference goes away.
5345 5353 */
5346 5354 ASSERT(zonecount > 1); /* must be > 1; can't destroy global zone */
5347 5355 zonecount--;
5348 5356 /* remove from active list and hash tables */
5349 5357 list_remove(&zone_active, zone);
5350 5358 (void) mod_hash_destroy(zonehashbyname,
5351 5359 (mod_hash_key_t)zone->zone_name);
5352 5360 (void) mod_hash_destroy(zonehashbyid,
5353 5361 (mod_hash_key_t)(uintptr_t)zone->zone_id);
5354 5362 if (zone->zone_flags & ZF_HASHED_LABEL)
5355 5363 (void) mod_hash_destroy(zonehashbylabel,
5356 5364 (mod_hash_key_t)zone->zone_slabel);
5357 5365 mutex_exit(&zonehash_lock);
5358 5366
5359 5367 /*
5360 5368 * Release the root vnode; we're not using it anymore. Nor should any
5361 5369 * other thread that might access it exist.
5362 5370 */
5363 5371 if (zone->zone_rootvp != NULL) {
5364 5372 VN_RELE(zone->zone_rootvp);
5365 5373 zone->zone_rootvp = NULL;
5366 5374 }
5367 5375
5368 5376 /* add to deathrow list */
5369 5377 mutex_enter(&zone_deathrow_lock);
5370 5378 list_insert_tail(&zone_deathrow, zone);
5371 5379 mutex_exit(&zone_deathrow_lock);
5372 5380
5373 5381 /*
5374 5382 * Drop last reference (which was added by zsched()), this will
5375 5383 * free the zone unless there are outstanding cred references.
5376 5384 */
5377 5385 zone_rele(zone);
5378 5386 return (0);
5379 5387 }
5380 5388
5381 5389 /*
5382 5390 * Systemcall entry point for zone_getattr(2).
5383 5391 */
5384 5392 static ssize_t
5385 5393 zone_getattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
5386 5394 {
5387 5395 size_t size;
5388 5396 int error = 0, err;
5389 5397 zone_t *zone;
5390 5398 char *zonepath;
5391 5399 char *outstr;
5392 5400 zone_status_t zone_status;
5393 5401 pid_t initpid;
5394 5402 boolean_t global = (curzone == global_zone);
5395 5403 boolean_t inzone = (curzone->zone_id == zoneid);
5396 5404 ushort_t flags;
5397 5405 zone_net_data_t *zbuf;
5398 5406
5399 5407 mutex_enter(&zonehash_lock);
5400 5408 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5401 5409 mutex_exit(&zonehash_lock);
5402 5410 return (set_errno(EINVAL));
5403 5411 }
5404 5412 zone_status = zone_status_get(zone);
5405 5413 if (zone_status < ZONE_IS_INITIALIZED) {
5406 5414 mutex_exit(&zonehash_lock);
5407 5415 return (set_errno(EINVAL));
5408 5416 }
5409 5417 zone_hold(zone);
5410 5418 mutex_exit(&zonehash_lock);
5411 5419
5412 5420 /*
5413 5421 * If not in the global zone, don't show information about other zones,
5414 5422 * unless the system is labeled and the local zone's label dominates
5415 5423 * the other zone.
5416 5424 */
5417 5425 if (!zone_list_access(zone)) {
5418 5426 zone_rele(zone);
5419 5427 return (set_errno(EINVAL));
5420 5428 }
5421 5429
5422 5430 switch (attr) {
5423 5431 case ZONE_ATTR_ROOT:
5424 5432 if (global) {
5425 5433 /*
5426 5434 * Copy the path to trim the trailing "/" (except for
5427 5435 * the global zone).
5428 5436 */
5429 5437 if (zone != global_zone)
5430 5438 size = zone->zone_rootpathlen - 1;
5431 5439 else
5432 5440 size = zone->zone_rootpathlen;
5433 5441 zonepath = kmem_alloc(size, KM_SLEEP);
5434 5442 bcopy(zone->zone_rootpath, zonepath, size);
5435 5443 zonepath[size - 1] = '\0';
5436 5444 } else {
5437 5445 if (inzone || !is_system_labeled()) {
5438 5446 /*
5439 5447 * Caller is not in the global zone.
5440 5448 * if the query is on the current zone
5441 5449 * or the system is not labeled,
5442 5450 * just return faked-up path for current zone.
5443 5451 */
5444 5452 zonepath = "/";
5445 5453 size = 2;
5446 5454 } else {
5447 5455 /*
5448 5456 * Return related path for current zone.
5449 5457 */
5450 5458 int prefix_len = strlen(zone_prefix);
5451 5459 int zname_len = strlen(zone->zone_name);
5452 5460
5453 5461 size = prefix_len + zname_len + 1;
5454 5462 zonepath = kmem_alloc(size, KM_SLEEP);
5455 5463 bcopy(zone_prefix, zonepath, prefix_len);
5456 5464 bcopy(zone->zone_name, zonepath +
5457 5465 prefix_len, zname_len);
5458 5466 zonepath[size - 1] = '\0';
5459 5467 }
5460 5468 }
5461 5469 if (bufsize > size)
5462 5470 bufsize = size;
5463 5471 if (buf != NULL) {
5464 5472 err = copyoutstr(zonepath, buf, bufsize, NULL);
5465 5473 if (err != 0 && err != ENAMETOOLONG)
5466 5474 error = EFAULT;
5467 5475 }
5468 5476 if (global || (is_system_labeled() && !inzone))
5469 5477 kmem_free(zonepath, size);
5470 5478 break;
5471 5479
5472 5480 case ZONE_ATTR_NAME:
5473 5481 size = strlen(zone->zone_name) + 1;
5474 5482 if (bufsize > size)
5475 5483 bufsize = size;
5476 5484 if (buf != NULL) {
5477 5485 err = copyoutstr(zone->zone_name, buf, bufsize, NULL);
5478 5486 if (err != 0 && err != ENAMETOOLONG)
5479 5487 error = EFAULT;
5480 5488 }
5481 5489 break;
5482 5490
5483 5491 case ZONE_ATTR_STATUS:
5484 5492 /*
5485 5493 * Since we're not holding zonehash_lock, the zone status
5486 5494 * may be anything; leave it up to userland to sort it out.
5487 5495 */
5488 5496 size = sizeof (zone_status);
5489 5497 if (bufsize > size)
5490 5498 bufsize = size;
5491 5499 zone_status = zone_status_get(zone);
5492 5500 if (buf != NULL &&
5493 5501 copyout(&zone_status, buf, bufsize) != 0)
5494 5502 error = EFAULT;
5495 5503 break;
5496 5504 case ZONE_ATTR_FLAGS:
5497 5505 size = sizeof (zone->zone_flags);
5498 5506 if (bufsize > size)
5499 5507 bufsize = size;
5500 5508 flags = zone->zone_flags;
5501 5509 if (buf != NULL &&
5502 5510 copyout(&flags, buf, bufsize) != 0)
5503 5511 error = EFAULT;
5504 5512 break;
5505 5513 case ZONE_ATTR_PRIVSET:
5506 5514 size = sizeof (priv_set_t);
5507 5515 if (bufsize > size)
5508 5516 bufsize = size;
5509 5517 if (buf != NULL &&
5510 5518 copyout(zone->zone_privset, buf, bufsize) != 0)
5511 5519 error = EFAULT;
5512 5520 break;
5513 5521 case ZONE_ATTR_UNIQID:
5514 5522 size = sizeof (zone->zone_uniqid);
5515 5523 if (bufsize > size)
5516 5524 bufsize = size;
5517 5525 if (buf != NULL &&
5518 5526 copyout(&zone->zone_uniqid, buf, bufsize) != 0)
5519 5527 error = EFAULT;
5520 5528 break;
5521 5529 case ZONE_ATTR_POOLID:
5522 5530 {
5523 5531 pool_t *pool;
5524 5532 poolid_t poolid;
5525 5533
5526 5534 if (pool_lock_intr() != 0) {
5527 5535 error = EINTR;
5528 5536 break;
5529 5537 }
5530 5538 pool = zone_pool_get(zone);
5531 5539 poolid = pool->pool_id;
5532 5540 pool_unlock();
5533 5541 size = sizeof (poolid);
5534 5542 if (bufsize > size)
5535 5543 bufsize = size;
5536 5544 if (buf != NULL && copyout(&poolid, buf, size) != 0)
5537 5545 error = EFAULT;
5538 5546 }
5539 5547 break;
5540 5548 case ZONE_ATTR_SLBL:
5541 5549 size = sizeof (bslabel_t);
5542 5550 if (bufsize > size)
5543 5551 bufsize = size;
5544 5552 if (zone->zone_slabel == NULL)
5545 5553 error = EINVAL;
5546 5554 else if (buf != NULL &&
5547 5555 copyout(label2bslabel(zone->zone_slabel), buf,
5548 5556 bufsize) != 0)
5549 5557 error = EFAULT;
5550 5558 break;
5551 5559 case ZONE_ATTR_INITPID:
5552 5560 size = sizeof (initpid);
5553 5561 if (bufsize > size)
5554 5562 bufsize = size;
5555 5563 initpid = zone->zone_proc_initpid;
5556 5564 if (initpid == -1) {
5557 5565 error = ESRCH;
5558 5566 break;
5559 5567 }
5560 5568 if (buf != NULL &&
5561 5569 copyout(&initpid, buf, bufsize) != 0)
5562 5570 error = EFAULT;
5563 5571 break;
5564 5572 case ZONE_ATTR_BRAND:
5565 5573 size = strlen(zone->zone_brand->b_name) + 1;
5566 5574
5567 5575 if (bufsize > size)
5568 5576 bufsize = size;
5569 5577 if (buf != NULL) {
5570 5578 err = copyoutstr(zone->zone_brand->b_name, buf,
5571 5579 bufsize, NULL);
5572 5580 if (err != 0 && err != ENAMETOOLONG)
5573 5581 error = EFAULT;
5574 5582 }
5575 5583 break;
5576 5584 case ZONE_ATTR_INITNAME:
5577 5585 size = strlen(zone->zone_initname) + 1;
5578 5586 if (bufsize > size)
5579 5587 bufsize = size;
5580 5588 if (buf != NULL) {
5581 5589 err = copyoutstr(zone->zone_initname, buf, bufsize,
5582 5590 NULL);
5583 5591 if (err != 0 && err != ENAMETOOLONG)
5584 5592 error = EFAULT;
5585 5593 }
5586 5594 break;
5587 5595 case ZONE_ATTR_BOOTARGS:
5588 5596 if (zone->zone_bootargs == NULL)
5589 5597 outstr = "";
5590 5598 else
5591 5599 outstr = zone->zone_bootargs;
5592 5600 size = strlen(outstr) + 1;
5593 5601 if (bufsize > size)
5594 5602 bufsize = size;
5595 5603 if (buf != NULL) {
5596 5604 err = copyoutstr(outstr, buf, bufsize, NULL);
5597 5605 if (err != 0 && err != ENAMETOOLONG)
5598 5606 error = EFAULT;
5599 5607 }
5600 5608 break;
5601 5609 case ZONE_ATTR_PHYS_MCAP:
5602 5610 size = sizeof (zone->zone_phys_mcap);
5603 5611 if (bufsize > size)
5604 5612 bufsize = size;
5605 5613 if (buf != NULL &&
5606 5614 copyout(&zone->zone_phys_mcap, buf, bufsize) != 0)
5607 5615 error = EFAULT;
5608 5616 break;
5609 5617 case ZONE_ATTR_SCHED_CLASS:
5610 5618 mutex_enter(&class_lock);
5611 5619
5612 5620 if (zone->zone_defaultcid >= loaded_classes)
5613 5621 outstr = "";
5614 5622 else
5615 5623 outstr = sclass[zone->zone_defaultcid].cl_name;
5616 5624 size = strlen(outstr) + 1;
5617 5625 if (bufsize > size)
5618 5626 bufsize = size;
5619 5627 if (buf != NULL) {
5620 5628 err = copyoutstr(outstr, buf, bufsize, NULL);
5621 5629 if (err != 0 && err != ENAMETOOLONG)
5622 5630 error = EFAULT;
5623 5631 }
5624 5632
5625 5633 mutex_exit(&class_lock);
5626 5634 break;
5627 5635 case ZONE_ATTR_HOSTID:
5628 5636 if (zone->zone_hostid != HW_INVALID_HOSTID &&
5629 5637 bufsize == sizeof (zone->zone_hostid)) {
5630 5638 size = sizeof (zone->zone_hostid);
5631 5639 if (buf != NULL && copyout(&zone->zone_hostid, buf,
5632 5640 bufsize) != 0)
5633 5641 error = EFAULT;
5634 5642 } else {
5635 5643 error = EINVAL;
5636 5644 }
5637 5645 break;
5638 5646 case ZONE_ATTR_FS_ALLOWED:
5639 5647 if (zone->zone_fs_allowed == NULL)
5640 5648 outstr = "";
5641 5649 else
5642 5650 outstr = zone->zone_fs_allowed;
5643 5651 size = strlen(outstr) + 1;
5644 5652 if (bufsize > size)
5645 5653 bufsize = size;
5646 5654 if (buf != NULL) {
5647 5655 err = copyoutstr(outstr, buf, bufsize, NULL);
5648 5656 if (err != 0 && err != ENAMETOOLONG)
5649 5657 error = EFAULT;
5650 5658 }
5651 5659 break;
5652 5660 case ZONE_ATTR_SECFLAGS:
5653 5661 size = sizeof (zone->zone_secflags);
5654 5662 if (bufsize > size)
5655 5663 bufsize = size;
5656 5664 if ((err = copyout(&zone->zone_secflags, buf, bufsize)) != 0)
5657 5665 error = EFAULT;
5658 5666 break;
5659 5667 case ZONE_ATTR_NETWORK:
5660 5668 bufsize = MIN(bufsize, PIPE_BUF + sizeof (zone_net_data_t));
5661 5669 size = bufsize;
5662 5670 zbuf = kmem_alloc(bufsize, KM_SLEEP);
5663 5671 if (copyin(buf, zbuf, bufsize) != 0) {
5664 5672 error = EFAULT;
5665 5673 } else {
5666 5674 error = zone_get_network(zoneid, zbuf);
5667 5675 if (error == 0 && copyout(zbuf, buf, bufsize) != 0)
5668 5676 error = EFAULT;
5669 5677 }
5670 5678 kmem_free(zbuf, bufsize);
5671 5679 break;
5672 5680 default:
5673 5681 if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone)) {
5674 5682 size = bufsize;
5675 5683 error = ZBROP(zone)->b_getattr(zone, attr, buf, &size);
5676 5684 } else {
5677 5685 error = EINVAL;
5678 5686 }
5679 5687 }
5680 5688 zone_rele(zone);
5681 5689
5682 5690 if (error)
5683 5691 return (set_errno(error));
5684 5692 return ((ssize_t)size);
5685 5693 }
5686 5694
5687 5695 /*
5688 5696 * Systemcall entry point for zone_setattr(2).
5689 5697 */
5690 5698 /*ARGSUSED*/
5691 5699 static int
5692 5700 zone_setattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize)
5693 5701 {
5694 5702 zone_t *zone;
5695 5703 zone_status_t zone_status;
5696 5704 int err = -1;
5697 5705 zone_net_data_t *zbuf;
5698 5706
5699 5707 if (secpolicy_zone_config(CRED()) != 0)
5700 5708 return (set_errno(EPERM));
5701 5709
5702 5710 /*
5703 5711 * Only the ZONE_ATTR_PHYS_MCAP attribute can be set on the
5704 5712 * global zone.
5705 5713 */
5706 5714 if (zoneid == GLOBAL_ZONEID && attr != ZONE_ATTR_PHYS_MCAP) {
5707 5715 return (set_errno(EINVAL));
5708 5716 }
5709 5717
5710 5718 mutex_enter(&zonehash_lock);
5711 5719 if ((zone = zone_find_all_by_id(zoneid)) == NULL) {
5712 5720 mutex_exit(&zonehash_lock);
5713 5721 return (set_errno(EINVAL));
5714 5722 }
5715 5723 zone_hold(zone);
5716 5724 mutex_exit(&zonehash_lock);
5717 5725
5718 5726 /*
5719 5727 * At present most attributes can only be set on non-running,
5720 5728 * non-global zones.
5721 5729 */
5722 5730 zone_status = zone_status_get(zone);
5723 5731 if (attr != ZONE_ATTR_PHYS_MCAP && zone_status > ZONE_IS_READY) {
5724 5732 err = EINVAL;
5725 5733 goto done;
5726 5734 }
5727 5735
5728 5736 switch (attr) {
5729 5737 case ZONE_ATTR_INITNAME:
5730 5738 err = zone_set_initname(zone, (const char *)buf);
5731 5739 break;
5732 5740 case ZONE_ATTR_INITNORESTART:
5733 5741 zone->zone_restart_init = B_FALSE;
5734 5742 err = 0;
5735 5743 break;
5736 5744 case ZONE_ATTR_BOOTARGS:
5737 5745 err = zone_set_bootargs(zone, (const char *)buf);
5738 5746 break;
5739 5747 case ZONE_ATTR_BRAND:
5740 5748 err = zone_set_brand(zone, (const char *)buf);
5741 5749 break;
5742 5750 case ZONE_ATTR_FS_ALLOWED:
5743 5751 err = zone_set_fs_allowed(zone, (const char *)buf);
5744 5752 break;
5745 5753 case ZONE_ATTR_SECFLAGS:
5746 5754 err = zone_set_secflags(zone, (psecflags_t *)buf);
5747 5755 break;
5748 5756 case ZONE_ATTR_PHYS_MCAP:
5749 5757 err = zone_set_phys_mcap(zone, (const uint64_t *)buf);
5750 5758 break;
5751 5759 case ZONE_ATTR_SCHED_CLASS:
5752 5760 err = zone_set_sched_class(zone, (const char *)buf);
5753 5761 break;
5754 5762 case ZONE_ATTR_HOSTID:
5755 5763 if (bufsize == sizeof (zone->zone_hostid)) {
5756 5764 if (copyin(buf, &zone->zone_hostid, bufsize) == 0)
5757 5765 err = 0;
5758 5766 else
5759 5767 err = EFAULT;
5760 5768 } else {
5761 5769 err = EINVAL;
5762 5770 }
5763 5771 break;
5764 5772 case ZONE_ATTR_NETWORK:
5765 5773 if (bufsize > (PIPE_BUF + sizeof (zone_net_data_t))) {
5766 5774 err = EINVAL;
5767 5775 break;
5768 5776 }
5769 5777 zbuf = kmem_alloc(bufsize, KM_SLEEP);
5770 5778 if (copyin(buf, zbuf, bufsize) != 0) {
5771 5779 kmem_free(zbuf, bufsize);
5772 5780 err = EFAULT;
5773 5781 break;
5774 5782 }
5775 5783 err = zone_set_network(zoneid, zbuf);
5776 5784 kmem_free(zbuf, bufsize);
5777 5785 break;
5778 5786 default:
5779 5787 if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone))
5780 5788 err = ZBROP(zone)->b_setattr(zone, attr, buf, bufsize);
5781 5789 else
5782 5790 err = EINVAL;
5783 5791 }
5784 5792
5785 5793 done:
5786 5794 zone_rele(zone);
5787 5795 ASSERT(err != -1);
5788 5796 return (err != 0 ? set_errno(err) : 0);
5789 5797 }
5790 5798
5791 5799 /*
5792 5800 * Return zero if the process has at least one vnode mapped in to its
5793 5801 * address space which shouldn't be allowed to change zones.
5794 5802 *
5795 5803 * Also return zero if the process has any shared mappings which reserve
5796 5804 * swap. This is because the counting for zone.max-swap does not allow swap
5797 5805 * reservation to be shared between zones. zone swap reservation is counted
5798 5806 * on zone->zone_max_swap.
5799 5807 */
5800 5808 static int
5801 5809 as_can_change_zones(void)
5802 5810 {
5803 5811 proc_t *pp = curproc;
5804 5812 struct seg *seg;
5805 5813 struct as *as = pp->p_as;
5806 5814 vnode_t *vp;
5807 5815 int allow = 1;
5808 5816
5809 5817 ASSERT(pp->p_as != &kas);
5810 5818 AS_LOCK_ENTER(as, RW_READER);
5811 5819 for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) {
5812 5820
5813 5821 /*
5814 5822 * Cannot enter zone with shared anon memory which
5815 5823 * reserves swap. See comment above.
5816 5824 */
5817 5825 if (seg_can_change_zones(seg) == B_FALSE) {
5818 5826 allow = 0;
5819 5827 break;
5820 5828 }
5821 5829 /*
5822 5830 * if we can't get a backing vnode for this segment then skip
5823 5831 * it.
5824 5832 */
5825 5833 vp = NULL;
5826 5834 if (SEGOP_GETVP(seg, seg->s_base, &vp) != 0 || vp == NULL)
5827 5835 continue;
5828 5836 if (!vn_can_change_zones(vp)) { /* bail on first match */
5829 5837 allow = 0;
5830 5838 break;
5831 5839 }
5832 5840 }
5833 5841 AS_LOCK_EXIT(as);
5834 5842 return (allow);
5835 5843 }
5836 5844
5837 5845 /*
5838 5846 * Count swap reserved by curproc's address space
5839 5847 */
5840 5848 static size_t
5841 5849 as_swresv(void)
5842 5850 {
5843 5851 proc_t *pp = curproc;
5844 5852 struct seg *seg;
5845 5853 struct as *as = pp->p_as;
5846 5854 size_t swap = 0;
5847 5855
5848 5856 ASSERT(pp->p_as != &kas);
5849 5857 ASSERT(AS_WRITE_HELD(as));
5850 5858 for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg))
5851 5859 swap += seg_swresv(seg);
5852 5860
5853 5861 return (swap);
5854 5862 }
5855 5863
5856 5864 /*
5857 5865 * Systemcall entry point for zone_enter().
5858 5866 *
5859 5867 * The current process is injected into said zone. In the process
5860 5868 * it will change its project membership, privileges, rootdir/cwd,
5861 5869 * zone-wide rctls, and pool association to match those of the zone.
5862 5870 *
5863 5871 * The first zone_enter() called while the zone is in the ZONE_IS_READY
5864 5872 * state will transition it to ZONE_IS_RUNNING. Processes may only
5865 5873 * enter a zone that is "ready" or "running".
5866 5874 */
5867 5875 static int
5868 5876 zone_enter(zoneid_t zoneid)
5869 5877 {
5870 5878 zone_t *zone;
5871 5879 vnode_t *vp;
5872 5880 proc_t *pp = curproc;
5873 5881 contract_t *ct;
5874 5882 cont_process_t *ctp;
5875 5883 task_t *tk, *oldtk;
5876 5884 kproject_t *zone_proj0;
5877 5885 cred_t *cr, *newcr;
5878 5886 pool_t *oldpool, *newpool;
5879 5887 sess_t *sp;
5880 5888 uid_t uid;
5881 5889 zone_status_t status;
5882 5890 int err = 0;
5883 5891 rctl_entity_p_t e;
5884 5892 size_t swap;
5885 5893 kthread_id_t t;
5886 5894
5887 5895 if (secpolicy_zone_config(CRED()) != 0)
5888 5896 return (set_errno(EPERM));
5889 5897 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID)
5890 5898 return (set_errno(EINVAL));
5891 5899
5892 5900 /*
5893 5901 * Stop all lwps so we don't need to hold a lock to look at
5894 5902 * curproc->p_zone. This needs to happen before we grab any
5895 5903 * locks to avoid deadlock (another lwp in the process could
5896 5904 * be waiting for the held lock).
5897 5905 */
5898 5906 if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK))
5899 5907 return (set_errno(EINTR));
5900 5908
5901 5909 /*
5902 5910 * Make sure we're not changing zones with files open or mapped in
5903 5911 * to our address space which shouldn't be changing zones.
5904 5912 */
5905 5913 if (!files_can_change_zones()) {
5906 5914 err = EBADF;
5907 5915 goto out;
5908 5916 }
5909 5917 if (!as_can_change_zones()) {
5910 5918 err = EFAULT;
5911 5919 goto out;
5912 5920 }
5913 5921
5914 5922 mutex_enter(&zonehash_lock);
5915 5923 if (pp->p_zone != global_zone) {
5916 5924 mutex_exit(&zonehash_lock);
5917 5925 err = EINVAL;
5918 5926 goto out;
5919 5927 }
5920 5928
5921 5929 zone = zone_find_all_by_id(zoneid);
5922 5930 if (zone == NULL) {
5923 5931 mutex_exit(&zonehash_lock);
5924 5932 err = EINVAL;
5925 5933 goto out;
5926 5934 }
5927 5935
5928 5936 /*
5929 5937 * To prevent processes in a zone from holding contracts on
5930 5938 * extrazonal resources, and to avoid process contract
5931 5939 * memberships which span zones, contract holders and processes
5932 5940 * which aren't the sole members of their encapsulating process
5933 5941 * contracts are not allowed to zone_enter.
5934 5942 */
5935 5943 ctp = pp->p_ct_process;
5936 5944 ct = &ctp->conp_contract;
5937 5945 mutex_enter(&ct->ct_lock);
5938 5946 mutex_enter(&pp->p_lock);
5939 5947 if ((avl_numnodes(&pp->p_ct_held) != 0) || (ctp->conp_nmembers != 1)) {
5940 5948 mutex_exit(&pp->p_lock);
5941 5949 mutex_exit(&ct->ct_lock);
5942 5950 mutex_exit(&zonehash_lock);
5943 5951 err = EINVAL;
5944 5952 goto out;
5945 5953 }
5946 5954
5947 5955 /*
5948 5956 * Moreover, we don't allow processes whose encapsulating
5949 5957 * process contracts have inherited extrazonal contracts.
5950 5958 * While it would be easier to eliminate all process contracts
5951 5959 * with inherited contracts, we need to be able to give a
5952 5960 * restarted init (or other zone-penetrating process) its
5953 5961 * predecessor's contracts.
5954 5962 */
5955 5963 if (ctp->conp_ninherited != 0) {
5956 5964 contract_t *next;
5957 5965 for (next = list_head(&ctp->conp_inherited); next;
5958 5966 next = list_next(&ctp->conp_inherited, next)) {
5959 5967 if (contract_getzuniqid(next) != zone->zone_uniqid) {
5960 5968 mutex_exit(&pp->p_lock);
5961 5969 mutex_exit(&ct->ct_lock);
5962 5970 mutex_exit(&zonehash_lock);
5963 5971 err = EINVAL;
5964 5972 goto out;
5965 5973 }
5966 5974 }
5967 5975 }
5968 5976
5969 5977 mutex_exit(&pp->p_lock);
5970 5978 mutex_exit(&ct->ct_lock);
5971 5979
5972 5980 status = zone_status_get(zone);
5973 5981 if (status < ZONE_IS_READY || status >= ZONE_IS_SHUTTING_DOWN) {
5974 5982 /*
5975 5983 * Can't join
5976 5984 */
5977 5985 mutex_exit(&zonehash_lock);
5978 5986 err = EINVAL;
5979 5987 goto out;
5980 5988 }
5981 5989
5982 5990 /*
5983 5991 * Make sure new priv set is within the permitted set for caller
5984 5992 */
5985 5993 if (!priv_issubset(zone->zone_privset, &CR_OPPRIV(CRED()))) {
5986 5994 mutex_exit(&zonehash_lock);
5987 5995 err = EPERM;
5988 5996 goto out;
5989 5997 }
5990 5998 /*
5991 5999 * We want to momentarily drop zonehash_lock while we optimistically
5992 6000 * bind curproc to the pool it should be running in. This is safe
5993 6001 * since the zone can't disappear (we have a hold on it).
5994 6002 */
5995 6003 zone_hold(zone);
5996 6004 mutex_exit(&zonehash_lock);
5997 6005
5998 6006 /*
5999 6007 * Grab pool_lock to keep the pools configuration from changing
6000 6008 * and to stop ourselves from getting rebound to another pool
6001 6009 * until we join the zone.
6002 6010 */
6003 6011 if (pool_lock_intr() != 0) {
6004 6012 zone_rele(zone);
6005 6013 err = EINTR;
6006 6014 goto out;
6007 6015 }
6008 6016 ASSERT(secpolicy_pool(CRED()) == 0);
6009 6017 /*
6010 6018 * Bind ourselves to the pool currently associated with the zone.
6011 6019 */
6012 6020 oldpool = curproc->p_pool;
6013 6021 newpool = zone_pool_get(zone);
6014 6022 if (pool_state == POOL_ENABLED && newpool != oldpool &&
6015 6023 (err = pool_do_bind(newpool, P_PID, P_MYID,
6016 6024 POOL_BIND_ALL)) != 0) {
6017 6025 pool_unlock();
6018 6026 zone_rele(zone);
6019 6027 goto out;
6020 6028 }
6021 6029
6022 6030 /*
6023 6031 * Grab cpu_lock now; we'll need it later when we call
6024 6032 * task_join().
6025 6033 */
6026 6034 mutex_enter(&cpu_lock);
6027 6035 mutex_enter(&zonehash_lock);
6028 6036 /*
6029 6037 * Make sure the zone hasn't moved on since we dropped zonehash_lock.
6030 6038 */
6031 6039 if (zone_status_get(zone) >= ZONE_IS_SHUTTING_DOWN) {
6032 6040 /*
6033 6041 * Can't join anymore.
6034 6042 */
6035 6043 mutex_exit(&zonehash_lock);
6036 6044 mutex_exit(&cpu_lock);
6037 6045 if (pool_state == POOL_ENABLED &&
6038 6046 newpool != oldpool)
6039 6047 (void) pool_do_bind(oldpool, P_PID, P_MYID,
6040 6048 POOL_BIND_ALL);
6041 6049 pool_unlock();
6042 6050 zone_rele(zone);
6043 6051 err = EINVAL;
6044 6052 goto out;
6045 6053 }
6046 6054
6047 6055 /*
6048 6056 * a_lock must be held while transfering locked memory and swap
6049 6057 * reservation from the global zone to the non global zone because
6050 6058 * asynchronous faults on the processes' address space can lock
6051 6059 * memory and reserve swap via MCL_FUTURE and MAP_NORESERVE
6052 6060 * segments respectively.
6053 6061 */
6054 6062 AS_LOCK_ENTER(pp->p_as, RW_WRITER);
6055 6063 swap = as_swresv();
6056 6064 mutex_enter(&pp->p_lock);
6057 6065 zone_proj0 = zone->zone_zsched->p_task->tk_proj;
6058 6066 /* verify that we do not exceed and task or lwp limits */
6059 6067 mutex_enter(&zone->zone_nlwps_lock);
6060 6068 /* add new lwps to zone and zone's proj0 */
6061 6069 zone_proj0->kpj_nlwps += pp->p_lwpcnt;
6062 6070 zone->zone_nlwps += pp->p_lwpcnt;
6063 6071 /* add 1 task to zone's proj0 */
6064 6072 zone_proj0->kpj_ntasks += 1;
6065 6073
6066 6074 zone_proj0->kpj_nprocs++;
6067 6075 zone->zone_nprocs++;
6068 6076 mutex_exit(&zone->zone_nlwps_lock);
6069 6077
6070 6078 mutex_enter(&zone->zone_mem_lock);
6071 6079 zone->zone_locked_mem += pp->p_locked_mem;
6072 6080 zone_proj0->kpj_data.kpd_locked_mem += pp->p_locked_mem;
6073 6081 zone->zone_max_swap += swap;
6074 6082 mutex_exit(&zone->zone_mem_lock);
6075 6083
6076 6084 mutex_enter(&(zone_proj0->kpj_data.kpd_crypto_lock));
6077 6085 zone_proj0->kpj_data.kpd_crypto_mem += pp->p_crypto_mem;
6078 6086 mutex_exit(&(zone_proj0->kpj_data.kpd_crypto_lock));
6079 6087
6080 6088 /* remove lwps and process from proc's old zone and old project */
6081 6089 mutex_enter(&pp->p_zone->zone_nlwps_lock);
6082 6090 pp->p_zone->zone_nlwps -= pp->p_lwpcnt;
6083 6091 pp->p_task->tk_proj->kpj_nlwps -= pp->p_lwpcnt;
6084 6092 pp->p_task->tk_proj->kpj_nprocs--;
6085 6093 pp->p_zone->zone_nprocs--;
6086 6094 mutex_exit(&pp->p_zone->zone_nlwps_lock);
6087 6095
6088 6096 mutex_enter(&pp->p_zone->zone_mem_lock);
6089 6097 pp->p_zone->zone_locked_mem -= pp->p_locked_mem;
6090 6098 pp->p_task->tk_proj->kpj_data.kpd_locked_mem -= pp->p_locked_mem;
6091 6099 pp->p_zone->zone_max_swap -= swap;
6092 6100 mutex_exit(&pp->p_zone->zone_mem_lock);
6093 6101
6094 6102 mutex_enter(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock));
6095 6103 pp->p_task->tk_proj->kpj_data.kpd_crypto_mem -= pp->p_crypto_mem;
6096 6104 mutex_exit(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock));
6097 6105
6098 6106 pp->p_flag |= SZONETOP;
6099 6107 pp->p_zone = zone;
6100 6108 mutex_exit(&pp->p_lock);
6101 6109 AS_LOCK_EXIT(pp->p_as);
6102 6110
6103 6111 /*
6104 6112 * Joining the zone cannot fail from now on.
6105 6113 *
6106 6114 * This means that a lot of the following code can be commonized and
6107 6115 * shared with zsched().
6108 6116 */
6109 6117
6110 6118 /*
6111 6119 * If the process contract fmri was inherited, we need to
6112 6120 * flag this so that any contract status will not leak
6113 6121 * extra zone information, svc_fmri in this case
6114 6122 */
6115 6123 if (ctp->conp_svc_ctid != ct->ct_id) {
6116 6124 mutex_enter(&ct->ct_lock);
6117 6125 ctp->conp_svc_zone_enter = ct->ct_id;
6118 6126 mutex_exit(&ct->ct_lock);
6119 6127 }
6120 6128
6121 6129 /*
6122 6130 * Reset the encapsulating process contract's zone.
6123 6131 */
6124 6132 ASSERT(ct->ct_mzuniqid == GLOBAL_ZONEUNIQID);
6125 6133 contract_setzuniqid(ct, zone->zone_uniqid);
6126 6134
6127 6135 /*
6128 6136 * Create a new task and associate the process with the project keyed
6129 6137 * by (projid,zoneid).
6130 6138 *
6131 6139 * We might as well be in project 0; the global zone's projid doesn't
6132 6140 * make much sense in a zone anyhow.
6133 6141 *
6134 6142 * This also increments zone_ntasks, and returns with p_lock held.
6135 6143 */
6136 6144 tk = task_create(0, zone);
6137 6145 oldtk = task_join(tk, 0);
6138 6146 mutex_exit(&cpu_lock);
6139 6147
6140 6148 /*
6141 6149 * call RCTLOP_SET functions on this proc
6142 6150 */
6143 6151 e.rcep_p.zone = zone;
6144 6152 e.rcep_t = RCENTITY_ZONE;
6145 6153 (void) rctl_set_dup(NULL, NULL, pp, &e, zone->zone_rctls, NULL,
6146 6154 RCD_CALLBACK);
6147 6155 mutex_exit(&pp->p_lock);
6148 6156
6149 6157 /*
6150 6158 * We don't need to hold any of zsched's locks here; not only do we know
6151 6159 * the process and zone aren't going away, we know its session isn't
6152 6160 * changing either.
6153 6161 *
6154 6162 * By joining zsched's session here, we mimic the behavior in the
6155 6163 * global zone of init's sid being the pid of sched. We extend this
6156 6164 * to all zlogin-like zone_enter()'ing processes as well.
6157 6165 */
6158 6166 mutex_enter(&pidlock);
6159 6167 sp = zone->zone_zsched->p_sessp;
6160 6168 sess_hold(zone->zone_zsched);
6161 6169 mutex_enter(&pp->p_lock);
6162 6170 pgexit(pp);
6163 6171 sess_rele(pp->p_sessp, B_TRUE);
6164 6172 pp->p_sessp = sp;
6165 6173 pgjoin(pp, zone->zone_zsched->p_pidp);
6166 6174
6167 6175 /*
6168 6176 * If any threads are scheduled to be placed on zone wait queue they
6169 6177 * should abandon the idea since the wait queue is changing.
6170 6178 * We need to be holding pidlock & p_lock to do this.
6171 6179 */
6172 6180 if ((t = pp->p_tlist) != NULL) {
6173 6181 do {
6174 6182 thread_lock(t);
6175 6183 /*
6176 6184 * Kick this thread so that it doesn't sit
6177 6185 * on a wrong wait queue.
6178 6186 */
6179 6187 if (ISWAITING(t))
6180 6188 setrun_locked(t);
6181 6189
6182 6190 if (t->t_schedflag & TS_ANYWAITQ)
6183 6191 t->t_schedflag &= ~ TS_ANYWAITQ;
6184 6192
6185 6193 thread_unlock(t);
6186 6194 } while ((t = t->t_forw) != pp->p_tlist);
6187 6195 }
6188 6196
6189 6197 /*
6190 6198 * If there is a default scheduling class for the zone and it is not
6191 6199 * the class we are currently in, change all of the threads in the
6192 6200 * process to the new class. We need to be holding pidlock & p_lock
6193 6201 * when we call parmsset so this is a good place to do it.
6194 6202 */
6195 6203 if (zone->zone_defaultcid > 0 &&
6196 6204 zone->zone_defaultcid != curthread->t_cid) {
6197 6205 pcparms_t pcparms;
6198 6206
6199 6207 pcparms.pc_cid = zone->zone_defaultcid;
6200 6208 pcparms.pc_clparms[0] = 0;
6201 6209
6202 6210 /*
6203 6211 * If setting the class fails, we still want to enter the zone.
6204 6212 */
6205 6213 if ((t = pp->p_tlist) != NULL) {
6206 6214 do {
6207 6215 (void) parmsset(&pcparms, t);
6208 6216 } while ((t = t->t_forw) != pp->p_tlist);
6209 6217 }
6210 6218 }
6211 6219
6212 6220 mutex_exit(&pp->p_lock);
6213 6221 mutex_exit(&pidlock);
6214 6222
6215 6223 mutex_exit(&zonehash_lock);
6216 6224 /*
6217 6225 * We're firmly in the zone; let pools progress.
6218 6226 */
6219 6227 pool_unlock();
6220 6228 task_rele(oldtk);
6221 6229 /*
6222 6230 * We don't need to retain a hold on the zone since we already
6223 6231 * incremented zone_ntasks, so the zone isn't going anywhere.
6224 6232 */
6225 6233 zone_rele(zone);
6226 6234
6227 6235 /*
6228 6236 * Chroot
6229 6237 */
6230 6238 vp = zone->zone_rootvp;
6231 6239 zone_chdir(vp, &PTOU(pp)->u_cdir, pp);
6232 6240 zone_chdir(vp, &PTOU(pp)->u_rdir, pp);
6233 6241
6234 6242 /*
6235 6243 * Change process security flags. Note that the _effective_ flags
6236 6244 * cannot change
6237 6245 */
6238 6246 secflags_copy(&pp->p_secflags.psf_lower,
6239 6247 &zone->zone_secflags.psf_lower);
6240 6248 secflags_copy(&pp->p_secflags.psf_upper,
6241 6249 &zone->zone_secflags.psf_upper);
6242 6250 secflags_copy(&pp->p_secflags.psf_inherit,
6243 6251 &zone->zone_secflags.psf_inherit);
6244 6252
6245 6253 /*
6246 6254 * Change process credentials
6247 6255 */
6248 6256 newcr = cralloc();
6249 6257 mutex_enter(&pp->p_crlock);
6250 6258 cr = pp->p_cred;
6251 6259 crcopy_to(cr, newcr);
6252 6260 crsetzone(newcr, zone);
6253 6261 pp->p_cred = newcr;
6254 6262
6255 6263 /*
6256 6264 * Restrict all process privilege sets to zone limit
6257 6265 */
6258 6266 priv_intersect(zone->zone_privset, &CR_PPRIV(newcr));
6259 6267 priv_intersect(zone->zone_privset, &CR_EPRIV(newcr));
6260 6268 priv_intersect(zone->zone_privset, &CR_IPRIV(newcr));
6261 6269 priv_intersect(zone->zone_privset, &CR_LPRIV(newcr));
6262 6270 mutex_exit(&pp->p_crlock);
6263 6271 crset(pp, newcr);
6264 6272
6265 6273 /*
6266 6274 * Adjust upcount to reflect zone entry.
6267 6275 */
6268 6276 uid = crgetruid(newcr);
6269 6277 mutex_enter(&pidlock);
6270 6278 upcount_dec(uid, GLOBAL_ZONEID);
6271 6279 upcount_inc(uid, zoneid);
6272 6280 mutex_exit(&pidlock);
6273 6281
6274 6282 /*
6275 6283 * Set up core file path and content.
6276 6284 */
6277 6285 set_core_defaults();
6278 6286
6279 6287 out:
6280 6288 /*
6281 6289 * Let the other lwps continue.
6282 6290 */
6283 6291 mutex_enter(&pp->p_lock);
6284 6292 if (curthread != pp->p_agenttp)
6285 6293 continuelwps(pp);
6286 6294 mutex_exit(&pp->p_lock);
6287 6295
6288 6296 return (err != 0 ? set_errno(err) : 0);
6289 6297 }
6290 6298
6291 6299 /*
6292 6300 * Systemcall entry point for zone_list(2).
6293 6301 *
6294 6302 * Processes running in a (non-global) zone only see themselves.
6295 6303 * On labeled systems, they see all zones whose label they dominate.
6296 6304 */
6297 6305 static int
6298 6306 zone_list(zoneid_t *zoneidlist, uint_t *numzones)
6299 6307 {
6300 6308 zoneid_t *zoneids;
6301 6309 zone_t *zone, *myzone;
6302 6310 uint_t user_nzones, real_nzones;
6303 6311 uint_t domi_nzones;
6304 6312 int error;
6305 6313
6306 6314 if (copyin(numzones, &user_nzones, sizeof (uint_t)) != 0)
6307 6315 return (set_errno(EFAULT));
6308 6316
6309 6317 myzone = curproc->p_zone;
6310 6318 if (myzone != global_zone) {
6311 6319 bslabel_t *mybslab;
6312 6320
6313 6321 if (!is_system_labeled()) {
6314 6322 /* just return current zone */
6315 6323 real_nzones = domi_nzones = 1;
6316 6324 zoneids = kmem_alloc(sizeof (zoneid_t), KM_SLEEP);
6317 6325 zoneids[0] = myzone->zone_id;
6318 6326 } else {
6319 6327 /* return all zones that are dominated */
6320 6328 mutex_enter(&zonehash_lock);
6321 6329 real_nzones = zonecount;
6322 6330 domi_nzones = 0;
6323 6331 if (real_nzones > 0) {
6324 6332 zoneids = kmem_alloc(real_nzones *
6325 6333 sizeof (zoneid_t), KM_SLEEP);
6326 6334 mybslab = label2bslabel(myzone->zone_slabel);
6327 6335 for (zone = list_head(&zone_active);
6328 6336 zone != NULL;
6329 6337 zone = list_next(&zone_active, zone)) {
6330 6338 if (zone->zone_id == GLOBAL_ZONEID)
6331 6339 continue;
6332 6340 if (zone != myzone &&
6333 6341 (zone->zone_flags & ZF_IS_SCRATCH))
6334 6342 continue;
6335 6343 /*
6336 6344 * Note that a label always dominates
6337 6345 * itself, so myzone is always included
6338 6346 * in the list.
6339 6347 */
6340 6348 if (bldominates(mybslab,
6341 6349 label2bslabel(zone->zone_slabel))) {
6342 6350 zoneids[domi_nzones++] =
6343 6351 zone->zone_id;
6344 6352 }
6345 6353 }
6346 6354 }
6347 6355 mutex_exit(&zonehash_lock);
6348 6356 }
6349 6357 } else {
6350 6358 mutex_enter(&zonehash_lock);
6351 6359 real_nzones = zonecount;
6352 6360 domi_nzones = 0;
6353 6361 if (real_nzones > 0) {
6354 6362 zoneids = kmem_alloc(real_nzones * sizeof (zoneid_t),
6355 6363 KM_SLEEP);
6356 6364 for (zone = list_head(&zone_active); zone != NULL;
6357 6365 zone = list_next(&zone_active, zone))
6358 6366 zoneids[domi_nzones++] = zone->zone_id;
6359 6367 ASSERT(domi_nzones == real_nzones);
6360 6368 }
6361 6369 mutex_exit(&zonehash_lock);
6362 6370 }
6363 6371
6364 6372 /*
6365 6373 * If user has allocated space for fewer entries than we found, then
6366 6374 * return only up to their limit. Either way, tell them exactly how
6367 6375 * many we found.
6368 6376 */
6369 6377 if (domi_nzones < user_nzones)
6370 6378 user_nzones = domi_nzones;
6371 6379 error = 0;
6372 6380 if (copyout(&domi_nzones, numzones, sizeof (uint_t)) != 0) {
6373 6381 error = EFAULT;
6374 6382 } else if (zoneidlist != NULL && user_nzones != 0) {
6375 6383 if (copyout(zoneids, zoneidlist,
6376 6384 user_nzones * sizeof (zoneid_t)) != 0)
6377 6385 error = EFAULT;
6378 6386 }
6379 6387
6380 6388 if (real_nzones > 0)
6381 6389 kmem_free(zoneids, real_nzones * sizeof (zoneid_t));
6382 6390
6383 6391 if (error != 0)
6384 6392 return (set_errno(error));
6385 6393 else
6386 6394 return (0);
6387 6395 }
6388 6396
6389 6397 /*
6390 6398 * Systemcall entry point for zone_lookup(2).
6391 6399 *
6392 6400 * Non-global zones are only able to see themselves and (on labeled systems)
6393 6401 * the zones they dominate.
6394 6402 */
6395 6403 static zoneid_t
6396 6404 zone_lookup(const char *zone_name)
6397 6405 {
6398 6406 char *kname;
6399 6407 zone_t *zone;
6400 6408 zoneid_t zoneid;
6401 6409 int err;
6402 6410
6403 6411 if (zone_name == NULL) {
6404 6412 /* return caller's zone id */
6405 6413 return (getzoneid());
6406 6414 }
6407 6415
6408 6416 kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP);
6409 6417 if ((err = copyinstr(zone_name, kname, ZONENAME_MAX, NULL)) != 0) {
6410 6418 kmem_free(kname, ZONENAME_MAX);
6411 6419 return (set_errno(err));
6412 6420 }
6413 6421
6414 6422 mutex_enter(&zonehash_lock);
6415 6423 zone = zone_find_all_by_name(kname);
6416 6424 kmem_free(kname, ZONENAME_MAX);
6417 6425 /*
6418 6426 * In a non-global zone, can only lookup global and own name.
6419 6427 * In Trusted Extensions zone label dominance rules apply.
6420 6428 */
6421 6429 if (zone == NULL ||
6422 6430 zone_status_get(zone) < ZONE_IS_READY ||
6423 6431 !zone_list_access(zone)) {
6424 6432 mutex_exit(&zonehash_lock);
6425 6433 return (set_errno(EINVAL));
6426 6434 } else {
6427 6435 zoneid = zone->zone_id;
6428 6436 mutex_exit(&zonehash_lock);
6429 6437 return (zoneid);
6430 6438 }
6431 6439 }
6432 6440
6433 6441 static int
6434 6442 zone_version(int *version_arg)
6435 6443 {
6436 6444 int version = ZONE_SYSCALL_API_VERSION;
6437 6445
6438 6446 if (copyout(&version, version_arg, sizeof (int)) != 0)
6439 6447 return (set_errno(EFAULT));
6440 6448 return (0);
6441 6449 }
6442 6450
6443 6451 /* ARGSUSED */
6444 6452 long
6445 6453 zone(int cmd, void *arg1, void *arg2, void *arg3, void *arg4)
6446 6454 {
6447 6455 zone_def zs;
6448 6456 int err;
6449 6457
6450 6458 switch (cmd) {
6451 6459 case ZONE_CREATE:
6452 6460 if (get_udatamodel() == DATAMODEL_NATIVE) {
6453 6461 if (copyin(arg1, &zs, sizeof (zone_def))) {
6454 6462 return (set_errno(EFAULT));
6455 6463 }
6456 6464 } else {
6457 6465 #ifdef _SYSCALL32_IMPL
6458 6466 zone_def32 zs32;
6459 6467
6460 6468 if (copyin(arg1, &zs32, sizeof (zone_def32))) {
6461 6469 return (set_errno(EFAULT));
6462 6470 }
6463 6471 zs.zone_name =
6464 6472 (const char *)(unsigned long)zs32.zone_name;
6465 6473 zs.zone_root =
6466 6474 (const char *)(unsigned long)zs32.zone_root;
6467 6475 zs.zone_privs =
6468 6476 (const struct priv_set *)
6469 6477 (unsigned long)zs32.zone_privs;
6470 6478 zs.zone_privssz = zs32.zone_privssz;
6471 6479 zs.rctlbuf = (caddr_t)(unsigned long)zs32.rctlbuf;
6472 6480 zs.rctlbufsz = zs32.rctlbufsz;
6473 6481 zs.zfsbuf = (caddr_t)(unsigned long)zs32.zfsbuf;
6474 6482 zs.zfsbufsz = zs32.zfsbufsz;
6475 6483 zs.extended_error =
6476 6484 (int *)(unsigned long)zs32.extended_error;
6477 6485 zs.match = zs32.match;
6478 6486 zs.doi = zs32.doi;
6479 6487 zs.label = (const bslabel_t *)(uintptr_t)zs32.label;
6480 6488 zs.flags = zs32.flags;
6481 6489 #else
6482 6490 panic("get_udatamodel() returned bogus result\n");
6483 6491 #endif
6484 6492 }
6485 6493
6486 6494 return (zone_create(zs.zone_name, zs.zone_root,
6487 6495 zs.zone_privs, zs.zone_privssz,
6488 6496 (caddr_t)zs.rctlbuf, zs.rctlbufsz,
6489 6497 (caddr_t)zs.zfsbuf, zs.zfsbufsz,
6490 6498 zs.extended_error, zs.match, zs.doi,
6491 6499 zs.label, zs.flags));
6492 6500 case ZONE_BOOT:
6493 6501 return (zone_boot((zoneid_t)(uintptr_t)arg1));
6494 6502 case ZONE_DESTROY:
6495 6503 return (zone_destroy((zoneid_t)(uintptr_t)arg1));
6496 6504 case ZONE_GETATTR:
6497 6505 return (zone_getattr((zoneid_t)(uintptr_t)arg1,
6498 6506 (int)(uintptr_t)arg2, arg3, (size_t)arg4));
6499 6507 case ZONE_SETATTR:
6500 6508 return (zone_setattr((zoneid_t)(uintptr_t)arg1,
6501 6509 (int)(uintptr_t)arg2, arg3, (size_t)arg4));
6502 6510 case ZONE_ENTER:
6503 6511 return (zone_enter((zoneid_t)(uintptr_t)arg1));
6504 6512 case ZONE_LIST:
6505 6513 return (zone_list((zoneid_t *)arg1, (uint_t *)arg2));
6506 6514 case ZONE_SHUTDOWN:
6507 6515 return (zone_shutdown((zoneid_t)(uintptr_t)arg1));
6508 6516 case ZONE_LOOKUP:
6509 6517 return (zone_lookup((const char *)arg1));
6510 6518 case ZONE_VERSION:
6511 6519 return (zone_version((int *)arg1));
6512 6520 case ZONE_ADD_DATALINK:
6513 6521 return (zone_add_datalink((zoneid_t)(uintptr_t)arg1,
6514 6522 (datalink_id_t)(uintptr_t)arg2));
6515 6523 case ZONE_DEL_DATALINK:
6516 6524 return (zone_remove_datalink((zoneid_t)(uintptr_t)arg1,
6517 6525 (datalink_id_t)(uintptr_t)arg2));
6518 6526 case ZONE_CHECK_DATALINK: {
6519 6527 zoneid_t zoneid;
6520 6528 boolean_t need_copyout;
6521 6529
6522 6530 if (copyin(arg1, &zoneid, sizeof (zoneid)) != 0)
6523 6531 return (EFAULT);
6524 6532 need_copyout = (zoneid == ALL_ZONES);
6525 6533 err = zone_check_datalink(&zoneid,
6526 6534 (datalink_id_t)(uintptr_t)arg2);
6527 6535 if (err == 0 && need_copyout) {
6528 6536 if (copyout(&zoneid, arg1, sizeof (zoneid)) != 0)
6529 6537 err = EFAULT;
6530 6538 }
6531 6539 return (err == 0 ? 0 : set_errno(err));
6532 6540 }
6533 6541 case ZONE_LIST_DATALINK:
6534 6542 return (zone_list_datalink((zoneid_t)(uintptr_t)arg1,
6535 6543 (int *)arg2, (datalink_id_t *)(uintptr_t)arg3));
6536 6544 default:
6537 6545 return (set_errno(EINVAL));
6538 6546 }
6539 6547 }
6540 6548
6541 6549 struct zarg {
6542 6550 zone_t *zone;
6543 6551 zone_cmd_arg_t arg;
6544 6552 };
6545 6553
6546 6554 static int
6547 6555 zone_lookup_door(const char *zone_name, door_handle_t *doorp)
6548 6556 {
6549 6557 char *buf;
6550 6558 size_t buflen;
6551 6559 int error;
6552 6560
6553 6561 buflen = sizeof (ZONE_DOOR_PATH) + strlen(zone_name);
6554 6562 buf = kmem_alloc(buflen, KM_SLEEP);
6555 6563 (void) snprintf(buf, buflen, ZONE_DOOR_PATH, zone_name);
6556 6564 error = door_ki_open(buf, doorp);
6557 6565 kmem_free(buf, buflen);
6558 6566 return (error);
6559 6567 }
6560 6568
6561 6569 static void
6562 6570 zone_release_door(door_handle_t *doorp)
6563 6571 {
6564 6572 door_ki_rele(*doorp);
6565 6573 *doorp = NULL;
6566 6574 }
6567 6575
6568 6576 static void
6569 6577 zone_ki_call_zoneadmd(struct zarg *zargp)
6570 6578 {
6571 6579 door_handle_t door = NULL;
6572 6580 door_arg_t darg, save_arg;
6573 6581 char *zone_name;
6574 6582 size_t zone_namelen;
6575 6583 zoneid_t zoneid;
6576 6584 zone_t *zone;
6577 6585 zone_cmd_arg_t arg;
6578 6586 uint64_t uniqid;
6579 6587 size_t size;
6580 6588 int error;
6581 6589 int retry;
6582 6590
6583 6591 zone = zargp->zone;
6584 6592 arg = zargp->arg;
6585 6593 kmem_free(zargp, sizeof (*zargp));
6586 6594
6587 6595 zone_namelen = strlen(zone->zone_name) + 1;
6588 6596 zone_name = kmem_alloc(zone_namelen, KM_SLEEP);
6589 6597 bcopy(zone->zone_name, zone_name, zone_namelen);
6590 6598 zoneid = zone->zone_id;
6591 6599 uniqid = zone->zone_uniqid;
6592 6600 /*
6593 6601 * zoneadmd may be down, but at least we can empty out the zone.
6594 6602 * We can ignore the return value of zone_empty() since we're called
6595 6603 * from a kernel thread and know we won't be delivered any signals.
6596 6604 */
6597 6605 ASSERT(curproc == &p0);
6598 6606 (void) zone_empty(zone);
6599 6607 ASSERT(zone_status_get(zone) >= ZONE_IS_EMPTY);
6600 6608 zone_rele(zone);
6601 6609
6602 6610 size = sizeof (arg);
6603 6611 darg.rbuf = (char *)&arg;
6604 6612 darg.data_ptr = (char *)&arg;
6605 6613 darg.rsize = size;
6606 6614 darg.data_size = size;
6607 6615 darg.desc_ptr = NULL;
6608 6616 darg.desc_num = 0;
6609 6617
6610 6618 save_arg = darg;
6611 6619 /*
6612 6620 * Since we're not holding a reference to the zone, any number of
6613 6621 * things can go wrong, including the zone disappearing before we get a
6614 6622 * chance to talk to zoneadmd.
6615 6623 */
6616 6624 for (retry = 0; /* forever */; retry++) {
6617 6625 if (door == NULL &&
6618 6626 (error = zone_lookup_door(zone_name, &door)) != 0) {
6619 6627 goto next;
6620 6628 }
6621 6629 ASSERT(door != NULL);
6622 6630
6623 6631 if ((error = door_ki_upcall_limited(door, &darg, NULL,
6624 6632 SIZE_MAX, 0)) == 0) {
6625 6633 break;
6626 6634 }
6627 6635 switch (error) {
6628 6636 case EINTR:
6629 6637 /* FALLTHROUGH */
6630 6638 case EAGAIN: /* process may be forking */
6631 6639 /*
6632 6640 * Back off for a bit
6633 6641 */
6634 6642 break;
6635 6643 case EBADF:
6636 6644 zone_release_door(&door);
6637 6645 if (zone_lookup_door(zone_name, &door) != 0) {
6638 6646 /*
6639 6647 * zoneadmd may be dead, but it may come back to
6640 6648 * life later.
6641 6649 */
6642 6650 break;
6643 6651 }
6644 6652 break;
6645 6653 default:
6646 6654 cmn_err(CE_WARN,
6647 6655 "zone_ki_call_zoneadmd: door_ki_upcall error %d\n",
6648 6656 error);
6649 6657 goto out;
6650 6658 }
6651 6659 next:
6652 6660 /*
6653 6661 * If this isn't the same zone_t that we originally had in mind,
6654 6662 * then this is the same as if two kadmin requests come in at
6655 6663 * the same time: the first one wins. This means we lose, so we
6656 6664 * bail.
6657 6665 */
6658 6666 if ((zone = zone_find_by_id(zoneid)) == NULL) {
6659 6667 /*
6660 6668 * Problem is solved.
6661 6669 */
6662 6670 break;
6663 6671 }
6664 6672 if (zone->zone_uniqid != uniqid) {
6665 6673 /*
6666 6674 * zoneid recycled
6667 6675 */
6668 6676 zone_rele(zone);
6669 6677 break;
6670 6678 }
6671 6679 /*
6672 6680 * We could zone_status_timedwait(), but there doesn't seem to
6673 6681 * be much point in doing that (plus, it would mean that
6674 6682 * zone_free() isn't called until this thread exits).
6675 6683 */
6676 6684 zone_rele(zone);
6677 6685 delay(hz);
6678 6686 darg = save_arg;
6679 6687 }
6680 6688 out:
6681 6689 if (door != NULL) {
6682 6690 zone_release_door(&door);
6683 6691 }
6684 6692 kmem_free(zone_name, zone_namelen);
6685 6693 thread_exit();
6686 6694 }
6687 6695
6688 6696 /*
6689 6697 * Entry point for uadmin() to tell the zone to go away or reboot. Analog to
6690 6698 * kadmin(). The caller is a process in the zone.
6691 6699 *
6692 6700 * In order to shutdown the zone, we will hand off control to zoneadmd
6693 6701 * (running in the global zone) via a door. We do a half-hearted job at
6694 6702 * killing all processes in the zone, create a kernel thread to contact
6695 6703 * zoneadmd, and make note of the "uniqid" of the zone. The uniqid is
6696 6704 * a form of generation number used to let zoneadmd (as well as
6697 6705 * zone_destroy()) know exactly which zone they're re talking about.
6698 6706 */
6699 6707 int
6700 6708 zone_kadmin(int cmd, int fcn, const char *mdep, cred_t *credp)
6701 6709 {
6702 6710 struct zarg *zargp;
6703 6711 zone_cmd_t zcmd;
6704 6712 zone_t *zone;
6705 6713
6706 6714 zone = curproc->p_zone;
6707 6715 ASSERT(getzoneid() != GLOBAL_ZONEID);
6708 6716
6709 6717 switch (cmd) {
6710 6718 case A_SHUTDOWN:
6711 6719 switch (fcn) {
6712 6720 case AD_HALT:
6713 6721 case AD_POWEROFF:
6714 6722 zcmd = Z_HALT;
6715 6723 break;
6716 6724 case AD_BOOT:
6717 6725 zcmd = Z_REBOOT;
6718 6726 break;
6719 6727 case AD_IBOOT:
6720 6728 case AD_SBOOT:
6721 6729 case AD_SIBOOT:
6722 6730 case AD_NOSYNC:
6723 6731 return (ENOTSUP);
6724 6732 default:
6725 6733 return (EINVAL);
6726 6734 }
6727 6735 break;
6728 6736 case A_REBOOT:
6729 6737 zcmd = Z_REBOOT;
6730 6738 break;
6731 6739 case A_FTRACE:
6732 6740 case A_REMOUNT:
6733 6741 case A_FREEZE:
6734 6742 case A_DUMP:
6735 6743 case A_CONFIG:
6736 6744 return (ENOTSUP);
6737 6745 default:
6738 6746 ASSERT(cmd != A_SWAPCTL); /* handled by uadmin() */
6739 6747 return (EINVAL);
6740 6748 }
6741 6749
6742 6750 if (secpolicy_zone_admin(credp, B_FALSE))
6743 6751 return (EPERM);
6744 6752 mutex_enter(&zone_status_lock);
6745 6753
6746 6754 /*
6747 6755 * zone_status can't be ZONE_IS_EMPTY or higher since curproc
6748 6756 * is in the zone.
6749 6757 */
6750 6758 ASSERT(zone_status_get(zone) < ZONE_IS_EMPTY);
6751 6759 if (zone_status_get(zone) > ZONE_IS_RUNNING) {
6752 6760 /*
6753 6761 * This zone is already on its way down.
6754 6762 */
6755 6763 mutex_exit(&zone_status_lock);
6756 6764 return (0);
6757 6765 }
6758 6766 /*
6759 6767 * Prevent future zone_enter()s
6760 6768 */
6761 6769 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN);
6762 6770 mutex_exit(&zone_status_lock);
6763 6771
6764 6772 /*
6765 6773 * Kill everyone now and call zoneadmd later.
6766 6774 * zone_ki_call_zoneadmd() will do a more thorough job of this
6767 6775 * later.
6768 6776 */
6769 6777 killall(zone->zone_id);
6770 6778 /*
6771 6779 * Now, create the thread to contact zoneadmd and do the rest of the
6772 6780 * work. This thread can't be created in our zone otherwise
6773 6781 * zone_destroy() would deadlock.
6774 6782 */
6775 6783 zargp = kmem_zalloc(sizeof (*zargp), KM_SLEEP);
6776 6784 zargp->arg.cmd = zcmd;
6777 6785 zargp->arg.uniqid = zone->zone_uniqid;
6778 6786 zargp->zone = zone;
6779 6787 (void) strcpy(zargp->arg.locale, "C");
6780 6788 /* mdep was already copied in for us by uadmin */
6781 6789 if (mdep != NULL)
6782 6790 (void) strlcpy(zargp->arg.bootbuf, mdep,
6783 6791 sizeof (zargp->arg.bootbuf));
6784 6792 zone_hold(zone);
6785 6793
6786 6794 (void) thread_create(NULL, 0, zone_ki_call_zoneadmd, zargp, 0, &p0,
6787 6795 TS_RUN, minclsyspri);
6788 6796 exit(CLD_EXITED, 0);
6789 6797
6790 6798 return (EINVAL);
6791 6799 }
6792 6800
6793 6801 /*
6794 6802 * Entry point so kadmin(A_SHUTDOWN, ...) can set the global zone's
6795 6803 * status to ZONE_IS_SHUTTING_DOWN.
6796 6804 *
6797 6805 * This function also shuts down all running zones to ensure that they won't
6798 6806 * fork new processes.
6799 6807 */
6800 6808 void
6801 6809 zone_shutdown_global(void)
6802 6810 {
6803 6811 zone_t *current_zonep;
6804 6812
6805 6813 ASSERT(INGLOBALZONE(curproc));
6806 6814 mutex_enter(&zonehash_lock);
6807 6815 mutex_enter(&zone_status_lock);
6808 6816
6809 6817 /* Modify the global zone's status first. */
6810 6818 ASSERT(zone_status_get(global_zone) == ZONE_IS_RUNNING);
6811 6819 zone_status_set(global_zone, ZONE_IS_SHUTTING_DOWN);
6812 6820
6813 6821 /*
6814 6822 * Now change the states of all running zones to ZONE_IS_SHUTTING_DOWN.
6815 6823 * We don't mark all zones with ZONE_IS_SHUTTING_DOWN because doing so
6816 6824 * could cause assertions to fail (e.g., assertions about a zone's
6817 6825 * state during initialization, readying, or booting) or produce races.
6818 6826 * We'll let threads continue to initialize and ready new zones: they'll
6819 6827 * fail to boot the new zones when they see that the global zone is
6820 6828 * shutting down.
6821 6829 */
6822 6830 for (current_zonep = list_head(&zone_active); current_zonep != NULL;
6823 6831 current_zonep = list_next(&zone_active, current_zonep)) {
6824 6832 if (zone_status_get(current_zonep) == ZONE_IS_RUNNING)
6825 6833 zone_status_set(current_zonep, ZONE_IS_SHUTTING_DOWN);
6826 6834 }
6827 6835 mutex_exit(&zone_status_lock);
6828 6836 mutex_exit(&zonehash_lock);
6829 6837 }
6830 6838
6831 6839 /*
6832 6840 * Returns true if the named dataset is visible in the current zone.
6833 6841 * The 'write' parameter is set to 1 if the dataset is also writable.
6834 6842 */
6835 6843 int
6836 6844 zone_dataset_visible(const char *dataset, int *write)
6837 6845 {
6838 6846 static int zfstype = -1;
6839 6847 zone_dataset_t *zd;
6840 6848 size_t len;
6841 6849 zone_t *zone = curproc->p_zone;
6842 6850 const char *name = NULL;
6843 6851 vfs_t *vfsp = NULL;
6844 6852
6845 6853 if (dataset[0] == '\0')
6846 6854 return (0);
6847 6855
6848 6856 /*
6849 6857 * Walk the list once, looking for datasets which match exactly, or
6850 6858 * specify a dataset underneath an exported dataset. If found, return
6851 6859 * true and note that it is writable.
6852 6860 */
6853 6861 for (zd = list_head(&zone->zone_datasets); zd != NULL;
6854 6862 zd = list_next(&zone->zone_datasets, zd)) {
6855 6863
6856 6864 len = strlen(zd->zd_dataset);
6857 6865 if (strlen(dataset) >= len &&
6858 6866 bcmp(dataset, zd->zd_dataset, len) == 0 &&
6859 6867 (dataset[len] == '\0' || dataset[len] == '/' ||
6860 6868 dataset[len] == '@')) {
6861 6869 if (write)
6862 6870 *write = 1;
6863 6871 return (1);
6864 6872 }
6865 6873 }
6866 6874
6867 6875 /*
6868 6876 * Walk the list a second time, searching for datasets which are parents
6869 6877 * of exported datasets. These should be visible, but read-only.
6870 6878 *
6871 6879 * Note that we also have to support forms such as 'pool/dataset/', with
6872 6880 * a trailing slash.
6873 6881 */
6874 6882 for (zd = list_head(&zone->zone_datasets); zd != NULL;
6875 6883 zd = list_next(&zone->zone_datasets, zd)) {
6876 6884
6877 6885 len = strlen(dataset);
6878 6886 if (dataset[len - 1] == '/')
6879 6887 len--; /* Ignore trailing slash */
6880 6888 if (len < strlen(zd->zd_dataset) &&
6881 6889 bcmp(dataset, zd->zd_dataset, len) == 0 &&
6882 6890 zd->zd_dataset[len] == '/') {
6883 6891 if (write)
6884 6892 *write = 0;
6885 6893 return (1);
6886 6894 }
6887 6895 }
6888 6896
6889 6897 /*
6890 6898 * We reach here if the given dataset is not found in the zone_dataset
6891 6899 * list. Check if this dataset was added as a filesystem (ie. "add fs")
6892 6900 * instead of delegation. For this we search for the dataset in the
6893 6901 * zone_vfslist of this zone. If found, return true and note that it is
6894 6902 * not writable.
6895 6903 */
6896 6904
6897 6905 /*
6898 6906 * Initialize zfstype if it is not initialized yet.
6899 6907 */
6900 6908 if (zfstype == -1) {
6901 6909 struct vfssw *vswp = vfs_getvfssw("zfs");
6902 6910 zfstype = vswp - vfssw;
6903 6911 vfs_unrefvfssw(vswp);
6904 6912 }
6905 6913
6906 6914 vfs_list_read_lock();
6907 6915 vfsp = zone->zone_vfslist;
6908 6916 do {
6909 6917 ASSERT(vfsp);
6910 6918 if (vfsp->vfs_fstype == zfstype) {
6911 6919 name = refstr_value(vfsp->vfs_resource);
6912 6920
6913 6921 /*
6914 6922 * Check if we have an exact match.
6915 6923 */
6916 6924 if (strcmp(dataset, name) == 0) {
6917 6925 vfs_list_unlock();
6918 6926 if (write)
6919 6927 *write = 0;
6920 6928 return (1);
6921 6929 }
6922 6930 /*
6923 6931 * We need to check if we are looking for parents of
6924 6932 * a dataset. These should be visible, but read-only.
6925 6933 */
6926 6934 len = strlen(dataset);
6927 6935 if (dataset[len - 1] == '/')
6928 6936 len--;
6929 6937
6930 6938 if (len < strlen(name) &&
6931 6939 bcmp(dataset, name, len) == 0 && name[len] == '/') {
6932 6940 vfs_list_unlock();
6933 6941 if (write)
6934 6942 *write = 0;
6935 6943 return (1);
6936 6944 }
6937 6945 }
6938 6946 vfsp = vfsp->vfs_zone_next;
6939 6947 } while (vfsp != zone->zone_vfslist);
6940 6948
6941 6949 vfs_list_unlock();
6942 6950 return (0);
6943 6951 }
6944 6952
6945 6953 /*
6946 6954 * zone_find_by_any_path() -
6947 6955 *
6948 6956 * kernel-private routine similar to zone_find_by_path(), but which
6949 6957 * effectively compares against zone paths rather than zonerootpath
6950 6958 * (i.e., the last component of zonerootpaths, which should be "root/",
6951 6959 * are not compared.) This is done in order to accurately identify all
6952 6960 * paths, whether zone-visible or not, including those which are parallel
6953 6961 * to /root/, such as /dev/, /home/, etc...
6954 6962 *
6955 6963 * If the specified path does not fall under any zone path then global
6956 6964 * zone is returned.
6957 6965 *
6958 6966 * The treat_abs parameter indicates whether the path should be treated as
6959 6967 * an absolute path although it does not begin with "/". (This supports
6960 6968 * nfs mount syntax such as host:any/path.)
6961 6969 *
6962 6970 * The caller is responsible for zone_rele of the returned zone.
6963 6971 */
6964 6972 zone_t *
6965 6973 zone_find_by_any_path(const char *path, boolean_t treat_abs)
6966 6974 {
6967 6975 zone_t *zone;
6968 6976 int path_offset = 0;
6969 6977
6970 6978 if (path == NULL) {
6971 6979 zone_hold(global_zone);
6972 6980 return (global_zone);
6973 6981 }
6974 6982
6975 6983 if (*path != '/') {
6976 6984 ASSERT(treat_abs);
6977 6985 path_offset = 1;
6978 6986 }
6979 6987
6980 6988 mutex_enter(&zonehash_lock);
6981 6989 for (zone = list_head(&zone_active); zone != NULL;
6982 6990 zone = list_next(&zone_active, zone)) {
6983 6991 char *c;
6984 6992 size_t pathlen;
6985 6993 char *rootpath_start;
6986 6994
6987 6995 if (zone == global_zone) /* skip global zone */
6988 6996 continue;
6989 6997
6990 6998 /* scan backwards to find start of last component */
6991 6999 c = zone->zone_rootpath + zone->zone_rootpathlen - 2;
6992 7000 do {
6993 7001 c--;
6994 7002 } while (*c != '/');
6995 7003
6996 7004 pathlen = c - zone->zone_rootpath + 1 - path_offset;
6997 7005 rootpath_start = (zone->zone_rootpath + path_offset);
6998 7006 if (strncmp(path, rootpath_start, pathlen) == 0)
6999 7007 break;
7000 7008 }
7001 7009 if (zone == NULL)
7002 7010 zone = global_zone;
7003 7011 zone_hold(zone);
7004 7012 mutex_exit(&zonehash_lock);
7005 7013 return (zone);
7006 7014 }
7007 7015
7008 7016 /*
7009 7017 * Finds a zone_dl_t with the given linkid in the given zone. Returns the
7010 7018 * zone_dl_t pointer if found, and NULL otherwise.
7011 7019 */
7012 7020 static zone_dl_t *
7013 7021 zone_find_dl(zone_t *zone, datalink_id_t linkid)
7014 7022 {
7015 7023 zone_dl_t *zdl;
7016 7024
7017 7025 ASSERT(mutex_owned(&zone->zone_lock));
7018 7026 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7019 7027 zdl = list_next(&zone->zone_dl_list, zdl)) {
7020 7028 if (zdl->zdl_id == linkid)
7021 7029 break;
7022 7030 }
7023 7031 return (zdl);
7024 7032 }
7025 7033
7026 7034 static boolean_t
7027 7035 zone_dl_exists(zone_t *zone, datalink_id_t linkid)
7028 7036 {
7029 7037 boolean_t exists;
7030 7038
7031 7039 mutex_enter(&zone->zone_lock);
7032 7040 exists = (zone_find_dl(zone, linkid) != NULL);
7033 7041 mutex_exit(&zone->zone_lock);
7034 7042 return (exists);
7035 7043 }
7036 7044
7037 7045 /*
7038 7046 * Add an data link name for the zone.
7039 7047 */
7040 7048 static int
7041 7049 zone_add_datalink(zoneid_t zoneid, datalink_id_t linkid)
7042 7050 {
7043 7051 zone_dl_t *zdl;
7044 7052 zone_t *zone;
7045 7053 zone_t *thiszone;
7046 7054
7047 7055 if ((thiszone = zone_find_by_id(zoneid)) == NULL)
7048 7056 return (set_errno(ENXIO));
7049 7057
7050 7058 /* Verify that the datalink ID doesn't already belong to a zone. */
7051 7059 mutex_enter(&zonehash_lock);
7052 7060 for (zone = list_head(&zone_active); zone != NULL;
7053 7061 zone = list_next(&zone_active, zone)) {
7054 7062 if (zone_dl_exists(zone, linkid)) {
7055 7063 mutex_exit(&zonehash_lock);
7056 7064 zone_rele(thiszone);
7057 7065 return (set_errno((zone == thiszone) ? EEXIST : EPERM));
7058 7066 }
7059 7067 }
7060 7068
7061 7069 zdl = kmem_zalloc(sizeof (*zdl), KM_SLEEP);
7062 7070 zdl->zdl_id = linkid;
7063 7071 zdl->zdl_net = NULL;
7064 7072 mutex_enter(&thiszone->zone_lock);
7065 7073 list_insert_head(&thiszone->zone_dl_list, zdl);
7066 7074 mutex_exit(&thiszone->zone_lock);
7067 7075 mutex_exit(&zonehash_lock);
7068 7076 zone_rele(thiszone);
7069 7077 return (0);
7070 7078 }
7071 7079
7072 7080 static int
7073 7081 zone_remove_datalink(zoneid_t zoneid, datalink_id_t linkid)
7074 7082 {
7075 7083 zone_dl_t *zdl;
7076 7084 zone_t *zone;
7077 7085 int err = 0;
7078 7086
7079 7087 if ((zone = zone_find_by_id(zoneid)) == NULL)
7080 7088 return (set_errno(EINVAL));
7081 7089
7082 7090 mutex_enter(&zone->zone_lock);
7083 7091 if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
7084 7092 err = ENXIO;
7085 7093 } else {
7086 7094 list_remove(&zone->zone_dl_list, zdl);
7087 7095 nvlist_free(zdl->zdl_net);
7088 7096 kmem_free(zdl, sizeof (zone_dl_t));
7089 7097 }
7090 7098 mutex_exit(&zone->zone_lock);
7091 7099 zone_rele(zone);
7092 7100 return (err == 0 ? 0 : set_errno(err));
7093 7101 }
7094 7102
7095 7103 /*
7096 7104 * Using the zoneidp as ALL_ZONES, we can lookup which zone has been assigned
7097 7105 * the linkid. Otherwise we just check if the specified zoneidp has been
7098 7106 * assigned the supplied linkid.
7099 7107 */
7100 7108 int
7101 7109 zone_check_datalink(zoneid_t *zoneidp, datalink_id_t linkid)
7102 7110 {
7103 7111 zone_t *zone;
7104 7112 int err = ENXIO;
7105 7113
7106 7114 if (*zoneidp != ALL_ZONES) {
7107 7115 if ((zone = zone_find_by_id(*zoneidp)) != NULL) {
7108 7116 if (zone_dl_exists(zone, linkid))
7109 7117 err = 0;
7110 7118 zone_rele(zone);
7111 7119 }
7112 7120 return (err);
7113 7121 }
7114 7122
7115 7123 mutex_enter(&zonehash_lock);
7116 7124 for (zone = list_head(&zone_active); zone != NULL;
7117 7125 zone = list_next(&zone_active, zone)) {
7118 7126 if (zone_dl_exists(zone, linkid)) {
7119 7127 *zoneidp = zone->zone_id;
7120 7128 err = 0;
7121 7129 break;
7122 7130 }
7123 7131 }
7124 7132 mutex_exit(&zonehash_lock);
7125 7133 return (err);
7126 7134 }
7127 7135
7128 7136 /*
7129 7137 * Get the list of datalink IDs assigned to a zone.
7130 7138 *
7131 7139 * On input, *nump is the number of datalink IDs that can fit in the supplied
7132 7140 * idarray. Upon return, *nump is either set to the number of datalink IDs
7133 7141 * that were placed in the array if the array was large enough, or to the
7134 7142 * number of datalink IDs that the function needs to place in the array if the
7135 7143 * array is too small.
7136 7144 */
7137 7145 static int
7138 7146 zone_list_datalink(zoneid_t zoneid, int *nump, datalink_id_t *idarray)
7139 7147 {
7140 7148 uint_t num, dlcount;
7141 7149 zone_t *zone;
7142 7150 zone_dl_t *zdl;
7143 7151 datalink_id_t *idptr = idarray;
7144 7152
7145 7153 if (copyin(nump, &dlcount, sizeof (dlcount)) != 0)
7146 7154 return (set_errno(EFAULT));
7147 7155 if ((zone = zone_find_by_id(zoneid)) == NULL)
7148 7156 return (set_errno(ENXIO));
7149 7157
7150 7158 num = 0;
7151 7159 mutex_enter(&zone->zone_lock);
7152 7160 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7153 7161 zdl = list_next(&zone->zone_dl_list, zdl)) {
7154 7162 /*
7155 7163 * If the list is bigger than what the caller supplied, just
7156 7164 * count, don't do copyout.
7157 7165 */
7158 7166 if (++num > dlcount)
7159 7167 continue;
7160 7168 if (copyout(&zdl->zdl_id, idptr, sizeof (*idptr)) != 0) {
7161 7169 mutex_exit(&zone->zone_lock);
7162 7170 zone_rele(zone);
7163 7171 return (set_errno(EFAULT));
7164 7172 }
7165 7173 idptr++;
7166 7174 }
7167 7175 mutex_exit(&zone->zone_lock);
7168 7176 zone_rele(zone);
7169 7177
7170 7178 /* Increased or decreased, caller should be notified. */
7171 7179 if (num != dlcount) {
7172 7180 if (copyout(&num, nump, sizeof (num)) != 0)
7173 7181 return (set_errno(EFAULT));
7174 7182 }
7175 7183 return (0);
7176 7184 }
7177 7185
7178 7186 /*
7179 7187 * Public interface for looking up a zone by zoneid. It's a customized version
7180 7188 * for netstack_zone_create(). It can only be called from the zsd create
7181 7189 * callbacks, since it doesn't have reference on the zone structure hence if
7182 7190 * it is called elsewhere the zone could disappear after the zonehash_lock
7183 7191 * is dropped.
7184 7192 *
7185 7193 * Furthermore it
7186 7194 * 1. Doesn't check the status of the zone.
7187 7195 * 2. It will be called even before zone_init is called, in that case the
7188 7196 * address of zone0 is returned directly, and netstack_zone_create()
7189 7197 * will only assign a value to zone0.zone_netstack, won't break anything.
7190 7198 * 3. Returns without the zone being held.
7191 7199 */
7192 7200 zone_t *
7193 7201 zone_find_by_id_nolock(zoneid_t zoneid)
7194 7202 {
7195 7203 zone_t *zone;
7196 7204
7197 7205 mutex_enter(&zonehash_lock);
7198 7206 if (zonehashbyid == NULL)
7199 7207 zone = &zone0;
7200 7208 else
7201 7209 zone = zone_find_all_by_id(zoneid);
7202 7210 mutex_exit(&zonehash_lock);
7203 7211 return (zone);
7204 7212 }
7205 7213
7206 7214 /*
7207 7215 * Walk the datalinks for a given zone
7208 7216 */
7209 7217 int
7210 7218 zone_datalink_walk(zoneid_t zoneid, int (*cb)(datalink_id_t, void *),
7211 7219 void *data)
7212 7220 {
7213 7221 zone_t *zone;
7214 7222 zone_dl_t *zdl;
7215 7223 datalink_id_t *idarray;
7216 7224 uint_t idcount = 0;
7217 7225 int i, ret = 0;
7218 7226
7219 7227 if ((zone = zone_find_by_id(zoneid)) == NULL)
7220 7228 return (ENOENT);
7221 7229
7222 7230 /*
7223 7231 * We first build an array of linkid's so that we can walk these and
7224 7232 * execute the callback with the zone_lock dropped.
7225 7233 */
7226 7234 mutex_enter(&zone->zone_lock);
7227 7235 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7228 7236 zdl = list_next(&zone->zone_dl_list, zdl)) {
7229 7237 idcount++;
7230 7238 }
7231 7239
7232 7240 if (idcount == 0) {
7233 7241 mutex_exit(&zone->zone_lock);
7234 7242 zone_rele(zone);
7235 7243 return (0);
7236 7244 }
7237 7245
7238 7246 idarray = kmem_alloc(sizeof (datalink_id_t) * idcount, KM_NOSLEEP);
7239 7247 if (idarray == NULL) {
7240 7248 mutex_exit(&zone->zone_lock);
7241 7249 zone_rele(zone);
7242 7250 return (ENOMEM);
7243 7251 }
7244 7252
7245 7253 for (i = 0, zdl = list_head(&zone->zone_dl_list); zdl != NULL;
7246 7254 i++, zdl = list_next(&zone->zone_dl_list, zdl)) {
7247 7255 idarray[i] = zdl->zdl_id;
7248 7256 }
7249 7257
7250 7258 mutex_exit(&zone->zone_lock);
7251 7259
7252 7260 for (i = 0; i < idcount && ret == 0; i++) {
7253 7261 if ((ret = (*cb)(idarray[i], data)) != 0)
7254 7262 break;
7255 7263 }
7256 7264
7257 7265 zone_rele(zone);
7258 7266 kmem_free(idarray, sizeof (datalink_id_t) * idcount);
7259 7267 return (ret);
7260 7268 }
7261 7269
7262 7270 static char *
7263 7271 zone_net_type2name(int type)
7264 7272 {
7265 7273 switch (type) {
7266 7274 case ZONE_NETWORK_ADDRESS:
7267 7275 return (ZONE_NET_ADDRNAME);
7268 7276 case ZONE_NETWORK_DEFROUTER:
7269 7277 return (ZONE_NET_RTRNAME);
7270 7278 default:
7271 7279 return (NULL);
7272 7280 }
7273 7281 }
7274 7282
7275 7283 static int
7276 7284 zone_set_network(zoneid_t zoneid, zone_net_data_t *znbuf)
7277 7285 {
7278 7286 zone_t *zone;
7279 7287 zone_dl_t *zdl;
7280 7288 nvlist_t *nvl;
7281 7289 int err = 0;
7282 7290 uint8_t *new = NULL;
7283 7291 char *nvname;
7284 7292 int bufsize;
7285 7293 datalink_id_t linkid = znbuf->zn_linkid;
7286 7294
7287 7295 if (secpolicy_zone_config(CRED()) != 0)
7288 7296 return (set_errno(EPERM));
7289 7297
7290 7298 if (zoneid == GLOBAL_ZONEID)
7291 7299 return (set_errno(EINVAL));
7292 7300
7293 7301 nvname = zone_net_type2name(znbuf->zn_type);
7294 7302 bufsize = znbuf->zn_len;
7295 7303 new = znbuf->zn_val;
7296 7304 if (nvname == NULL)
7297 7305 return (set_errno(EINVAL));
7298 7306
7299 7307 if ((zone = zone_find_by_id(zoneid)) == NULL) {
7300 7308 return (set_errno(EINVAL));
7301 7309 }
7302 7310
7303 7311 mutex_enter(&zone->zone_lock);
7304 7312 if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
7305 7313 err = ENXIO;
7306 7314 goto done;
7307 7315 }
7308 7316 if ((nvl = zdl->zdl_net) == NULL) {
7309 7317 if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP)) {
7310 7318 err = ENOMEM;
7311 7319 goto done;
7312 7320 } else {
7313 7321 zdl->zdl_net = nvl;
7314 7322 }
7315 7323 }
7316 7324 if (nvlist_exists(nvl, nvname)) {
7317 7325 err = EINVAL;
7318 7326 goto done;
7319 7327 }
7320 7328 err = nvlist_add_uint8_array(nvl, nvname, new, bufsize);
7321 7329 ASSERT(err == 0);
7322 7330 done:
7323 7331 mutex_exit(&zone->zone_lock);
7324 7332 zone_rele(zone);
7325 7333 if (err != 0)
7326 7334 return (set_errno(err));
7327 7335 else
7328 7336 return (0);
7329 7337 }
7330 7338
7331 7339 static int
7332 7340 zone_get_network(zoneid_t zoneid, zone_net_data_t *znbuf)
7333 7341 {
7334 7342 zone_t *zone;
7335 7343 zone_dl_t *zdl;
7336 7344 nvlist_t *nvl;
7337 7345 uint8_t *ptr;
7338 7346 uint_t psize;
7339 7347 int err = 0;
7340 7348 char *nvname;
7341 7349 int bufsize;
7342 7350 void *buf;
7343 7351 datalink_id_t linkid = znbuf->zn_linkid;
7344 7352
7345 7353 if (zoneid == GLOBAL_ZONEID)
7346 7354 return (set_errno(EINVAL));
7347 7355
7348 7356 nvname = zone_net_type2name(znbuf->zn_type);
7349 7357 bufsize = znbuf->zn_len;
7350 7358 buf = znbuf->zn_val;
7351 7359
7352 7360 if (nvname == NULL)
7353 7361 return (set_errno(EINVAL));
7354 7362 if ((zone = zone_find_by_id(zoneid)) == NULL)
7355 7363 return (set_errno(EINVAL));
7356 7364
7357 7365 mutex_enter(&zone->zone_lock);
7358 7366 if ((zdl = zone_find_dl(zone, linkid)) == NULL) {
7359 7367 err = ENXIO;
7360 7368 goto done;
7361 7369 }
7362 7370 if ((nvl = zdl->zdl_net) == NULL || !nvlist_exists(nvl, nvname)) {
7363 7371 err = ENOENT;
7364 7372 goto done;
7365 7373 }
7366 7374 err = nvlist_lookup_uint8_array(nvl, nvname, &ptr, &psize);
7367 7375 ASSERT(err == 0);
7368 7376
7369 7377 if (psize > bufsize) {
7370 7378 err = ENOBUFS;
7371 7379 goto done;
7372 7380 }
7373 7381 znbuf->zn_len = psize;
7374 7382 bcopy(ptr, buf, psize);
7375 7383 done:
7376 7384 mutex_exit(&zone->zone_lock);
7377 7385 zone_rele(zone);
7378 7386 if (err != 0)
7379 7387 return (set_errno(err));
7380 7388 else
7381 7389 return (0);
7382 7390 }
|
↓ open down ↓ |
2826 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX