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