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 * Copyright (c) 2016 by Delphix. All rights reserved. 26 * Copyright 2018 OmniOS Community Edition (OmniOSce) Association. 27 */ 28 29 /* 30 * Zones 31 * 32 * A zone is a named collection of processes, namespace constraints, 33 * and other system resources which comprise a secure and manageable 34 * application containment facility. 35 * 36 * Zones (represented by the reference counted zone_t) are tracked in 37 * the kernel in the zonehash. Elsewhere in the kernel, Zone IDs 38 * (zoneid_t) are used to track zone association. Zone IDs are 39 * dynamically generated when the zone is created; if a persistent 40 * identifier is needed (core files, accounting logs, audit trail, 41 * etc.), the zone name should be used. 42 * 43 * 44 * Global Zone: 45 * 46 * The global zone (zoneid 0) is automatically associated with all 47 * system resources that have not been bound to a user-created zone. 48 * This means that even systems where zones are not in active use 49 * have a global zone, and all processes, mounts, etc. are 50 * associated with that zone. The global zone is generally 51 * unconstrained in terms of privileges and access, though the usual 52 * credential and privilege based restrictions apply. 53 * 54 * 55 * Zone States: 56 * 57 * The states in which a zone may be in and the transitions are as 58 * follows: 59 * 60 * ZONE_IS_UNINITIALIZED: primordial state for a zone. The partially 61 * initialized zone is added to the list of active zones on the system but 62 * isn't accessible. 63 * 64 * ZONE_IS_INITIALIZED: Initialization complete except the ZSD callbacks are 65 * not yet completed. Not possible to enter the zone, but attributes can 66 * be retrieved. 67 * 68 * ZONE_IS_READY: zsched (the kernel dummy process for a zone) is 69 * ready. The zone is made visible after the ZSD constructor callbacks are 70 * executed. A zone remains in this state until it transitions into 71 * the ZONE_IS_BOOTING state as a result of a call to zone_boot(). 72 * 73 * ZONE_IS_BOOTING: in this shortlived-state, zsched attempts to start 74 * init. Should that fail, the zone proceeds to the ZONE_IS_SHUTTING_DOWN 75 * state. 76 * 77 * ZONE_IS_RUNNING: The zone is open for business: zsched has 78 * successfully started init. A zone remains in this state until 79 * zone_shutdown() is called. 80 * 81 * ZONE_IS_SHUTTING_DOWN: zone_shutdown() has been called, the system is 82 * killing all processes running in the zone. The zone remains 83 * in this state until there are no more user processes running in the zone. 84 * zone_create(), zone_enter(), and zone_destroy() on this zone will fail. 85 * Since zone_shutdown() is restartable, it may be called successfully 86 * multiple times for the same zone_t. Setting of the zone's state to 87 * ZONE_IS_SHUTTING_DOWN is synchronized with mounts, so VOP_MOUNT() may check 88 * the zone's status without worrying about it being a moving target. 89 * 90 * ZONE_IS_EMPTY: zone_shutdown() has been called, and there 91 * are no more user processes in the zone. The zone remains in this 92 * state until there are no more kernel threads associated with the 93 * zone. zone_create(), zone_enter(), and zone_destroy() on this zone will 94 * fail. 95 * 96 * ZONE_IS_DOWN: All kernel threads doing work on behalf of the zone 97 * have exited. zone_shutdown() returns. Henceforth it is not possible to 98 * join the zone or create kernel threads therein. 99 * 100 * ZONE_IS_DYING: zone_destroy() has been called on the zone; zone 101 * remains in this state until zsched exits. Calls to zone_find_by_*() 102 * return NULL from now on. 103 * 104 * ZONE_IS_DEAD: zsched has exited (zone_ntasks == 0). There are no 105 * processes or threads doing work on behalf of the zone. The zone is 106 * removed from the list of active zones. zone_destroy() returns, and 107 * the zone can be recreated. 108 * 109 * ZONE_IS_FREE (internal state): zone_ref goes to 0, ZSD destructor 110 * callbacks are executed, and all memory associated with the zone is 111 * freed. 112 * 113 * Threads can wait for the zone to enter a requested state by using 114 * zone_status_wait() or zone_status_timedwait() with the desired 115 * state passed in as an argument. Zone state transitions are 116 * uni-directional; it is not possible to move back to an earlier state. 117 * 118 * 119 * Zone-Specific Data: 120 * 121 * Subsystems needing to maintain zone-specific data can store that 122 * data using the ZSD mechanism. This provides a zone-specific data 123 * store, similar to thread-specific data (see pthread_getspecific(3C) 124 * or the TSD code in uts/common/disp/thread.c. Also, ZSD can be used 125 * to register callbacks to be invoked when a zone is created, shut 126 * down, or destroyed. This can be used to initialize zone-specific 127 * data for new zones and to clean up when zones go away. 128 * 129 * 130 * Data Structures: 131 * 132 * The per-zone structure (zone_t) is reference counted, and freed 133 * when all references are released. zone_hold and zone_rele can be 134 * used to adjust the reference count. In addition, reference counts 135 * associated with the cred_t structure are tracked separately using 136 * zone_cred_hold and zone_cred_rele. 137 * 138 * Pointers to active zone_t's are stored in two hash tables; one 139 * for searching by id, the other for searching by name. Lookups 140 * can be performed on either basis, using zone_find_by_id and 141 * zone_find_by_name. Both return zone_t pointers with the zone 142 * held, so zone_rele should be called when the pointer is no longer 143 * needed. Zones can also be searched by path; zone_find_by_path 144 * returns the zone with which a path name is associated (global 145 * zone if the path is not within some other zone's file system 146 * hierarchy). This currently requires iterating through each zone, 147 * so it is slower than an id or name search via a hash table. 148 * 149 * 150 * Locking: 151 * 152 * zonehash_lock: This is a top-level global lock used to protect the 153 * zone hash tables and lists. Zones cannot be created or destroyed 154 * while this lock is held. 155 * zone_status_lock: This is a global lock protecting zone state. 156 * Zones cannot change state while this lock is held. It also 157 * protects the list of kernel threads associated with a zone. 158 * zone_lock: This is a per-zone lock used to protect several fields of 159 * the zone_t (see <sys/zone.h> for details). In addition, holding 160 * this lock means that the zone cannot go away. 161 * zone_nlwps_lock: This is a per-zone lock used to protect the fields 162 * related to the zone.max-lwps rctl. 163 * zone_mem_lock: This is a per-zone lock used to protect the fields 164 * related to the zone.max-locked-memory and zone.max-swap rctls. 165 * zone_rctl_lock: This is a per-zone lock used to protect other rctls, 166 * currently just max_lofi 167 * zsd_key_lock: This is a global lock protecting the key state for ZSD. 168 * zone_deathrow_lock: This is a global lock protecting the "deathrow" 169 * list (a list of zones in the ZONE_IS_DEAD state). 170 * 171 * Ordering requirements: 172 * pool_lock --> cpu_lock --> zonehash_lock --> zone_status_lock --> 173 * zone_lock --> zsd_key_lock --> pidlock --> p_lock 174 * 175 * When taking zone_mem_lock or zone_nlwps_lock, the lock ordering is: 176 * zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_mem_lock 177 * zonehash_lock --> a_lock --> pidlock --> p_lock --> zone_nlwps_lock 178 * 179 * Blocking memory allocations are permitted while holding any of the 180 * zone locks. 181 * 182 * 183 * System Call Interface: 184 * 185 * The zone subsystem can be managed and queried from user level with 186 * the following system calls (all subcodes of the primary "zone" 187 * system call): 188 * - zone_create: creates a zone with selected attributes (name, 189 * root path, privileges, resource controls, ZFS datasets) 190 * - zone_enter: allows the current process to enter a zone 191 * - zone_getattr: reports attributes of a zone 192 * - zone_setattr: set attributes of a zone 193 * - zone_boot: set 'init' running for the zone 194 * - zone_list: lists all zones active in the system 195 * - zone_lookup: looks up zone id based on name 196 * - zone_shutdown: initiates shutdown process (see states above) 197 * - zone_destroy: completes shutdown process (see states above) 198 * 199 */ 200 201 #include <sys/priv_impl.h> 202 #include <sys/cred.h> 203 #include <c2/audit.h> 204 #include <sys/debug.h> 205 #include <sys/file.h> 206 #include <sys/kmem.h> 207 #include <sys/kstat.h> 208 #include <sys/mutex.h> 209 #include <sys/note.h> 210 #include <sys/pathname.h> 211 #include <sys/proc.h> 212 #include <sys/project.h> 213 #include <sys/sysevent.h> 214 #include <sys/task.h> 215 #include <sys/systm.h> 216 #include <sys/types.h> 217 #include <sys/utsname.h> 218 #include <sys/vnode.h> 219 #include <sys/vfs.h> 220 #include <sys/systeminfo.h> 221 #include <sys/policy.h> 222 #include <sys/cred_impl.h> 223 #include <sys/contract_impl.h> 224 #include <sys/contract/process_impl.h> 225 #include <sys/class.h> 226 #include <sys/pool.h> 227 #include <sys/pool_pset.h> 228 #include <sys/pset.h> 229 #include <sys/strlog.h> 230 #include <sys/sysmacros.h> 231 #include <sys/callb.h> 232 #include <sys/vmparam.h> 233 #include <sys/corectl.h> 234 #include <sys/ipc_impl.h> 235 #include <sys/klpd.h> 236 237 #include <sys/door.h> 238 #include <sys/cpuvar.h> 239 #include <sys/sdt.h> 240 241 #include <sys/uadmin.h> 242 #include <sys/session.h> 243 #include <sys/cmn_err.h> 244 #include <sys/modhash.h> 245 #include <sys/sunddi.h> 246 #include <sys/nvpair.h> 247 #include <sys/rctl.h> 248 #include <sys/fss.h> 249 #include <sys/brand.h> 250 #include <sys/zone.h> 251 #include <net/if.h> 252 #include <sys/cpucaps.h> 253 #include <vm/seg.h> 254 #include <sys/mac.h> 255 256 /* 257 * This constant specifies the number of seconds that threads waiting for 258 * subsystems to release a zone's general-purpose references will wait before 259 * they log the zone's reference counts. The constant's value shouldn't 260 * be so small that reference counts are unnecessarily reported for zones 261 * whose references are slowly released. On the other hand, it shouldn't be so 262 * large that users reboot their systems out of frustration over hung zones 263 * before the system logs the zones' reference counts. 264 */ 265 #define ZONE_DESTROY_TIMEOUT_SECS 60 266 267 /* List of data link IDs which are accessible from the zone */ 268 typedef struct zone_dl { 269 datalink_id_t zdl_id; 270 nvlist_t *zdl_net; 271 list_node_t zdl_linkage; 272 } zone_dl_t; 273 274 /* 275 * cv used to signal that all references to the zone have been released. This 276 * needs to be global since there may be multiple waiters, and the first to 277 * wake up will free the zone_t, hence we cannot use zone->zone_cv. 278 */ 279 static kcondvar_t zone_destroy_cv; 280 /* 281 * Lock used to serialize access to zone_cv. This could have been per-zone, 282 * but then we'd need another lock for zone_destroy_cv, and why bother? 283 */ 284 static kmutex_t zone_status_lock; 285 286 /* 287 * ZSD-related global variables. 288 */ 289 static kmutex_t zsd_key_lock; /* protects the following two */ 290 /* 291 * The next caller of zone_key_create() will be assigned a key of ++zsd_keyval. 292 */ 293 static zone_key_t zsd_keyval = 0; 294 /* 295 * Global list of registered keys. We use this when a new zone is created. 296 */ 297 static list_t zsd_registered_keys; 298 299 int zone_hash_size = 256; 300 static mod_hash_t *zonehashbyname, *zonehashbyid, *zonehashbylabel; 301 static kmutex_t zonehash_lock; 302 static uint_t zonecount; 303 static id_space_t *zoneid_space; 304 305 /* 306 * The global zone (aka zone0) is the all-seeing, all-knowing zone in which the 307 * kernel proper runs, and which manages all other zones. 308 * 309 * Although not declared as static, the variable "zone0" should not be used 310 * except for by code that needs to reference the global zone early on in boot, 311 * before it is fully initialized. All other consumers should use 312 * 'global_zone'. 313 */ 314 zone_t zone0; 315 zone_t *global_zone = NULL; /* Set when the global zone is initialized */ 316 317 /* 318 * List of active zones, protected by zonehash_lock. 319 */ 320 static list_t zone_active; 321 322 /* 323 * List of destroyed zones that still have outstanding cred references. 324 * Used for debugging. Uses a separate lock to avoid lock ordering 325 * problems in zone_free. 326 */ 327 static list_t zone_deathrow; 328 static kmutex_t zone_deathrow_lock; 329 330 /* number of zones is limited by virtual interface limit in IP */ 331 uint_t maxzones = 8192; 332 333 /* Event channel to sent zone state change notifications */ 334 evchan_t *zone_event_chan; 335 336 /* 337 * This table holds the mapping from kernel zone states to 338 * states visible in the state notification API. 339 * The idea is that we only expose "obvious" states and 340 * do not expose states which are just implementation details. 341 */ 342 const char *zone_status_table[] = { 343 ZONE_EVENT_UNINITIALIZED, /* uninitialized */ 344 ZONE_EVENT_INITIALIZED, /* initialized */ 345 ZONE_EVENT_READY, /* ready */ 346 ZONE_EVENT_READY, /* booting */ 347 ZONE_EVENT_RUNNING, /* running */ 348 ZONE_EVENT_SHUTTING_DOWN, /* shutting_down */ 349 ZONE_EVENT_SHUTTING_DOWN, /* empty */ 350 ZONE_EVENT_SHUTTING_DOWN, /* down */ 351 ZONE_EVENT_SHUTTING_DOWN, /* dying */ 352 ZONE_EVENT_UNINITIALIZED, /* dead */ 353 }; 354 355 /* 356 * This array contains the names of the subsystems listed in zone_ref_subsys_t 357 * (see sys/zone.h). 358 */ 359 static char *zone_ref_subsys_names[] = { 360 "NFS", /* ZONE_REF_NFS */ 361 "NFSv4", /* ZONE_REF_NFSV4 */ 362 "SMBFS", /* ZONE_REF_SMBFS */ 363 "MNTFS", /* ZONE_REF_MNTFS */ 364 "LOFI", /* ZONE_REF_LOFI */ 365 "VFS", /* ZONE_REF_VFS */ 366 "IPC" /* ZONE_REF_IPC */ 367 }; 368 369 /* 370 * This isn't static so lint doesn't complain. 371 */ 372 rctl_hndl_t rc_zone_cpu_shares; 373 rctl_hndl_t rc_zone_locked_mem; 374 rctl_hndl_t rc_zone_max_swap; 375 rctl_hndl_t rc_zone_max_lofi; 376 rctl_hndl_t rc_zone_cpu_cap; 377 rctl_hndl_t rc_zone_nlwps; 378 rctl_hndl_t rc_zone_nprocs; 379 rctl_hndl_t rc_zone_shmmax; 380 rctl_hndl_t rc_zone_shmmni; 381 rctl_hndl_t rc_zone_semmni; 382 rctl_hndl_t rc_zone_msgmni; 383 384 const char * const zone_default_initname = "/sbin/init"; 385 static char * const zone_prefix = "/zone/"; 386 static int zone_shutdown(zoneid_t zoneid); 387 static int zone_add_datalink(zoneid_t, datalink_id_t); 388 static int zone_remove_datalink(zoneid_t, datalink_id_t); 389 static int zone_list_datalink(zoneid_t, int *, datalink_id_t *); 390 static int zone_set_network(zoneid_t, zone_net_data_t *); 391 static int zone_get_network(zoneid_t, zone_net_data_t *); 392 393 typedef boolean_t zsd_applyfn_t(kmutex_t *, boolean_t, zone_t *, zone_key_t); 394 395 static void zsd_apply_all_zones(zsd_applyfn_t *, zone_key_t); 396 static void zsd_apply_all_keys(zsd_applyfn_t *, zone_t *); 397 static boolean_t zsd_apply_create(kmutex_t *, boolean_t, zone_t *, zone_key_t); 398 static boolean_t zsd_apply_shutdown(kmutex_t *, boolean_t, zone_t *, 399 zone_key_t); 400 static boolean_t zsd_apply_destroy(kmutex_t *, boolean_t, zone_t *, zone_key_t); 401 static boolean_t zsd_wait_for_creator(zone_t *, struct zsd_entry *, 402 kmutex_t *); 403 static boolean_t zsd_wait_for_inprogress(zone_t *, struct zsd_entry *, 404 kmutex_t *); 405 406 /* 407 * Bump this number when you alter the zone syscall interfaces; this is 408 * because we need to have support for previous API versions in libc 409 * to support patching; libc calls into the kernel to determine this number. 410 * 411 * Version 1 of the API is the version originally shipped with Solaris 10 412 * Version 2 alters the zone_create system call in order to support more 413 * arguments by moving the args into a structure; and to do better 414 * error reporting when zone_create() fails. 415 * Version 3 alters the zone_create system call in order to support the 416 * import of ZFS datasets to zones. 417 * Version 4 alters the zone_create system call in order to support 418 * Trusted Extensions. 419 * Version 5 alters the zone_boot system call, and converts its old 420 * bootargs parameter to be set by the zone_setattr API instead. 421 * Version 6 adds the flag argument to zone_create. 422 */ 423 static const int ZONE_SYSCALL_API_VERSION = 6; 424 425 /* 426 * Certain filesystems (such as NFS and autofs) need to know which zone 427 * the mount is being placed in. Because of this, we need to be able to 428 * ensure that a zone isn't in the process of being created/destroyed such 429 * that nfs_mount() thinks it is in the global/NGZ zone, while by the time 430 * it gets added the list of mounted zones, it ends up on the wrong zone's 431 * mount list. Since a zone can't reside on an NFS file system, we don't 432 * have to worry about the zonepath itself. 433 * 434 * The following functions: block_mounts()/resume_mounts() and 435 * mount_in_progress()/mount_completed() are used by zones and the VFS 436 * layer (respectively) to synchronize zone state transitions and new 437 * mounts within a zone. This syncronization is on a per-zone basis, so 438 * activity for one zone will not interfere with activity for another zone. 439 * 440 * The semantics are like a reader-reader lock such that there may 441 * either be multiple mounts (or zone state transitions, if that weren't 442 * serialized by zonehash_lock) in progress at the same time, but not 443 * both. 444 * 445 * We use cv's so the user can ctrl-C out of the operation if it's 446 * taking too long. 447 * 448 * The semantics are such that there is unfair bias towards the 449 * "current" operation. This means that zone halt may starve if 450 * there is a rapid succession of new mounts coming in to the zone. 451 */ 452 /* 453 * Prevent new mounts from progressing to the point of calling 454 * VFS_MOUNT(). If there are already mounts in this "region", wait for 455 * them to complete. 456 */ 457 static int 458 block_mounts(zone_t *zp) 459 { 460 int retval = 0; 461 462 /* 463 * Since it may block for a long time, block_mounts() shouldn't be 464 * called with zonehash_lock held. 465 */ 466 ASSERT(MUTEX_NOT_HELD(&zonehash_lock)); 467 mutex_enter(&zp->zone_mount_lock); 468 while (zp->zone_mounts_in_progress > 0) { 469 if (cv_wait_sig(&zp->zone_mount_cv, &zp->zone_mount_lock) == 0) 470 goto signaled; 471 } 472 /* 473 * A negative value of mounts_in_progress indicates that mounts 474 * have been blocked by (-mounts_in_progress) different callers 475 * (remotely possible if two threads enter zone_shutdown at the same 476 * time). 477 */ 478 zp->zone_mounts_in_progress--; 479 retval = 1; 480 signaled: 481 mutex_exit(&zp->zone_mount_lock); 482 return (retval); 483 } 484 485 /* 486 * The VFS layer may progress with new mounts as far as we're concerned. 487 * Allow them to progress if we were the last obstacle. 488 */ 489 static void 490 resume_mounts(zone_t *zp) 491 { 492 mutex_enter(&zp->zone_mount_lock); 493 if (++zp->zone_mounts_in_progress == 0) 494 cv_broadcast(&zp->zone_mount_cv); 495 mutex_exit(&zp->zone_mount_lock); 496 } 497 498 /* 499 * The VFS layer is busy with a mount; this zone should wait until all 500 * of its mounts are completed to progress. 501 */ 502 void 503 mount_in_progress(zone_t *zp) 504 { 505 mutex_enter(&zp->zone_mount_lock); 506 while (zp->zone_mounts_in_progress < 0) 507 cv_wait(&zp->zone_mount_cv, &zp->zone_mount_lock); 508 zp->zone_mounts_in_progress++; 509 mutex_exit(&zp->zone_mount_lock); 510 } 511 512 /* 513 * VFS is done with one mount; wake up any waiting block_mounts() 514 * callers if this is the last mount. 515 */ 516 void 517 mount_completed(zone_t *zp) 518 { 519 mutex_enter(&zp->zone_mount_lock); 520 if (--zp->zone_mounts_in_progress == 0) 521 cv_broadcast(&zp->zone_mount_cv); 522 mutex_exit(&zp->zone_mount_lock); 523 } 524 525 /* 526 * ZSD routines. 527 * 528 * Zone Specific Data (ZSD) is modeled after Thread Specific Data as 529 * defined by the pthread_key_create() and related interfaces. 530 * 531 * Kernel subsystems may register one or more data items and/or 532 * callbacks to be executed when a zone is created, shutdown, or 533 * destroyed. 534 * 535 * Unlike the thread counterpart, destructor callbacks will be executed 536 * even if the data pointer is NULL and/or there are no constructor 537 * callbacks, so it is the responsibility of such callbacks to check for 538 * NULL data values if necessary. 539 * 540 * The locking strategy and overall picture is as follows: 541 * 542 * When someone calls zone_key_create(), a template ZSD entry is added to the 543 * global list "zsd_registered_keys", protected by zsd_key_lock. While 544 * holding that lock all the existing zones are marked as 545 * ZSD_CREATE_NEEDED and a copy of the ZSD entry added to the per-zone 546 * zone_zsd list (protected by zone_lock). The global list is updated first 547 * (under zone_key_lock) to make sure that newly created zones use the 548 * most recent list of keys. Then under zonehash_lock we walk the zones 549 * and mark them. Similar locking is used in zone_key_delete(). 550 * 551 * The actual create, shutdown, and destroy callbacks are done without 552 * holding any lock. And zsd_flags are used to ensure that the operations 553 * completed so that when zone_key_create (and zone_create) is done, as well as 554 * zone_key_delete (and zone_destroy) is done, all the necessary callbacks 555 * are completed. 556 * 557 * When new zones are created constructor callbacks for all registered ZSD 558 * entries will be called. That also uses the above two phases of marking 559 * what needs to be done, and then running the callbacks without holding 560 * any locks. 561 * 562 * The framework does not provide any locking around zone_getspecific() and 563 * zone_setspecific() apart from that needed for internal consistency, so 564 * callers interested in atomic "test-and-set" semantics will need to provide 565 * their own locking. 566 */ 567 568 /* 569 * Helper function to find the zsd_entry associated with the key in the 570 * given list. 571 */ 572 static struct zsd_entry * 573 zsd_find(list_t *l, zone_key_t key) 574 { 575 struct zsd_entry *zsd; 576 577 for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) { 578 if (zsd->zsd_key == key) { 579 return (zsd); 580 } 581 } 582 return (NULL); 583 } 584 585 /* 586 * Helper function to find the zsd_entry associated with the key in the 587 * given list. Move it to the front of the list. 588 */ 589 static struct zsd_entry * 590 zsd_find_mru(list_t *l, zone_key_t key) 591 { 592 struct zsd_entry *zsd; 593 594 for (zsd = list_head(l); zsd != NULL; zsd = list_next(l, zsd)) { 595 if (zsd->zsd_key == key) { 596 /* 597 * Move to head of list to keep list in MRU order. 598 */ 599 if (zsd != list_head(l)) { 600 list_remove(l, zsd); 601 list_insert_head(l, zsd); 602 } 603 return (zsd); 604 } 605 } 606 return (NULL); 607 } 608 609 void 610 zone_key_create(zone_key_t *keyp, void *(*create)(zoneid_t), 611 void (*shutdown)(zoneid_t, void *), void (*destroy)(zoneid_t, void *)) 612 { 613 struct zsd_entry *zsdp; 614 struct zsd_entry *t; 615 struct zone *zone; 616 zone_key_t key; 617 618 zsdp = kmem_zalloc(sizeof (*zsdp), KM_SLEEP); 619 zsdp->zsd_data = NULL; 620 zsdp->zsd_create = create; 621 zsdp->zsd_shutdown = shutdown; 622 zsdp->zsd_destroy = destroy; 623 624 /* 625 * Insert in global list of callbacks. Makes future zone creations 626 * see it. 627 */ 628 mutex_enter(&zsd_key_lock); 629 key = zsdp->zsd_key = ++zsd_keyval; 630 ASSERT(zsd_keyval != 0); 631 list_insert_tail(&zsd_registered_keys, zsdp); 632 mutex_exit(&zsd_key_lock); 633 634 /* 635 * Insert for all existing zones and mark them as needing 636 * a create callback. 637 */ 638 mutex_enter(&zonehash_lock); /* stop the world */ 639 for (zone = list_head(&zone_active); zone != NULL; 640 zone = list_next(&zone_active, zone)) { 641 zone_status_t status; 642 643 mutex_enter(&zone->zone_lock); 644 645 /* Skip zones that are on the way down or not yet up */ 646 status = zone_status_get(zone); 647 if (status >= ZONE_IS_DOWN || 648 status == ZONE_IS_UNINITIALIZED) { 649 mutex_exit(&zone->zone_lock); 650 continue; 651 } 652 653 t = zsd_find_mru(&zone->zone_zsd, key); 654 if (t != NULL) { 655 /* 656 * A zsd_configure already inserted it after 657 * we dropped zsd_key_lock above. 658 */ 659 mutex_exit(&zone->zone_lock); 660 continue; 661 } 662 t = kmem_zalloc(sizeof (*t), KM_SLEEP); 663 t->zsd_key = key; 664 t->zsd_create = create; 665 t->zsd_shutdown = shutdown; 666 t->zsd_destroy = destroy; 667 if (create != NULL) { 668 t->zsd_flags = ZSD_CREATE_NEEDED; 669 DTRACE_PROBE2(zsd__create__needed, 670 zone_t *, zone, zone_key_t, key); 671 } 672 list_insert_tail(&zone->zone_zsd, t); 673 mutex_exit(&zone->zone_lock); 674 } 675 mutex_exit(&zonehash_lock); 676 677 if (create != NULL) { 678 /* Now call the create callback for this key */ 679 zsd_apply_all_zones(zsd_apply_create, key); 680 } 681 /* 682 * It is safe for consumers to use the key now, make it 683 * globally visible. Specifically zone_getspecific() will 684 * always successfully return the zone specific data associated 685 * with the key. 686 */ 687 *keyp = key; 688 689 } 690 691 /* 692 * Function called when a module is being unloaded, or otherwise wishes 693 * to unregister its ZSD key and callbacks. 694 * 695 * Remove from the global list and determine the functions that need to 696 * be called under a global lock. Then call the functions without 697 * holding any locks. Finally free up the zone_zsd entries. (The apply 698 * functions need to access the zone_zsd entries to find zsd_data etc.) 699 */ 700 int 701 zone_key_delete(zone_key_t key) 702 { 703 struct zsd_entry *zsdp = NULL; 704 zone_t *zone; 705 706 mutex_enter(&zsd_key_lock); 707 zsdp = zsd_find_mru(&zsd_registered_keys, key); 708 if (zsdp == NULL) { 709 mutex_exit(&zsd_key_lock); 710 return (-1); 711 } 712 list_remove(&zsd_registered_keys, zsdp); 713 mutex_exit(&zsd_key_lock); 714 715 mutex_enter(&zonehash_lock); 716 for (zone = list_head(&zone_active); zone != NULL; 717 zone = list_next(&zone_active, zone)) { 718 struct zsd_entry *del; 719 720 mutex_enter(&zone->zone_lock); 721 del = zsd_find_mru(&zone->zone_zsd, key); 722 if (del == NULL) { 723 /* 724 * Somebody else got here first e.g the zone going 725 * away. 726 */ 727 mutex_exit(&zone->zone_lock); 728 continue; 729 } 730 ASSERT(del->zsd_shutdown == zsdp->zsd_shutdown); 731 ASSERT(del->zsd_destroy == zsdp->zsd_destroy); 732 if (del->zsd_shutdown != NULL && 733 (del->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) { 734 del->zsd_flags |= ZSD_SHUTDOWN_NEEDED; 735 DTRACE_PROBE2(zsd__shutdown__needed, 736 zone_t *, zone, zone_key_t, key); 737 } 738 if (del->zsd_destroy != NULL && 739 (del->zsd_flags & ZSD_DESTROY_ALL) == 0) { 740 del->zsd_flags |= ZSD_DESTROY_NEEDED; 741 DTRACE_PROBE2(zsd__destroy__needed, 742 zone_t *, zone, zone_key_t, key); 743 } 744 mutex_exit(&zone->zone_lock); 745 } 746 mutex_exit(&zonehash_lock); 747 kmem_free(zsdp, sizeof (*zsdp)); 748 749 /* Now call the shutdown and destroy callback for this key */ 750 zsd_apply_all_zones(zsd_apply_shutdown, key); 751 zsd_apply_all_zones(zsd_apply_destroy, key); 752 753 /* Now we can free up the zsdp structures in each zone */ 754 mutex_enter(&zonehash_lock); 755 for (zone = list_head(&zone_active); zone != NULL; 756 zone = list_next(&zone_active, zone)) { 757 struct zsd_entry *del; 758 759 mutex_enter(&zone->zone_lock); 760 del = zsd_find(&zone->zone_zsd, key); 761 if (del != NULL) { 762 list_remove(&zone->zone_zsd, del); 763 ASSERT(!(del->zsd_flags & ZSD_ALL_INPROGRESS)); 764 kmem_free(del, sizeof (*del)); 765 } 766 mutex_exit(&zone->zone_lock); 767 } 768 mutex_exit(&zonehash_lock); 769 770 return (0); 771 } 772 773 /* 774 * ZSD counterpart of pthread_setspecific(). 775 * 776 * Since all zsd callbacks, including those with no create function, 777 * have an entry in zone_zsd, if the key is registered it is part of 778 * the zone_zsd list. 779 * Return an error if the key wasn't registerd. 780 */ 781 int 782 zone_setspecific(zone_key_t key, zone_t *zone, const void *data) 783 { 784 struct zsd_entry *t; 785 786 mutex_enter(&zone->zone_lock); 787 t = zsd_find_mru(&zone->zone_zsd, key); 788 if (t != NULL) { 789 /* 790 * Replace old value with new 791 */ 792 t->zsd_data = (void *)data; 793 mutex_exit(&zone->zone_lock); 794 return (0); 795 } 796 mutex_exit(&zone->zone_lock); 797 return (-1); 798 } 799 800 /* 801 * ZSD counterpart of pthread_getspecific(). 802 */ 803 void * 804 zone_getspecific(zone_key_t key, zone_t *zone) 805 { 806 struct zsd_entry *t; 807 void *data; 808 809 mutex_enter(&zone->zone_lock); 810 t = zsd_find_mru(&zone->zone_zsd, key); 811 data = (t == NULL ? NULL : t->zsd_data); 812 mutex_exit(&zone->zone_lock); 813 return (data); 814 } 815 816 /* 817 * Function used to initialize a zone's list of ZSD callbacks and data 818 * when the zone is being created. The callbacks are initialized from 819 * the template list (zsd_registered_keys). The constructor callback is 820 * executed later (once the zone exists and with locks dropped). 821 */ 822 static void 823 zone_zsd_configure(zone_t *zone) 824 { 825 struct zsd_entry *zsdp; 826 struct zsd_entry *t; 827 828 ASSERT(MUTEX_HELD(&zonehash_lock)); 829 ASSERT(list_head(&zone->zone_zsd) == NULL); 830 mutex_enter(&zone->zone_lock); 831 mutex_enter(&zsd_key_lock); 832 for (zsdp = list_head(&zsd_registered_keys); zsdp != NULL; 833 zsdp = list_next(&zsd_registered_keys, zsdp)) { 834 /* 835 * Since this zone is ZONE_IS_UNCONFIGURED, zone_key_create 836 * should not have added anything to it. 837 */ 838 ASSERT(zsd_find(&zone->zone_zsd, zsdp->zsd_key) == NULL); 839 840 t = kmem_zalloc(sizeof (*t), KM_SLEEP); 841 t->zsd_key = zsdp->zsd_key; 842 t->zsd_create = zsdp->zsd_create; 843 t->zsd_shutdown = zsdp->zsd_shutdown; 844 t->zsd_destroy = zsdp->zsd_destroy; 845 if (zsdp->zsd_create != NULL) { 846 t->zsd_flags = ZSD_CREATE_NEEDED; 847 DTRACE_PROBE2(zsd__create__needed, 848 zone_t *, zone, zone_key_t, zsdp->zsd_key); 849 } 850 list_insert_tail(&zone->zone_zsd, t); 851 } 852 mutex_exit(&zsd_key_lock); 853 mutex_exit(&zone->zone_lock); 854 } 855 856 enum zsd_callback_type { ZSD_CREATE, ZSD_SHUTDOWN, ZSD_DESTROY }; 857 858 /* 859 * Helper function to execute shutdown or destructor callbacks. 860 */ 861 static void 862 zone_zsd_callbacks(zone_t *zone, enum zsd_callback_type ct) 863 { 864 struct zsd_entry *t; 865 866 ASSERT(ct == ZSD_SHUTDOWN || ct == ZSD_DESTROY); 867 ASSERT(ct != ZSD_SHUTDOWN || zone_status_get(zone) >= ZONE_IS_EMPTY); 868 ASSERT(ct != ZSD_DESTROY || zone_status_get(zone) >= ZONE_IS_DOWN); 869 870 /* 871 * Run the callback solely based on what is registered for the zone 872 * in zone_zsd. The global list can change independently of this 873 * as keys are registered and unregistered and we don't register new 874 * callbacks for a zone that is in the process of going away. 875 */ 876 mutex_enter(&zone->zone_lock); 877 for (t = list_head(&zone->zone_zsd); t != NULL; 878 t = list_next(&zone->zone_zsd, t)) { 879 zone_key_t key = t->zsd_key; 880 881 /* Skip if no callbacks registered */ 882 883 if (ct == ZSD_SHUTDOWN) { 884 if (t->zsd_shutdown != NULL && 885 (t->zsd_flags & ZSD_SHUTDOWN_ALL) == 0) { 886 t->zsd_flags |= ZSD_SHUTDOWN_NEEDED; 887 DTRACE_PROBE2(zsd__shutdown__needed, 888 zone_t *, zone, zone_key_t, key); 889 } 890 } else { 891 if (t->zsd_destroy != NULL && 892 (t->zsd_flags & ZSD_DESTROY_ALL) == 0) { 893 t->zsd_flags |= ZSD_DESTROY_NEEDED; 894 DTRACE_PROBE2(zsd__destroy__needed, 895 zone_t *, zone, zone_key_t, key); 896 } 897 } 898 } 899 mutex_exit(&zone->zone_lock); 900 901 /* Now call the shutdown and destroy callback for this key */ 902 zsd_apply_all_keys(zsd_apply_shutdown, zone); 903 zsd_apply_all_keys(zsd_apply_destroy, zone); 904 905 } 906 907 /* 908 * Called when the zone is going away; free ZSD-related memory, and 909 * destroy the zone_zsd list. 910 */ 911 static void 912 zone_free_zsd(zone_t *zone) 913 { 914 struct zsd_entry *t, *next; 915 916 /* 917 * Free all the zsd_entry's we had on this zone. 918 */ 919 mutex_enter(&zone->zone_lock); 920 for (t = list_head(&zone->zone_zsd); t != NULL; t = next) { 921 next = list_next(&zone->zone_zsd, t); 922 list_remove(&zone->zone_zsd, t); 923 ASSERT(!(t->zsd_flags & ZSD_ALL_INPROGRESS)); 924 kmem_free(t, sizeof (*t)); 925 } 926 list_destroy(&zone->zone_zsd); 927 mutex_exit(&zone->zone_lock); 928 929 } 930 931 /* 932 * Apply a function to all zones for particular key value. 933 * 934 * The applyfn has to drop zonehash_lock if it does some work, and 935 * then reacquire it before it returns. 936 * When the lock is dropped we don't follow list_next even 937 * if it is possible to do so without any hazards. This is 938 * because we want the design to allow for the list of zones 939 * to change in any arbitrary way during the time the 940 * lock was dropped. 941 * 942 * It is safe to restart the loop at list_head since the applyfn 943 * changes the zsd_flags as it does work, so a subsequent 944 * pass through will have no effect in applyfn, hence the loop will terminate 945 * in at worst O(N^2). 946 */ 947 static void 948 zsd_apply_all_zones(zsd_applyfn_t *applyfn, zone_key_t key) 949 { 950 zone_t *zone; 951 952 mutex_enter(&zonehash_lock); 953 zone = list_head(&zone_active); 954 while (zone != NULL) { 955 if ((applyfn)(&zonehash_lock, B_FALSE, zone, key)) { 956 /* Lock dropped - restart at head */ 957 zone = list_head(&zone_active); 958 } else { 959 zone = list_next(&zone_active, zone); 960 } 961 } 962 mutex_exit(&zonehash_lock); 963 } 964 965 /* 966 * Apply a function to all keys for a particular zone. 967 * 968 * The applyfn has to drop zonehash_lock if it does some work, and 969 * then reacquire it before it returns. 970 * When the lock is dropped we don't follow list_next even 971 * if it is possible to do so without any hazards. This is 972 * because we want the design to allow for the list of zsd callbacks 973 * to change in any arbitrary way during the time the 974 * lock was dropped. 975 * 976 * It is safe to restart the loop at list_head since the applyfn 977 * changes the zsd_flags as it does work, so a subsequent 978 * pass through will have no effect in applyfn, hence the loop will terminate 979 * in at worst O(N^2). 980 */ 981 static void 982 zsd_apply_all_keys(zsd_applyfn_t *applyfn, zone_t *zone) 983 { 984 struct zsd_entry *t; 985 986 mutex_enter(&zone->zone_lock); 987 t = list_head(&zone->zone_zsd); 988 while (t != NULL) { 989 if ((applyfn)(NULL, B_TRUE, zone, t->zsd_key)) { 990 /* Lock dropped - restart at head */ 991 t = list_head(&zone->zone_zsd); 992 } else { 993 t = list_next(&zone->zone_zsd, t); 994 } 995 } 996 mutex_exit(&zone->zone_lock); 997 } 998 999 /* 1000 * Call the create function for the zone and key if CREATE_NEEDED 1001 * is set. 1002 * If some other thread gets here first and sets CREATE_INPROGRESS, then 1003 * we wait for that thread to complete so that we can ensure that 1004 * all the callbacks are done when we've looped over all zones/keys. 1005 * 1006 * When we call the create function, we drop the global held by the 1007 * caller, and return true to tell the caller it needs to re-evalute the 1008 * state. 1009 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock 1010 * remains held on exit. 1011 */ 1012 static boolean_t 1013 zsd_apply_create(kmutex_t *lockp, boolean_t zone_lock_held, 1014 zone_t *zone, zone_key_t key) 1015 { 1016 void *result; 1017 struct zsd_entry *t; 1018 boolean_t dropped; 1019 1020 if (lockp != NULL) { 1021 ASSERT(MUTEX_HELD(lockp)); 1022 } 1023 if (zone_lock_held) { 1024 ASSERT(MUTEX_HELD(&zone->zone_lock)); 1025 } else { 1026 mutex_enter(&zone->zone_lock); 1027 } 1028 1029 t = zsd_find(&zone->zone_zsd, key); 1030 if (t == NULL) { 1031 /* 1032 * Somebody else got here first e.g the zone going 1033 * away. 1034 */ 1035 if (!zone_lock_held) 1036 mutex_exit(&zone->zone_lock); 1037 return (B_FALSE); 1038 } 1039 dropped = B_FALSE; 1040 if (zsd_wait_for_inprogress(zone, t, lockp)) 1041 dropped = B_TRUE; 1042 1043 if (t->zsd_flags & ZSD_CREATE_NEEDED) { 1044 t->zsd_flags &= ~ZSD_CREATE_NEEDED; 1045 t->zsd_flags |= ZSD_CREATE_INPROGRESS; 1046 DTRACE_PROBE2(zsd__create__inprogress, 1047 zone_t *, zone, zone_key_t, key); 1048 mutex_exit(&zone->zone_lock); 1049 if (lockp != NULL) 1050 mutex_exit(lockp); 1051 1052 dropped = B_TRUE; 1053 ASSERT(t->zsd_create != NULL); 1054 DTRACE_PROBE2(zsd__create__start, 1055 zone_t *, zone, zone_key_t, key); 1056 1057 result = (*t->zsd_create)(zone->zone_id); 1058 1059 DTRACE_PROBE2(zsd__create__end, 1060 zone_t *, zone, voidn *, result); 1061 1062 ASSERT(result != NULL); 1063 if (lockp != NULL) 1064 mutex_enter(lockp); 1065 mutex_enter(&zone->zone_lock); 1066 t->zsd_data = result; 1067 t->zsd_flags &= ~ZSD_CREATE_INPROGRESS; 1068 t->zsd_flags |= ZSD_CREATE_COMPLETED; 1069 cv_broadcast(&t->zsd_cv); 1070 DTRACE_PROBE2(zsd__create__completed, 1071 zone_t *, zone, zone_key_t, key); 1072 } 1073 if (!zone_lock_held) 1074 mutex_exit(&zone->zone_lock); 1075 return (dropped); 1076 } 1077 1078 /* 1079 * Call the shutdown function for the zone and key if SHUTDOWN_NEEDED 1080 * is set. 1081 * If some other thread gets here first and sets *_INPROGRESS, then 1082 * we wait for that thread to complete so that we can ensure that 1083 * all the callbacks are done when we've looped over all zones/keys. 1084 * 1085 * When we call the shutdown function, we drop the global held by the 1086 * caller, and return true to tell the caller it needs to re-evalute the 1087 * state. 1088 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock 1089 * remains held on exit. 1090 */ 1091 static boolean_t 1092 zsd_apply_shutdown(kmutex_t *lockp, boolean_t zone_lock_held, 1093 zone_t *zone, zone_key_t key) 1094 { 1095 struct zsd_entry *t; 1096 void *data; 1097 boolean_t dropped; 1098 1099 if (lockp != NULL) { 1100 ASSERT(MUTEX_HELD(lockp)); 1101 } 1102 if (zone_lock_held) { 1103 ASSERT(MUTEX_HELD(&zone->zone_lock)); 1104 } else { 1105 mutex_enter(&zone->zone_lock); 1106 } 1107 1108 t = zsd_find(&zone->zone_zsd, key); 1109 if (t == NULL) { 1110 /* 1111 * Somebody else got here first e.g the zone going 1112 * away. 1113 */ 1114 if (!zone_lock_held) 1115 mutex_exit(&zone->zone_lock); 1116 return (B_FALSE); 1117 } 1118 dropped = B_FALSE; 1119 if (zsd_wait_for_creator(zone, t, lockp)) 1120 dropped = B_TRUE; 1121 1122 if (zsd_wait_for_inprogress(zone, t, lockp)) 1123 dropped = B_TRUE; 1124 1125 if (t->zsd_flags & ZSD_SHUTDOWN_NEEDED) { 1126 t->zsd_flags &= ~ZSD_SHUTDOWN_NEEDED; 1127 t->zsd_flags |= ZSD_SHUTDOWN_INPROGRESS; 1128 DTRACE_PROBE2(zsd__shutdown__inprogress, 1129 zone_t *, zone, zone_key_t, key); 1130 mutex_exit(&zone->zone_lock); 1131 if (lockp != NULL) 1132 mutex_exit(lockp); 1133 dropped = B_TRUE; 1134 1135 ASSERT(t->zsd_shutdown != NULL); 1136 data = t->zsd_data; 1137 1138 DTRACE_PROBE2(zsd__shutdown__start, 1139 zone_t *, zone, zone_key_t, key); 1140 1141 (t->zsd_shutdown)(zone->zone_id, data); 1142 DTRACE_PROBE2(zsd__shutdown__end, 1143 zone_t *, zone, zone_key_t, key); 1144 1145 if (lockp != NULL) 1146 mutex_enter(lockp); 1147 mutex_enter(&zone->zone_lock); 1148 t->zsd_flags &= ~ZSD_SHUTDOWN_INPROGRESS; 1149 t->zsd_flags |= ZSD_SHUTDOWN_COMPLETED; 1150 cv_broadcast(&t->zsd_cv); 1151 DTRACE_PROBE2(zsd__shutdown__completed, 1152 zone_t *, zone, zone_key_t, key); 1153 } 1154 if (!zone_lock_held) 1155 mutex_exit(&zone->zone_lock); 1156 return (dropped); 1157 } 1158 1159 /* 1160 * Call the destroy function for the zone and key if DESTROY_NEEDED 1161 * is set. 1162 * If some other thread gets here first and sets *_INPROGRESS, then 1163 * we wait for that thread to complete so that we can ensure that 1164 * all the callbacks are done when we've looped over all zones/keys. 1165 * 1166 * When we call the destroy function, we drop the global held by the 1167 * caller, and return true to tell the caller it needs to re-evalute the 1168 * state. 1169 * If the caller holds zone_lock then zone_lock_held is set, and zone_lock 1170 * remains held on exit. 1171 */ 1172 static boolean_t 1173 zsd_apply_destroy(kmutex_t *lockp, boolean_t zone_lock_held, 1174 zone_t *zone, zone_key_t key) 1175 { 1176 struct zsd_entry *t; 1177 void *data; 1178 boolean_t dropped; 1179 1180 if (lockp != NULL) { 1181 ASSERT(MUTEX_HELD(lockp)); 1182 } 1183 if (zone_lock_held) { 1184 ASSERT(MUTEX_HELD(&zone->zone_lock)); 1185 } else { 1186 mutex_enter(&zone->zone_lock); 1187 } 1188 1189 t = zsd_find(&zone->zone_zsd, key); 1190 if (t == NULL) { 1191 /* 1192 * Somebody else got here first e.g the zone going 1193 * away. 1194 */ 1195 if (!zone_lock_held) 1196 mutex_exit(&zone->zone_lock); 1197 return (B_FALSE); 1198 } 1199 dropped = B_FALSE; 1200 if (zsd_wait_for_creator(zone, t, lockp)) 1201 dropped = B_TRUE; 1202 1203 if (zsd_wait_for_inprogress(zone, t, lockp)) 1204 dropped = B_TRUE; 1205 1206 if (t->zsd_flags & ZSD_DESTROY_NEEDED) { 1207 t->zsd_flags &= ~ZSD_DESTROY_NEEDED; 1208 t->zsd_flags |= ZSD_DESTROY_INPROGRESS; 1209 DTRACE_PROBE2(zsd__destroy__inprogress, 1210 zone_t *, zone, zone_key_t, key); 1211 mutex_exit(&zone->zone_lock); 1212 if (lockp != NULL) 1213 mutex_exit(lockp); 1214 dropped = B_TRUE; 1215 1216 ASSERT(t->zsd_destroy != NULL); 1217 data = t->zsd_data; 1218 DTRACE_PROBE2(zsd__destroy__start, 1219 zone_t *, zone, zone_key_t, key); 1220 1221 (t->zsd_destroy)(zone->zone_id, data); 1222 DTRACE_PROBE2(zsd__destroy__end, 1223 zone_t *, zone, zone_key_t, key); 1224 1225 if (lockp != NULL) 1226 mutex_enter(lockp); 1227 mutex_enter(&zone->zone_lock); 1228 t->zsd_data = NULL; 1229 t->zsd_flags &= ~ZSD_DESTROY_INPROGRESS; 1230 t->zsd_flags |= ZSD_DESTROY_COMPLETED; 1231 cv_broadcast(&t->zsd_cv); 1232 DTRACE_PROBE2(zsd__destroy__completed, 1233 zone_t *, zone, zone_key_t, key); 1234 } 1235 if (!zone_lock_held) 1236 mutex_exit(&zone->zone_lock); 1237 return (dropped); 1238 } 1239 1240 /* 1241 * Wait for any CREATE_NEEDED flag to be cleared. 1242 * Returns true if lockp was temporarily dropped while waiting. 1243 */ 1244 static boolean_t 1245 zsd_wait_for_creator(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp) 1246 { 1247 boolean_t dropped = B_FALSE; 1248 1249 while (t->zsd_flags & ZSD_CREATE_NEEDED) { 1250 DTRACE_PROBE2(zsd__wait__for__creator, 1251 zone_t *, zone, struct zsd_entry *, t); 1252 if (lockp != NULL) { 1253 dropped = B_TRUE; 1254 mutex_exit(lockp); 1255 } 1256 cv_wait(&t->zsd_cv, &zone->zone_lock); 1257 if (lockp != NULL) { 1258 /* First drop zone_lock to preserve order */ 1259 mutex_exit(&zone->zone_lock); 1260 mutex_enter(lockp); 1261 mutex_enter(&zone->zone_lock); 1262 } 1263 } 1264 return (dropped); 1265 } 1266 1267 /* 1268 * Wait for any INPROGRESS flag to be cleared. 1269 * Returns true if lockp was temporarily dropped while waiting. 1270 */ 1271 static boolean_t 1272 zsd_wait_for_inprogress(zone_t *zone, struct zsd_entry *t, kmutex_t *lockp) 1273 { 1274 boolean_t dropped = B_FALSE; 1275 1276 while (t->zsd_flags & ZSD_ALL_INPROGRESS) { 1277 DTRACE_PROBE2(zsd__wait__for__inprogress, 1278 zone_t *, zone, struct zsd_entry *, t); 1279 if (lockp != NULL) { 1280 dropped = B_TRUE; 1281 mutex_exit(lockp); 1282 } 1283 cv_wait(&t->zsd_cv, &zone->zone_lock); 1284 if (lockp != NULL) { 1285 /* First drop zone_lock to preserve order */ 1286 mutex_exit(&zone->zone_lock); 1287 mutex_enter(lockp); 1288 mutex_enter(&zone->zone_lock); 1289 } 1290 } 1291 return (dropped); 1292 } 1293 1294 /* 1295 * Frees memory associated with the zone dataset list. 1296 */ 1297 static void 1298 zone_free_datasets(zone_t *zone) 1299 { 1300 zone_dataset_t *t, *next; 1301 1302 for (t = list_head(&zone->zone_datasets); t != NULL; t = next) { 1303 next = list_next(&zone->zone_datasets, t); 1304 list_remove(&zone->zone_datasets, t); 1305 kmem_free(t->zd_dataset, strlen(t->zd_dataset) + 1); 1306 kmem_free(t, sizeof (*t)); 1307 } 1308 list_destroy(&zone->zone_datasets); 1309 } 1310 1311 /* 1312 * zone.cpu-shares resource control support. 1313 */ 1314 /*ARGSUSED*/ 1315 static rctl_qty_t 1316 zone_cpu_shares_usage(rctl_t *rctl, struct proc *p) 1317 { 1318 ASSERT(MUTEX_HELD(&p->p_lock)); 1319 return (p->p_zone->zone_shares); 1320 } 1321 1322 /*ARGSUSED*/ 1323 static int 1324 zone_cpu_shares_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, 1325 rctl_qty_t nv) 1326 { 1327 ASSERT(MUTEX_HELD(&p->p_lock)); 1328 ASSERT(e->rcep_t == RCENTITY_ZONE); 1329 if (e->rcep_p.zone == NULL) 1330 return (0); 1331 1332 e->rcep_p.zone->zone_shares = nv; 1333 return (0); 1334 } 1335 1336 static rctl_ops_t zone_cpu_shares_ops = { 1337 rcop_no_action, 1338 zone_cpu_shares_usage, 1339 zone_cpu_shares_set, 1340 rcop_no_test 1341 }; 1342 1343 /* 1344 * zone.cpu-cap resource control support. 1345 */ 1346 /*ARGSUSED*/ 1347 static rctl_qty_t 1348 zone_cpu_cap_get(rctl_t *rctl, struct proc *p) 1349 { 1350 ASSERT(MUTEX_HELD(&p->p_lock)); 1351 return (cpucaps_zone_get(p->p_zone)); 1352 } 1353 1354 /*ARGSUSED*/ 1355 static int 1356 zone_cpu_cap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, 1357 rctl_qty_t nv) 1358 { 1359 zone_t *zone = e->rcep_p.zone; 1360 1361 ASSERT(MUTEX_HELD(&p->p_lock)); 1362 ASSERT(e->rcep_t == RCENTITY_ZONE); 1363 1364 if (zone == NULL) 1365 return (0); 1366 1367 /* 1368 * set cap to the new value. 1369 */ 1370 return (cpucaps_zone_set(zone, nv)); 1371 } 1372 1373 static rctl_ops_t zone_cpu_cap_ops = { 1374 rcop_no_action, 1375 zone_cpu_cap_get, 1376 zone_cpu_cap_set, 1377 rcop_no_test 1378 }; 1379 1380 /*ARGSUSED*/ 1381 static rctl_qty_t 1382 zone_lwps_usage(rctl_t *r, proc_t *p) 1383 { 1384 rctl_qty_t nlwps; 1385 zone_t *zone = p->p_zone; 1386 1387 ASSERT(MUTEX_HELD(&p->p_lock)); 1388 1389 mutex_enter(&zone->zone_nlwps_lock); 1390 nlwps = zone->zone_nlwps; 1391 mutex_exit(&zone->zone_nlwps_lock); 1392 1393 return (nlwps); 1394 } 1395 1396 /*ARGSUSED*/ 1397 static int 1398 zone_lwps_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl, 1399 rctl_qty_t incr, uint_t flags) 1400 { 1401 rctl_qty_t nlwps; 1402 1403 ASSERT(MUTEX_HELD(&p->p_lock)); 1404 ASSERT(e->rcep_t == RCENTITY_ZONE); 1405 if (e->rcep_p.zone == NULL) 1406 return (0); 1407 ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock))); 1408 nlwps = e->rcep_p.zone->zone_nlwps; 1409 1410 if (nlwps + incr > rcntl->rcv_value) 1411 return (1); 1412 1413 return (0); 1414 } 1415 1416 /*ARGSUSED*/ 1417 static int 1418 zone_lwps_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv) 1419 { 1420 ASSERT(MUTEX_HELD(&p->p_lock)); 1421 ASSERT(e->rcep_t == RCENTITY_ZONE); 1422 if (e->rcep_p.zone == NULL) 1423 return (0); 1424 e->rcep_p.zone->zone_nlwps_ctl = nv; 1425 return (0); 1426 } 1427 1428 static rctl_ops_t zone_lwps_ops = { 1429 rcop_no_action, 1430 zone_lwps_usage, 1431 zone_lwps_set, 1432 zone_lwps_test, 1433 }; 1434 1435 /*ARGSUSED*/ 1436 static rctl_qty_t 1437 zone_procs_usage(rctl_t *r, proc_t *p) 1438 { 1439 rctl_qty_t nprocs; 1440 zone_t *zone = p->p_zone; 1441 1442 ASSERT(MUTEX_HELD(&p->p_lock)); 1443 1444 mutex_enter(&zone->zone_nlwps_lock); 1445 nprocs = zone->zone_nprocs; 1446 mutex_exit(&zone->zone_nlwps_lock); 1447 1448 return (nprocs); 1449 } 1450 1451 /*ARGSUSED*/ 1452 static int 1453 zone_procs_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rcntl, 1454 rctl_qty_t incr, uint_t flags) 1455 { 1456 rctl_qty_t nprocs; 1457 1458 ASSERT(MUTEX_HELD(&p->p_lock)); 1459 ASSERT(e->rcep_t == RCENTITY_ZONE); 1460 if (e->rcep_p.zone == NULL) 1461 return (0); 1462 ASSERT(MUTEX_HELD(&(e->rcep_p.zone->zone_nlwps_lock))); 1463 nprocs = e->rcep_p.zone->zone_nprocs; 1464 1465 if (nprocs + incr > rcntl->rcv_value) 1466 return (1); 1467 1468 return (0); 1469 } 1470 1471 /*ARGSUSED*/ 1472 static int 1473 zone_procs_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, rctl_qty_t nv) 1474 { 1475 ASSERT(MUTEX_HELD(&p->p_lock)); 1476 ASSERT(e->rcep_t == RCENTITY_ZONE); 1477 if (e->rcep_p.zone == NULL) 1478 return (0); 1479 e->rcep_p.zone->zone_nprocs_ctl = nv; 1480 return (0); 1481 } 1482 1483 static rctl_ops_t zone_procs_ops = { 1484 rcop_no_action, 1485 zone_procs_usage, 1486 zone_procs_set, 1487 zone_procs_test, 1488 }; 1489 1490 /*ARGSUSED*/ 1491 static rctl_qty_t 1492 zone_shmmax_usage(rctl_t *rctl, struct proc *p) 1493 { 1494 ASSERT(MUTEX_HELD(&p->p_lock)); 1495 return (p->p_zone->zone_shmmax); 1496 } 1497 1498 /*ARGSUSED*/ 1499 static int 1500 zone_shmmax_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval, 1501 rctl_qty_t incr, uint_t flags) 1502 { 1503 rctl_qty_t v; 1504 ASSERT(MUTEX_HELD(&p->p_lock)); 1505 ASSERT(e->rcep_t == RCENTITY_ZONE); 1506 v = e->rcep_p.zone->zone_shmmax + incr; 1507 if (v > rval->rcv_value) 1508 return (1); 1509 return (0); 1510 } 1511 1512 static rctl_ops_t zone_shmmax_ops = { 1513 rcop_no_action, 1514 zone_shmmax_usage, 1515 rcop_no_set, 1516 zone_shmmax_test 1517 }; 1518 1519 /*ARGSUSED*/ 1520 static rctl_qty_t 1521 zone_shmmni_usage(rctl_t *rctl, struct proc *p) 1522 { 1523 ASSERT(MUTEX_HELD(&p->p_lock)); 1524 return (p->p_zone->zone_ipc.ipcq_shmmni); 1525 } 1526 1527 /*ARGSUSED*/ 1528 static int 1529 zone_shmmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval, 1530 rctl_qty_t incr, uint_t flags) 1531 { 1532 rctl_qty_t v; 1533 ASSERT(MUTEX_HELD(&p->p_lock)); 1534 ASSERT(e->rcep_t == RCENTITY_ZONE); 1535 v = e->rcep_p.zone->zone_ipc.ipcq_shmmni + incr; 1536 if (v > rval->rcv_value) 1537 return (1); 1538 return (0); 1539 } 1540 1541 static rctl_ops_t zone_shmmni_ops = { 1542 rcop_no_action, 1543 zone_shmmni_usage, 1544 rcop_no_set, 1545 zone_shmmni_test 1546 }; 1547 1548 /*ARGSUSED*/ 1549 static rctl_qty_t 1550 zone_semmni_usage(rctl_t *rctl, struct proc *p) 1551 { 1552 ASSERT(MUTEX_HELD(&p->p_lock)); 1553 return (p->p_zone->zone_ipc.ipcq_semmni); 1554 } 1555 1556 /*ARGSUSED*/ 1557 static int 1558 zone_semmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval, 1559 rctl_qty_t incr, uint_t flags) 1560 { 1561 rctl_qty_t v; 1562 ASSERT(MUTEX_HELD(&p->p_lock)); 1563 ASSERT(e->rcep_t == RCENTITY_ZONE); 1564 v = e->rcep_p.zone->zone_ipc.ipcq_semmni + incr; 1565 if (v > rval->rcv_value) 1566 return (1); 1567 return (0); 1568 } 1569 1570 static rctl_ops_t zone_semmni_ops = { 1571 rcop_no_action, 1572 zone_semmni_usage, 1573 rcop_no_set, 1574 zone_semmni_test 1575 }; 1576 1577 /*ARGSUSED*/ 1578 static rctl_qty_t 1579 zone_msgmni_usage(rctl_t *rctl, struct proc *p) 1580 { 1581 ASSERT(MUTEX_HELD(&p->p_lock)); 1582 return (p->p_zone->zone_ipc.ipcq_msgmni); 1583 } 1584 1585 /*ARGSUSED*/ 1586 static int 1587 zone_msgmni_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, rctl_val_t *rval, 1588 rctl_qty_t incr, uint_t flags) 1589 { 1590 rctl_qty_t v; 1591 ASSERT(MUTEX_HELD(&p->p_lock)); 1592 ASSERT(e->rcep_t == RCENTITY_ZONE); 1593 v = e->rcep_p.zone->zone_ipc.ipcq_msgmni + incr; 1594 if (v > rval->rcv_value) 1595 return (1); 1596 return (0); 1597 } 1598 1599 static rctl_ops_t zone_msgmni_ops = { 1600 rcop_no_action, 1601 zone_msgmni_usage, 1602 rcop_no_set, 1603 zone_msgmni_test 1604 }; 1605 1606 /*ARGSUSED*/ 1607 static rctl_qty_t 1608 zone_locked_mem_usage(rctl_t *rctl, struct proc *p) 1609 { 1610 rctl_qty_t q; 1611 ASSERT(MUTEX_HELD(&p->p_lock)); 1612 mutex_enter(&p->p_zone->zone_mem_lock); 1613 q = p->p_zone->zone_locked_mem; 1614 mutex_exit(&p->p_zone->zone_mem_lock); 1615 return (q); 1616 } 1617 1618 /*ARGSUSED*/ 1619 static int 1620 zone_locked_mem_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, 1621 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags) 1622 { 1623 rctl_qty_t q; 1624 zone_t *z; 1625 1626 z = e->rcep_p.zone; 1627 ASSERT(MUTEX_HELD(&p->p_lock)); 1628 ASSERT(MUTEX_HELD(&z->zone_mem_lock)); 1629 q = z->zone_locked_mem; 1630 if (q + incr > rcntl->rcv_value) 1631 return (1); 1632 return (0); 1633 } 1634 1635 /*ARGSUSED*/ 1636 static int 1637 zone_locked_mem_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, 1638 rctl_qty_t nv) 1639 { 1640 ASSERT(MUTEX_HELD(&p->p_lock)); 1641 ASSERT(e->rcep_t == RCENTITY_ZONE); 1642 if (e->rcep_p.zone == NULL) 1643 return (0); 1644 e->rcep_p.zone->zone_locked_mem_ctl = nv; 1645 return (0); 1646 } 1647 1648 static rctl_ops_t zone_locked_mem_ops = { 1649 rcop_no_action, 1650 zone_locked_mem_usage, 1651 zone_locked_mem_set, 1652 zone_locked_mem_test 1653 }; 1654 1655 /*ARGSUSED*/ 1656 static rctl_qty_t 1657 zone_max_swap_usage(rctl_t *rctl, struct proc *p) 1658 { 1659 rctl_qty_t q; 1660 zone_t *z = p->p_zone; 1661 1662 ASSERT(MUTEX_HELD(&p->p_lock)); 1663 mutex_enter(&z->zone_mem_lock); 1664 q = z->zone_max_swap; 1665 mutex_exit(&z->zone_mem_lock); 1666 return (q); 1667 } 1668 1669 /*ARGSUSED*/ 1670 static int 1671 zone_max_swap_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, 1672 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags) 1673 { 1674 rctl_qty_t q; 1675 zone_t *z; 1676 1677 z = e->rcep_p.zone; 1678 ASSERT(MUTEX_HELD(&p->p_lock)); 1679 ASSERT(MUTEX_HELD(&z->zone_mem_lock)); 1680 q = z->zone_max_swap; 1681 if (q + incr > rcntl->rcv_value) 1682 return (1); 1683 return (0); 1684 } 1685 1686 /*ARGSUSED*/ 1687 static int 1688 zone_max_swap_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, 1689 rctl_qty_t nv) 1690 { 1691 ASSERT(MUTEX_HELD(&p->p_lock)); 1692 ASSERT(e->rcep_t == RCENTITY_ZONE); 1693 if (e->rcep_p.zone == NULL) 1694 return (0); 1695 e->rcep_p.zone->zone_max_swap_ctl = nv; 1696 return (0); 1697 } 1698 1699 static rctl_ops_t zone_max_swap_ops = { 1700 rcop_no_action, 1701 zone_max_swap_usage, 1702 zone_max_swap_set, 1703 zone_max_swap_test 1704 }; 1705 1706 /*ARGSUSED*/ 1707 static rctl_qty_t 1708 zone_max_lofi_usage(rctl_t *rctl, struct proc *p) 1709 { 1710 rctl_qty_t q; 1711 zone_t *z = p->p_zone; 1712 1713 ASSERT(MUTEX_HELD(&p->p_lock)); 1714 mutex_enter(&z->zone_rctl_lock); 1715 q = z->zone_max_lofi; 1716 mutex_exit(&z->zone_rctl_lock); 1717 return (q); 1718 } 1719 1720 /*ARGSUSED*/ 1721 static int 1722 zone_max_lofi_test(rctl_t *r, proc_t *p, rctl_entity_p_t *e, 1723 rctl_val_t *rcntl, rctl_qty_t incr, uint_t flags) 1724 { 1725 rctl_qty_t q; 1726 zone_t *z; 1727 1728 z = e->rcep_p.zone; 1729 ASSERT(MUTEX_HELD(&p->p_lock)); 1730 ASSERT(MUTEX_HELD(&z->zone_rctl_lock)); 1731 q = z->zone_max_lofi; 1732 if (q + incr > rcntl->rcv_value) 1733 return (1); 1734 return (0); 1735 } 1736 1737 /*ARGSUSED*/ 1738 static int 1739 zone_max_lofi_set(rctl_t *rctl, struct proc *p, rctl_entity_p_t *e, 1740 rctl_qty_t nv) 1741 { 1742 ASSERT(MUTEX_HELD(&p->p_lock)); 1743 ASSERT(e->rcep_t == RCENTITY_ZONE); 1744 if (e->rcep_p.zone == NULL) 1745 return (0); 1746 e->rcep_p.zone->zone_max_lofi_ctl = nv; 1747 return (0); 1748 } 1749 1750 static rctl_ops_t zone_max_lofi_ops = { 1751 rcop_no_action, 1752 zone_max_lofi_usage, 1753 zone_max_lofi_set, 1754 zone_max_lofi_test 1755 }; 1756 1757 /* 1758 * Helper function to brand the zone with a unique ID. 1759 */ 1760 static void 1761 zone_uniqid(zone_t *zone) 1762 { 1763 static uint64_t uniqid = 0; 1764 1765 ASSERT(MUTEX_HELD(&zonehash_lock)); 1766 zone->zone_uniqid = uniqid++; 1767 } 1768 1769 /* 1770 * Returns a held pointer to the "kcred" for the specified zone. 1771 */ 1772 struct cred * 1773 zone_get_kcred(zoneid_t zoneid) 1774 { 1775 zone_t *zone; 1776 cred_t *cr; 1777 1778 if ((zone = zone_find_by_id(zoneid)) == NULL) 1779 return (NULL); 1780 cr = zone->zone_kcred; 1781 crhold(cr); 1782 zone_rele(zone); 1783 return (cr); 1784 } 1785 1786 static int 1787 zone_lockedmem_kstat_update(kstat_t *ksp, int rw) 1788 { 1789 zone_t *zone = ksp->ks_private; 1790 zone_kstat_t *zk = ksp->ks_data; 1791 1792 if (rw == KSTAT_WRITE) 1793 return (EACCES); 1794 1795 zk->zk_usage.value.ui64 = zone->zone_locked_mem; 1796 zk->zk_value.value.ui64 = zone->zone_locked_mem_ctl; 1797 return (0); 1798 } 1799 1800 static int 1801 zone_nprocs_kstat_update(kstat_t *ksp, int rw) 1802 { 1803 zone_t *zone = ksp->ks_private; 1804 zone_kstat_t *zk = ksp->ks_data; 1805 1806 if (rw == KSTAT_WRITE) 1807 return (EACCES); 1808 1809 zk->zk_usage.value.ui64 = zone->zone_nprocs; 1810 zk->zk_value.value.ui64 = zone->zone_nprocs_ctl; 1811 return (0); 1812 } 1813 1814 static int 1815 zone_swapresv_kstat_update(kstat_t *ksp, int rw) 1816 { 1817 zone_t *zone = ksp->ks_private; 1818 zone_kstat_t *zk = ksp->ks_data; 1819 1820 if (rw == KSTAT_WRITE) 1821 return (EACCES); 1822 1823 zk->zk_usage.value.ui64 = zone->zone_max_swap; 1824 zk->zk_value.value.ui64 = zone->zone_max_swap_ctl; 1825 return (0); 1826 } 1827 1828 static kstat_t * 1829 zone_kstat_create_common(zone_t *zone, char *name, 1830 int (*updatefunc) (kstat_t *, int)) 1831 { 1832 kstat_t *ksp; 1833 zone_kstat_t *zk; 1834 1835 ksp = rctl_kstat_create_zone(zone, name, KSTAT_TYPE_NAMED, 1836 sizeof (zone_kstat_t) / sizeof (kstat_named_t), 1837 KSTAT_FLAG_VIRTUAL); 1838 1839 if (ksp == NULL) 1840 return (NULL); 1841 1842 zk = ksp->ks_data = kmem_alloc(sizeof (zone_kstat_t), KM_SLEEP); 1843 ksp->ks_data_size += strlen(zone->zone_name) + 1; 1844 kstat_named_init(&zk->zk_zonename, "zonename", KSTAT_DATA_STRING); 1845 kstat_named_setstr(&zk->zk_zonename, zone->zone_name); 1846 kstat_named_init(&zk->zk_usage, "usage", KSTAT_DATA_UINT64); 1847 kstat_named_init(&zk->zk_value, "value", KSTAT_DATA_UINT64); 1848 ksp->ks_update = updatefunc; 1849 ksp->ks_private = zone; 1850 kstat_install(ksp); 1851 return (ksp); 1852 } 1853 1854 1855 static int 1856 zone_mcap_kstat_update(kstat_t *ksp, int rw) 1857 { 1858 zone_t *zone = ksp->ks_private; 1859 zone_mcap_kstat_t *zmp = ksp->ks_data; 1860 1861 if (rw == KSTAT_WRITE) 1862 return (EACCES); 1863 1864 zmp->zm_pgpgin.value.ui64 = zone->zone_pgpgin; 1865 zmp->zm_anonpgin.value.ui64 = zone->zone_anonpgin; 1866 zmp->zm_execpgin.value.ui64 = zone->zone_execpgin; 1867 zmp->zm_fspgin.value.ui64 = zone->zone_fspgin; 1868 zmp->zm_anon_alloc_fail.value.ui64 = zone->zone_anon_alloc_fail; 1869 1870 return (0); 1871 } 1872 1873 static kstat_t * 1874 zone_mcap_kstat_create(zone_t *zone) 1875 { 1876 kstat_t *ksp; 1877 zone_mcap_kstat_t *zmp; 1878 1879 if ((ksp = kstat_create_zone("memory_cap", zone->zone_id, 1880 zone->zone_name, "zone_memory_cap", KSTAT_TYPE_NAMED, 1881 sizeof (zone_mcap_kstat_t) / sizeof (kstat_named_t), 1882 KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL) 1883 return (NULL); 1884 1885 if (zone->zone_id != GLOBAL_ZONEID) 1886 kstat_zone_add(ksp, GLOBAL_ZONEID); 1887 1888 zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_mcap_kstat_t), KM_SLEEP); 1889 ksp->ks_data_size += strlen(zone->zone_name) + 1; 1890 ksp->ks_lock = &zone->zone_mcap_lock; 1891 zone->zone_mcap_stats = zmp; 1892 1893 /* The kstat "name" field is not large enough for a full zonename */ 1894 kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING); 1895 kstat_named_setstr(&zmp->zm_zonename, zone->zone_name); 1896 kstat_named_init(&zmp->zm_pgpgin, "pgpgin", KSTAT_DATA_UINT64); 1897 kstat_named_init(&zmp->zm_anonpgin, "anonpgin", KSTAT_DATA_UINT64); 1898 kstat_named_init(&zmp->zm_execpgin, "execpgin", KSTAT_DATA_UINT64); 1899 kstat_named_init(&zmp->zm_fspgin, "fspgin", KSTAT_DATA_UINT64); 1900 kstat_named_init(&zmp->zm_anon_alloc_fail, "anon_alloc_fail", 1901 KSTAT_DATA_UINT64); 1902 1903 ksp->ks_update = zone_mcap_kstat_update; 1904 ksp->ks_private = zone; 1905 1906 kstat_install(ksp); 1907 return (ksp); 1908 } 1909 1910 static int 1911 zone_misc_kstat_update(kstat_t *ksp, int rw) 1912 { 1913 zone_t *zone = ksp->ks_private; 1914 zone_misc_kstat_t *zmp = ksp->ks_data; 1915 hrtime_t hrtime; 1916 uint64_t tmp; 1917 1918 if (rw == KSTAT_WRITE) 1919 return (EACCES); 1920 1921 tmp = cpu_uarray_sum(zone->zone_ustate, ZONE_USTATE_STIME); 1922 hrtime = UINT64_OVERFLOW_TO_INT64(tmp); 1923 scalehrtime(&hrtime); 1924 zmp->zm_stime.value.ui64 = hrtime; 1925 1926 tmp = cpu_uarray_sum(zone->zone_ustate, ZONE_USTATE_UTIME); 1927 hrtime = UINT64_OVERFLOW_TO_INT64(tmp); 1928 scalehrtime(&hrtime); 1929 zmp->zm_utime.value.ui64 = hrtime; 1930 1931 tmp = cpu_uarray_sum(zone->zone_ustate, ZONE_USTATE_WTIME); 1932 hrtime = UINT64_OVERFLOW_TO_INT64(tmp); 1933 scalehrtime(&hrtime); 1934 zmp->zm_wtime.value.ui64 = hrtime; 1935 1936 zmp->zm_avenrun1.value.ui32 = zone->zone_avenrun[0]; 1937 zmp->zm_avenrun5.value.ui32 = zone->zone_avenrun[1]; 1938 zmp->zm_avenrun15.value.ui32 = zone->zone_avenrun[2]; 1939 1940 zmp->zm_ffcap.value.ui32 = zone->zone_ffcap; 1941 zmp->zm_ffnoproc.value.ui32 = zone->zone_ffnoproc; 1942 zmp->zm_ffnomem.value.ui32 = zone->zone_ffnomem; 1943 zmp->zm_ffmisc.value.ui32 = zone->zone_ffmisc; 1944 1945 zmp->zm_nested_intp.value.ui32 = zone->zone_nested_intp; 1946 1947 zmp->zm_init_pid.value.ui32 = zone->zone_proc_initpid; 1948 zmp->zm_boot_time.value.ui64 = (uint64_t)zone->zone_boot_time; 1949 1950 return (0); 1951 } 1952 1953 static kstat_t * 1954 zone_misc_kstat_create(zone_t *zone) 1955 { 1956 kstat_t *ksp; 1957 zone_misc_kstat_t *zmp; 1958 1959 if ((ksp = kstat_create_zone("zones", zone->zone_id, 1960 zone->zone_name, "zone_misc", KSTAT_TYPE_NAMED, 1961 sizeof (zone_misc_kstat_t) / sizeof (kstat_named_t), 1962 KSTAT_FLAG_VIRTUAL, zone->zone_id)) == NULL) 1963 return (NULL); 1964 1965 if (zone->zone_id != GLOBAL_ZONEID) 1966 kstat_zone_add(ksp, GLOBAL_ZONEID); 1967 1968 zmp = ksp->ks_data = kmem_zalloc(sizeof (zone_misc_kstat_t), KM_SLEEP); 1969 ksp->ks_data_size += strlen(zone->zone_name) + 1; 1970 ksp->ks_lock = &zone->zone_misc_lock; 1971 zone->zone_misc_stats = zmp; 1972 1973 /* The kstat "name" field is not large enough for a full zonename */ 1974 kstat_named_init(&zmp->zm_zonename, "zonename", KSTAT_DATA_STRING); 1975 kstat_named_setstr(&zmp->zm_zonename, zone->zone_name); 1976 kstat_named_init(&zmp->zm_utime, "nsec_user", KSTAT_DATA_UINT64); 1977 kstat_named_init(&zmp->zm_stime, "nsec_sys", KSTAT_DATA_UINT64); 1978 kstat_named_init(&zmp->zm_wtime, "nsec_waitrq", KSTAT_DATA_UINT64); 1979 kstat_named_init(&zmp->zm_avenrun1, "avenrun_1min", KSTAT_DATA_UINT32); 1980 kstat_named_init(&zmp->zm_avenrun5, "avenrun_5min", KSTAT_DATA_UINT32); 1981 kstat_named_init(&zmp->zm_avenrun15, "avenrun_15min", 1982 KSTAT_DATA_UINT32); 1983 kstat_named_init(&zmp->zm_ffcap, "forkfail_cap", KSTAT_DATA_UINT32); 1984 kstat_named_init(&zmp->zm_ffnoproc, "forkfail_noproc", 1985 KSTAT_DATA_UINT32); 1986 kstat_named_init(&zmp->zm_ffnomem, "forkfail_nomem", KSTAT_DATA_UINT32); 1987 kstat_named_init(&zmp->zm_ffmisc, "forkfail_misc", KSTAT_DATA_UINT32); 1988 kstat_named_init(&zmp->zm_nested_intp, "nested_interp", 1989 KSTAT_DATA_UINT32); 1990 kstat_named_init(&zmp->zm_init_pid, "init_pid", KSTAT_DATA_UINT32); 1991 kstat_named_init(&zmp->zm_boot_time, "boot_time", KSTAT_DATA_UINT64); 1992 1993 ksp->ks_update = zone_misc_kstat_update; 1994 ksp->ks_private = zone; 1995 1996 kstat_install(ksp); 1997 return (ksp); 1998 } 1999 2000 static void 2001 zone_kstat_create(zone_t *zone) 2002 { 2003 zone->zone_lockedmem_kstat = zone_kstat_create_common(zone, 2004 "lockedmem", zone_lockedmem_kstat_update); 2005 zone->zone_swapresv_kstat = zone_kstat_create_common(zone, 2006 "swapresv", zone_swapresv_kstat_update); 2007 zone->zone_nprocs_kstat = zone_kstat_create_common(zone, 2008 "nprocs", zone_nprocs_kstat_update); 2009 2010 if ((zone->zone_mcap_ksp = zone_mcap_kstat_create(zone)) == NULL) { 2011 zone->zone_mcap_stats = kmem_zalloc( 2012 sizeof (zone_mcap_kstat_t), KM_SLEEP); 2013 } 2014 2015 if ((zone->zone_misc_ksp = zone_misc_kstat_create(zone)) == NULL) { 2016 zone->zone_misc_stats = kmem_zalloc( 2017 sizeof (zone_misc_kstat_t), KM_SLEEP); 2018 } 2019 } 2020 2021 static void 2022 zone_kstat_delete_common(kstat_t **pkstat, size_t datasz) 2023 { 2024 void *data; 2025 2026 if (*pkstat != NULL) { 2027 data = (*pkstat)->ks_data; 2028 kstat_delete(*pkstat); 2029 kmem_free(data, datasz); 2030 *pkstat = NULL; 2031 } 2032 } 2033 2034 static void 2035 zone_kstat_delete(zone_t *zone) 2036 { 2037 zone_kstat_delete_common(&zone->zone_lockedmem_kstat, 2038 sizeof (zone_kstat_t)); 2039 zone_kstat_delete_common(&zone->zone_swapresv_kstat, 2040 sizeof (zone_kstat_t)); 2041 zone_kstat_delete_common(&zone->zone_nprocs_kstat, 2042 sizeof (zone_kstat_t)); 2043 zone_kstat_delete_common(&zone->zone_mcap_ksp, 2044 sizeof (zone_mcap_kstat_t)); 2045 zone_kstat_delete_common(&zone->zone_misc_ksp, 2046 sizeof (zone_misc_kstat_t)); 2047 } 2048 2049 /* 2050 * Called very early on in boot to initialize the ZSD list so that 2051 * zone_key_create() can be called before zone_init(). It also initializes 2052 * portions of zone0 which may be used before zone_init() is called. The 2053 * variable "global_zone" will be set when zone0 is fully initialized by 2054 * zone_init(). 2055 */ 2056 void 2057 zone_zsd_init(void) 2058 { 2059 mutex_init(&zonehash_lock, NULL, MUTEX_DEFAULT, NULL); 2060 mutex_init(&zsd_key_lock, NULL, MUTEX_DEFAULT, NULL); 2061 list_create(&zsd_registered_keys, sizeof (struct zsd_entry), 2062 offsetof(struct zsd_entry, zsd_linkage)); 2063 list_create(&zone_active, sizeof (zone_t), 2064 offsetof(zone_t, zone_linkage)); 2065 list_create(&zone_deathrow, sizeof (zone_t), 2066 offsetof(zone_t, zone_linkage)); 2067 2068 mutex_init(&zone0.zone_lock, NULL, MUTEX_DEFAULT, NULL); 2069 mutex_init(&zone0.zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL); 2070 mutex_init(&zone0.zone_mem_lock, NULL, MUTEX_DEFAULT, NULL); 2071 zone0.zone_shares = 1; 2072 zone0.zone_nlwps = 0; 2073 zone0.zone_nlwps_ctl = INT_MAX; 2074 zone0.zone_nprocs = 0; 2075 zone0.zone_nprocs_ctl = INT_MAX; 2076 zone0.zone_locked_mem = 0; 2077 zone0.zone_locked_mem_ctl = UINT64_MAX; 2078 ASSERT(zone0.zone_max_swap == 0); 2079 zone0.zone_max_swap_ctl = UINT64_MAX; 2080 zone0.zone_max_lofi = 0; 2081 zone0.zone_max_lofi_ctl = UINT64_MAX; 2082 zone0.zone_shmmax = 0; 2083 zone0.zone_ipc.ipcq_shmmni = 0; 2084 zone0.zone_ipc.ipcq_semmni = 0; 2085 zone0.zone_ipc.ipcq_msgmni = 0; 2086 zone0.zone_name = GLOBAL_ZONENAME; 2087 zone0.zone_nodename = utsname.nodename; 2088 zone0.zone_domain = srpc_domain; 2089 zone0.zone_hostid = HW_INVALID_HOSTID; 2090 zone0.zone_fs_allowed = NULL; 2091 psecflags_default(&zone0.zone_secflags); 2092 zone0.zone_ref = 1; 2093 zone0.zone_id = GLOBAL_ZONEID; 2094 zone0.zone_status = ZONE_IS_RUNNING; 2095 zone0.zone_rootpath = "/"; 2096 zone0.zone_rootpathlen = 2; 2097 zone0.zone_psetid = ZONE_PS_INVAL; 2098 zone0.zone_ncpus = 0; 2099 zone0.zone_ncpus_online = 0; 2100 zone0.zone_proc_initpid = 1; 2101 zone0.zone_initname = initname; 2102 zone0.zone_lockedmem_kstat = NULL; 2103 zone0.zone_swapresv_kstat = NULL; 2104 zone0.zone_nprocs_kstat = NULL; 2105 2106 list_create(&zone0.zone_ref_list, sizeof (zone_ref_t), 2107 offsetof(zone_ref_t, zref_linkage)); 2108 list_create(&zone0.zone_zsd, sizeof (struct zsd_entry), 2109 offsetof(struct zsd_entry, zsd_linkage)); 2110 list_insert_head(&zone_active, &zone0); 2111 2112 /* 2113 * The root filesystem is not mounted yet, so zone_rootvp cannot be set 2114 * to anything meaningful. It is assigned to be 'rootdir' in 2115 * vfs_mountroot(). 2116 */ 2117 zone0.zone_rootvp = NULL; 2118 zone0.zone_vfslist = NULL; 2119 zone0.zone_bootargs = initargs; 2120 zone0.zone_privset = kmem_alloc(sizeof (priv_set_t), KM_SLEEP); 2121 /* 2122 * The global zone has all privileges 2123 */ 2124 priv_fillset(zone0.zone_privset); 2125 /* 2126 * Add p0 to the global zone 2127 */ 2128 zone0.zone_zsched = &p0; 2129 p0.p_zone = &zone0; 2130 } 2131 2132 /* 2133 * Compute a hash value based on the contents of the label and the DOI. The 2134 * hash algorithm is somewhat arbitrary, but is based on the observation that 2135 * humans will likely pick labels that differ by amounts that work out to be 2136 * multiples of the number of hash chains, and thus stirring in some primes 2137 * should help. 2138 */ 2139 static uint_t 2140 hash_bylabel(void *hdata, mod_hash_key_t key) 2141 { 2142 const ts_label_t *lab = (ts_label_t *)key; 2143 const uint32_t *up, *ue; 2144 uint_t hash; 2145 int i; 2146 2147 _NOTE(ARGUNUSED(hdata)); 2148 2149 hash = lab->tsl_doi + (lab->tsl_doi << 1); 2150 /* we depend on alignment of label, but not representation */ 2151 up = (const uint32_t *)&lab->tsl_label; 2152 ue = up + sizeof (lab->tsl_label) / sizeof (*up); 2153 i = 1; 2154 while (up < ue) { 2155 /* using 2^n + 1, 1 <= n <= 16 as source of many primes */ 2156 hash += *up + (*up << ((i % 16) + 1)); 2157 up++; 2158 i++; 2159 } 2160 return (hash); 2161 } 2162 2163 /* 2164 * All that mod_hash cares about here is zero (equal) versus non-zero (not 2165 * equal). This may need to be changed if less than / greater than is ever 2166 * needed. 2167 */ 2168 static int 2169 hash_labelkey_cmp(mod_hash_key_t key1, mod_hash_key_t key2) 2170 { 2171 ts_label_t *lab1 = (ts_label_t *)key1; 2172 ts_label_t *lab2 = (ts_label_t *)key2; 2173 2174 return (label_equal(lab1, lab2) ? 0 : 1); 2175 } 2176 2177 /* 2178 * Called by main() to initialize the zones framework. 2179 */ 2180 void 2181 zone_init(void) 2182 { 2183 rctl_dict_entry_t *rde; 2184 rctl_val_t *dval; 2185 rctl_set_t *set; 2186 rctl_alloc_gp_t *gp; 2187 rctl_entity_p_t e; 2188 int res; 2189 2190 ASSERT(curproc == &p0); 2191 2192 /* 2193 * Create ID space for zone IDs. ID 0 is reserved for the 2194 * global zone. 2195 */ 2196 zoneid_space = id_space_create("zoneid_space", 1, MAX_ZONEID); 2197 2198 /* 2199 * Initialize generic zone resource controls, if any. 2200 */ 2201 rc_zone_cpu_shares = rctl_register("zone.cpu-shares", 2202 RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_NEVER | 2203 RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT | RCTL_GLOBAL_SYSLOG_NEVER, 2204 FSS_MAXSHARES, FSS_MAXSHARES, &zone_cpu_shares_ops); 2205 2206 rc_zone_cpu_cap = rctl_register("zone.cpu-cap", 2207 RCENTITY_ZONE, RCTL_GLOBAL_SIGNAL_NEVER | RCTL_GLOBAL_DENY_ALWAYS | 2208 RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT |RCTL_GLOBAL_SYSLOG_NEVER | 2209 RCTL_GLOBAL_INFINITE, 2210 MAXCAP, MAXCAP, &zone_cpu_cap_ops); 2211 2212 rc_zone_nlwps = rctl_register("zone.max-lwps", RCENTITY_ZONE, 2213 RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT, 2214 INT_MAX, INT_MAX, &zone_lwps_ops); 2215 2216 rc_zone_nprocs = rctl_register("zone.max-processes", RCENTITY_ZONE, 2217 RCTL_GLOBAL_NOACTION | RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT, 2218 INT_MAX, INT_MAX, &zone_procs_ops); 2219 2220 /* 2221 * System V IPC resource controls 2222 */ 2223 rc_zone_msgmni = rctl_register("zone.max-msg-ids", 2224 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC | 2225 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_msgmni_ops); 2226 2227 rc_zone_semmni = rctl_register("zone.max-sem-ids", 2228 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC | 2229 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_semmni_ops); 2230 2231 rc_zone_shmmni = rctl_register("zone.max-shm-ids", 2232 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC | 2233 RCTL_GLOBAL_COUNT, IPC_IDS_MAX, IPC_IDS_MAX, &zone_shmmni_ops); 2234 2235 rc_zone_shmmax = rctl_register("zone.max-shm-memory", 2236 RCENTITY_ZONE, RCTL_GLOBAL_DENY_ALWAYS | RCTL_GLOBAL_NOBASIC | 2237 RCTL_GLOBAL_BYTES, UINT64_MAX, UINT64_MAX, &zone_shmmax_ops); 2238 2239 /* 2240 * Create a rctl_val with PRIVILEGED, NOACTION, value = 1. Then attach 2241 * this at the head of the rctl_dict_entry for ``zone.cpu-shares''. 2242 */ 2243 dval = kmem_cache_alloc(rctl_val_cache, KM_SLEEP); 2244 bzero(dval, sizeof (rctl_val_t)); 2245 dval->rcv_value = 1; 2246 dval->rcv_privilege = RCPRIV_PRIVILEGED; 2247 dval->rcv_flagaction = RCTL_LOCAL_NOACTION; 2248 dval->rcv_action_recip_pid = -1; 2249 2250 rde = rctl_dict_lookup("zone.cpu-shares"); 2251 (void) rctl_val_list_insert(&rde->rcd_default_value, dval); 2252 2253 rc_zone_locked_mem = rctl_register("zone.max-locked-memory", 2254 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES | 2255 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX, 2256 &zone_locked_mem_ops); 2257 2258 rc_zone_max_swap = rctl_register("zone.max-swap", 2259 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_BYTES | 2260 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX, 2261 &zone_max_swap_ops); 2262 2263 rc_zone_max_lofi = rctl_register("zone.max-lofi", 2264 RCENTITY_ZONE, RCTL_GLOBAL_NOBASIC | RCTL_GLOBAL_COUNT | 2265 RCTL_GLOBAL_DENY_ALWAYS, UINT64_MAX, UINT64_MAX, 2266 &zone_max_lofi_ops); 2267 2268 /* 2269 * Initialize the ``global zone''. 2270 */ 2271 set = rctl_set_create(); 2272 gp = rctl_set_init_prealloc(RCENTITY_ZONE); 2273 mutex_enter(&p0.p_lock); 2274 e.rcep_p.zone = &zone0; 2275 e.rcep_t = RCENTITY_ZONE; 2276 zone0.zone_rctls = rctl_set_init(RCENTITY_ZONE, &p0, &e, set, 2277 gp); 2278 2279 zone0.zone_nlwps = p0.p_lwpcnt; 2280 zone0.zone_nprocs = 1; 2281 zone0.zone_ntasks = 1; 2282 mutex_exit(&p0.p_lock); 2283 zone0.zone_restart_init = B_TRUE; 2284 zone0.zone_reboot_on_init_exit = B_FALSE; 2285 zone0.zone_restart_init_0 = B_FALSE; 2286 zone0.zone_brand = &native_brand; 2287 rctl_prealloc_destroy(gp); 2288 /* 2289 * pool_default hasn't been initialized yet, so we let pool_init() 2290 * take care of making sure the global zone is in the default pool. 2291 */ 2292 2293 /* 2294 * Initialize global zone kstats 2295 */ 2296 zone_kstat_create(&zone0); 2297 2298 /* 2299 * Initialize zone label. 2300 * mlp are initialized when tnzonecfg is loaded. 2301 */ 2302 zone0.zone_slabel = l_admin_low; 2303 rw_init(&zone0.zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL); 2304 label_hold(l_admin_low); 2305 2306 /* 2307 * Initialise the lock for the database structure used by mntfs. 2308 */ 2309 rw_init(&zone0.zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL); 2310 2311 zone0.zone_ustate = cpu_uarray_zalloc(ZONE_USTATE_MAX, KM_SLEEP); 2312 2313 mutex_enter(&zonehash_lock); 2314 zone_uniqid(&zone0); 2315 ASSERT(zone0.zone_uniqid == GLOBAL_ZONEUNIQID); 2316 2317 zonehashbyid = mod_hash_create_idhash("zone_by_id", zone_hash_size, 2318 mod_hash_null_valdtor); 2319 zonehashbyname = mod_hash_create_strhash("zone_by_name", 2320 zone_hash_size, mod_hash_null_valdtor); 2321 /* 2322 * maintain zonehashbylabel only for labeled systems 2323 */ 2324 if (is_system_labeled()) 2325 zonehashbylabel = mod_hash_create_extended("zone_by_label", 2326 zone_hash_size, mod_hash_null_keydtor, 2327 mod_hash_null_valdtor, hash_bylabel, NULL, 2328 hash_labelkey_cmp, KM_SLEEP); 2329 zonecount = 1; 2330 2331 (void) mod_hash_insert(zonehashbyid, (mod_hash_key_t)GLOBAL_ZONEID, 2332 (mod_hash_val_t)&zone0); 2333 (void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)zone0.zone_name, 2334 (mod_hash_val_t)&zone0); 2335 if (is_system_labeled()) { 2336 zone0.zone_flags |= ZF_HASHED_LABEL; 2337 (void) mod_hash_insert(zonehashbylabel, 2338 (mod_hash_key_t)zone0.zone_slabel, (mod_hash_val_t)&zone0); 2339 } 2340 mutex_exit(&zonehash_lock); 2341 2342 /* 2343 * We avoid setting zone_kcred until now, since kcred is initialized 2344 * sometime after zone_zsd_init() and before zone_init(). 2345 */ 2346 zone0.zone_kcred = kcred; 2347 /* 2348 * The global zone is fully initialized (except for zone_rootvp which 2349 * will be set when the root filesystem is mounted). 2350 */ 2351 global_zone = &zone0; 2352 2353 /* 2354 * Setup an event channel to send zone status change notifications on 2355 */ 2356 res = sysevent_evc_bind(ZONE_EVENT_CHANNEL, &zone_event_chan, 2357 EVCH_CREAT); 2358 2359 if (res) 2360 panic("Sysevent_evc_bind failed during zone setup.\n"); 2361 2362 } 2363 2364 static void 2365 zone_free(zone_t *zone) 2366 { 2367 ASSERT(zone != global_zone); 2368 ASSERT(zone->zone_ntasks == 0); 2369 ASSERT(zone->zone_nlwps == 0); 2370 ASSERT(zone->zone_nprocs == 0); 2371 ASSERT(zone->zone_cred_ref == 0); 2372 ASSERT(zone->zone_kcred == NULL); 2373 ASSERT(zone_status_get(zone) == ZONE_IS_DEAD || 2374 zone_status_get(zone) == ZONE_IS_UNINITIALIZED); 2375 ASSERT(list_is_empty(&zone->zone_ref_list)); 2376 2377 /* 2378 * Remove any zone caps. 2379 */ 2380 cpucaps_zone_remove(zone); 2381 2382 ASSERT(zone->zone_cpucap == NULL); 2383 2384 /* remove from deathrow list */ 2385 if (zone_status_get(zone) == ZONE_IS_DEAD) { 2386 ASSERT(zone->zone_ref == 0); 2387 mutex_enter(&zone_deathrow_lock); 2388 list_remove(&zone_deathrow, zone); 2389 mutex_exit(&zone_deathrow_lock); 2390 } 2391 2392 list_destroy(&zone->zone_ref_list); 2393 zone_free_zsd(zone); 2394 zone_free_datasets(zone); 2395 list_destroy(&zone->zone_dl_list); 2396 2397 cpu_uarray_free(zone->zone_ustate); 2398 2399 if (zone->zone_rootvp != NULL) 2400 VN_RELE(zone->zone_rootvp); 2401 if (zone->zone_rootpath) 2402 kmem_free(zone->zone_rootpath, zone->zone_rootpathlen); 2403 if (zone->zone_name != NULL) 2404 kmem_free(zone->zone_name, ZONENAME_MAX); 2405 if (zone->zone_slabel != NULL) 2406 label_rele(zone->zone_slabel); 2407 if (zone->zone_nodename != NULL) 2408 kmem_free(zone->zone_nodename, _SYS_NMLN); 2409 if (zone->zone_domain != NULL) 2410 kmem_free(zone->zone_domain, _SYS_NMLN); 2411 if (zone->zone_privset != NULL) 2412 kmem_free(zone->zone_privset, sizeof (priv_set_t)); 2413 if (zone->zone_rctls != NULL) 2414 rctl_set_free(zone->zone_rctls); 2415 if (zone->zone_bootargs != NULL) 2416 strfree(zone->zone_bootargs); 2417 if (zone->zone_initname != NULL) 2418 strfree(zone->zone_initname); 2419 if (zone->zone_fs_allowed != NULL) 2420 strfree(zone->zone_fs_allowed); 2421 if (zone->zone_pfexecd != NULL) 2422 klpd_freelist(&zone->zone_pfexecd); 2423 id_free(zoneid_space, zone->zone_id); 2424 mutex_destroy(&zone->zone_lock); 2425 cv_destroy(&zone->zone_cv); 2426 rw_destroy(&zone->zone_mlps.mlpl_rwlock); 2427 rw_destroy(&zone->zone_mntfs_db_lock); 2428 kmem_free(zone, sizeof (zone_t)); 2429 } 2430 2431 /* 2432 * See block comment at the top of this file for information about zone 2433 * status values. 2434 */ 2435 /* 2436 * Convenience function for setting zone status. 2437 */ 2438 static void 2439 zone_status_set(zone_t *zone, zone_status_t status) 2440 { 2441 2442 nvlist_t *nvl = NULL; 2443 ASSERT(MUTEX_HELD(&zone_status_lock)); 2444 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE && 2445 status >= zone_status_get(zone)); 2446 2447 if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) || 2448 nvlist_add_string(nvl, ZONE_CB_NAME, zone->zone_name) || 2449 nvlist_add_string(nvl, ZONE_CB_NEWSTATE, 2450 zone_status_table[status]) || 2451 nvlist_add_string(nvl, ZONE_CB_OLDSTATE, 2452 zone_status_table[zone->zone_status]) || 2453 nvlist_add_int32(nvl, ZONE_CB_ZONEID, zone->zone_id) || 2454 nvlist_add_uint64(nvl, ZONE_CB_TIMESTAMP, (uint64_t)gethrtime()) || 2455 sysevent_evc_publish(zone_event_chan, ZONE_EVENT_STATUS_CLASS, 2456 ZONE_EVENT_STATUS_SUBCLASS, "sun.com", "kernel", nvl, EVCH_SLEEP)) { 2457 #ifdef DEBUG 2458 (void) printf( 2459 "Failed to allocate and send zone state change event.\n"); 2460 #endif 2461 } 2462 nvlist_free(nvl); 2463 2464 zone->zone_status = status; 2465 2466 cv_broadcast(&zone->zone_cv); 2467 } 2468 2469 /* 2470 * Public function to retrieve the zone status. The zone status may 2471 * change after it is retrieved. 2472 */ 2473 zone_status_t 2474 zone_status_get(zone_t *zone) 2475 { 2476 return (zone->zone_status); 2477 } 2478 2479 static int 2480 zone_set_bootargs(zone_t *zone, const char *zone_bootargs) 2481 { 2482 char *buf = kmem_zalloc(BOOTARGS_MAX, KM_SLEEP); 2483 int err = 0; 2484 2485 ASSERT(zone != global_zone); 2486 if ((err = copyinstr(zone_bootargs, buf, BOOTARGS_MAX, NULL)) != 0) 2487 goto done; /* EFAULT or ENAMETOOLONG */ 2488 2489 if (zone->zone_bootargs != NULL) 2490 strfree(zone->zone_bootargs); 2491 2492 zone->zone_bootargs = strdup(buf); 2493 2494 done: 2495 kmem_free(buf, BOOTARGS_MAX); 2496 return (err); 2497 } 2498 2499 static int 2500 zone_set_brand(zone_t *zone, const char *brand) 2501 { 2502 struct brand_attr *attrp; 2503 brand_t *bp; 2504 2505 attrp = kmem_alloc(sizeof (struct brand_attr), KM_SLEEP); 2506 if (copyin(brand, attrp, sizeof (struct brand_attr)) != 0) { 2507 kmem_free(attrp, sizeof (struct brand_attr)); 2508 return (EFAULT); 2509 } 2510 2511 bp = brand_register_zone(attrp); 2512 kmem_free(attrp, sizeof (struct brand_attr)); 2513 if (bp == NULL) 2514 return (EINVAL); 2515 2516 /* 2517 * This is the only place where a zone can change it's brand. 2518 * We already need to hold zone_status_lock to check the zone 2519 * status, so we'll just use that lock to serialize zone 2520 * branding requests as well. 2521 */ 2522 mutex_enter(&zone_status_lock); 2523 2524 /* Re-Branding is not allowed and the zone can't be booted yet */ 2525 if ((ZONE_IS_BRANDED(zone)) || 2526 (zone_status_get(zone) >= ZONE_IS_BOOTING)) { 2527 mutex_exit(&zone_status_lock); 2528 brand_unregister_zone(bp); 2529 return (EINVAL); 2530 } 2531 2532 /* set up the brand specific data */ 2533 zone->zone_brand = bp; 2534 ZBROP(zone)->b_init_brand_data(zone); 2535 2536 mutex_exit(&zone_status_lock); 2537 return (0); 2538 } 2539 2540 static int 2541 zone_set_secflags(zone_t *zone, const psecflags_t *zone_secflags) 2542 { 2543 int err = 0; 2544 psecflags_t psf; 2545 2546 ASSERT(zone != global_zone); 2547 2548 if ((err = copyin(zone_secflags, &psf, sizeof (psf))) != 0) 2549 return (err); 2550 2551 if (zone_status_get(zone) > ZONE_IS_READY) 2552 return (EINVAL); 2553 2554 if (!psecflags_validate(&psf)) 2555 return (EINVAL); 2556 2557 (void) memcpy(&zone->zone_secflags, &psf, sizeof (psf)); 2558 2559 /* Set security flags on the zone's zsched */ 2560 (void) memcpy(&zone->zone_zsched->p_secflags, &zone->zone_secflags, 2561 sizeof (zone->zone_zsched->p_secflags)); 2562 2563 return (0); 2564 } 2565 2566 static int 2567 zone_set_fs_allowed(zone_t *zone, const char *zone_fs_allowed) 2568 { 2569 char *buf = kmem_zalloc(ZONE_FS_ALLOWED_MAX, KM_SLEEP); 2570 int err = 0; 2571 2572 ASSERT(zone != global_zone); 2573 if ((err = copyinstr(zone_fs_allowed, buf, 2574 ZONE_FS_ALLOWED_MAX, NULL)) != 0) 2575 goto done; 2576 2577 if (zone->zone_fs_allowed != NULL) 2578 strfree(zone->zone_fs_allowed); 2579 2580 zone->zone_fs_allowed = strdup(buf); 2581 2582 done: 2583 kmem_free(buf, ZONE_FS_ALLOWED_MAX); 2584 return (err); 2585 } 2586 2587 static int 2588 zone_set_initname(zone_t *zone, const char *zone_initname) 2589 { 2590 char initname[INITNAME_SZ]; 2591 size_t len; 2592 int err = 0; 2593 2594 ASSERT(zone != global_zone); 2595 if ((err = copyinstr(zone_initname, initname, INITNAME_SZ, &len)) != 0) 2596 return (err); /* EFAULT or ENAMETOOLONG */ 2597 2598 if (zone->zone_initname != NULL) 2599 strfree(zone->zone_initname); 2600 2601 zone->zone_initname = kmem_alloc(strlen(initname) + 1, KM_SLEEP); 2602 (void) strcpy(zone->zone_initname, initname); 2603 return (0); 2604 } 2605 2606 static int 2607 zone_set_phys_mcap(zone_t *zone, const uint64_t *zone_mcap) 2608 { 2609 uint64_t mcap; 2610 int err = 0; 2611 2612 if ((err = copyin(zone_mcap, &mcap, sizeof (uint64_t))) == 0) 2613 zone->zone_phys_mcap = mcap; 2614 2615 return (err); 2616 } 2617 2618 static int 2619 zone_set_sched_class(zone_t *zone, const char *new_class) 2620 { 2621 char sched_class[PC_CLNMSZ]; 2622 id_t classid; 2623 int err; 2624 2625 ASSERT(zone != global_zone); 2626 if ((err = copyinstr(new_class, sched_class, PC_CLNMSZ, NULL)) != 0) 2627 return (err); /* EFAULT or ENAMETOOLONG */ 2628 2629 if (getcid(sched_class, &classid) != 0 || CLASS_KERNEL(classid)) 2630 return (set_errno(EINVAL)); 2631 zone->zone_defaultcid = classid; 2632 ASSERT(zone->zone_defaultcid > 0 && 2633 zone->zone_defaultcid < loaded_classes); 2634 2635 return (0); 2636 } 2637 2638 /* 2639 * Block indefinitely waiting for (zone_status >= status) 2640 */ 2641 void 2642 zone_status_wait(zone_t *zone, zone_status_t status) 2643 { 2644 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE); 2645 2646 mutex_enter(&zone_status_lock); 2647 while (zone->zone_status < status) { 2648 cv_wait(&zone->zone_cv, &zone_status_lock); 2649 } 2650 mutex_exit(&zone_status_lock); 2651 } 2652 2653 /* 2654 * Private CPR-safe version of zone_status_wait(). 2655 */ 2656 static void 2657 zone_status_wait_cpr(zone_t *zone, zone_status_t status, char *str) 2658 { 2659 callb_cpr_t cprinfo; 2660 2661 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE); 2662 2663 CALLB_CPR_INIT(&cprinfo, &zone_status_lock, callb_generic_cpr, 2664 str); 2665 mutex_enter(&zone_status_lock); 2666 while (zone->zone_status < status) { 2667 CALLB_CPR_SAFE_BEGIN(&cprinfo); 2668 cv_wait(&zone->zone_cv, &zone_status_lock); 2669 CALLB_CPR_SAFE_END(&cprinfo, &zone_status_lock); 2670 } 2671 /* 2672 * zone_status_lock is implicitly released by the following. 2673 */ 2674 CALLB_CPR_EXIT(&cprinfo); 2675 } 2676 2677 /* 2678 * Block until zone enters requested state or signal is received. Return (0) 2679 * if signaled, non-zero otherwise. 2680 */ 2681 int 2682 zone_status_wait_sig(zone_t *zone, zone_status_t status) 2683 { 2684 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE); 2685 2686 mutex_enter(&zone_status_lock); 2687 while (zone->zone_status < status) { 2688 if (!cv_wait_sig(&zone->zone_cv, &zone_status_lock)) { 2689 mutex_exit(&zone_status_lock); 2690 return (0); 2691 } 2692 } 2693 mutex_exit(&zone_status_lock); 2694 return (1); 2695 } 2696 2697 /* 2698 * Block until the zone enters the requested state or the timeout expires, 2699 * whichever happens first. Return (-1) if operation timed out, time remaining 2700 * otherwise. 2701 */ 2702 clock_t 2703 zone_status_timedwait(zone_t *zone, clock_t tim, zone_status_t status) 2704 { 2705 clock_t timeleft = 0; 2706 2707 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE); 2708 2709 mutex_enter(&zone_status_lock); 2710 while (zone->zone_status < status && timeleft != -1) { 2711 timeleft = cv_timedwait(&zone->zone_cv, &zone_status_lock, tim); 2712 } 2713 mutex_exit(&zone_status_lock); 2714 return (timeleft); 2715 } 2716 2717 /* 2718 * Block until the zone enters the requested state, the current process is 2719 * signaled, or the timeout expires, whichever happens first. Return (-1) if 2720 * operation timed out, 0 if signaled, time remaining otherwise. 2721 */ 2722 clock_t 2723 zone_status_timedwait_sig(zone_t *zone, clock_t tim, zone_status_t status) 2724 { 2725 clock_t timeleft = tim - ddi_get_lbolt(); 2726 2727 ASSERT(status > ZONE_MIN_STATE && status <= ZONE_MAX_STATE); 2728 2729 mutex_enter(&zone_status_lock); 2730 while (zone->zone_status < status) { 2731 timeleft = cv_timedwait_sig(&zone->zone_cv, &zone_status_lock, 2732 tim); 2733 if (timeleft <= 0) 2734 break; 2735 } 2736 mutex_exit(&zone_status_lock); 2737 return (timeleft); 2738 } 2739 2740 /* 2741 * Zones have two reference counts: one for references from credential 2742 * structures (zone_cred_ref), and one (zone_ref) for everything else. 2743 * This is so we can allow a zone to be rebooted while there are still 2744 * outstanding cred references, since certain drivers cache dblks (which 2745 * implicitly results in cached creds). We wait for zone_ref to drop to 2746 * 0 (actually 1), but not zone_cred_ref. The zone structure itself is 2747 * later freed when the zone_cred_ref drops to 0, though nothing other 2748 * than the zone id and privilege set should be accessed once the zone 2749 * is "dead". 2750 * 2751 * A debugging flag, zone_wait_for_cred, can be set to a non-zero value 2752 * to force halt/reboot to block waiting for the zone_cred_ref to drop 2753 * to 0. This can be useful to flush out other sources of cached creds 2754 * that may be less innocuous than the driver case. 2755 * 2756 * Zones also provide a tracked reference counting mechanism in which zone 2757 * references are represented by "crumbs" (zone_ref structures). Crumbs help 2758 * debuggers determine the sources of leaked zone references. See 2759 * zone_hold_ref() and zone_rele_ref() below for more information. 2760 */ 2761 2762 int zone_wait_for_cred = 0; 2763 2764 static void 2765 zone_hold_locked(zone_t *z) 2766 { 2767 ASSERT(MUTEX_HELD(&z->zone_lock)); 2768 z->zone_ref++; 2769 ASSERT(z->zone_ref != 0); 2770 } 2771 2772 /* 2773 * Increment the specified zone's reference count. The zone's zone_t structure 2774 * will not be freed as long as the zone's reference count is nonzero. 2775 * Decrement the zone's reference count via zone_rele(). 2776 * 2777 * NOTE: This function should only be used to hold zones for short periods of 2778 * time. Use zone_hold_ref() if the zone must be held for a long time. 2779 */ 2780 void 2781 zone_hold(zone_t *z) 2782 { 2783 mutex_enter(&z->zone_lock); 2784 zone_hold_locked(z); 2785 mutex_exit(&z->zone_lock); 2786 } 2787 2788 /* 2789 * If the non-cred ref count drops to 1 and either the cred ref count 2790 * is 0 or we aren't waiting for cred references, the zone is ready to 2791 * be destroyed. 2792 */ 2793 #define ZONE_IS_UNREF(zone) ((zone)->zone_ref == 1 && \ 2794 (!zone_wait_for_cred || (zone)->zone_cred_ref == 0)) 2795 2796 /* 2797 * Common zone reference release function invoked by zone_rele() and 2798 * zone_rele_ref(). If subsys is ZONE_REF_NUM_SUBSYS, then the specified 2799 * zone's subsystem-specific reference counters are not affected by the 2800 * release. If ref is not NULL, then the zone_ref_t to which it refers is 2801 * removed from the specified zone's reference list. ref must be non-NULL iff 2802 * subsys is not ZONE_REF_NUM_SUBSYS. 2803 */ 2804 static void 2805 zone_rele_common(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys) 2806 { 2807 boolean_t wakeup; 2808 2809 mutex_enter(&z->zone_lock); 2810 ASSERT(z->zone_ref != 0); 2811 z->zone_ref--; 2812 if (subsys != ZONE_REF_NUM_SUBSYS) { 2813 ASSERT(z->zone_subsys_ref[subsys] != 0); 2814 z->zone_subsys_ref[subsys]--; 2815 list_remove(&z->zone_ref_list, ref); 2816 } 2817 if (z->zone_ref == 0 && z->zone_cred_ref == 0) { 2818 /* no more refs, free the structure */ 2819 mutex_exit(&z->zone_lock); 2820 zone_free(z); 2821 return; 2822 } 2823 /* signal zone_destroy so the zone can finish halting */ 2824 wakeup = (ZONE_IS_UNREF(z) && zone_status_get(z) >= ZONE_IS_DEAD); 2825 mutex_exit(&z->zone_lock); 2826 2827 if (wakeup) { 2828 /* 2829 * Grabbing zonehash_lock here effectively synchronizes with 2830 * zone_destroy() to avoid missed signals. 2831 */ 2832 mutex_enter(&zonehash_lock); 2833 cv_broadcast(&zone_destroy_cv); 2834 mutex_exit(&zonehash_lock); 2835 } 2836 } 2837 2838 /* 2839 * Decrement the specified zone's reference count. The specified zone will 2840 * cease to exist after this function returns if the reference count drops to 2841 * zero. This function should be paired with zone_hold(). 2842 */ 2843 void 2844 zone_rele(zone_t *z) 2845 { 2846 zone_rele_common(z, NULL, ZONE_REF_NUM_SUBSYS); 2847 } 2848 2849 /* 2850 * Initialize a zone reference structure. This function must be invoked for 2851 * a reference structure before the structure is passed to zone_hold_ref(). 2852 */ 2853 void 2854 zone_init_ref(zone_ref_t *ref) 2855 { 2856 ref->zref_zone = NULL; 2857 list_link_init(&ref->zref_linkage); 2858 } 2859 2860 /* 2861 * Acquire a reference to zone z. The caller must specify the 2862 * zone_ref_subsys_t constant associated with its subsystem. The specified 2863 * zone_ref_t structure will represent a reference to the specified zone. Use 2864 * zone_rele_ref() to release the reference. 2865 * 2866 * The referenced zone_t structure will not be freed as long as the zone_t's 2867 * zone_status field is not ZONE_IS_DEAD and the zone has outstanding 2868 * references. 2869 * 2870 * NOTE: The zone_ref_t structure must be initialized before it is used. 2871 * See zone_init_ref() above. 2872 */ 2873 void 2874 zone_hold_ref(zone_t *z, zone_ref_t *ref, zone_ref_subsys_t subsys) 2875 { 2876 ASSERT(subsys >= 0 && subsys < ZONE_REF_NUM_SUBSYS); 2877 2878 /* 2879 * Prevent consumers from reusing a reference structure before 2880 * releasing it. 2881 */ 2882 VERIFY(ref->zref_zone == NULL); 2883 2884 ref->zref_zone = z; 2885 mutex_enter(&z->zone_lock); 2886 zone_hold_locked(z); 2887 z->zone_subsys_ref[subsys]++; 2888 ASSERT(z->zone_subsys_ref[subsys] != 0); 2889 list_insert_head(&z->zone_ref_list, ref); 2890 mutex_exit(&z->zone_lock); 2891 } 2892 2893 /* 2894 * Release the zone reference represented by the specified zone_ref_t. 2895 * The reference is invalid after it's released; however, the zone_ref_t 2896 * structure can be reused without having to invoke zone_init_ref(). 2897 * subsys should be the same value that was passed to zone_hold_ref() 2898 * when the reference was acquired. 2899 */ 2900 void 2901 zone_rele_ref(zone_ref_t *ref, zone_ref_subsys_t subsys) 2902 { 2903 zone_rele_common(ref->zref_zone, ref, subsys); 2904 2905 /* 2906 * Set the zone_ref_t's zref_zone field to NULL to generate panics 2907 * when consumers dereference the reference. This helps us catch 2908 * consumers who use released references. Furthermore, this lets 2909 * consumers reuse the zone_ref_t structure without having to 2910 * invoke zone_init_ref(). 2911 */ 2912 ref->zref_zone = NULL; 2913 } 2914 2915 void 2916 zone_cred_hold(zone_t *z) 2917 { 2918 mutex_enter(&z->zone_lock); 2919 z->zone_cred_ref++; 2920 ASSERT(z->zone_cred_ref != 0); 2921 mutex_exit(&z->zone_lock); 2922 } 2923 2924 void 2925 zone_cred_rele(zone_t *z) 2926 { 2927 boolean_t wakeup; 2928 2929 mutex_enter(&z->zone_lock); 2930 ASSERT(z->zone_cred_ref != 0); 2931 z->zone_cred_ref--; 2932 if (z->zone_ref == 0 && z->zone_cred_ref == 0) { 2933 /* no more refs, free the structure */ 2934 mutex_exit(&z->zone_lock); 2935 zone_free(z); 2936 return; 2937 } 2938 /* 2939 * If zone_destroy is waiting for the cred references to drain 2940 * out, and they have, signal it. 2941 */ 2942 wakeup = (zone_wait_for_cred && ZONE_IS_UNREF(z) && 2943 zone_status_get(z) >= ZONE_IS_DEAD); 2944 mutex_exit(&z->zone_lock); 2945 2946 if (wakeup) { 2947 /* 2948 * Grabbing zonehash_lock here effectively synchronizes with 2949 * zone_destroy() to avoid missed signals. 2950 */ 2951 mutex_enter(&zonehash_lock); 2952 cv_broadcast(&zone_destroy_cv); 2953 mutex_exit(&zonehash_lock); 2954 } 2955 } 2956 2957 void 2958 zone_task_hold(zone_t *z) 2959 { 2960 mutex_enter(&z->zone_lock); 2961 z->zone_ntasks++; 2962 ASSERT(z->zone_ntasks != 0); 2963 mutex_exit(&z->zone_lock); 2964 } 2965 2966 void 2967 zone_task_rele(zone_t *zone) 2968 { 2969 uint_t refcnt; 2970 2971 mutex_enter(&zone->zone_lock); 2972 ASSERT(zone->zone_ntasks != 0); 2973 refcnt = --zone->zone_ntasks; 2974 if (refcnt > 1) { /* Common case */ 2975 mutex_exit(&zone->zone_lock); 2976 return; 2977 } 2978 zone_hold_locked(zone); /* so we can use the zone_t later */ 2979 mutex_exit(&zone->zone_lock); 2980 if (refcnt == 1) { 2981 /* 2982 * See if the zone is shutting down. 2983 */ 2984 mutex_enter(&zone_status_lock); 2985 if (zone_status_get(zone) != ZONE_IS_SHUTTING_DOWN) { 2986 goto out; 2987 } 2988 2989 /* 2990 * Make sure the ntasks didn't change since we 2991 * dropped zone_lock. 2992 */ 2993 mutex_enter(&zone->zone_lock); 2994 if (refcnt != zone->zone_ntasks) { 2995 mutex_exit(&zone->zone_lock); 2996 goto out; 2997 } 2998 mutex_exit(&zone->zone_lock); 2999 3000 /* 3001 * No more user processes in the zone. The zone is empty. 3002 */ 3003 zone_status_set(zone, ZONE_IS_EMPTY); 3004 goto out; 3005 } 3006 3007 ASSERT(refcnt == 0); 3008 /* 3009 * zsched has exited; the zone is dead. 3010 */ 3011 zone->zone_zsched = NULL; /* paranoia */ 3012 mutex_enter(&zone_status_lock); 3013 zone_status_set(zone, ZONE_IS_DEAD); 3014 out: 3015 mutex_exit(&zone_status_lock); 3016 zone_rele(zone); 3017 } 3018 3019 zoneid_t 3020 getzoneid(void) 3021 { 3022 return (curproc->p_zone->zone_id); 3023 } 3024 3025 /* 3026 * Internal versions of zone_find_by_*(). These don't zone_hold() or 3027 * check the validity of a zone's state. 3028 */ 3029 static zone_t * 3030 zone_find_all_by_id(zoneid_t zoneid) 3031 { 3032 mod_hash_val_t hv; 3033 zone_t *zone = NULL; 3034 3035 ASSERT(MUTEX_HELD(&zonehash_lock)); 3036 3037 if (mod_hash_find(zonehashbyid, 3038 (mod_hash_key_t)(uintptr_t)zoneid, &hv) == 0) 3039 zone = (zone_t *)hv; 3040 return (zone); 3041 } 3042 3043 static zone_t * 3044 zone_find_all_by_label(const ts_label_t *label) 3045 { 3046 mod_hash_val_t hv; 3047 zone_t *zone = NULL; 3048 3049 ASSERT(MUTEX_HELD(&zonehash_lock)); 3050 3051 /* 3052 * zonehashbylabel is not maintained for unlabeled systems 3053 */ 3054 if (!is_system_labeled()) 3055 return (NULL); 3056 if (mod_hash_find(zonehashbylabel, (mod_hash_key_t)label, &hv) == 0) 3057 zone = (zone_t *)hv; 3058 return (zone); 3059 } 3060 3061 static zone_t * 3062 zone_find_all_by_name(char *name) 3063 { 3064 mod_hash_val_t hv; 3065 zone_t *zone = NULL; 3066 3067 ASSERT(MUTEX_HELD(&zonehash_lock)); 3068 3069 if (mod_hash_find(zonehashbyname, (mod_hash_key_t)name, &hv) == 0) 3070 zone = (zone_t *)hv; 3071 return (zone); 3072 } 3073 3074 /* 3075 * Public interface for looking up a zone by zoneid. Only returns the zone if 3076 * it is fully initialized, and has not yet begun the zone_destroy() sequence. 3077 * Caller must call zone_rele() once it is done with the zone. 3078 * 3079 * The zone may begin the zone_destroy() sequence immediately after this 3080 * function returns, but may be safely used until zone_rele() is called. 3081 */ 3082 zone_t * 3083 zone_find_by_id(zoneid_t zoneid) 3084 { 3085 zone_t *zone; 3086 zone_status_t status; 3087 3088 mutex_enter(&zonehash_lock); 3089 if ((zone = zone_find_all_by_id(zoneid)) == NULL) { 3090 mutex_exit(&zonehash_lock); 3091 return (NULL); 3092 } 3093 status = zone_status_get(zone); 3094 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) { 3095 /* 3096 * For all practical purposes the zone doesn't exist. 3097 */ 3098 mutex_exit(&zonehash_lock); 3099 return (NULL); 3100 } 3101 zone_hold(zone); 3102 mutex_exit(&zonehash_lock); 3103 return (zone); 3104 } 3105 3106 /* 3107 * Similar to zone_find_by_id, but using zone label as the key. 3108 */ 3109 zone_t * 3110 zone_find_by_label(const ts_label_t *label) 3111 { 3112 zone_t *zone; 3113 zone_status_t status; 3114 3115 mutex_enter(&zonehash_lock); 3116 if ((zone = zone_find_all_by_label(label)) == NULL) { 3117 mutex_exit(&zonehash_lock); 3118 return (NULL); 3119 } 3120 3121 status = zone_status_get(zone); 3122 if (status > ZONE_IS_DOWN) { 3123 /* 3124 * For all practical purposes the zone doesn't exist. 3125 */ 3126 mutex_exit(&zonehash_lock); 3127 return (NULL); 3128 } 3129 zone_hold(zone); 3130 mutex_exit(&zonehash_lock); 3131 return (zone); 3132 } 3133 3134 /* 3135 * Similar to zone_find_by_id, but using zone name as the key. 3136 */ 3137 zone_t * 3138 zone_find_by_name(char *name) 3139 { 3140 zone_t *zone; 3141 zone_status_t status; 3142 3143 mutex_enter(&zonehash_lock); 3144 if ((zone = zone_find_all_by_name(name)) == NULL) { 3145 mutex_exit(&zonehash_lock); 3146 return (NULL); 3147 } 3148 status = zone_status_get(zone); 3149 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) { 3150 /* 3151 * For all practical purposes the zone doesn't exist. 3152 */ 3153 mutex_exit(&zonehash_lock); 3154 return (NULL); 3155 } 3156 zone_hold(zone); 3157 mutex_exit(&zonehash_lock); 3158 return (zone); 3159 } 3160 3161 /* 3162 * Similar to zone_find_by_id(), using the path as a key. For instance, 3163 * if there is a zone "foo" rooted at /foo/root, and the path argument 3164 * is "/foo/root/proc", it will return the held zone_t corresponding to 3165 * zone "foo". 3166 * 3167 * zone_find_by_path() always returns a non-NULL value, since at the 3168 * very least every path will be contained in the global zone. 3169 * 3170 * As with the other zone_find_by_*() functions, the caller is 3171 * responsible for zone_rele()ing the return value of this function. 3172 */ 3173 zone_t * 3174 zone_find_by_path(const char *path) 3175 { 3176 zone_t *zone; 3177 zone_t *zret = NULL; 3178 zone_status_t status; 3179 3180 if (path == NULL) { 3181 /* 3182 * Call from rootconf(). 3183 */ 3184 zone_hold(global_zone); 3185 return (global_zone); 3186 } 3187 ASSERT(*path == '/'); 3188 mutex_enter(&zonehash_lock); 3189 for (zone = list_head(&zone_active); zone != NULL; 3190 zone = list_next(&zone_active, zone)) { 3191 if (ZONE_PATH_VISIBLE(path, zone)) 3192 zret = zone; 3193 } 3194 ASSERT(zret != NULL); 3195 status = zone_status_get(zret); 3196 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) { 3197 /* 3198 * Zone practically doesn't exist. 3199 */ 3200 zret = global_zone; 3201 } 3202 zone_hold(zret); 3203 mutex_exit(&zonehash_lock); 3204 return (zret); 3205 } 3206 3207 /* 3208 * Public interface for updating per-zone load averages. Called once per 3209 * second. 3210 * 3211 * Based on loadavg_update(), genloadavg() and calcloadavg() from clock.c. 3212 */ 3213 void 3214 zone_loadavg_update(void) 3215 { 3216 zone_t *zp; 3217 zone_status_t status; 3218 struct loadavg_s *lavg; 3219 hrtime_t zone_total; 3220 uint64_t tmp; 3221 int i; 3222 hrtime_t hr_avg; 3223 int nrun; 3224 static int64_t f[3] = { 135, 27, 9 }; 3225 int64_t q, r; 3226 3227 mutex_enter(&zonehash_lock); 3228 for (zp = list_head(&zone_active); zp != NULL; 3229 zp = list_next(&zone_active, zp)) { 3230 mutex_enter(&zp->zone_lock); 3231 3232 /* Skip zones that are on the way down or not yet up */ 3233 status = zone_status_get(zp); 3234 if (status < ZONE_IS_READY || status >= ZONE_IS_DOWN) { 3235 /* For all practical purposes the zone doesn't exist. */ 3236 mutex_exit(&zp->zone_lock); 3237 continue; 3238 } 3239 3240 /* 3241 * Update the 10 second moving average data in zone_loadavg. 3242 */ 3243 lavg = &zp->zone_loadavg; 3244 3245 tmp = cpu_uarray_sum_all(zp->zone_ustate); 3246 zone_total = UINT64_OVERFLOW_TO_INT64(tmp); 3247 3248 scalehrtime(&zone_total); 3249 3250 /* The zone_total should always be increasing. */ 3251 lavg->lg_loads[lavg->lg_cur] = (zone_total > lavg->lg_total) ? 3252 zone_total - lavg->lg_total : 0; 3253 lavg->lg_cur = (lavg->lg_cur + 1) % S_LOADAVG_SZ; 3254 /* lg_total holds the prev. 1 sec. total */ 3255 lavg->lg_total = zone_total; 3256 3257 /* 3258 * To simplify the calculation, we don't calculate the load avg. 3259 * until the zone has been up for at least 10 seconds and our 3260 * moving average is thus full. 3261 */ 3262 if ((lavg->lg_len + 1) < S_LOADAVG_SZ) { 3263 lavg->lg_len++; 3264 mutex_exit(&zp->zone_lock); 3265 continue; 3266 } 3267 3268 /* Now calculate the 1min, 5min, 15 min load avg. */ 3269 hr_avg = 0; 3270 for (i = 0; i < S_LOADAVG_SZ; i++) 3271 hr_avg += lavg->lg_loads[i]; 3272 hr_avg = hr_avg / S_LOADAVG_SZ; 3273 nrun = hr_avg / (NANOSEC / LGRP_LOADAVG_IN_THREAD_MAX); 3274 3275 /* Compute load avg. See comment in calcloadavg() */ 3276 for (i = 0; i < 3; i++) { 3277 q = (zp->zone_hp_avenrun[i] >> 16) << 7; 3278 r = (zp->zone_hp_avenrun[i] & 0xffff) << 7; 3279 zp->zone_hp_avenrun[i] += 3280 ((nrun - q) * f[i] - ((r * f[i]) >> 16)) >> 4; 3281 3282 /* avenrun[] can only hold 31 bits of load avg. */ 3283 if (zp->zone_hp_avenrun[i] < 3284 ((uint64_t)1<<(31+16-FSHIFT))) 3285 zp->zone_avenrun[i] = (int32_t) 3286 (zp->zone_hp_avenrun[i] >> (16 - FSHIFT)); 3287 else 3288 zp->zone_avenrun[i] = 0x7fffffff; 3289 } 3290 3291 mutex_exit(&zp->zone_lock); 3292 } 3293 mutex_exit(&zonehash_lock); 3294 } 3295 3296 /* 3297 * Get the number of cpus visible to this zone. The system-wide global 3298 * 'ncpus' is returned if pools are disabled, the caller is in the 3299 * global zone, or a NULL zone argument is passed in. 3300 */ 3301 int 3302 zone_ncpus_get(zone_t *zone) 3303 { 3304 int myncpus = zone == NULL ? 0 : zone->zone_ncpus; 3305 3306 return (myncpus != 0 ? myncpus : ncpus); 3307 } 3308 3309 /* 3310 * Get the number of online cpus visible to this zone. The system-wide 3311 * global 'ncpus_online' is returned if pools are disabled, the caller 3312 * is in the global zone, or a NULL zone argument is passed in. 3313 */ 3314 int 3315 zone_ncpus_online_get(zone_t *zone) 3316 { 3317 int myncpus_online = zone == NULL ? 0 : zone->zone_ncpus_online; 3318 3319 return (myncpus_online != 0 ? myncpus_online : ncpus_online); 3320 } 3321 3322 /* 3323 * Return the pool to which the zone is currently bound. 3324 */ 3325 pool_t * 3326 zone_pool_get(zone_t *zone) 3327 { 3328 ASSERT(pool_lock_held()); 3329 3330 return (zone->zone_pool); 3331 } 3332 3333 /* 3334 * Set the zone's pool pointer and update the zone's visibility to match 3335 * the resources in the new pool. 3336 */ 3337 void 3338 zone_pool_set(zone_t *zone, pool_t *pool) 3339 { 3340 ASSERT(pool_lock_held()); 3341 ASSERT(MUTEX_HELD(&cpu_lock)); 3342 3343 zone->zone_pool = pool; 3344 zone_pset_set(zone, pool->pool_pset->pset_id); 3345 } 3346 3347 /* 3348 * Return the cached value of the id of the processor set to which the 3349 * zone is currently bound. The value will be ZONE_PS_INVAL if the pools 3350 * facility is disabled. 3351 */ 3352 psetid_t 3353 zone_pset_get(zone_t *zone) 3354 { 3355 ASSERT(MUTEX_HELD(&cpu_lock)); 3356 3357 return (zone->zone_psetid); 3358 } 3359 3360 /* 3361 * Set the cached value of the id of the processor set to which the zone 3362 * is currently bound. Also update the zone's visibility to match the 3363 * resources in the new processor set. 3364 */ 3365 void 3366 zone_pset_set(zone_t *zone, psetid_t newpsetid) 3367 { 3368 psetid_t oldpsetid; 3369 3370 ASSERT(MUTEX_HELD(&cpu_lock)); 3371 oldpsetid = zone_pset_get(zone); 3372 3373 if (oldpsetid == newpsetid) 3374 return; 3375 /* 3376 * Global zone sees all. 3377 */ 3378 if (zone != global_zone) { 3379 zone->zone_psetid = newpsetid; 3380 if (newpsetid != ZONE_PS_INVAL) 3381 pool_pset_visibility_add(newpsetid, zone); 3382 if (oldpsetid != ZONE_PS_INVAL) 3383 pool_pset_visibility_remove(oldpsetid, zone); 3384 } 3385 /* 3386 * Disabling pools, so we should start using the global values 3387 * for ncpus and ncpus_online. 3388 */ 3389 if (newpsetid == ZONE_PS_INVAL) { 3390 zone->zone_ncpus = 0; 3391 zone->zone_ncpus_online = 0; 3392 } 3393 } 3394 3395 /* 3396 * Walk the list of active zones and issue the provided callback for 3397 * each of them. 3398 * 3399 * Caller must not be holding any locks that may be acquired under 3400 * zonehash_lock. See comment at the beginning of the file for a list of 3401 * common locks and their interactions with zones. 3402 */ 3403 int 3404 zone_walk(int (*cb)(zone_t *, void *), void *data) 3405 { 3406 zone_t *zone; 3407 int ret = 0; 3408 zone_status_t status; 3409 3410 mutex_enter(&zonehash_lock); 3411 for (zone = list_head(&zone_active); zone != NULL; 3412 zone = list_next(&zone_active, zone)) { 3413 /* 3414 * Skip zones that shouldn't be externally visible. 3415 */ 3416 status = zone_status_get(zone); 3417 if (status < ZONE_IS_READY || status > ZONE_IS_DOWN) 3418 continue; 3419 /* 3420 * Bail immediately if any callback invocation returns a 3421 * non-zero value. 3422 */ 3423 ret = (*cb)(zone, data); 3424 if (ret != 0) 3425 break; 3426 } 3427 mutex_exit(&zonehash_lock); 3428 return (ret); 3429 } 3430 3431 static int 3432 zone_set_root(zone_t *zone, const char *upath) 3433 { 3434 vnode_t *vp; 3435 int trycount; 3436 int error = 0; 3437 char *path; 3438 struct pathname upn, pn; 3439 size_t pathlen; 3440 3441 if ((error = pn_get((char *)upath, UIO_USERSPACE, &upn)) != 0) 3442 return (error); 3443 3444 pn_alloc(&pn); 3445 3446 /* prevent infinite loop */ 3447 trycount = 10; 3448 for (;;) { 3449 if (--trycount <= 0) { 3450 error = ESTALE; 3451 goto out; 3452 } 3453 3454 if ((error = lookuppn(&upn, &pn, FOLLOW, NULLVPP, &vp)) == 0) { 3455 /* 3456 * VOP_ACCESS() may cover 'vp' with a new 3457 * filesystem, if 'vp' is an autoFS vnode. 3458 * Get the new 'vp' if so. 3459 */ 3460 if ((error = 3461 VOP_ACCESS(vp, VEXEC, 0, CRED(), NULL)) == 0 && 3462 (!vn_ismntpt(vp) || 3463 (error = traverse(&vp)) == 0)) { 3464 pathlen = pn.pn_pathlen + 2; 3465 path = kmem_alloc(pathlen, KM_SLEEP); 3466 (void) strncpy(path, pn.pn_path, 3467 pn.pn_pathlen + 1); 3468 path[pathlen - 2] = '/'; 3469 path[pathlen - 1] = '\0'; 3470 pn_free(&pn); 3471 pn_free(&upn); 3472 3473 /* Success! */ 3474 break; 3475 } 3476 VN_RELE(vp); 3477 } 3478 if (error != ESTALE) 3479 goto out; 3480 } 3481 3482 ASSERT(error == 0); 3483 zone->zone_rootvp = vp; /* we hold a reference to vp */ 3484 zone->zone_rootpath = path; 3485 zone->zone_rootpathlen = pathlen; 3486 if (pathlen > 5 && strcmp(path + pathlen - 5, "/lu/") == 0) 3487 zone->zone_flags |= ZF_IS_SCRATCH; 3488 return (0); 3489 3490 out: 3491 pn_free(&pn); 3492 pn_free(&upn); 3493 return (error); 3494 } 3495 3496 #define isalnum(c) (((c) >= '0' && (c) <= '9') || \ 3497 ((c) >= 'a' && (c) <= 'z') || \ 3498 ((c) >= 'A' && (c) <= 'Z')) 3499 3500 static int 3501 zone_set_name(zone_t *zone, const char *uname) 3502 { 3503 char *kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP); 3504 size_t len; 3505 int i, err; 3506 3507 if ((err = copyinstr(uname, kname, ZONENAME_MAX, &len)) != 0) { 3508 kmem_free(kname, ZONENAME_MAX); 3509 return (err); /* EFAULT or ENAMETOOLONG */ 3510 } 3511 3512 /* must be less than ZONENAME_MAX */ 3513 if (len == ZONENAME_MAX && kname[ZONENAME_MAX - 1] != '\0') { 3514 kmem_free(kname, ZONENAME_MAX); 3515 return (EINVAL); 3516 } 3517 3518 /* 3519 * Name must start with an alphanumeric and must contain only 3520 * alphanumerics, '-', '_' and '.'. 3521 */ 3522 if (!isalnum(kname[0])) { 3523 kmem_free(kname, ZONENAME_MAX); 3524 return (EINVAL); 3525 } 3526 for (i = 1; i < len - 1; i++) { 3527 if (!isalnum(kname[i]) && kname[i] != '-' && kname[i] != '_' && 3528 kname[i] != '.') { 3529 kmem_free(kname, ZONENAME_MAX); 3530 return (EINVAL); 3531 } 3532 } 3533 3534 zone->zone_name = kname; 3535 return (0); 3536 } 3537 3538 /* 3539 * Gets the 32-bit hostid of the specified zone as an unsigned int. If 'zonep' 3540 * is NULL or it points to a zone with no hostid emulation, then the machine's 3541 * hostid (i.e., the global zone's hostid) is returned. This function returns 3542 * zero if neither the zone nor the host machine (global zone) have hostids. It 3543 * returns HW_INVALID_HOSTID if the function attempts to return the machine's 3544 * hostid and the machine's hostid is invalid. 3545 */ 3546 uint32_t 3547 zone_get_hostid(zone_t *zonep) 3548 { 3549 unsigned long machine_hostid; 3550 3551 if (zonep == NULL || zonep->zone_hostid == HW_INVALID_HOSTID) { 3552 if (ddi_strtoul(hw_serial, NULL, 10, &machine_hostid) != 0) 3553 return (HW_INVALID_HOSTID); 3554 return ((uint32_t)machine_hostid); 3555 } 3556 return (zonep->zone_hostid); 3557 } 3558 3559 /* 3560 * Similar to thread_create(), but makes sure the thread is in the appropriate 3561 * zone's zsched process (curproc->p_zone->zone_zsched) before returning. 3562 */ 3563 /*ARGSUSED*/ 3564 kthread_t * 3565 zthread_create( 3566 caddr_t stk, 3567 size_t stksize, 3568 void (*proc)(), 3569 void *arg, 3570 size_t len, 3571 pri_t pri) 3572 { 3573 kthread_t *t; 3574 zone_t *zone = curproc->p_zone; 3575 proc_t *pp = zone->zone_zsched; 3576 3577 zone_hold(zone); /* Reference to be dropped when thread exits */ 3578 3579 /* 3580 * No-one should be trying to create threads if the zone is shutting 3581 * down and there aren't any kernel threads around. See comment 3582 * in zthread_exit(). 3583 */ 3584 ASSERT(!(zone->zone_kthreads == NULL && 3585 zone_status_get(zone) >= ZONE_IS_EMPTY)); 3586 /* 3587 * Create a thread, but don't let it run until we've finished setting 3588 * things up. 3589 */ 3590 t = thread_create(stk, stksize, proc, arg, len, pp, TS_STOPPED, pri); 3591 ASSERT(t->t_forw == NULL); 3592 mutex_enter(&zone_status_lock); 3593 if (zone->zone_kthreads == NULL) { 3594 t->t_forw = t->t_back = t; 3595 } else { 3596 kthread_t *tx = zone->zone_kthreads; 3597 3598 t->t_forw = tx; 3599 t->t_back = tx->t_back; 3600 tx->t_back->t_forw = t; 3601 tx->t_back = t; 3602 } 3603 zone->zone_kthreads = t; 3604 mutex_exit(&zone_status_lock); 3605 3606 mutex_enter(&pp->p_lock); 3607 t->t_proc_flag |= TP_ZTHREAD; 3608 project_rele(t->t_proj); 3609 t->t_proj = project_hold(pp->p_task->tk_proj); 3610 3611 /* 3612 * Setup complete, let it run. 3613 */ 3614 thread_lock(t); 3615 t->t_schedflag |= TS_ALLSTART; 3616 setrun_locked(t); 3617 thread_unlock(t); 3618 3619 mutex_exit(&pp->p_lock); 3620 3621 return (t); 3622 } 3623 3624 /* 3625 * Similar to thread_exit(). Must be called by threads created via 3626 * zthread_exit(). 3627 */ 3628 void 3629 zthread_exit(void) 3630 { 3631 kthread_t *t = curthread; 3632 proc_t *pp = curproc; 3633 zone_t *zone = pp->p_zone; 3634 3635 mutex_enter(&zone_status_lock); 3636 3637 /* 3638 * Reparent to p0 3639 */ 3640 kpreempt_disable(); 3641 mutex_enter(&pp->p_lock); 3642 t->t_proc_flag &= ~TP_ZTHREAD; 3643 t->t_procp = &p0; 3644 hat_thread_exit(t); 3645 mutex_exit(&pp->p_lock); 3646 kpreempt_enable(); 3647 3648 if (t->t_back == t) { 3649 ASSERT(t->t_forw == t); 3650 /* 3651 * If the zone is empty, once the thread count 3652 * goes to zero no further kernel threads can be 3653 * created. This is because if the creator is a process 3654 * in the zone, then it must have exited before the zone 3655 * state could be set to ZONE_IS_EMPTY. 3656 * Otherwise, if the creator is a kernel thread in the 3657 * zone, the thread count is non-zero. 3658 * 3659 * This really means that non-zone kernel threads should 3660 * not create zone kernel threads. 3661 */ 3662 zone->zone_kthreads = NULL; 3663 if (zone_status_get(zone) == ZONE_IS_EMPTY) { 3664 zone_status_set(zone, ZONE_IS_DOWN); 3665 /* 3666 * Remove any CPU caps on this zone. 3667 */ 3668 cpucaps_zone_remove(zone); 3669 } 3670 } else { 3671 t->t_forw->t_back = t->t_back; 3672 t->t_back->t_forw = t->t_forw; 3673 if (zone->zone_kthreads == t) 3674 zone->zone_kthreads = t->t_forw; 3675 } 3676 mutex_exit(&zone_status_lock); 3677 zone_rele(zone); 3678 thread_exit(); 3679 /* NOTREACHED */ 3680 } 3681 3682 static void 3683 zone_chdir(vnode_t *vp, vnode_t **vpp, proc_t *pp) 3684 { 3685 vnode_t *oldvp; 3686 3687 /* we're going to hold a reference here to the directory */ 3688 VN_HOLD(vp); 3689 3690 /* update abs cwd/root path see c2/audit.c */ 3691 if (AU_AUDITING()) 3692 audit_chdirec(vp, vpp); 3693 3694 mutex_enter(&pp->p_lock); 3695 oldvp = *vpp; 3696 *vpp = vp; 3697 mutex_exit(&pp->p_lock); 3698 if (oldvp != NULL) 3699 VN_RELE(oldvp); 3700 } 3701 3702 /* 3703 * Convert an rctl value represented by an nvlist_t into an rctl_val_t. 3704 */ 3705 static int 3706 nvlist2rctlval(nvlist_t *nvl, rctl_val_t *rv) 3707 { 3708 nvpair_t *nvp = NULL; 3709 boolean_t priv_set = B_FALSE; 3710 boolean_t limit_set = B_FALSE; 3711 boolean_t action_set = B_FALSE; 3712 3713 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) { 3714 const char *name; 3715 uint64_t ui64; 3716 3717 name = nvpair_name(nvp); 3718 if (nvpair_type(nvp) != DATA_TYPE_UINT64) 3719 return (EINVAL); 3720 (void) nvpair_value_uint64(nvp, &ui64); 3721 if (strcmp(name, "privilege") == 0) { 3722 /* 3723 * Currently only privileged values are allowed, but 3724 * this may change in the future. 3725 */ 3726 if (ui64 != RCPRIV_PRIVILEGED) 3727 return (EINVAL); 3728 rv->rcv_privilege = ui64; 3729 priv_set = B_TRUE; 3730 } else if (strcmp(name, "limit") == 0) { 3731 rv->rcv_value = ui64; 3732 limit_set = B_TRUE; 3733 } else if (strcmp(name, "action") == 0) { 3734 if (ui64 != RCTL_LOCAL_NOACTION && 3735 ui64 != RCTL_LOCAL_DENY) 3736 return (EINVAL); 3737 rv->rcv_flagaction = ui64; 3738 action_set = B_TRUE; 3739 } else { 3740 return (EINVAL); 3741 } 3742 } 3743 3744 if (!(priv_set && limit_set && action_set)) 3745 return (EINVAL); 3746 rv->rcv_action_signal = 0; 3747 rv->rcv_action_recipient = NULL; 3748 rv->rcv_action_recip_pid = -1; 3749 rv->rcv_firing_time = 0; 3750 3751 return (0); 3752 } 3753 3754 /* 3755 * Non-global zone version of start_init. 3756 */ 3757 void 3758 zone_start_init(void) 3759 { 3760 proc_t *p = ttoproc(curthread); 3761 zone_t *z = p->p_zone; 3762 3763 ASSERT(!INGLOBALZONE(curproc)); 3764 3765 /* 3766 * For all purposes (ZONE_ATTR_INITPID and restart_init), 3767 * storing just the pid of init is sufficient. 3768 */ 3769 z->zone_proc_initpid = p->p_pid; 3770 3771 /* 3772 * We maintain zone_boot_err so that we can return the cause of the 3773 * failure back to the caller of the zone_boot syscall. 3774 */ 3775 p->p_zone->zone_boot_err = start_init_common(); 3776 3777 /* 3778 * We will prevent booting zones from becoming running zones if the 3779 * global zone is shutting down. 3780 */ 3781 mutex_enter(&zone_status_lock); 3782 if (z->zone_boot_err != 0 || zone_status_get(global_zone) >= 3783 ZONE_IS_SHUTTING_DOWN) { 3784 /* 3785 * Make sure we are still in the booting state-- we could have 3786 * raced and already be shutting down, or even further along. 3787 */ 3788 if (zone_status_get(z) == ZONE_IS_BOOTING) { 3789 zone_status_set(z, ZONE_IS_SHUTTING_DOWN); 3790 } 3791 mutex_exit(&zone_status_lock); 3792 /* It's gone bad, dispose of the process */ 3793 if (proc_exit(CLD_EXITED, z->zone_boot_err) != 0) { 3794 mutex_enter(&p->p_lock); 3795 ASSERT(p->p_flag & SEXITLWPS); 3796 lwp_exit(); 3797 } 3798 } else { 3799 if (zone_status_get(z) == ZONE_IS_BOOTING) 3800 zone_status_set(z, ZONE_IS_RUNNING); 3801 mutex_exit(&zone_status_lock); 3802 /* cause the process to return to userland. */ 3803 lwp_rtt(); 3804 } 3805 } 3806 3807 struct zsched_arg { 3808 zone_t *zone; 3809 nvlist_t *nvlist; 3810 }; 3811 3812 /* 3813 * Per-zone "sched" workalike. The similarity to "sched" doesn't have 3814 * anything to do with scheduling, but rather with the fact that 3815 * per-zone kernel threads are parented to zsched, just like regular 3816 * kernel threads are parented to sched (p0). 3817 * 3818 * zsched is also responsible for launching init for the zone. 3819 */ 3820 static void 3821 zsched(void *arg) 3822 { 3823 struct zsched_arg *za = arg; 3824 proc_t *pp = curproc; 3825 proc_t *initp = proc_init; 3826 zone_t *zone = za->zone; 3827 cred_t *cr, *oldcred; 3828 rctl_set_t *set; 3829 rctl_alloc_gp_t *gp; 3830 contract_t *ct = NULL; 3831 task_t *tk, *oldtk; 3832 rctl_entity_p_t e; 3833 kproject_t *pj; 3834 3835 nvlist_t *nvl = za->nvlist; 3836 nvpair_t *nvp = NULL; 3837 3838 bcopy("zsched", PTOU(pp)->u_psargs, sizeof ("zsched")); 3839 bcopy("zsched", PTOU(pp)->u_comm, sizeof ("zsched")); 3840 PTOU(pp)->u_argc = 0; 3841 PTOU(pp)->u_argv = 0; 3842 PTOU(pp)->u_envp = 0; 3843 PTOU(pp)->u_commpagep = 0; 3844 closeall(P_FINFO(pp)); 3845 3846 /* 3847 * We are this zone's "zsched" process. As the zone isn't generally 3848 * visible yet we don't need to grab any locks before initializing its 3849 * zone_proc pointer. 3850 */ 3851 zone_hold(zone); /* this hold is released by zone_destroy() */ 3852 zone->zone_zsched = pp; 3853 mutex_enter(&pp->p_lock); 3854 pp->p_zone = zone; 3855 mutex_exit(&pp->p_lock); 3856 3857 /* 3858 * Disassociate process from its 'parent'; parent ourselves to init 3859 * (pid 1) and change other values as needed. 3860 */ 3861 sess_create(); 3862 3863 mutex_enter(&pidlock); 3864 proc_detach(pp); 3865 pp->p_ppid = 1; 3866 pp->p_flag |= SZONETOP; 3867 pp->p_ancpid = 1; 3868 pp->p_parent = initp; 3869 pp->p_psibling = NULL; 3870 if (initp->p_child) 3871 initp->p_child->p_psibling = pp; 3872 pp->p_sibling = initp->p_child; 3873 initp->p_child = pp; 3874 3875 /* Decrement what newproc() incremented. */ 3876 upcount_dec(crgetruid(CRED()), GLOBAL_ZONEID); 3877 /* 3878 * Our credentials are about to become kcred-like, so we don't care 3879 * about the caller's ruid. 3880 */ 3881 upcount_inc(crgetruid(kcred), zone->zone_id); 3882 mutex_exit(&pidlock); 3883 3884 /* 3885 * getting out of global zone, so decrement lwp and process counts 3886 */ 3887 pj = pp->p_task->tk_proj; 3888 mutex_enter(&global_zone->zone_nlwps_lock); 3889 pj->kpj_nlwps -= pp->p_lwpcnt; 3890 global_zone->zone_nlwps -= pp->p_lwpcnt; 3891 pj->kpj_nprocs--; 3892 global_zone->zone_nprocs--; 3893 mutex_exit(&global_zone->zone_nlwps_lock); 3894 3895 /* 3896 * Decrement locked memory counts on old zone and project. 3897 */ 3898 mutex_enter(&global_zone->zone_mem_lock); 3899 global_zone->zone_locked_mem -= pp->p_locked_mem; 3900 pj->kpj_data.kpd_locked_mem -= pp->p_locked_mem; 3901 mutex_exit(&global_zone->zone_mem_lock); 3902 3903 /* 3904 * Create and join a new task in project '0' of this zone. 3905 * 3906 * We don't need to call holdlwps() since we know we're the only lwp in 3907 * this process. 3908 * 3909 * task_join() returns with p_lock held. 3910 */ 3911 tk = task_create(0, zone); 3912 mutex_enter(&cpu_lock); 3913 oldtk = task_join(tk, 0); 3914 3915 pj = pp->p_task->tk_proj; 3916 3917 mutex_enter(&zone->zone_mem_lock); 3918 zone->zone_locked_mem += pp->p_locked_mem; 3919 pj->kpj_data.kpd_locked_mem += pp->p_locked_mem; 3920 mutex_exit(&zone->zone_mem_lock); 3921 3922 /* 3923 * add lwp and process counts to zsched's zone, and increment 3924 * project's task and process count due to the task created in 3925 * the above task_create. 3926 */ 3927 mutex_enter(&zone->zone_nlwps_lock); 3928 pj->kpj_nlwps += pp->p_lwpcnt; 3929 pj->kpj_ntasks += 1; 3930 zone->zone_nlwps += pp->p_lwpcnt; 3931 pj->kpj_nprocs++; 3932 zone->zone_nprocs++; 3933 mutex_exit(&zone->zone_nlwps_lock); 3934 3935 mutex_exit(&curproc->p_lock); 3936 mutex_exit(&cpu_lock); 3937 task_rele(oldtk); 3938 3939 /* 3940 * The process was created by a process in the global zone, hence the 3941 * credentials are wrong. We might as well have kcred-ish credentials. 3942 */ 3943 cr = zone->zone_kcred; 3944 crhold(cr); 3945 mutex_enter(&pp->p_crlock); 3946 oldcred = pp->p_cred; 3947 pp->p_cred = cr; 3948 mutex_exit(&pp->p_crlock); 3949 crfree(oldcred); 3950 3951 /* 3952 * Hold credentials again (for thread) 3953 */ 3954 crhold(cr); 3955 3956 /* 3957 * p_lwpcnt can't change since this is a kernel process. 3958 */ 3959 crset(pp, cr); 3960 3961 /* 3962 * Chroot 3963 */ 3964 zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_cdir, pp); 3965 zone_chdir(zone->zone_rootvp, &PTOU(pp)->u_rdir, pp); 3966 3967 /* 3968 * Initialize zone's rctl set. 3969 */ 3970 set = rctl_set_create(); 3971 gp = rctl_set_init_prealloc(RCENTITY_ZONE); 3972 mutex_enter(&pp->p_lock); 3973 e.rcep_p.zone = zone; 3974 e.rcep_t = RCENTITY_ZONE; 3975 zone->zone_rctls = rctl_set_init(RCENTITY_ZONE, pp, &e, set, gp); 3976 mutex_exit(&pp->p_lock); 3977 rctl_prealloc_destroy(gp); 3978 3979 /* 3980 * Apply the rctls passed in to zone_create(). This is basically a list 3981 * assignment: all of the old values are removed and the new ones 3982 * inserted. That is, if an empty list is passed in, all values are 3983 * removed. 3984 */ 3985 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) { 3986 rctl_dict_entry_t *rde; 3987 rctl_hndl_t hndl; 3988 char *name; 3989 nvlist_t **nvlarray; 3990 uint_t i, nelem; 3991 int error; /* For ASSERT()s */ 3992 3993 name = nvpair_name(nvp); 3994 hndl = rctl_hndl_lookup(name); 3995 ASSERT(hndl != -1); 3996 rde = rctl_dict_lookup_hndl(hndl); 3997 ASSERT(rde != NULL); 3998 3999 for (; /* ever */; ) { 4000 rctl_val_t oval; 4001 4002 mutex_enter(&pp->p_lock); 4003 error = rctl_local_get(hndl, NULL, &oval, pp); 4004 mutex_exit(&pp->p_lock); 4005 ASSERT(error == 0); /* Can't fail for RCTL_FIRST */ 4006 ASSERT(oval.rcv_privilege != RCPRIV_BASIC); 4007 if (oval.rcv_privilege == RCPRIV_SYSTEM) 4008 break; 4009 mutex_enter(&pp->p_lock); 4010 error = rctl_local_delete(hndl, &oval, pp); 4011 mutex_exit(&pp->p_lock); 4012 ASSERT(error == 0); 4013 } 4014 error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem); 4015 ASSERT(error == 0); 4016 for (i = 0; i < nelem; i++) { 4017 rctl_val_t *nvalp; 4018 4019 nvalp = kmem_cache_alloc(rctl_val_cache, KM_SLEEP); 4020 error = nvlist2rctlval(nvlarray[i], nvalp); 4021 ASSERT(error == 0); 4022 /* 4023 * rctl_local_insert can fail if the value being 4024 * inserted is a duplicate; this is OK. 4025 */ 4026 mutex_enter(&pp->p_lock); 4027 if (rctl_local_insert(hndl, nvalp, pp) != 0) 4028 kmem_cache_free(rctl_val_cache, nvalp); 4029 mutex_exit(&pp->p_lock); 4030 } 4031 } 4032 4033 /* 4034 * Tell the world that we're done setting up. 4035 * 4036 * At this point we want to set the zone status to ZONE_IS_INITIALIZED 4037 * and atomically set the zone's processor set visibility. Once 4038 * we drop pool_lock() this zone will automatically get updated 4039 * to reflect any future changes to the pools configuration. 4040 * 4041 * Note that after we drop the locks below (zonehash_lock in 4042 * particular) other operations such as a zone_getattr call can 4043 * now proceed and observe the zone. That is the reason for doing a 4044 * state transition to the INITIALIZED state. 4045 */ 4046 pool_lock(); 4047 mutex_enter(&cpu_lock); 4048 mutex_enter(&zonehash_lock); 4049 zone_uniqid(zone); 4050 zone_zsd_configure(zone); 4051 if (pool_state == POOL_ENABLED) 4052 zone_pset_set(zone, pool_default->pool_pset->pset_id); 4053 mutex_enter(&zone_status_lock); 4054 ASSERT(zone_status_get(zone) == ZONE_IS_UNINITIALIZED); 4055 zone_status_set(zone, ZONE_IS_INITIALIZED); 4056 mutex_exit(&zone_status_lock); 4057 mutex_exit(&zonehash_lock); 4058 mutex_exit(&cpu_lock); 4059 pool_unlock(); 4060 4061 /* Now call the create callback for this key */ 4062 zsd_apply_all_keys(zsd_apply_create, zone); 4063 4064 /* The callbacks are complete. Mark ZONE_IS_READY */ 4065 mutex_enter(&zone_status_lock); 4066 ASSERT(zone_status_get(zone) == ZONE_IS_INITIALIZED); 4067 zone_status_set(zone, ZONE_IS_READY); 4068 mutex_exit(&zone_status_lock); 4069 4070 /* 4071 * Once we see the zone transition to the ZONE_IS_BOOTING state, 4072 * we launch init, and set the state to running. 4073 */ 4074 zone_status_wait_cpr(zone, ZONE_IS_BOOTING, "zsched"); 4075 4076 if (zone_status_get(zone) == ZONE_IS_BOOTING) { 4077 id_t cid; 4078 4079 /* 4080 * Ok, this is a little complicated. We need to grab the 4081 * zone's pool's scheduling class ID; note that by now, we 4082 * are already bound to a pool if we need to be (zoneadmd 4083 * will have done that to us while we're in the READY 4084 * state). *But* the scheduling class for the zone's 'init' 4085 * must be explicitly passed to newproc, which doesn't 4086 * respect pool bindings. 4087 * 4088 * We hold the pool_lock across the call to newproc() to 4089 * close the obvious race: the pool's scheduling class 4090 * could change before we manage to create the LWP with 4091 * classid 'cid'. 4092 */ 4093 pool_lock(); 4094 if (zone->zone_defaultcid > 0) 4095 cid = zone->zone_defaultcid; 4096 else 4097 cid = pool_get_class(zone->zone_pool); 4098 if (cid == -1) 4099 cid = defaultcid; 4100 4101 /* 4102 * If this fails, zone_boot will ultimately fail. The 4103 * state of the zone will be set to SHUTTING_DOWN-- userland 4104 * will have to tear down the zone, and fail, or try again. 4105 */ 4106 if ((zone->zone_boot_err = newproc(zone_start_init, NULL, cid, 4107 minclsyspri - 1, &ct, 0)) != 0) { 4108 mutex_enter(&zone_status_lock); 4109 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN); 4110 mutex_exit(&zone_status_lock); 4111 } else { 4112 zone->zone_boot_time = gethrestime_sec(); 4113 } 4114 4115 pool_unlock(); 4116 } 4117 4118 /* 4119 * Wait for zone_destroy() to be called. This is what we spend 4120 * most of our life doing. 4121 */ 4122 zone_status_wait_cpr(zone, ZONE_IS_DYING, "zsched"); 4123 4124 if (ct) 4125 /* 4126 * At this point the process contract should be empty. 4127 * (Though if it isn't, it's not the end of the world.) 4128 */ 4129 VERIFY(contract_abandon(ct, curproc, B_TRUE) == 0); 4130 4131 /* 4132 * Allow kcred to be freed when all referring processes 4133 * (including this one) go away. We can't just do this in 4134 * zone_free because we need to wait for the zone_cred_ref to 4135 * drop to 0 before calling zone_free, and the existence of 4136 * zone_kcred will prevent that. Thus, we call crfree here to 4137 * balance the crdup in zone_create. The crhold calls earlier 4138 * in zsched will be dropped when the thread and process exit. 4139 */ 4140 crfree(zone->zone_kcred); 4141 zone->zone_kcred = NULL; 4142 4143 exit(CLD_EXITED, 0); 4144 } 4145 4146 /* 4147 * Helper function to determine if there are any submounts of the 4148 * provided path. Used to make sure the zone doesn't "inherit" any 4149 * mounts from before it is created. 4150 */ 4151 static uint_t 4152 zone_mount_count(const char *rootpath) 4153 { 4154 vfs_t *vfsp; 4155 uint_t count = 0; 4156 size_t rootpathlen = strlen(rootpath); 4157 4158 /* 4159 * Holding zonehash_lock prevents race conditions with 4160 * vfs_list_add()/vfs_list_remove() since we serialize with 4161 * zone_find_by_path(). 4162 */ 4163 ASSERT(MUTEX_HELD(&zonehash_lock)); 4164 /* 4165 * The rootpath must end with a '/' 4166 */ 4167 ASSERT(rootpath[rootpathlen - 1] == '/'); 4168 4169 /* 4170 * This intentionally does not count the rootpath itself if that 4171 * happens to be a mount point. 4172 */ 4173 vfs_list_read_lock(); 4174 vfsp = rootvfs; 4175 do { 4176 if (strncmp(rootpath, refstr_value(vfsp->vfs_mntpt), 4177 rootpathlen) == 0) 4178 count++; 4179 vfsp = vfsp->vfs_next; 4180 } while (vfsp != rootvfs); 4181 vfs_list_unlock(); 4182 return (count); 4183 } 4184 4185 /* 4186 * Helper function to make sure that a zone created on 'rootpath' 4187 * wouldn't end up containing other zones' rootpaths. 4188 */ 4189 static boolean_t 4190 zone_is_nested(const char *rootpath) 4191 { 4192 zone_t *zone; 4193 size_t rootpathlen = strlen(rootpath); 4194 size_t len; 4195 4196 ASSERT(MUTEX_HELD(&zonehash_lock)); 4197 4198 /* 4199 * zone_set_root() appended '/' and '\0' at the end of rootpath 4200 */ 4201 if ((rootpathlen <= 3) && (rootpath[0] == '/') && 4202 (rootpath[1] == '/') && (rootpath[2] == '\0')) 4203 return (B_TRUE); 4204 4205 for (zone = list_head(&zone_active); zone != NULL; 4206 zone = list_next(&zone_active, zone)) { 4207 if (zone == global_zone) 4208 continue; 4209 len = strlen(zone->zone_rootpath); 4210 if (strncmp(rootpath, zone->zone_rootpath, 4211 MIN(rootpathlen, len)) == 0) 4212 return (B_TRUE); 4213 } 4214 return (B_FALSE); 4215 } 4216 4217 static int 4218 zone_set_privset(zone_t *zone, const priv_set_t *zone_privs, 4219 size_t zone_privssz) 4220 { 4221 priv_set_t *privs; 4222 4223 if (zone_privssz < sizeof (priv_set_t)) 4224 return (ENOMEM); 4225 4226 privs = kmem_alloc(sizeof (priv_set_t), KM_SLEEP); 4227 4228 if (copyin(zone_privs, privs, sizeof (priv_set_t))) { 4229 kmem_free(privs, sizeof (priv_set_t)); 4230 return (EFAULT); 4231 } 4232 4233 zone->zone_privset = privs; 4234 return (0); 4235 } 4236 4237 /* 4238 * We make creative use of nvlists to pass in rctls from userland. The list is 4239 * a list of the following structures: 4240 * 4241 * (name = rctl_name, value = nvpair_list_array) 4242 * 4243 * Where each element of the nvpair_list_array is of the form: 4244 * 4245 * [(name = "privilege", value = RCPRIV_PRIVILEGED), 4246 * (name = "limit", value = uint64_t), 4247 * (name = "action", value = (RCTL_LOCAL_NOACTION || RCTL_LOCAL_DENY))] 4248 */ 4249 static int 4250 parse_rctls(caddr_t ubuf, size_t buflen, nvlist_t **nvlp) 4251 { 4252 nvpair_t *nvp = NULL; 4253 nvlist_t *nvl = NULL; 4254 char *kbuf; 4255 int error; 4256 rctl_val_t rv; 4257 4258 *nvlp = NULL; 4259 4260 if (buflen == 0) 4261 return (0); 4262 4263 if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL) 4264 return (ENOMEM); 4265 if (copyin(ubuf, kbuf, buflen)) { 4266 error = EFAULT; 4267 goto out; 4268 } 4269 if (nvlist_unpack(kbuf, buflen, &nvl, KM_SLEEP) != 0) { 4270 /* 4271 * nvl may have been allocated/free'd, but the value set to 4272 * non-NULL, so we reset it here. 4273 */ 4274 nvl = NULL; 4275 error = EINVAL; 4276 goto out; 4277 } 4278 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) { 4279 rctl_dict_entry_t *rde; 4280 rctl_hndl_t hndl; 4281 nvlist_t **nvlarray; 4282 uint_t i, nelem; 4283 char *name; 4284 4285 error = EINVAL; 4286 name = nvpair_name(nvp); 4287 if (strncmp(nvpair_name(nvp), "zone.", sizeof ("zone.") - 1) 4288 != 0 || nvpair_type(nvp) != DATA_TYPE_NVLIST_ARRAY) { 4289 goto out; 4290 } 4291 if ((hndl = rctl_hndl_lookup(name)) == -1) { 4292 goto out; 4293 } 4294 rde = rctl_dict_lookup_hndl(hndl); 4295 error = nvpair_value_nvlist_array(nvp, &nvlarray, &nelem); 4296 ASSERT(error == 0); 4297 for (i = 0; i < nelem; i++) { 4298 if (error = nvlist2rctlval(nvlarray[i], &rv)) 4299 goto out; 4300 } 4301 if (rctl_invalid_value(rde, &rv)) { 4302 error = EINVAL; 4303 goto out; 4304 } 4305 } 4306 error = 0; 4307 *nvlp = nvl; 4308 out: 4309 kmem_free(kbuf, buflen); 4310 if (error && nvl != NULL) 4311 nvlist_free(nvl); 4312 return (error); 4313 } 4314 4315 int 4316 zone_create_error(int er_error, int er_ext, int *er_out) 4317 { 4318 if (er_out != NULL) { 4319 if (copyout(&er_ext, er_out, sizeof (int))) { 4320 return (set_errno(EFAULT)); 4321 } 4322 } 4323 return (set_errno(er_error)); 4324 } 4325 4326 static int 4327 zone_set_label(zone_t *zone, const bslabel_t *lab, uint32_t doi) 4328 { 4329 ts_label_t *tsl; 4330 bslabel_t blab; 4331 4332 /* Get label from user */ 4333 if (copyin(lab, &blab, sizeof (blab)) != 0) 4334 return (EFAULT); 4335 tsl = labelalloc(&blab, doi, KM_NOSLEEP); 4336 if (tsl == NULL) 4337 return (ENOMEM); 4338 4339 zone->zone_slabel = tsl; 4340 return (0); 4341 } 4342 4343 /* 4344 * Parses a comma-separated list of ZFS datasets into a per-zone dictionary. 4345 */ 4346 static int 4347 parse_zfs(zone_t *zone, caddr_t ubuf, size_t buflen) 4348 { 4349 char *kbuf; 4350 char *dataset, *next; 4351 zone_dataset_t *zd; 4352 size_t len; 4353 4354 if (ubuf == NULL || buflen == 0) 4355 return (0); 4356 4357 if ((kbuf = kmem_alloc(buflen, KM_NOSLEEP)) == NULL) 4358 return (ENOMEM); 4359 4360 if (copyin(ubuf, kbuf, buflen) != 0) { 4361 kmem_free(kbuf, buflen); 4362 return (EFAULT); 4363 } 4364 4365 dataset = next = kbuf; 4366 for (;;) { 4367 zd = kmem_alloc(sizeof (zone_dataset_t), KM_SLEEP); 4368 4369 next = strchr(dataset, ','); 4370 4371 if (next == NULL) 4372 len = strlen(dataset); 4373 else 4374 len = next - dataset; 4375 4376 zd->zd_dataset = kmem_alloc(len + 1, KM_SLEEP); 4377 bcopy(dataset, zd->zd_dataset, len); 4378 zd->zd_dataset[len] = '\0'; 4379 4380 list_insert_head(&zone->zone_datasets, zd); 4381 4382 if (next == NULL) 4383 break; 4384 4385 dataset = next + 1; 4386 } 4387 4388 kmem_free(kbuf, buflen); 4389 return (0); 4390 } 4391 4392 /* 4393 * System call to create/initialize a new zone named 'zone_name', rooted 4394 * at 'zone_root', with a zone-wide privilege limit set of 'zone_privs', 4395 * and initialized with the zone-wide rctls described in 'rctlbuf', and 4396 * with labeling set by 'match', 'doi', and 'label'. 4397 * 4398 * If extended error is non-null, we may use it to return more detailed 4399 * error information. 4400 */ 4401 static zoneid_t 4402 zone_create(const char *zone_name, const char *zone_root, 4403 const priv_set_t *zone_privs, size_t zone_privssz, 4404 caddr_t rctlbuf, size_t rctlbufsz, 4405 caddr_t zfsbuf, size_t zfsbufsz, int *extended_error, 4406 int match, uint32_t doi, const bslabel_t *label, 4407 int flags) 4408 { 4409 struct zsched_arg zarg; 4410 nvlist_t *rctls = NULL; 4411 proc_t *pp = curproc; 4412 zone_t *zone, *ztmp; 4413 zoneid_t zoneid, start = GLOBAL_ZONEID; 4414 int error; 4415 int error2 = 0; 4416 char *str; 4417 cred_t *zkcr; 4418 boolean_t insert_label_hash; 4419 4420 if (secpolicy_zone_config(CRED()) != 0) 4421 return (set_errno(EPERM)); 4422 4423 /* can't boot zone from within chroot environment */ 4424 if (PTOU(pp)->u_rdir != NULL && PTOU(pp)->u_rdir != rootdir) 4425 return (zone_create_error(ENOTSUP, ZE_CHROOTED, 4426 extended_error)); 4427 /* 4428 * As the first step of zone creation, we want to allocate a zoneid. 4429 * This allocation is complicated by the fact that netstacks use the 4430 * zoneid to determine their stackid, but netstacks themselves are 4431 * freed asynchronously with respect to zone destruction. This means 4432 * that a netstack reference leak (or in principle, an extraordinarily 4433 * long netstack reference hold) could result in a zoneid being 4434 * allocated that in fact corresponds to a stackid from an active 4435 * (referenced) netstack -- unleashing all sorts of havoc when that 4436 * netstack is actually (re)used. (In the abstract, we might wish a 4437 * zoneid to not be deallocated until its last referencing netstack 4438 * has been released, but netstacks lack a backpointer into their 4439 * referencing zone -- and changing them to have such a pointer would 4440 * be substantial, to put it euphemistically.) To avoid this, we 4441 * detect this condition on allocation: if we have allocated a zoneid 4442 * that corresponds to a netstack that's still in use, we warn about 4443 * it (as it is much more likely to be a reference leak than an actual 4444 * netstack reference), free it, and allocate another. That these 4445 * identifers are allocated out of an ID space assures that we won't 4446 * see the identifier we just allocated. 4447 */ 4448 for (;;) { 4449 zoneid = id_alloc(zoneid_space); 4450 4451 if (!netstack_inuse_by_stackid(zoneid_to_netstackid(zoneid))) 4452 break; 4453 4454 id_free(zoneid_space, zoneid); 4455 4456 if (start == GLOBAL_ZONEID) { 4457 start = zoneid; 4458 } else if (zoneid == start) { 4459 /* 4460 * We have managed to iterate over the entire available 4461 * zoneid space -- there are no identifiers available, 4462 * presumably due to some number of leaked netstack 4463 * references. While it's in principle possible for us 4464 * to continue to try, it seems wiser to give up at 4465 * this point to warn and fail explicitly with a 4466 * distinctive error. 4467 */ 4468 cmn_err(CE_WARN, "zone_create() failed: all available " 4469 "zone IDs have netstacks still in use"); 4470 return (set_errno(ENFILE)); 4471 } 4472 4473 cmn_err(CE_WARN, "unable to reuse zone ID %d; " 4474 "netstack still in use", zoneid); 4475 } 4476 4477 zone = kmem_zalloc(sizeof (zone_t), KM_SLEEP); 4478 zone->zone_id = zoneid; 4479 zone->zone_status = ZONE_IS_UNINITIALIZED; 4480 zone->zone_pool = pool_default; 4481 zone->zone_pool_mod = gethrtime(); 4482 zone->zone_psetid = ZONE_PS_INVAL; 4483 zone->zone_ncpus = 0; 4484 zone->zone_ncpus_online = 0; 4485 zone->zone_restart_init = B_TRUE; 4486 zone->zone_reboot_on_init_exit = B_FALSE; 4487 zone->zone_restart_init_0 = B_FALSE; 4488 zone->zone_brand = &native_brand; 4489 zone->zone_initname = NULL; 4490 mutex_init(&zone->zone_lock, NULL, MUTEX_DEFAULT, NULL); 4491 mutex_init(&zone->zone_nlwps_lock, NULL, MUTEX_DEFAULT, NULL); 4492 mutex_init(&zone->zone_mem_lock, NULL, MUTEX_DEFAULT, NULL); 4493 cv_init(&zone->zone_cv, NULL, CV_DEFAULT, NULL); 4494 list_create(&zone->zone_ref_list, sizeof (zone_ref_t), 4495 offsetof(zone_ref_t, zref_linkage)); 4496 list_create(&zone->zone_zsd, sizeof (struct zsd_entry), 4497 offsetof(struct zsd_entry, zsd_linkage)); 4498 list_create(&zone->zone_datasets, sizeof (zone_dataset_t), 4499 offsetof(zone_dataset_t, zd_linkage)); 4500 list_create(&zone->zone_dl_list, sizeof (zone_dl_t), 4501 offsetof(zone_dl_t, zdl_linkage)); 4502 rw_init(&zone->zone_mlps.mlpl_rwlock, NULL, RW_DEFAULT, NULL); 4503 rw_init(&zone->zone_mntfs_db_lock, NULL, RW_DEFAULT, NULL); 4504 4505 if (flags & ZCF_NET_EXCL) { 4506 zone->zone_flags |= ZF_NET_EXCL; 4507 } 4508 4509 if ((error = zone_set_name(zone, zone_name)) != 0) { 4510 zone_free(zone); 4511 return (zone_create_error(error, 0, extended_error)); 4512 } 4513 4514 if ((error = zone_set_root(zone, zone_root)) != 0) { 4515 zone_free(zone); 4516 return (zone_create_error(error, 0, extended_error)); 4517 } 4518 if ((error = zone_set_privset(zone, zone_privs, zone_privssz)) != 0) { 4519 zone_free(zone); 4520 return (zone_create_error(error, 0, extended_error)); 4521 } 4522 4523 /* initialize node name to be the same as zone name */ 4524 zone->zone_nodename = kmem_alloc(_SYS_NMLN, KM_SLEEP); 4525 (void) strncpy(zone->zone_nodename, zone->zone_name, _SYS_NMLN); 4526 zone->zone_nodename[_SYS_NMLN - 1] = '\0'; 4527 4528 zone->zone_domain = kmem_alloc(_SYS_NMLN, KM_SLEEP); 4529 zone->zone_domain[0] = '\0'; 4530 zone->zone_hostid = HW_INVALID_HOSTID; 4531 zone->zone_shares = 1; 4532 zone->zone_shmmax = 0; 4533 zone->zone_ipc.ipcq_shmmni = 0; 4534 zone->zone_ipc.ipcq_semmni = 0; 4535 zone->zone_ipc.ipcq_msgmni = 0; 4536 zone->zone_bootargs = NULL; 4537 zone->zone_fs_allowed = NULL; 4538 4539 psecflags_default(&zone->zone_secflags); 4540 4541 zone->zone_initname = 4542 kmem_alloc(strlen(zone_default_initname) + 1, KM_SLEEP); 4543 (void) strcpy(zone->zone_initname, zone_default_initname); 4544 zone->zone_nlwps = 0; 4545 zone->zone_nlwps_ctl = INT_MAX; 4546 zone->zone_nprocs = 0; 4547 zone->zone_nprocs_ctl = INT_MAX; 4548 zone->zone_locked_mem = 0; 4549 zone->zone_locked_mem_ctl = UINT64_MAX; 4550 zone->zone_max_swap = 0; 4551 zone->zone_max_swap_ctl = UINT64_MAX; 4552 zone->zone_max_lofi = 0; 4553 zone->zone_max_lofi_ctl = UINT64_MAX; 4554 zone0.zone_lockedmem_kstat = NULL; 4555 zone0.zone_swapresv_kstat = NULL; 4556 4557 zone->zone_ustate = cpu_uarray_zalloc(ZONE_USTATE_MAX, KM_SLEEP); 4558 4559 /* 4560 * Zsched initializes the rctls. 4561 */ 4562 zone->zone_rctls = NULL; 4563 4564 if ((error = parse_rctls(rctlbuf, rctlbufsz, &rctls)) != 0) { 4565 zone_free(zone); 4566 return (zone_create_error(error, 0, extended_error)); 4567 } 4568 4569 if ((error = parse_zfs(zone, zfsbuf, zfsbufsz)) != 0) { 4570 zone_free(zone); 4571 return (set_errno(error)); 4572 } 4573 4574 /* 4575 * Read in the trusted system parameters: 4576 * match flag and sensitivity label. 4577 */ 4578 zone->zone_match = match; 4579 if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) { 4580 /* Fail if requested to set doi to anything but system's doi */ 4581 if (doi != 0 && doi != default_doi) { 4582 zone_free(zone); 4583 return (set_errno(EINVAL)); 4584 } 4585 /* Always apply system's doi to the zone */ 4586 error = zone_set_label(zone, label, default_doi); 4587 if (error != 0) { 4588 zone_free(zone); 4589 return (set_errno(error)); 4590 } 4591 insert_label_hash = B_TRUE; 4592 } else { 4593 /* all zones get an admin_low label if system is not labeled */ 4594 zone->zone_slabel = l_admin_low; 4595 label_hold(l_admin_low); 4596 insert_label_hash = B_FALSE; 4597 } 4598 4599 /* 4600 * Stop all lwps since that's what normally happens as part of fork(). 4601 * This needs to happen before we grab any locks to avoid deadlock 4602 * (another lwp in the process could be waiting for the held lock). 4603 */ 4604 if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK)) { 4605 zone_free(zone); 4606 nvlist_free(rctls); 4607 return (zone_create_error(error, 0, extended_error)); 4608 } 4609 4610 if (block_mounts(zone) == 0) { 4611 mutex_enter(&pp->p_lock); 4612 if (curthread != pp->p_agenttp) 4613 continuelwps(pp); 4614 mutex_exit(&pp->p_lock); 4615 zone_free(zone); 4616 nvlist_free(rctls); 4617 return (zone_create_error(error, 0, extended_error)); 4618 } 4619 4620 /* 4621 * Set up credential for kernel access. After this, any errors 4622 * should go through the dance in errout rather than calling 4623 * zone_free directly. 4624 */ 4625 zone->zone_kcred = crdup(kcred); 4626 crsetzone(zone->zone_kcred, zone); 4627 priv_intersect(zone->zone_privset, &CR_PPRIV(zone->zone_kcred)); 4628 priv_intersect(zone->zone_privset, &CR_EPRIV(zone->zone_kcred)); 4629 priv_intersect(zone->zone_privset, &CR_IPRIV(zone->zone_kcred)); 4630 priv_intersect(zone->zone_privset, &CR_LPRIV(zone->zone_kcred)); 4631 4632 mutex_enter(&zonehash_lock); 4633 /* 4634 * Make sure zone doesn't already exist. 4635 * 4636 * If the system and zone are labeled, 4637 * make sure no other zone exists that has the same label. 4638 */ 4639 if ((ztmp = zone_find_all_by_name(zone->zone_name)) != NULL || 4640 (insert_label_hash && 4641 (ztmp = zone_find_all_by_label(zone->zone_slabel)) != NULL)) { 4642 zone_status_t status; 4643 4644 status = zone_status_get(ztmp); 4645 if (status == ZONE_IS_READY || status == ZONE_IS_RUNNING) 4646 error = EEXIST; 4647 else 4648 error = EBUSY; 4649 4650 if (insert_label_hash) 4651 error2 = ZE_LABELINUSE; 4652 4653 goto errout; 4654 } 4655 4656 /* 4657 * Don't allow zone creations which would cause one zone's rootpath to 4658 * be accessible from that of another (non-global) zone. 4659 */ 4660 if (zone_is_nested(zone->zone_rootpath)) { 4661 error = EBUSY; 4662 goto errout; 4663 } 4664 4665 ASSERT(zonecount != 0); /* check for leaks */ 4666 if (zonecount + 1 > maxzones) { 4667 error = ENOMEM; 4668 goto errout; 4669 } 4670 4671 if (zone_mount_count(zone->zone_rootpath) != 0) { 4672 error = EBUSY; 4673 error2 = ZE_AREMOUNTS; 4674 goto errout; 4675 } 4676 4677 /* 4678 * Zone is still incomplete, but we need to drop all locks while 4679 * zsched() initializes this zone's kernel process. We 4680 * optimistically add the zone to the hashtable and associated 4681 * lists so a parallel zone_create() doesn't try to create the 4682 * same zone. 4683 */ 4684 zonecount++; 4685 (void) mod_hash_insert(zonehashbyid, 4686 (mod_hash_key_t)(uintptr_t)zone->zone_id, 4687 (mod_hash_val_t)(uintptr_t)zone); 4688 str = kmem_alloc(strlen(zone->zone_name) + 1, KM_SLEEP); 4689 (void) strcpy(str, zone->zone_name); 4690 (void) mod_hash_insert(zonehashbyname, (mod_hash_key_t)str, 4691 (mod_hash_val_t)(uintptr_t)zone); 4692 if (insert_label_hash) { 4693 (void) mod_hash_insert(zonehashbylabel, 4694 (mod_hash_key_t)zone->zone_slabel, (mod_hash_val_t)zone); 4695 zone->zone_flags |= ZF_HASHED_LABEL; 4696 } 4697 4698 /* 4699 * Insert into active list. At this point there are no 'hold's 4700 * on the zone, but everyone else knows not to use it, so we can 4701 * continue to use it. zsched() will do a zone_hold() if the 4702 * newproc() is successful. 4703 */ 4704 list_insert_tail(&zone_active, zone); 4705 mutex_exit(&zonehash_lock); 4706 4707 zarg.zone = zone; 4708 zarg.nvlist = rctls; 4709 /* 4710 * The process, task, and project rctls are probably wrong; 4711 * we need an interface to get the default values of all rctls, 4712 * and initialize zsched appropriately. I'm not sure that that 4713 * makes much of a difference, though. 4714 */ 4715 error = newproc(zsched, (void *)&zarg, syscid, minclsyspri, NULL, 0); 4716 if (error != 0) { 4717 /* 4718 * We need to undo all globally visible state. 4719 */ 4720 mutex_enter(&zonehash_lock); 4721 list_remove(&zone_active, zone); 4722 if (zone->zone_flags & ZF_HASHED_LABEL) { 4723 ASSERT(zone->zone_slabel != NULL); 4724 (void) mod_hash_destroy(zonehashbylabel, 4725 (mod_hash_key_t)zone->zone_slabel); 4726 } 4727 (void) mod_hash_destroy(zonehashbyname, 4728 (mod_hash_key_t)(uintptr_t)zone->zone_name); 4729 (void) mod_hash_destroy(zonehashbyid, 4730 (mod_hash_key_t)(uintptr_t)zone->zone_id); 4731 ASSERT(zonecount > 1); 4732 zonecount--; 4733 goto errout; 4734 } 4735 4736 /* 4737 * Zone creation can't fail from now on. 4738 */ 4739 4740 /* 4741 * Create zone kstats 4742 */ 4743 zone_kstat_create(zone); 4744 4745 /* 4746 * Let the other lwps continue. 4747 */ 4748 mutex_enter(&pp->p_lock); 4749 if (curthread != pp->p_agenttp) 4750 continuelwps(pp); 4751 mutex_exit(&pp->p_lock); 4752 4753 /* 4754 * Wait for zsched to finish initializing the zone. 4755 */ 4756 zone_status_wait(zone, ZONE_IS_READY); 4757 /* 4758 * The zone is fully visible, so we can let mounts progress. 4759 */ 4760 resume_mounts(zone); 4761 nvlist_free(rctls); 4762 4763 return (zoneid); 4764 4765 errout: 4766 mutex_exit(&zonehash_lock); 4767 /* 4768 * Let the other lwps continue. 4769 */ 4770 mutex_enter(&pp->p_lock); 4771 if (curthread != pp->p_agenttp) 4772 continuelwps(pp); 4773 mutex_exit(&pp->p_lock); 4774 4775 resume_mounts(zone); 4776 nvlist_free(rctls); 4777 /* 4778 * There is currently one reference to the zone, a cred_ref from 4779 * zone_kcred. To free the zone, we call crfree, which will call 4780 * zone_cred_rele, which will call zone_free. 4781 */ 4782 ASSERT(zone->zone_cred_ref == 1); 4783 ASSERT(zone->zone_kcred->cr_ref == 1); 4784 ASSERT(zone->zone_ref == 0); 4785 zkcr = zone->zone_kcred; 4786 zone->zone_kcred = NULL; 4787 crfree(zkcr); /* triggers call to zone_free */ 4788 return (zone_create_error(error, error2, extended_error)); 4789 } 4790 4791 /* 4792 * Cause the zone to boot. This is pretty simple, since we let zoneadmd do 4793 * the heavy lifting. initname is the path to the program to launch 4794 * at the "top" of the zone; if this is NULL, we use the system default, 4795 * which is stored at zone_default_initname. 4796 */ 4797 static int 4798 zone_boot(zoneid_t zoneid) 4799 { 4800 int err; 4801 zone_t *zone; 4802 4803 if (secpolicy_zone_config(CRED()) != 0) 4804 return (set_errno(EPERM)); 4805 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID) 4806 return (set_errno(EINVAL)); 4807 4808 mutex_enter(&zonehash_lock); 4809 /* 4810 * Look for zone under hash lock to prevent races with calls to 4811 * zone_shutdown, zone_destroy, etc. 4812 */ 4813 if ((zone = zone_find_all_by_id(zoneid)) == NULL) { 4814 mutex_exit(&zonehash_lock); 4815 return (set_errno(EINVAL)); 4816 } 4817 4818 mutex_enter(&zone_status_lock); 4819 if (zone_status_get(zone) != ZONE_IS_READY) { 4820 mutex_exit(&zone_status_lock); 4821 mutex_exit(&zonehash_lock); 4822 return (set_errno(EINVAL)); 4823 } 4824 zone_status_set(zone, ZONE_IS_BOOTING); 4825 mutex_exit(&zone_status_lock); 4826 4827 zone_hold(zone); /* so we can use the zone_t later */ 4828 mutex_exit(&zonehash_lock); 4829 4830 if (zone_status_wait_sig(zone, ZONE_IS_RUNNING) == 0) { 4831 zone_rele(zone); 4832 return (set_errno(EINTR)); 4833 } 4834 4835 /* 4836 * Boot (starting init) might have failed, in which case the zone 4837 * will go to the SHUTTING_DOWN state; an appropriate errno will 4838 * be placed in zone->zone_boot_err, and so we return that. 4839 */ 4840 err = zone->zone_boot_err; 4841 zone_rele(zone); 4842 return (err ? set_errno(err) : 0); 4843 } 4844 4845 /* 4846 * Kills all user processes in the zone, waiting for them all to exit 4847 * before returning. 4848 */ 4849 static int 4850 zone_empty(zone_t *zone) 4851 { 4852 int waitstatus; 4853 4854 /* 4855 * We need to drop zonehash_lock before killing all 4856 * processes, otherwise we'll deadlock with zone_find_* 4857 * which can be called from the exit path. 4858 */ 4859 ASSERT(MUTEX_NOT_HELD(&zonehash_lock)); 4860 while ((waitstatus = zone_status_timedwait_sig(zone, 4861 ddi_get_lbolt() + hz, ZONE_IS_EMPTY)) == -1) { 4862 killall(zone->zone_id); 4863 } 4864 /* 4865 * return EINTR if we were signaled 4866 */ 4867 if (waitstatus == 0) 4868 return (EINTR); 4869 return (0); 4870 } 4871 4872 /* 4873 * This function implements the policy for zone visibility. 4874 * 4875 * In standard Solaris, a non-global zone can only see itself. 4876 * 4877 * In Trusted Extensions, a labeled zone can lookup any zone whose label 4878 * it dominates. For this test, the label of the global zone is treated as 4879 * admin_high so it is special-cased instead of being checked for dominance. 4880 * 4881 * Returns true if zone attributes are viewable, false otherwise. 4882 */ 4883 static boolean_t 4884 zone_list_access(zone_t *zone) 4885 { 4886 4887 if (curproc->p_zone == global_zone || 4888 curproc->p_zone == zone) { 4889 return (B_TRUE); 4890 } else if (is_system_labeled() && !(zone->zone_flags & ZF_IS_SCRATCH)) { 4891 bslabel_t *curproc_label; 4892 bslabel_t *zone_label; 4893 4894 curproc_label = label2bslabel(curproc->p_zone->zone_slabel); 4895 zone_label = label2bslabel(zone->zone_slabel); 4896 4897 if (zone->zone_id != GLOBAL_ZONEID && 4898 bldominates(curproc_label, zone_label)) { 4899 return (B_TRUE); 4900 } else { 4901 return (B_FALSE); 4902 } 4903 } else { 4904 return (B_FALSE); 4905 } 4906 } 4907 4908 /* 4909 * Systemcall to start the zone's halt sequence. By the time this 4910 * function successfully returns, all user processes and kernel threads 4911 * executing in it will have exited, ZSD shutdown callbacks executed, 4912 * and the zone status set to ZONE_IS_DOWN. 4913 * 4914 * It is possible that the call will interrupt itself if the caller is the 4915 * parent of any process running in the zone, and doesn't have SIGCHLD blocked. 4916 */ 4917 static int 4918 zone_shutdown(zoneid_t zoneid) 4919 { 4920 int error; 4921 zone_t *zone; 4922 zone_status_t status; 4923 4924 if (secpolicy_zone_config(CRED()) != 0) 4925 return (set_errno(EPERM)); 4926 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID) 4927 return (set_errno(EINVAL)); 4928 4929 mutex_enter(&zonehash_lock); 4930 /* 4931 * Look for zone under hash lock to prevent races with other 4932 * calls to zone_shutdown and zone_destroy. 4933 */ 4934 if ((zone = zone_find_all_by_id(zoneid)) == NULL) { 4935 mutex_exit(&zonehash_lock); 4936 return (set_errno(EINVAL)); 4937 } 4938 4939 /* 4940 * We have to drop zonehash_lock before calling block_mounts. 4941 * Hold the zone so we can continue to use the zone_t. 4942 */ 4943 zone_hold(zone); 4944 mutex_exit(&zonehash_lock); 4945 4946 /* 4947 * Block mounts so that VFS_MOUNT() can get an accurate view of 4948 * the zone's status with regards to ZONE_IS_SHUTTING down. 4949 * 4950 * e.g. NFS can fail the mount if it determines that the zone 4951 * has already begun the shutdown sequence. 4952 * 4953 */ 4954 if (block_mounts(zone) == 0) { 4955 zone_rele(zone); 4956 return (set_errno(EINTR)); 4957 } 4958 4959 mutex_enter(&zonehash_lock); 4960 mutex_enter(&zone_status_lock); 4961 status = zone_status_get(zone); 4962 /* 4963 * Fail if the zone isn't fully initialized yet. 4964 */ 4965 if (status < ZONE_IS_READY) { 4966 mutex_exit(&zone_status_lock); 4967 mutex_exit(&zonehash_lock); 4968 resume_mounts(zone); 4969 zone_rele(zone); 4970 return (set_errno(EINVAL)); 4971 } 4972 /* 4973 * If conditions required for zone_shutdown() to return have been met, 4974 * return success. 4975 */ 4976 if (status >= ZONE_IS_DOWN) { 4977 mutex_exit(&zone_status_lock); 4978 mutex_exit(&zonehash_lock); 4979 resume_mounts(zone); 4980 zone_rele(zone); 4981 return (0); 4982 } 4983 /* 4984 * If zone_shutdown() hasn't been called before, go through the motions. 4985 * If it has, there's nothing to do but wait for the kernel threads to 4986 * drain. 4987 */ 4988 if (status < ZONE_IS_EMPTY) { 4989 uint_t ntasks; 4990 4991 mutex_enter(&zone->zone_lock); 4992 if ((ntasks = zone->zone_ntasks) != 1) { 4993 /* 4994 * There's still stuff running. 4995 */ 4996 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN); 4997 } 4998 mutex_exit(&zone->zone_lock); 4999 if (ntasks == 1) { 5000 /* 5001 * The only way to create another task is through 5002 * zone_enter(), which will block until we drop 5003 * zonehash_lock. The zone is empty. 5004 */ 5005 if (zone->zone_kthreads == NULL) { 5006 /* 5007 * Skip ahead to ZONE_IS_DOWN 5008 */ 5009 zone_status_set(zone, ZONE_IS_DOWN); 5010 } else { 5011 zone_status_set(zone, ZONE_IS_EMPTY); 5012 } 5013 } 5014 } 5015 mutex_exit(&zone_status_lock); 5016 mutex_exit(&zonehash_lock); 5017 resume_mounts(zone); 5018 5019 if (error = zone_empty(zone)) { 5020 zone_rele(zone); 5021 return (set_errno(error)); 5022 } 5023 /* 5024 * After the zone status goes to ZONE_IS_DOWN this zone will no 5025 * longer be notified of changes to the pools configuration, so 5026 * in order to not end up with a stale pool pointer, we point 5027 * ourselves at the default pool and remove all resource 5028 * visibility. This is especially important as the zone_t may 5029 * languish on the deathrow for a very long time waiting for 5030 * cred's to drain out. 5031 * 5032 * This rebinding of the zone can happen multiple times 5033 * (presumably due to interrupted or parallel systemcalls) 5034 * without any adverse effects. 5035 */ 5036 if (pool_lock_intr() != 0) { 5037 zone_rele(zone); 5038 return (set_errno(EINTR)); 5039 } 5040 if (pool_state == POOL_ENABLED) { 5041 mutex_enter(&cpu_lock); 5042 zone_pool_set(zone, pool_default); 5043 /* 5044 * The zone no longer needs to be able to see any cpus. 5045 */ 5046 zone_pset_set(zone, ZONE_PS_INVAL); 5047 mutex_exit(&cpu_lock); 5048 } 5049 pool_unlock(); 5050 5051 /* 5052 * ZSD shutdown callbacks can be executed multiple times, hence 5053 * it is safe to not be holding any locks across this call. 5054 */ 5055 zone_zsd_callbacks(zone, ZSD_SHUTDOWN); 5056 5057 mutex_enter(&zone_status_lock); 5058 if (zone->zone_kthreads == NULL && zone_status_get(zone) < ZONE_IS_DOWN) 5059 zone_status_set(zone, ZONE_IS_DOWN); 5060 mutex_exit(&zone_status_lock); 5061 5062 /* 5063 * Wait for kernel threads to drain. 5064 */ 5065 if (!zone_status_wait_sig(zone, ZONE_IS_DOWN)) { 5066 zone_rele(zone); 5067 return (set_errno(EINTR)); 5068 } 5069 5070 /* 5071 * Zone can be become down/destroyable even if the above wait 5072 * returns EINTR, so any code added here may never execute. 5073 * (i.e. don't add code here) 5074 */ 5075 5076 zone_rele(zone); 5077 return (0); 5078 } 5079 5080 /* 5081 * Log the specified zone's reference counts. The caller should not be 5082 * holding the zone's zone_lock. 5083 */ 5084 static void 5085 zone_log_refcounts(zone_t *zone) 5086 { 5087 char *buffer; 5088 char *buffer_position; 5089 uint32_t buffer_size; 5090 uint32_t index; 5091 uint_t ref; 5092 uint_t cred_ref; 5093 5094 /* 5095 * Construct a string representing the subsystem-specific reference 5096 * counts. The counts are printed in ascending order by index into the 5097 * zone_t::zone_subsys_ref array. The list will be surrounded by 5098 * square brackets [] and will only contain nonzero reference counts. 5099 * 5100 * The buffer will hold two square bracket characters plus ten digits, 5101 * one colon, one space, one comma, and some characters for a 5102 * subsystem name per subsystem-specific reference count. (Unsigned 32- 5103 * bit integers have at most ten decimal digits.) The last 5104 * reference count's comma is replaced by the closing square 5105 * bracket and a NULL character to terminate the string. 5106 * 5107 * NOTE: We have to grab the zone's zone_lock to create a consistent 5108 * snapshot of the zone's reference counters. 5109 * 5110 * First, figure out how much space the string buffer will need. 5111 * The buffer's size is stored in buffer_size. 5112 */ 5113 buffer_size = 2; /* for the square brackets */ 5114 mutex_enter(&zone->zone_lock); 5115 zone->zone_flags |= ZF_REFCOUNTS_LOGGED; 5116 ref = zone->zone_ref; 5117 cred_ref = zone->zone_cred_ref; 5118 for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index) 5119 if (zone->zone_subsys_ref[index] != 0) 5120 buffer_size += strlen(zone_ref_subsys_names[index]) + 5121 13; 5122 if (buffer_size == 2) { 5123 /* 5124 * No subsystems had nonzero reference counts. Don't bother 5125 * with allocating a buffer; just log the general-purpose and 5126 * credential reference counts. 5127 */ 5128 mutex_exit(&zone->zone_lock); 5129 (void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE, 5130 "Zone '%s' (ID: %d) is shutting down, but %u zone " 5131 "references and %u credential references are still extant", 5132 zone->zone_name, zone->zone_id, ref, cred_ref); 5133 return; 5134 } 5135 5136 /* 5137 * buffer_size contains the exact number of characters that the 5138 * buffer will need. Allocate the buffer and fill it with nonzero 5139 * subsystem-specific reference counts. Surround the results with 5140 * square brackets afterwards. 5141 */ 5142 buffer = kmem_alloc(buffer_size, KM_SLEEP); 5143 buffer_position = &buffer[1]; 5144 for (index = 0; index < ZONE_REF_NUM_SUBSYS; ++index) { 5145 /* 5146 * NOTE: The DDI's version of sprintf() returns a pointer to 5147 * the modified buffer rather than the number of bytes written 5148 * (as in snprintf(3C)). This is unfortunate and annoying. 5149 * Therefore, we'll use snprintf() with INT_MAX to get the 5150 * number of bytes written. Using INT_MAX is safe because 5151 * the buffer is perfectly sized for the data: we'll never 5152 * overrun the buffer. 5153 */ 5154 if (zone->zone_subsys_ref[index] != 0) 5155 buffer_position += snprintf(buffer_position, INT_MAX, 5156 "%s: %u,", zone_ref_subsys_names[index], 5157 zone->zone_subsys_ref[index]); 5158 } 5159 mutex_exit(&zone->zone_lock); 5160 buffer[0] = '['; 5161 ASSERT((uintptr_t)(buffer_position - buffer) < buffer_size); 5162 ASSERT(buffer_position[0] == '\0' && buffer_position[-1] == ','); 5163 buffer_position[-1] = ']'; 5164 5165 /* 5166 * Log the reference counts and free the message buffer. 5167 */ 5168 (void) strlog(0, 0, 1, SL_CONSOLE | SL_NOTE, 5169 "Zone '%s' (ID: %d) is shutting down, but %u zone references and " 5170 "%u credential references are still extant %s", zone->zone_name, 5171 zone->zone_id, ref, cred_ref, buffer); 5172 kmem_free(buffer, buffer_size); 5173 } 5174 5175 /* 5176 * Systemcall entry point to finalize the zone halt process. The caller 5177 * must have already successfully called zone_shutdown(). 5178 * 5179 * Upon successful completion, the zone will have been fully destroyed: 5180 * zsched will have exited, destructor callbacks executed, and the zone 5181 * removed from the list of active zones. 5182 */ 5183 static int 5184 zone_destroy(zoneid_t zoneid) 5185 { 5186 uint64_t uniqid; 5187 zone_t *zone; 5188 zone_status_t status; 5189 clock_t wait_time; 5190 boolean_t log_refcounts; 5191 5192 if (secpolicy_zone_config(CRED()) != 0) 5193 return (set_errno(EPERM)); 5194 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID) 5195 return (set_errno(EINVAL)); 5196 5197 mutex_enter(&zonehash_lock); 5198 /* 5199 * Look for zone under hash lock to prevent races with other 5200 * calls to zone_destroy. 5201 */ 5202 if ((zone = zone_find_all_by_id(zoneid)) == NULL) { 5203 mutex_exit(&zonehash_lock); 5204 return (set_errno(EINVAL)); 5205 } 5206 5207 if (zone_mount_count(zone->zone_rootpath) != 0) { 5208 mutex_exit(&zonehash_lock); 5209 return (set_errno(EBUSY)); 5210 } 5211 mutex_enter(&zone_status_lock); 5212 status = zone_status_get(zone); 5213 if (status < ZONE_IS_DOWN) { 5214 mutex_exit(&zone_status_lock); 5215 mutex_exit(&zonehash_lock); 5216 return (set_errno(EBUSY)); 5217 } else if (status == ZONE_IS_DOWN) { 5218 zone_status_set(zone, ZONE_IS_DYING); /* Tell zsched to exit */ 5219 } 5220 mutex_exit(&zone_status_lock); 5221 zone_hold(zone); 5222 mutex_exit(&zonehash_lock); 5223 5224 /* 5225 * wait for zsched to exit 5226 */ 5227 zone_status_wait(zone, ZONE_IS_DEAD); 5228 zone_zsd_callbacks(zone, ZSD_DESTROY); 5229 zone->zone_netstack = NULL; 5230 uniqid = zone->zone_uniqid; 5231 zone_rele(zone); 5232 zone = NULL; /* potentially free'd */ 5233 5234 log_refcounts = B_FALSE; 5235 wait_time = SEC_TO_TICK(ZONE_DESTROY_TIMEOUT_SECS); 5236 mutex_enter(&zonehash_lock); 5237 for (; /* ever */; ) { 5238 boolean_t unref; 5239 boolean_t refs_have_been_logged; 5240 5241 if ((zone = zone_find_all_by_id(zoneid)) == NULL || 5242 zone->zone_uniqid != uniqid) { 5243 /* 5244 * The zone has gone away. Necessary conditions 5245 * are met, so we return success. 5246 */ 5247 mutex_exit(&zonehash_lock); 5248 return (0); 5249 } 5250 mutex_enter(&zone->zone_lock); 5251 unref = ZONE_IS_UNREF(zone); 5252 refs_have_been_logged = (zone->zone_flags & 5253 ZF_REFCOUNTS_LOGGED); 5254 mutex_exit(&zone->zone_lock); 5255 if (unref) { 5256 /* 5257 * There is only one reference to the zone -- that 5258 * added when the zone was added to the hashtables -- 5259 * and things will remain this way until we drop 5260 * zonehash_lock... we can go ahead and cleanup the 5261 * zone. 5262 */ 5263 break; 5264 } 5265 5266 /* 5267 * Wait for zone_rele_common() or zone_cred_rele() to signal 5268 * zone_destroy_cv. zone_destroy_cv is signaled only when 5269 * some zone's general-purpose reference count reaches one. 5270 * If ZONE_DESTROY_TIMEOUT_SECS seconds elapse while waiting 5271 * on zone_destroy_cv, then log the zone's reference counts and 5272 * continue to wait for zone_rele() and zone_cred_rele(). 5273 */ 5274 if (!refs_have_been_logged) { 5275 if (!log_refcounts) { 5276 /* 5277 * This thread hasn't timed out waiting on 5278 * zone_destroy_cv yet. Wait wait_time clock 5279 * ticks (initially ZONE_DESTROY_TIMEOUT_SECS 5280 * seconds) for the zone's references to clear. 5281 */ 5282 ASSERT(wait_time > 0); 5283 wait_time = cv_reltimedwait_sig( 5284 &zone_destroy_cv, &zonehash_lock, wait_time, 5285 TR_SEC); 5286 if (wait_time > 0) { 5287 /* 5288 * A thread in zone_rele() or 5289 * zone_cred_rele() signaled 5290 * zone_destroy_cv before this thread's 5291 * wait timed out. The zone might have 5292 * only one reference left; find out! 5293 */ 5294 continue; 5295 } else if (wait_time == 0) { 5296 /* The thread's process was signaled. */ 5297 mutex_exit(&zonehash_lock); 5298 return (set_errno(EINTR)); 5299 } 5300 5301 /* 5302 * The thread timed out while waiting on 5303 * zone_destroy_cv. Even though the thread 5304 * timed out, it has to check whether another 5305 * thread woke up from zone_destroy_cv and 5306 * destroyed the zone. 5307 * 5308 * If the zone still exists and has more than 5309 * one unreleased general-purpose reference, 5310 * then log the zone's reference counts. 5311 */ 5312 log_refcounts = B_TRUE; 5313 continue; 5314 } 5315 5316 /* 5317 * The thread already timed out on zone_destroy_cv while 5318 * waiting for subsystems to release the zone's last 5319 * general-purpose references. Log the zone's reference 5320 * counts and wait indefinitely on zone_destroy_cv. 5321 */ 5322 zone_log_refcounts(zone); 5323 } 5324 if (cv_wait_sig(&zone_destroy_cv, &zonehash_lock) == 0) { 5325 /* The thread's process was signaled. */ 5326 mutex_exit(&zonehash_lock); 5327 return (set_errno(EINTR)); 5328 } 5329 } 5330 5331 /* 5332 * Remove CPU cap for this zone now since we're not going to 5333 * fail below this point. 5334 */ 5335 cpucaps_zone_remove(zone); 5336 5337 /* Get rid of the zone's kstats */ 5338 zone_kstat_delete(zone); 5339 5340 /* remove the pfexecd doors */ 5341 if (zone->zone_pfexecd != NULL) { 5342 klpd_freelist(&zone->zone_pfexecd); 5343 zone->zone_pfexecd = NULL; 5344 } 5345 5346 /* free brand specific data */ 5347 if (ZONE_IS_BRANDED(zone)) 5348 ZBROP(zone)->b_free_brand_data(zone); 5349 5350 /* Say goodbye to brand framework. */ 5351 brand_unregister_zone(zone->zone_brand); 5352 5353 /* 5354 * It is now safe to let the zone be recreated; remove it from the 5355 * lists. The memory will not be freed until the last cred 5356 * reference goes away. 5357 */ 5358 ASSERT(zonecount > 1); /* must be > 1; can't destroy global zone */ 5359 zonecount--; 5360 /* remove from active list and hash tables */ 5361 list_remove(&zone_active, zone); 5362 (void) mod_hash_destroy(zonehashbyname, 5363 (mod_hash_key_t)zone->zone_name); 5364 (void) mod_hash_destroy(zonehashbyid, 5365 (mod_hash_key_t)(uintptr_t)zone->zone_id); 5366 if (zone->zone_flags & ZF_HASHED_LABEL) 5367 (void) mod_hash_destroy(zonehashbylabel, 5368 (mod_hash_key_t)zone->zone_slabel); 5369 mutex_exit(&zonehash_lock); 5370 5371 /* 5372 * Release the root vnode; we're not using it anymore. Nor should any 5373 * other thread that might access it exist. 5374 */ 5375 if (zone->zone_rootvp != NULL) { 5376 VN_RELE(zone->zone_rootvp); 5377 zone->zone_rootvp = NULL; 5378 } 5379 5380 /* add to deathrow list */ 5381 mutex_enter(&zone_deathrow_lock); 5382 list_insert_tail(&zone_deathrow, zone); 5383 mutex_exit(&zone_deathrow_lock); 5384 5385 /* 5386 * Drop last reference (which was added by zsched()), this will 5387 * free the zone unless there are outstanding cred references. 5388 */ 5389 zone_rele(zone); 5390 return (0); 5391 } 5392 5393 /* 5394 * Systemcall entry point for zone_getattr(2). 5395 */ 5396 static ssize_t 5397 zone_getattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize) 5398 { 5399 size_t size; 5400 int error = 0, err; 5401 zone_t *zone; 5402 char *zonepath; 5403 char *outstr; 5404 zone_status_t zone_status; 5405 pid_t initpid; 5406 boolean_t global = (curzone == global_zone); 5407 boolean_t inzone = (curzone->zone_id == zoneid); 5408 ushort_t flags; 5409 zone_net_data_t *zbuf; 5410 5411 mutex_enter(&zonehash_lock); 5412 if ((zone = zone_find_all_by_id(zoneid)) == NULL) { 5413 mutex_exit(&zonehash_lock); 5414 return (set_errno(EINVAL)); 5415 } 5416 zone_status = zone_status_get(zone); 5417 if (zone_status < ZONE_IS_INITIALIZED) { 5418 mutex_exit(&zonehash_lock); 5419 return (set_errno(EINVAL)); 5420 } 5421 zone_hold(zone); 5422 mutex_exit(&zonehash_lock); 5423 5424 /* 5425 * If not in the global zone, don't show information about other zones, 5426 * unless the system is labeled and the local zone's label dominates 5427 * the other zone. 5428 */ 5429 if (!zone_list_access(zone)) { 5430 zone_rele(zone); 5431 return (set_errno(EINVAL)); 5432 } 5433 5434 switch (attr) { 5435 case ZONE_ATTR_ROOT: 5436 if (global) { 5437 /* 5438 * Copy the path to trim the trailing "/" (except for 5439 * the global zone). 5440 */ 5441 if (zone != global_zone) 5442 size = zone->zone_rootpathlen - 1; 5443 else 5444 size = zone->zone_rootpathlen; 5445 zonepath = kmem_alloc(size, KM_SLEEP); 5446 bcopy(zone->zone_rootpath, zonepath, size); 5447 zonepath[size - 1] = '\0'; 5448 } else { 5449 if (inzone || !is_system_labeled()) { 5450 /* 5451 * Caller is not in the global zone. 5452 * if the query is on the current zone 5453 * or the system is not labeled, 5454 * just return faked-up path for current zone. 5455 */ 5456 zonepath = "/"; 5457 size = 2; 5458 } else { 5459 /* 5460 * Return related path for current zone. 5461 */ 5462 int prefix_len = strlen(zone_prefix); 5463 int zname_len = strlen(zone->zone_name); 5464 5465 size = prefix_len + zname_len + 1; 5466 zonepath = kmem_alloc(size, KM_SLEEP); 5467 bcopy(zone_prefix, zonepath, prefix_len); 5468 bcopy(zone->zone_name, zonepath + 5469 prefix_len, zname_len); 5470 zonepath[size - 1] = '\0'; 5471 } 5472 } 5473 if (bufsize > size) 5474 bufsize = size; 5475 if (buf != NULL) { 5476 err = copyoutstr(zonepath, buf, bufsize, NULL); 5477 if (err != 0 && err != ENAMETOOLONG) 5478 error = EFAULT; 5479 } 5480 if (global || (is_system_labeled() && !inzone)) 5481 kmem_free(zonepath, size); 5482 break; 5483 5484 case ZONE_ATTR_NAME: 5485 size = strlen(zone->zone_name) + 1; 5486 if (bufsize > size) 5487 bufsize = size; 5488 if (buf != NULL) { 5489 err = copyoutstr(zone->zone_name, buf, bufsize, NULL); 5490 if (err != 0 && err != ENAMETOOLONG) 5491 error = EFAULT; 5492 } 5493 break; 5494 5495 case ZONE_ATTR_STATUS: 5496 /* 5497 * Since we're not holding zonehash_lock, the zone status 5498 * may be anything; leave it up to userland to sort it out. 5499 */ 5500 size = sizeof (zone_status); 5501 if (bufsize > size) 5502 bufsize = size; 5503 zone_status = zone_status_get(zone); 5504 if (buf != NULL && 5505 copyout(&zone_status, buf, bufsize) != 0) 5506 error = EFAULT; 5507 break; 5508 case ZONE_ATTR_FLAGS: 5509 size = sizeof (zone->zone_flags); 5510 if (bufsize > size) 5511 bufsize = size; 5512 flags = zone->zone_flags; 5513 if (buf != NULL && 5514 copyout(&flags, buf, bufsize) != 0) 5515 error = EFAULT; 5516 break; 5517 case ZONE_ATTR_PRIVSET: 5518 size = sizeof (priv_set_t); 5519 if (bufsize > size) 5520 bufsize = size; 5521 if (buf != NULL && 5522 copyout(zone->zone_privset, buf, bufsize) != 0) 5523 error = EFAULT; 5524 break; 5525 case ZONE_ATTR_UNIQID: 5526 size = sizeof (zone->zone_uniqid); 5527 if (bufsize > size) 5528 bufsize = size; 5529 if (buf != NULL && 5530 copyout(&zone->zone_uniqid, buf, bufsize) != 0) 5531 error = EFAULT; 5532 break; 5533 case ZONE_ATTR_POOLID: 5534 { 5535 pool_t *pool; 5536 poolid_t poolid; 5537 5538 if (pool_lock_intr() != 0) { 5539 error = EINTR; 5540 break; 5541 } 5542 pool = zone_pool_get(zone); 5543 poolid = pool->pool_id; 5544 pool_unlock(); 5545 size = sizeof (poolid); 5546 if (bufsize > size) 5547 bufsize = size; 5548 if (buf != NULL && copyout(&poolid, buf, size) != 0) 5549 error = EFAULT; 5550 } 5551 break; 5552 case ZONE_ATTR_SLBL: 5553 size = sizeof (bslabel_t); 5554 if (bufsize > size) 5555 bufsize = size; 5556 if (zone->zone_slabel == NULL) 5557 error = EINVAL; 5558 else if (buf != NULL && 5559 copyout(label2bslabel(zone->zone_slabel), buf, 5560 bufsize) != 0) 5561 error = EFAULT; 5562 break; 5563 case ZONE_ATTR_INITPID: 5564 size = sizeof (initpid); 5565 if (bufsize > size) 5566 bufsize = size; 5567 initpid = zone->zone_proc_initpid; 5568 if (initpid == -1) { 5569 error = ESRCH; 5570 break; 5571 } 5572 if (buf != NULL && 5573 copyout(&initpid, buf, bufsize) != 0) 5574 error = EFAULT; 5575 break; 5576 case ZONE_ATTR_BRAND: 5577 size = strlen(zone->zone_brand->b_name) + 1; 5578 5579 if (bufsize > size) 5580 bufsize = size; 5581 if (buf != NULL) { 5582 err = copyoutstr(zone->zone_brand->b_name, buf, 5583 bufsize, NULL); 5584 if (err != 0 && err != ENAMETOOLONG) 5585 error = EFAULT; 5586 } 5587 break; 5588 case ZONE_ATTR_INITNAME: 5589 size = strlen(zone->zone_initname) + 1; 5590 if (bufsize > size) 5591 bufsize = size; 5592 if (buf != NULL) { 5593 err = copyoutstr(zone->zone_initname, buf, bufsize, 5594 NULL); 5595 if (err != 0 && err != ENAMETOOLONG) 5596 error = EFAULT; 5597 } 5598 break; 5599 case ZONE_ATTR_BOOTARGS: 5600 if (zone->zone_bootargs == NULL) 5601 outstr = ""; 5602 else 5603 outstr = zone->zone_bootargs; 5604 size = strlen(outstr) + 1; 5605 if (bufsize > size) 5606 bufsize = size; 5607 if (buf != NULL) { 5608 err = copyoutstr(outstr, buf, bufsize, NULL); 5609 if (err != 0 && err != ENAMETOOLONG) 5610 error = EFAULT; 5611 } 5612 break; 5613 case ZONE_ATTR_PHYS_MCAP: 5614 size = sizeof (zone->zone_phys_mcap); 5615 if (bufsize > size) 5616 bufsize = size; 5617 if (buf != NULL && 5618 copyout(&zone->zone_phys_mcap, buf, bufsize) != 0) 5619 error = EFAULT; 5620 break; 5621 case ZONE_ATTR_SCHED_CLASS: 5622 mutex_enter(&class_lock); 5623 5624 if (zone->zone_defaultcid >= loaded_classes) 5625 outstr = ""; 5626 else 5627 outstr = sclass[zone->zone_defaultcid].cl_name; 5628 size = strlen(outstr) + 1; 5629 if (bufsize > size) 5630 bufsize = size; 5631 if (buf != NULL) { 5632 err = copyoutstr(outstr, buf, bufsize, NULL); 5633 if (err != 0 && err != ENAMETOOLONG) 5634 error = EFAULT; 5635 } 5636 5637 mutex_exit(&class_lock); 5638 break; 5639 case ZONE_ATTR_HOSTID: 5640 if (zone->zone_hostid != HW_INVALID_HOSTID && 5641 bufsize == sizeof (zone->zone_hostid)) { 5642 size = sizeof (zone->zone_hostid); 5643 if (buf != NULL && copyout(&zone->zone_hostid, buf, 5644 bufsize) != 0) 5645 error = EFAULT; 5646 } else { 5647 error = EINVAL; 5648 } 5649 break; 5650 case ZONE_ATTR_FS_ALLOWED: 5651 if (zone->zone_fs_allowed == NULL) 5652 outstr = ""; 5653 else 5654 outstr = zone->zone_fs_allowed; 5655 size = strlen(outstr) + 1; 5656 if (bufsize > size) 5657 bufsize = size; 5658 if (buf != NULL) { 5659 err = copyoutstr(outstr, buf, bufsize, NULL); 5660 if (err != 0 && err != ENAMETOOLONG) 5661 error = EFAULT; 5662 } 5663 break; 5664 case ZONE_ATTR_SECFLAGS: 5665 size = sizeof (zone->zone_secflags); 5666 if (bufsize > size) 5667 bufsize = size; 5668 if ((err = copyout(&zone->zone_secflags, buf, bufsize)) != 0) 5669 error = EFAULT; 5670 break; 5671 case ZONE_ATTR_NETWORK: 5672 bufsize = MIN(bufsize, PIPE_BUF + sizeof (zone_net_data_t)); 5673 size = bufsize; 5674 zbuf = kmem_alloc(bufsize, KM_SLEEP); 5675 if (copyin(buf, zbuf, bufsize) != 0) { 5676 error = EFAULT; 5677 } else { 5678 error = zone_get_network(zoneid, zbuf); 5679 if (error == 0 && copyout(zbuf, buf, bufsize) != 0) 5680 error = EFAULT; 5681 } 5682 kmem_free(zbuf, bufsize); 5683 break; 5684 default: 5685 if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone)) { 5686 size = bufsize; 5687 error = ZBROP(zone)->b_getattr(zone, attr, buf, &size); 5688 } else { 5689 error = EINVAL; 5690 } 5691 } 5692 zone_rele(zone); 5693 5694 if (error) 5695 return (set_errno(error)); 5696 return ((ssize_t)size); 5697 } 5698 5699 /* 5700 * Systemcall entry point for zone_setattr(2). 5701 */ 5702 /*ARGSUSED*/ 5703 static int 5704 zone_setattr(zoneid_t zoneid, int attr, void *buf, size_t bufsize) 5705 { 5706 zone_t *zone; 5707 zone_status_t zone_status; 5708 int err = -1; 5709 zone_net_data_t *zbuf; 5710 5711 if (secpolicy_zone_config(CRED()) != 0) 5712 return (set_errno(EPERM)); 5713 5714 /* 5715 * Only the ZONE_ATTR_PHYS_MCAP attribute can be set on the 5716 * global zone. 5717 */ 5718 if (zoneid == GLOBAL_ZONEID && attr != ZONE_ATTR_PHYS_MCAP) { 5719 return (set_errno(EINVAL)); 5720 } 5721 5722 mutex_enter(&zonehash_lock); 5723 if ((zone = zone_find_all_by_id(zoneid)) == NULL) { 5724 mutex_exit(&zonehash_lock); 5725 return (set_errno(EINVAL)); 5726 } 5727 zone_hold(zone); 5728 mutex_exit(&zonehash_lock); 5729 5730 /* 5731 * At present most attributes can only be set on non-running, 5732 * non-global zones. 5733 */ 5734 zone_status = zone_status_get(zone); 5735 if (attr != ZONE_ATTR_PHYS_MCAP && zone_status > ZONE_IS_READY) { 5736 err = EINVAL; 5737 goto done; 5738 } 5739 5740 switch (attr) { 5741 case ZONE_ATTR_INITNAME: 5742 err = zone_set_initname(zone, (const char *)buf); 5743 break; 5744 case ZONE_ATTR_INITNORESTART: 5745 zone->zone_restart_init = B_FALSE; 5746 err = 0; 5747 break; 5748 case ZONE_ATTR_INITRESTART0: 5749 zone->zone_restart_init_0 = B_TRUE; 5750 err = 0; 5751 break; 5752 case ZONE_ATTR_INITREBOOT: 5753 zone->zone_reboot_on_init_exit = B_TRUE; 5754 err = 0; 5755 break; 5756 case ZONE_ATTR_BOOTARGS: 5757 err = zone_set_bootargs(zone, (const char *)buf); 5758 break; 5759 case ZONE_ATTR_BRAND: 5760 err = zone_set_brand(zone, (const char *)buf); 5761 break; 5762 case ZONE_ATTR_FS_ALLOWED: 5763 err = zone_set_fs_allowed(zone, (const char *)buf); 5764 break; 5765 case ZONE_ATTR_SECFLAGS: 5766 err = zone_set_secflags(zone, (psecflags_t *)buf); 5767 break; 5768 case ZONE_ATTR_PHYS_MCAP: 5769 err = zone_set_phys_mcap(zone, (const uint64_t *)buf); 5770 break; 5771 case ZONE_ATTR_SCHED_CLASS: 5772 err = zone_set_sched_class(zone, (const char *)buf); 5773 break; 5774 case ZONE_ATTR_HOSTID: 5775 if (bufsize == sizeof (zone->zone_hostid)) { 5776 if (copyin(buf, &zone->zone_hostid, bufsize) == 0) 5777 err = 0; 5778 else 5779 err = EFAULT; 5780 } else { 5781 err = EINVAL; 5782 } 5783 break; 5784 case ZONE_ATTR_NETWORK: 5785 if (bufsize > (PIPE_BUF + sizeof (zone_net_data_t))) { 5786 err = EINVAL; 5787 break; 5788 } 5789 zbuf = kmem_alloc(bufsize, KM_SLEEP); 5790 if (copyin(buf, zbuf, bufsize) != 0) { 5791 kmem_free(zbuf, bufsize); 5792 err = EFAULT; 5793 break; 5794 } 5795 err = zone_set_network(zoneid, zbuf); 5796 kmem_free(zbuf, bufsize); 5797 break; 5798 default: 5799 if ((attr >= ZONE_ATTR_BRAND_ATTRS) && ZONE_IS_BRANDED(zone)) 5800 err = ZBROP(zone)->b_setattr(zone, attr, buf, bufsize); 5801 else 5802 err = EINVAL; 5803 } 5804 5805 done: 5806 zone_rele(zone); 5807 ASSERT(err != -1); 5808 return (err != 0 ? set_errno(err) : 0); 5809 } 5810 5811 /* 5812 * Return zero if the process has at least one vnode mapped in to its 5813 * address space which shouldn't be allowed to change zones. 5814 * 5815 * Also return zero if the process has any shared mappings which reserve 5816 * swap. This is because the counting for zone.max-swap does not allow swap 5817 * reservation to be shared between zones. zone swap reservation is counted 5818 * on zone->zone_max_swap. 5819 */ 5820 static int 5821 as_can_change_zones(void) 5822 { 5823 proc_t *pp = curproc; 5824 struct seg *seg; 5825 struct as *as = pp->p_as; 5826 vnode_t *vp; 5827 int allow = 1; 5828 5829 ASSERT(pp->p_as != &kas); 5830 AS_LOCK_ENTER(as, RW_READER); 5831 for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) { 5832 5833 /* 5834 * Cannot enter zone with shared anon memory which 5835 * reserves swap. See comment above. 5836 */ 5837 if (seg_can_change_zones(seg) == B_FALSE) { 5838 allow = 0; 5839 break; 5840 } 5841 /* 5842 * if we can't get a backing vnode for this segment then skip 5843 * it. 5844 */ 5845 vp = NULL; 5846 if (SEGOP_GETVP(seg, seg->s_base, &vp) != 0 || vp == NULL) 5847 continue; 5848 if (!vn_can_change_zones(vp)) { /* bail on first match */ 5849 allow = 0; 5850 break; 5851 } 5852 } 5853 AS_LOCK_EXIT(as); 5854 return (allow); 5855 } 5856 5857 /* 5858 * Count swap reserved by curproc's address space 5859 */ 5860 static size_t 5861 as_swresv(void) 5862 { 5863 proc_t *pp = curproc; 5864 struct seg *seg; 5865 struct as *as = pp->p_as; 5866 size_t swap = 0; 5867 5868 ASSERT(pp->p_as != &kas); 5869 ASSERT(AS_WRITE_HELD(as)); 5870 for (seg = AS_SEGFIRST(as); seg != NULL; seg = AS_SEGNEXT(as, seg)) 5871 swap += seg_swresv(seg); 5872 5873 return (swap); 5874 } 5875 5876 /* 5877 * Systemcall entry point for zone_enter(). 5878 * 5879 * The current process is injected into said zone. In the process 5880 * it will change its project membership, privileges, rootdir/cwd, 5881 * zone-wide rctls, and pool association to match those of the zone. 5882 * 5883 * The first zone_enter() called while the zone is in the ZONE_IS_READY 5884 * state will transition it to ZONE_IS_RUNNING. Processes may only 5885 * enter a zone that is "ready" or "running". 5886 */ 5887 static int 5888 zone_enter(zoneid_t zoneid) 5889 { 5890 zone_t *zone; 5891 vnode_t *vp; 5892 proc_t *pp = curproc; 5893 contract_t *ct; 5894 cont_process_t *ctp; 5895 task_t *tk, *oldtk; 5896 kproject_t *zone_proj0; 5897 cred_t *cr, *newcr; 5898 pool_t *oldpool, *newpool; 5899 sess_t *sp; 5900 uid_t uid; 5901 zone_status_t status; 5902 int err = 0; 5903 rctl_entity_p_t e; 5904 size_t swap; 5905 kthread_id_t t; 5906 5907 if (secpolicy_zone_config(CRED()) != 0) 5908 return (set_errno(EPERM)); 5909 if (zoneid < MIN_USERZONEID || zoneid > MAX_ZONEID) 5910 return (set_errno(EINVAL)); 5911 5912 /* 5913 * Stop all lwps so we don't need to hold a lock to look at 5914 * curproc->p_zone. This needs to happen before we grab any 5915 * locks to avoid deadlock (another lwp in the process could 5916 * be waiting for the held lock). 5917 */ 5918 if (curthread != pp->p_agenttp && !holdlwps(SHOLDFORK)) 5919 return (set_errno(EINTR)); 5920 5921 /* 5922 * Make sure we're not changing zones with files open or mapped in 5923 * to our address space which shouldn't be changing zones. 5924 */ 5925 if (!files_can_change_zones()) { 5926 err = EBADF; 5927 goto out; 5928 } 5929 if (!as_can_change_zones()) { 5930 err = EFAULT; 5931 goto out; 5932 } 5933 5934 mutex_enter(&zonehash_lock); 5935 if (pp->p_zone != global_zone) { 5936 mutex_exit(&zonehash_lock); 5937 err = EINVAL; 5938 goto out; 5939 } 5940 5941 zone = zone_find_all_by_id(zoneid); 5942 if (zone == NULL) { 5943 mutex_exit(&zonehash_lock); 5944 err = EINVAL; 5945 goto out; 5946 } 5947 5948 /* 5949 * To prevent processes in a zone from holding contracts on 5950 * extrazonal resources, and to avoid process contract 5951 * memberships which span zones, contract holders and processes 5952 * which aren't the sole members of their encapsulating process 5953 * contracts are not allowed to zone_enter. 5954 */ 5955 ctp = pp->p_ct_process; 5956 ct = &ctp->conp_contract; 5957 mutex_enter(&ct->ct_lock); 5958 mutex_enter(&pp->p_lock); 5959 if ((avl_numnodes(&pp->p_ct_held) != 0) || (ctp->conp_nmembers != 1)) { 5960 mutex_exit(&pp->p_lock); 5961 mutex_exit(&ct->ct_lock); 5962 mutex_exit(&zonehash_lock); 5963 err = EINVAL; 5964 goto out; 5965 } 5966 5967 /* 5968 * Moreover, we don't allow processes whose encapsulating 5969 * process contracts have inherited extrazonal contracts. 5970 * While it would be easier to eliminate all process contracts 5971 * with inherited contracts, we need to be able to give a 5972 * restarted init (or other zone-penetrating process) its 5973 * predecessor's contracts. 5974 */ 5975 if (ctp->conp_ninherited != 0) { 5976 contract_t *next; 5977 for (next = list_head(&ctp->conp_inherited); next; 5978 next = list_next(&ctp->conp_inherited, next)) { 5979 if (contract_getzuniqid(next) != zone->zone_uniqid) { 5980 mutex_exit(&pp->p_lock); 5981 mutex_exit(&ct->ct_lock); 5982 mutex_exit(&zonehash_lock); 5983 err = EINVAL; 5984 goto out; 5985 } 5986 } 5987 } 5988 5989 mutex_exit(&pp->p_lock); 5990 mutex_exit(&ct->ct_lock); 5991 5992 status = zone_status_get(zone); 5993 if (status < ZONE_IS_READY || status >= ZONE_IS_SHUTTING_DOWN) { 5994 /* 5995 * Can't join 5996 */ 5997 mutex_exit(&zonehash_lock); 5998 err = EINVAL; 5999 goto out; 6000 } 6001 6002 /* 6003 * Make sure new priv set is within the permitted set for caller 6004 */ 6005 if (!priv_issubset(zone->zone_privset, &CR_OPPRIV(CRED()))) { 6006 mutex_exit(&zonehash_lock); 6007 err = EPERM; 6008 goto out; 6009 } 6010 /* 6011 * We want to momentarily drop zonehash_lock while we optimistically 6012 * bind curproc to the pool it should be running in. This is safe 6013 * since the zone can't disappear (we have a hold on it). 6014 */ 6015 zone_hold(zone); 6016 mutex_exit(&zonehash_lock); 6017 6018 /* 6019 * Grab pool_lock to keep the pools configuration from changing 6020 * and to stop ourselves from getting rebound to another pool 6021 * until we join the zone. 6022 */ 6023 if (pool_lock_intr() != 0) { 6024 zone_rele(zone); 6025 err = EINTR; 6026 goto out; 6027 } 6028 ASSERT(secpolicy_pool(CRED()) == 0); 6029 /* 6030 * Bind ourselves to the pool currently associated with the zone. 6031 */ 6032 oldpool = curproc->p_pool; 6033 newpool = zone_pool_get(zone); 6034 if (pool_state == POOL_ENABLED && newpool != oldpool && 6035 (err = pool_do_bind(newpool, P_PID, P_MYID, 6036 POOL_BIND_ALL)) != 0) { 6037 pool_unlock(); 6038 zone_rele(zone); 6039 goto out; 6040 } 6041 6042 /* 6043 * Grab cpu_lock now; we'll need it later when we call 6044 * task_join(). 6045 */ 6046 mutex_enter(&cpu_lock); 6047 mutex_enter(&zonehash_lock); 6048 /* 6049 * Make sure the zone hasn't moved on since we dropped zonehash_lock. 6050 */ 6051 if (zone_status_get(zone) >= ZONE_IS_SHUTTING_DOWN) { 6052 /* 6053 * Can't join anymore. 6054 */ 6055 mutex_exit(&zonehash_lock); 6056 mutex_exit(&cpu_lock); 6057 if (pool_state == POOL_ENABLED && 6058 newpool != oldpool) 6059 (void) pool_do_bind(oldpool, P_PID, P_MYID, 6060 POOL_BIND_ALL); 6061 pool_unlock(); 6062 zone_rele(zone); 6063 err = EINVAL; 6064 goto out; 6065 } 6066 6067 /* 6068 * a_lock must be held while transfering locked memory and swap 6069 * reservation from the global zone to the non global zone because 6070 * asynchronous faults on the processes' address space can lock 6071 * memory and reserve swap via MCL_FUTURE and MAP_NORESERVE 6072 * segments respectively. 6073 */ 6074 AS_LOCK_ENTER(pp->p_as, RW_WRITER); 6075 swap = as_swresv(); 6076 mutex_enter(&pp->p_lock); 6077 zone_proj0 = zone->zone_zsched->p_task->tk_proj; 6078 /* verify that we do not exceed and task or lwp limits */ 6079 mutex_enter(&zone->zone_nlwps_lock); 6080 /* add new lwps to zone and zone's proj0 */ 6081 zone_proj0->kpj_nlwps += pp->p_lwpcnt; 6082 zone->zone_nlwps += pp->p_lwpcnt; 6083 /* add 1 task to zone's proj0 */ 6084 zone_proj0->kpj_ntasks += 1; 6085 6086 zone_proj0->kpj_nprocs++; 6087 zone->zone_nprocs++; 6088 mutex_exit(&zone->zone_nlwps_lock); 6089 6090 mutex_enter(&zone->zone_mem_lock); 6091 zone->zone_locked_mem += pp->p_locked_mem; 6092 zone_proj0->kpj_data.kpd_locked_mem += pp->p_locked_mem; 6093 zone->zone_max_swap += swap; 6094 mutex_exit(&zone->zone_mem_lock); 6095 6096 mutex_enter(&(zone_proj0->kpj_data.kpd_crypto_lock)); 6097 zone_proj0->kpj_data.kpd_crypto_mem += pp->p_crypto_mem; 6098 mutex_exit(&(zone_proj0->kpj_data.kpd_crypto_lock)); 6099 6100 /* remove lwps and process from proc's old zone and old project */ 6101 mutex_enter(&pp->p_zone->zone_nlwps_lock); 6102 pp->p_zone->zone_nlwps -= pp->p_lwpcnt; 6103 pp->p_task->tk_proj->kpj_nlwps -= pp->p_lwpcnt; 6104 pp->p_task->tk_proj->kpj_nprocs--; 6105 pp->p_zone->zone_nprocs--; 6106 mutex_exit(&pp->p_zone->zone_nlwps_lock); 6107 6108 mutex_enter(&pp->p_zone->zone_mem_lock); 6109 pp->p_zone->zone_locked_mem -= pp->p_locked_mem; 6110 pp->p_task->tk_proj->kpj_data.kpd_locked_mem -= pp->p_locked_mem; 6111 pp->p_zone->zone_max_swap -= swap; 6112 mutex_exit(&pp->p_zone->zone_mem_lock); 6113 6114 mutex_enter(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock)); 6115 pp->p_task->tk_proj->kpj_data.kpd_crypto_mem -= pp->p_crypto_mem; 6116 mutex_exit(&(pp->p_task->tk_proj->kpj_data.kpd_crypto_lock)); 6117 6118 pp->p_flag |= SZONETOP; 6119 pp->p_zone = zone; 6120 mutex_exit(&pp->p_lock); 6121 AS_LOCK_EXIT(pp->p_as); 6122 6123 /* 6124 * Joining the zone cannot fail from now on. 6125 * 6126 * This means that a lot of the following code can be commonized and 6127 * shared with zsched(). 6128 */ 6129 6130 /* 6131 * If the process contract fmri was inherited, we need to 6132 * flag this so that any contract status will not leak 6133 * extra zone information, svc_fmri in this case 6134 */ 6135 if (ctp->conp_svc_ctid != ct->ct_id) { 6136 mutex_enter(&ct->ct_lock); 6137 ctp->conp_svc_zone_enter = ct->ct_id; 6138 mutex_exit(&ct->ct_lock); 6139 } 6140 6141 /* 6142 * Reset the encapsulating process contract's zone. 6143 */ 6144 ASSERT(ct->ct_mzuniqid == GLOBAL_ZONEUNIQID); 6145 contract_setzuniqid(ct, zone->zone_uniqid); 6146 6147 /* 6148 * Create a new task and associate the process with the project keyed 6149 * by (projid,zoneid). 6150 * 6151 * We might as well be in project 0; the global zone's projid doesn't 6152 * make much sense in a zone anyhow. 6153 * 6154 * This also increments zone_ntasks, and returns with p_lock held. 6155 */ 6156 tk = task_create(0, zone); 6157 oldtk = task_join(tk, 0); 6158 mutex_exit(&cpu_lock); 6159 6160 /* 6161 * call RCTLOP_SET functions on this proc 6162 */ 6163 e.rcep_p.zone = zone; 6164 e.rcep_t = RCENTITY_ZONE; 6165 (void) rctl_set_dup(NULL, NULL, pp, &e, zone->zone_rctls, NULL, 6166 RCD_CALLBACK); 6167 mutex_exit(&pp->p_lock); 6168 6169 /* 6170 * We don't need to hold any of zsched's locks here; not only do we know 6171 * the process and zone aren't going away, we know its session isn't 6172 * changing either. 6173 * 6174 * By joining zsched's session here, we mimic the behavior in the 6175 * global zone of init's sid being the pid of sched. We extend this 6176 * to all zlogin-like zone_enter()'ing processes as well. 6177 */ 6178 mutex_enter(&pidlock); 6179 sp = zone->zone_zsched->p_sessp; 6180 sess_hold(zone->zone_zsched); 6181 mutex_enter(&pp->p_lock); 6182 pgexit(pp); 6183 sess_rele(pp->p_sessp, B_TRUE); 6184 pp->p_sessp = sp; 6185 pgjoin(pp, zone->zone_zsched->p_pidp); 6186 6187 /* 6188 * If any threads are scheduled to be placed on zone wait queue they 6189 * should abandon the idea since the wait queue is changing. 6190 * We need to be holding pidlock & p_lock to do this. 6191 */ 6192 if ((t = pp->p_tlist) != NULL) { 6193 do { 6194 thread_lock(t); 6195 /* 6196 * Kick this thread so that it doesn't sit 6197 * on a wrong wait queue. 6198 */ 6199 if (ISWAITING(t)) 6200 setrun_locked(t); 6201 6202 if (t->t_schedflag & TS_ANYWAITQ) 6203 t->t_schedflag &= ~ TS_ANYWAITQ; 6204 6205 thread_unlock(t); 6206 } while ((t = t->t_forw) != pp->p_tlist); 6207 } 6208 6209 /* 6210 * If there is a default scheduling class for the zone and it is not 6211 * the class we are currently in, change all of the threads in the 6212 * process to the new class. We need to be holding pidlock & p_lock 6213 * when we call parmsset so this is a good place to do it. 6214 */ 6215 if (zone->zone_defaultcid > 0 && 6216 zone->zone_defaultcid != curthread->t_cid) { 6217 pcparms_t pcparms; 6218 6219 pcparms.pc_cid = zone->zone_defaultcid; 6220 pcparms.pc_clparms[0] = 0; 6221 6222 /* 6223 * If setting the class fails, we still want to enter the zone. 6224 */ 6225 if ((t = pp->p_tlist) != NULL) { 6226 do { 6227 (void) parmsset(&pcparms, t); 6228 } while ((t = t->t_forw) != pp->p_tlist); 6229 } 6230 } 6231 6232 mutex_exit(&pp->p_lock); 6233 mutex_exit(&pidlock); 6234 6235 mutex_exit(&zonehash_lock); 6236 /* 6237 * We're firmly in the zone; let pools progress. 6238 */ 6239 pool_unlock(); 6240 task_rele(oldtk); 6241 /* 6242 * We don't need to retain a hold on the zone since we already 6243 * incremented zone_ntasks, so the zone isn't going anywhere. 6244 */ 6245 zone_rele(zone); 6246 6247 /* 6248 * Chroot 6249 */ 6250 vp = zone->zone_rootvp; 6251 zone_chdir(vp, &PTOU(pp)->u_cdir, pp); 6252 zone_chdir(vp, &PTOU(pp)->u_rdir, pp); 6253 6254 /* 6255 * Change process security flags. Note that the _effective_ flags 6256 * cannot change 6257 */ 6258 secflags_copy(&pp->p_secflags.psf_lower, 6259 &zone->zone_secflags.psf_lower); 6260 secflags_copy(&pp->p_secflags.psf_upper, 6261 &zone->zone_secflags.psf_upper); 6262 secflags_copy(&pp->p_secflags.psf_inherit, 6263 &zone->zone_secflags.psf_inherit); 6264 6265 /* 6266 * Change process credentials 6267 */ 6268 newcr = cralloc(); 6269 mutex_enter(&pp->p_crlock); 6270 cr = pp->p_cred; 6271 crcopy_to(cr, newcr); 6272 crsetzone(newcr, zone); 6273 pp->p_cred = newcr; 6274 6275 /* 6276 * Restrict all process privilege sets to zone limit 6277 */ 6278 priv_intersect(zone->zone_privset, &CR_PPRIV(newcr)); 6279 priv_intersect(zone->zone_privset, &CR_EPRIV(newcr)); 6280 priv_intersect(zone->zone_privset, &CR_IPRIV(newcr)); 6281 priv_intersect(zone->zone_privset, &CR_LPRIV(newcr)); 6282 mutex_exit(&pp->p_crlock); 6283 crset(pp, newcr); 6284 6285 /* 6286 * Adjust upcount to reflect zone entry. 6287 */ 6288 uid = crgetruid(newcr); 6289 mutex_enter(&pidlock); 6290 upcount_dec(uid, GLOBAL_ZONEID); 6291 upcount_inc(uid, zoneid); 6292 mutex_exit(&pidlock); 6293 6294 /* 6295 * Set up core file path and content. 6296 */ 6297 set_core_defaults(); 6298 6299 out: 6300 /* 6301 * Let the other lwps continue. 6302 */ 6303 mutex_enter(&pp->p_lock); 6304 if (curthread != pp->p_agenttp) 6305 continuelwps(pp); 6306 mutex_exit(&pp->p_lock); 6307 6308 return (err != 0 ? set_errno(err) : 0); 6309 } 6310 6311 /* 6312 * Systemcall entry point for zone_list(2). 6313 * 6314 * Processes running in a (non-global) zone only see themselves. 6315 * On labeled systems, they see all zones whose label they dominate. 6316 */ 6317 static int 6318 zone_list(zoneid_t *zoneidlist, uint_t *numzones) 6319 { 6320 zoneid_t *zoneids; 6321 zone_t *zone, *myzone; 6322 uint_t user_nzones, real_nzones; 6323 uint_t domi_nzones; 6324 int error; 6325 6326 if (copyin(numzones, &user_nzones, sizeof (uint_t)) != 0) 6327 return (set_errno(EFAULT)); 6328 6329 myzone = curproc->p_zone; 6330 ASSERT(zonecount > 0); 6331 if (myzone != global_zone) { 6332 bslabel_t *mybslab; 6333 6334 if (!is_system_labeled()) { 6335 /* just return current zone */ 6336 real_nzones = domi_nzones = 1; 6337 zoneids = kmem_alloc(sizeof (zoneid_t), KM_SLEEP); 6338 zoneids[0] = myzone->zone_id; 6339 } else { 6340 /* return all zones that are dominated */ 6341 mutex_enter(&zonehash_lock); 6342 real_nzones = zonecount; 6343 domi_nzones = 0; 6344 zoneids = kmem_alloc(real_nzones * 6345 sizeof (zoneid_t), KM_SLEEP); 6346 mybslab = label2bslabel(myzone->zone_slabel); 6347 for (zone = list_head(&zone_active); 6348 zone != NULL; 6349 zone = list_next(&zone_active, zone)) { 6350 if (zone->zone_id == GLOBAL_ZONEID) 6351 continue; 6352 if (zone != myzone && 6353 (zone->zone_flags & ZF_IS_SCRATCH)) 6354 continue; 6355 /* 6356 * Note that a label always dominates 6357 * itself, so myzone is always included 6358 * in the list. 6359 */ 6360 if (bldominates(mybslab, 6361 label2bslabel(zone->zone_slabel))) { 6362 zoneids[domi_nzones++] = zone->zone_id; 6363 } 6364 } 6365 mutex_exit(&zonehash_lock); 6366 } 6367 } else { 6368 mutex_enter(&zonehash_lock); 6369 real_nzones = zonecount; 6370 domi_nzones = 0; 6371 zoneids = kmem_alloc(real_nzones * sizeof (zoneid_t), KM_SLEEP); 6372 for (zone = list_head(&zone_active); zone != NULL; 6373 zone = list_next(&zone_active, zone)) 6374 zoneids[domi_nzones++] = zone->zone_id; 6375 6376 ASSERT(domi_nzones == real_nzones); 6377 mutex_exit(&zonehash_lock); 6378 } 6379 6380 /* 6381 * If user has allocated space for fewer entries than we found, then 6382 * return only up to their limit. Either way, tell them exactly how 6383 * many we found. 6384 */ 6385 if (domi_nzones < user_nzones) 6386 user_nzones = domi_nzones; 6387 error = 0; 6388 if (copyout(&domi_nzones, numzones, sizeof (uint_t)) != 0) { 6389 error = EFAULT; 6390 } else if (zoneidlist != NULL && user_nzones != 0) { 6391 if (copyout(zoneids, zoneidlist, 6392 user_nzones * sizeof (zoneid_t)) != 0) 6393 error = EFAULT; 6394 } 6395 6396 kmem_free(zoneids, real_nzones * sizeof (zoneid_t)); 6397 6398 if (error != 0) 6399 return (set_errno(error)); 6400 else 6401 return (0); 6402 } 6403 6404 /* 6405 * Systemcall entry point for zone_lookup(2). 6406 * 6407 * Non-global zones are only able to see themselves and (on labeled systems) 6408 * the zones they dominate. 6409 */ 6410 static zoneid_t 6411 zone_lookup(const char *zone_name) 6412 { 6413 char *kname; 6414 zone_t *zone; 6415 zoneid_t zoneid; 6416 int err; 6417 6418 if (zone_name == NULL) { 6419 /* return caller's zone id */ 6420 return (getzoneid()); 6421 } 6422 6423 kname = kmem_zalloc(ZONENAME_MAX, KM_SLEEP); 6424 if ((err = copyinstr(zone_name, kname, ZONENAME_MAX, NULL)) != 0) { 6425 kmem_free(kname, ZONENAME_MAX); 6426 return (set_errno(err)); 6427 } 6428 6429 mutex_enter(&zonehash_lock); 6430 zone = zone_find_all_by_name(kname); 6431 kmem_free(kname, ZONENAME_MAX); 6432 /* 6433 * In a non-global zone, can only lookup global and own name. 6434 * In Trusted Extensions zone label dominance rules apply. 6435 */ 6436 if (zone == NULL || 6437 zone_status_get(zone) < ZONE_IS_READY || 6438 !zone_list_access(zone)) { 6439 mutex_exit(&zonehash_lock); 6440 return (set_errno(EINVAL)); 6441 } else { 6442 zoneid = zone->zone_id; 6443 mutex_exit(&zonehash_lock); 6444 return (zoneid); 6445 } 6446 } 6447 6448 static int 6449 zone_version(int *version_arg) 6450 { 6451 int version = ZONE_SYSCALL_API_VERSION; 6452 6453 if (copyout(&version, version_arg, sizeof (int)) != 0) 6454 return (set_errno(EFAULT)); 6455 return (0); 6456 } 6457 6458 /* ARGSUSED */ 6459 long 6460 zone(int cmd, void *arg1, void *arg2, void *arg3, void *arg4) 6461 { 6462 zone_def zs; 6463 int err; 6464 6465 switch (cmd) { 6466 case ZONE_CREATE: 6467 if (get_udatamodel() == DATAMODEL_NATIVE) { 6468 if (copyin(arg1, &zs, sizeof (zone_def))) { 6469 return (set_errno(EFAULT)); 6470 } 6471 } else { 6472 #ifdef _SYSCALL32_IMPL 6473 zone_def32 zs32; 6474 6475 if (copyin(arg1, &zs32, sizeof (zone_def32))) { 6476 return (set_errno(EFAULT)); 6477 } 6478 zs.zone_name = 6479 (const char *)(unsigned long)zs32.zone_name; 6480 zs.zone_root = 6481 (const char *)(unsigned long)zs32.zone_root; 6482 zs.zone_privs = 6483 (const struct priv_set *) 6484 (unsigned long)zs32.zone_privs; 6485 zs.zone_privssz = zs32.zone_privssz; 6486 zs.rctlbuf = (caddr_t)(unsigned long)zs32.rctlbuf; 6487 zs.rctlbufsz = zs32.rctlbufsz; 6488 zs.zfsbuf = (caddr_t)(unsigned long)zs32.zfsbuf; 6489 zs.zfsbufsz = zs32.zfsbufsz; 6490 zs.extended_error = 6491 (int *)(unsigned long)zs32.extended_error; 6492 zs.match = zs32.match; 6493 zs.doi = zs32.doi; 6494 zs.label = (const bslabel_t *)(uintptr_t)zs32.label; 6495 zs.flags = zs32.flags; 6496 #else 6497 panic("get_udatamodel() returned bogus result\n"); 6498 #endif 6499 } 6500 6501 return (zone_create(zs.zone_name, zs.zone_root, 6502 zs.zone_privs, zs.zone_privssz, 6503 (caddr_t)zs.rctlbuf, zs.rctlbufsz, 6504 (caddr_t)zs.zfsbuf, zs.zfsbufsz, 6505 zs.extended_error, zs.match, zs.doi, 6506 zs.label, zs.flags)); 6507 case ZONE_BOOT: 6508 return (zone_boot((zoneid_t)(uintptr_t)arg1)); 6509 case ZONE_DESTROY: 6510 return (zone_destroy((zoneid_t)(uintptr_t)arg1)); 6511 case ZONE_GETATTR: 6512 return (zone_getattr((zoneid_t)(uintptr_t)arg1, 6513 (int)(uintptr_t)arg2, arg3, (size_t)arg4)); 6514 case ZONE_SETATTR: 6515 return (zone_setattr((zoneid_t)(uintptr_t)arg1, 6516 (int)(uintptr_t)arg2, arg3, (size_t)arg4)); 6517 case ZONE_ENTER: 6518 return (zone_enter((zoneid_t)(uintptr_t)arg1)); 6519 case ZONE_LIST: 6520 return (zone_list((zoneid_t *)arg1, (uint_t *)arg2)); 6521 case ZONE_SHUTDOWN: 6522 return (zone_shutdown((zoneid_t)(uintptr_t)arg1)); 6523 case ZONE_LOOKUP: 6524 return (zone_lookup((const char *)arg1)); 6525 case ZONE_VERSION: 6526 return (zone_version((int *)arg1)); 6527 case ZONE_ADD_DATALINK: 6528 return (zone_add_datalink((zoneid_t)(uintptr_t)arg1, 6529 (datalink_id_t)(uintptr_t)arg2)); 6530 case ZONE_DEL_DATALINK: 6531 return (zone_remove_datalink((zoneid_t)(uintptr_t)arg1, 6532 (datalink_id_t)(uintptr_t)arg2)); 6533 case ZONE_CHECK_DATALINK: { 6534 zoneid_t zoneid; 6535 boolean_t need_copyout; 6536 6537 if (copyin(arg1, &zoneid, sizeof (zoneid)) != 0) 6538 return (EFAULT); 6539 need_copyout = (zoneid == ALL_ZONES); 6540 err = zone_check_datalink(&zoneid, 6541 (datalink_id_t)(uintptr_t)arg2); 6542 if (err == 0 && need_copyout) { 6543 if (copyout(&zoneid, arg1, sizeof (zoneid)) != 0) 6544 err = EFAULT; 6545 } 6546 return (err == 0 ? 0 : set_errno(err)); 6547 } 6548 case ZONE_LIST_DATALINK: 6549 return (zone_list_datalink((zoneid_t)(uintptr_t)arg1, 6550 (int *)arg2, (datalink_id_t *)(uintptr_t)arg3)); 6551 default: 6552 return (set_errno(EINVAL)); 6553 } 6554 } 6555 6556 struct zarg { 6557 zone_t *zone; 6558 zone_cmd_arg_t arg; 6559 }; 6560 6561 static int 6562 zone_lookup_door(const char *zone_name, door_handle_t *doorp) 6563 { 6564 char *buf; 6565 size_t buflen; 6566 int error; 6567 6568 buflen = sizeof (ZONE_DOOR_PATH) + strlen(zone_name); 6569 buf = kmem_alloc(buflen, KM_SLEEP); 6570 (void) snprintf(buf, buflen, ZONE_DOOR_PATH, zone_name); 6571 error = door_ki_open(buf, doorp); 6572 kmem_free(buf, buflen); 6573 return (error); 6574 } 6575 6576 static void 6577 zone_release_door(door_handle_t *doorp) 6578 { 6579 door_ki_rele(*doorp); 6580 *doorp = NULL; 6581 } 6582 6583 static void 6584 zone_ki_call_zoneadmd(struct zarg *zargp) 6585 { 6586 door_handle_t door = NULL; 6587 door_arg_t darg, save_arg; 6588 char *zone_name; 6589 size_t zone_namelen; 6590 zoneid_t zoneid; 6591 zone_t *zone; 6592 zone_cmd_arg_t arg; 6593 uint64_t uniqid; 6594 size_t size; 6595 int error; 6596 int retry; 6597 6598 zone = zargp->zone; 6599 arg = zargp->arg; 6600 kmem_free(zargp, sizeof (*zargp)); 6601 6602 zone_namelen = strlen(zone->zone_name) + 1; 6603 zone_name = kmem_alloc(zone_namelen, KM_SLEEP); 6604 bcopy(zone->zone_name, zone_name, zone_namelen); 6605 zoneid = zone->zone_id; 6606 uniqid = zone->zone_uniqid; 6607 /* 6608 * zoneadmd may be down, but at least we can empty out the zone. 6609 * We can ignore the return value of zone_empty() since we're called 6610 * from a kernel thread and know we won't be delivered any signals. 6611 */ 6612 ASSERT(curproc == &p0); 6613 (void) zone_empty(zone); 6614 ASSERT(zone_status_get(zone) >= ZONE_IS_EMPTY); 6615 zone_rele(zone); 6616 6617 size = sizeof (arg); 6618 darg.rbuf = (char *)&arg; 6619 darg.data_ptr = (char *)&arg; 6620 darg.rsize = size; 6621 darg.data_size = size; 6622 darg.desc_ptr = NULL; 6623 darg.desc_num = 0; 6624 6625 save_arg = darg; 6626 /* 6627 * Since we're not holding a reference to the zone, any number of 6628 * things can go wrong, including the zone disappearing before we get a 6629 * chance to talk to zoneadmd. 6630 */ 6631 for (retry = 0; /* forever */; retry++) { 6632 if (door == NULL && 6633 (error = zone_lookup_door(zone_name, &door)) != 0) { 6634 goto next; 6635 } 6636 ASSERT(door != NULL); 6637 6638 if ((error = door_ki_upcall_limited(door, &darg, NULL, 6639 SIZE_MAX, 0)) == 0) { 6640 break; 6641 } 6642 switch (error) { 6643 case EINTR: 6644 /* FALLTHROUGH */ 6645 case EAGAIN: /* process may be forking */ 6646 /* 6647 * Back off for a bit 6648 */ 6649 break; 6650 case EBADF: 6651 zone_release_door(&door); 6652 if (zone_lookup_door(zone_name, &door) != 0) { 6653 /* 6654 * zoneadmd may be dead, but it may come back to 6655 * life later. 6656 */ 6657 break; 6658 } 6659 break; 6660 default: 6661 cmn_err(CE_WARN, 6662 "zone_ki_call_zoneadmd: door_ki_upcall error %d\n", 6663 error); 6664 goto out; 6665 } 6666 next: 6667 /* 6668 * If this isn't the same zone_t that we originally had in mind, 6669 * then this is the same as if two kadmin requests come in at 6670 * the same time: the first one wins. This means we lose, so we 6671 * bail. 6672 */ 6673 if ((zone = zone_find_by_id(zoneid)) == NULL) { 6674 /* 6675 * Problem is solved. 6676 */ 6677 break; 6678 } 6679 if (zone->zone_uniqid != uniqid) { 6680 /* 6681 * zoneid recycled 6682 */ 6683 zone_rele(zone); 6684 break; 6685 } 6686 /* 6687 * We could zone_status_timedwait(), but there doesn't seem to 6688 * be much point in doing that (plus, it would mean that 6689 * zone_free() isn't called until this thread exits). 6690 */ 6691 zone_rele(zone); 6692 delay(hz); 6693 darg = save_arg; 6694 } 6695 out: 6696 if (door != NULL) { 6697 zone_release_door(&door); 6698 } 6699 kmem_free(zone_name, zone_namelen); 6700 thread_exit(); 6701 } 6702 6703 /* 6704 * Entry point for uadmin() to tell the zone to go away or reboot. Analog to 6705 * kadmin(). The caller is a process in the zone. 6706 * 6707 * In order to shutdown the zone, we will hand off control to zoneadmd 6708 * (running in the global zone) via a door. We do a half-hearted job at 6709 * killing all processes in the zone, create a kernel thread to contact 6710 * zoneadmd, and make note of the "uniqid" of the zone. The uniqid is 6711 * a form of generation number used to let zoneadmd (as well as 6712 * zone_destroy()) know exactly which zone they're re talking about. 6713 */ 6714 int 6715 zone_kadmin(int cmd, int fcn, const char *mdep, cred_t *credp) 6716 { 6717 struct zarg *zargp; 6718 zone_cmd_t zcmd; 6719 zone_t *zone; 6720 6721 zone = curproc->p_zone; 6722 ASSERT(getzoneid() != GLOBAL_ZONEID); 6723 6724 switch (cmd) { 6725 case A_SHUTDOWN: 6726 switch (fcn) { 6727 case AD_HALT: 6728 case AD_POWEROFF: 6729 zcmd = Z_HALT; 6730 break; 6731 case AD_BOOT: 6732 zcmd = Z_REBOOT; 6733 break; 6734 case AD_IBOOT: 6735 case AD_SBOOT: 6736 case AD_SIBOOT: 6737 case AD_NOSYNC: 6738 return (ENOTSUP); 6739 default: 6740 return (EINVAL); 6741 } 6742 break; 6743 case A_REBOOT: 6744 zcmd = Z_REBOOT; 6745 break; 6746 case A_FTRACE: 6747 case A_REMOUNT: 6748 case A_FREEZE: 6749 case A_DUMP: 6750 case A_CONFIG: 6751 return (ENOTSUP); 6752 default: 6753 ASSERT(cmd != A_SWAPCTL); /* handled by uadmin() */ 6754 return (EINVAL); 6755 } 6756 6757 if (secpolicy_zone_admin(credp, B_FALSE)) 6758 return (EPERM); 6759 mutex_enter(&zone_status_lock); 6760 6761 /* 6762 * zone_status can't be ZONE_IS_EMPTY or higher since curproc 6763 * is in the zone. 6764 */ 6765 ASSERT(zone_status_get(zone) < ZONE_IS_EMPTY); 6766 if (zone_status_get(zone) > ZONE_IS_RUNNING) { 6767 /* 6768 * This zone is already on its way down. 6769 */ 6770 mutex_exit(&zone_status_lock); 6771 return (0); 6772 } 6773 /* 6774 * Prevent future zone_enter()s 6775 */ 6776 zone_status_set(zone, ZONE_IS_SHUTTING_DOWN); 6777 mutex_exit(&zone_status_lock); 6778 6779 /* 6780 * Kill everyone now and call zoneadmd later. 6781 * zone_ki_call_zoneadmd() will do a more thorough job of this 6782 * later. 6783 */ 6784 killall(zone->zone_id); 6785 /* 6786 * Now, create the thread to contact zoneadmd and do the rest of the 6787 * work. This thread can't be created in our zone otherwise 6788 * zone_destroy() would deadlock. 6789 */ 6790 zargp = kmem_zalloc(sizeof (*zargp), KM_SLEEP); 6791 zargp->arg.cmd = zcmd; 6792 zargp->arg.uniqid = zone->zone_uniqid; 6793 zargp->zone = zone; 6794 (void) strcpy(zargp->arg.locale, "C"); 6795 /* mdep was already copied in for us by uadmin */ 6796 if (mdep != NULL) 6797 (void) strlcpy(zargp->arg.bootbuf, mdep, 6798 sizeof (zargp->arg.bootbuf)); 6799 zone_hold(zone); 6800 6801 (void) thread_create(NULL, 0, zone_ki_call_zoneadmd, zargp, 0, &p0, 6802 TS_RUN, minclsyspri); 6803 exit(CLD_EXITED, 0); 6804 6805 return (EINVAL); 6806 } 6807 6808 /* 6809 * Entry point so kadmin(A_SHUTDOWN, ...) can set the global zone's 6810 * status to ZONE_IS_SHUTTING_DOWN. 6811 * 6812 * This function also shuts down all running zones to ensure that they won't 6813 * fork new processes. 6814 */ 6815 void 6816 zone_shutdown_global(void) 6817 { 6818 zone_t *current_zonep; 6819 6820 ASSERT(INGLOBALZONE(curproc)); 6821 mutex_enter(&zonehash_lock); 6822 mutex_enter(&zone_status_lock); 6823 6824 /* Modify the global zone's status first. */ 6825 ASSERT(zone_status_get(global_zone) == ZONE_IS_RUNNING); 6826 zone_status_set(global_zone, ZONE_IS_SHUTTING_DOWN); 6827 6828 /* 6829 * Now change the states of all running zones to ZONE_IS_SHUTTING_DOWN. 6830 * We don't mark all zones with ZONE_IS_SHUTTING_DOWN because doing so 6831 * could cause assertions to fail (e.g., assertions about a zone's 6832 * state during initialization, readying, or booting) or produce races. 6833 * We'll let threads continue to initialize and ready new zones: they'll 6834 * fail to boot the new zones when they see that the global zone is 6835 * shutting down. 6836 */ 6837 for (current_zonep = list_head(&zone_active); current_zonep != NULL; 6838 current_zonep = list_next(&zone_active, current_zonep)) { 6839 if (zone_status_get(current_zonep) == ZONE_IS_RUNNING) 6840 zone_status_set(current_zonep, ZONE_IS_SHUTTING_DOWN); 6841 } 6842 mutex_exit(&zone_status_lock); 6843 mutex_exit(&zonehash_lock); 6844 } 6845 6846 /* 6847 * Returns true if the named dataset is visible in the current zone. 6848 * The 'write' parameter is set to 1 if the dataset is also writable. 6849 */ 6850 int 6851 zone_dataset_visible(const char *dataset, int *write) 6852 { 6853 static int zfstype = -1; 6854 zone_dataset_t *zd; 6855 size_t len; 6856 zone_t *zone = curproc->p_zone; 6857 const char *name = NULL; 6858 vfs_t *vfsp = NULL; 6859 6860 if (dataset[0] == '\0') 6861 return (0); 6862 6863 /* 6864 * Walk the list once, looking for datasets which match exactly, or 6865 * specify a dataset underneath an exported dataset. If found, return 6866 * true and note that it is writable. 6867 */ 6868 for (zd = list_head(&zone->zone_datasets); zd != NULL; 6869 zd = list_next(&zone->zone_datasets, zd)) { 6870 6871 len = strlen(zd->zd_dataset); 6872 if (strlen(dataset) >= len && 6873 bcmp(dataset, zd->zd_dataset, len) == 0 && 6874 (dataset[len] == '\0' || dataset[len] == '/' || 6875 dataset[len] == '@')) { 6876 if (write) 6877 *write = 1; 6878 return (1); 6879 } 6880 } 6881 6882 /* 6883 * Walk the list a second time, searching for datasets which are parents 6884 * of exported datasets. These should be visible, but read-only. 6885 * 6886 * Note that we also have to support forms such as 'pool/dataset/', with 6887 * a trailing slash. 6888 */ 6889 for (zd = list_head(&zone->zone_datasets); zd != NULL; 6890 zd = list_next(&zone->zone_datasets, zd)) { 6891 6892 len = strlen(dataset); 6893 if (dataset[len - 1] == '/') 6894 len--; /* Ignore trailing slash */ 6895 if (len < strlen(zd->zd_dataset) && 6896 bcmp(dataset, zd->zd_dataset, len) == 0 && 6897 zd->zd_dataset[len] == '/') { 6898 if (write) 6899 *write = 0; 6900 return (1); 6901 } 6902 } 6903 6904 /* 6905 * We reach here if the given dataset is not found in the zone_dataset 6906 * list. Check if this dataset was added as a filesystem (ie. "add fs") 6907 * instead of delegation. For this we search for the dataset in the 6908 * zone_vfslist of this zone. If found, return true and note that it is 6909 * not writable. 6910 */ 6911 6912 /* 6913 * Initialize zfstype if it is not initialized yet. 6914 */ 6915 if (zfstype == -1) { 6916 struct vfssw *vswp = vfs_getvfssw("zfs"); 6917 zfstype = vswp - vfssw; 6918 vfs_unrefvfssw(vswp); 6919 } 6920 6921 vfs_list_read_lock(); 6922 vfsp = zone->zone_vfslist; 6923 do { 6924 ASSERT(vfsp); 6925 if (vfsp->vfs_fstype == zfstype) { 6926 name = refstr_value(vfsp->vfs_resource); 6927 6928 /* 6929 * Check if we have an exact match. 6930 */ 6931 if (strcmp(dataset, name) == 0) { 6932 vfs_list_unlock(); 6933 if (write) 6934 *write = 0; 6935 return (1); 6936 } 6937 /* 6938 * We need to check if we are looking for parents of 6939 * a dataset. These should be visible, but read-only. 6940 */ 6941 len = strlen(dataset); 6942 if (dataset[len - 1] == '/') 6943 len--; 6944 6945 if (len < strlen(name) && 6946 bcmp(dataset, name, len) == 0 && name[len] == '/') { 6947 vfs_list_unlock(); 6948 if (write) 6949 *write = 0; 6950 return (1); 6951 } 6952 } 6953 vfsp = vfsp->vfs_zone_next; 6954 } while (vfsp != zone->zone_vfslist); 6955 6956 vfs_list_unlock(); 6957 return (0); 6958 } 6959 6960 /* 6961 * zone_find_by_any_path() - 6962 * 6963 * kernel-private routine similar to zone_find_by_path(), but which 6964 * effectively compares against zone paths rather than zonerootpath 6965 * (i.e., the last component of zonerootpaths, which should be "root/", 6966 * are not compared.) This is done in order to accurately identify all 6967 * paths, whether zone-visible or not, including those which are parallel 6968 * to /root/, such as /dev/, /home/, etc... 6969 * 6970 * If the specified path does not fall under any zone path then global 6971 * zone is returned. 6972 * 6973 * The treat_abs parameter indicates whether the path should be treated as 6974 * an absolute path although it does not begin with "/". (This supports 6975 * nfs mount syntax such as host:any/path.) 6976 * 6977 * The caller is responsible for zone_rele of the returned zone. 6978 */ 6979 zone_t * 6980 zone_find_by_any_path(const char *path, boolean_t treat_abs) 6981 { 6982 zone_t *zone; 6983 int path_offset = 0; 6984 6985 if (path == NULL) { 6986 zone_hold(global_zone); 6987 return (global_zone); 6988 } 6989 6990 if (*path != '/') { 6991 ASSERT(treat_abs); 6992 path_offset = 1; 6993 } 6994 6995 mutex_enter(&zonehash_lock); 6996 for (zone = list_head(&zone_active); zone != NULL; 6997 zone = list_next(&zone_active, zone)) { 6998 char *c; 6999 size_t pathlen; 7000 char *rootpath_start; 7001 7002 if (zone == global_zone) /* skip global zone */ 7003 continue; 7004 7005 /* scan backwards to find start of last component */ 7006 c = zone->zone_rootpath + zone->zone_rootpathlen - 2; 7007 do { 7008 c--; 7009 } while (*c != '/'); 7010 7011 pathlen = c - zone->zone_rootpath + 1 - path_offset; 7012 rootpath_start = (zone->zone_rootpath + path_offset); 7013 if (strncmp(path, rootpath_start, pathlen) == 0) 7014 break; 7015 } 7016 if (zone == NULL) 7017 zone = global_zone; 7018 zone_hold(zone); 7019 mutex_exit(&zonehash_lock); 7020 return (zone); 7021 } 7022 7023 /* 7024 * Finds a zone_dl_t with the given linkid in the given zone. Returns the 7025 * zone_dl_t pointer if found, and NULL otherwise. 7026 */ 7027 static zone_dl_t * 7028 zone_find_dl(zone_t *zone, datalink_id_t linkid) 7029 { 7030 zone_dl_t *zdl; 7031 7032 ASSERT(mutex_owned(&zone->zone_lock)); 7033 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL; 7034 zdl = list_next(&zone->zone_dl_list, zdl)) { 7035 if (zdl->zdl_id == linkid) 7036 break; 7037 } 7038 return (zdl); 7039 } 7040 7041 static boolean_t 7042 zone_dl_exists(zone_t *zone, datalink_id_t linkid) 7043 { 7044 boolean_t exists; 7045 7046 mutex_enter(&zone->zone_lock); 7047 exists = (zone_find_dl(zone, linkid) != NULL); 7048 mutex_exit(&zone->zone_lock); 7049 return (exists); 7050 } 7051 7052 /* 7053 * Add an data link name for the zone. 7054 */ 7055 static int 7056 zone_add_datalink(zoneid_t zoneid, datalink_id_t linkid) 7057 { 7058 zone_dl_t *zdl; 7059 zone_t *zone; 7060 zone_t *thiszone; 7061 7062 if ((thiszone = zone_find_by_id(zoneid)) == NULL) 7063 return (set_errno(ENXIO)); 7064 7065 /* Verify that the datalink ID doesn't already belong to a zone. */ 7066 mutex_enter(&zonehash_lock); 7067 for (zone = list_head(&zone_active); zone != NULL; 7068 zone = list_next(&zone_active, zone)) { 7069 if (zone_dl_exists(zone, linkid)) { 7070 mutex_exit(&zonehash_lock); 7071 zone_rele(thiszone); 7072 return (set_errno((zone == thiszone) ? EEXIST : EPERM)); 7073 } 7074 } 7075 7076 zdl = kmem_zalloc(sizeof (*zdl), KM_SLEEP); 7077 zdl->zdl_id = linkid; 7078 zdl->zdl_net = NULL; 7079 mutex_enter(&thiszone->zone_lock); 7080 list_insert_head(&thiszone->zone_dl_list, zdl); 7081 mutex_exit(&thiszone->zone_lock); 7082 mutex_exit(&zonehash_lock); 7083 zone_rele(thiszone); 7084 return (0); 7085 } 7086 7087 static int 7088 zone_remove_datalink(zoneid_t zoneid, datalink_id_t linkid) 7089 { 7090 zone_dl_t *zdl; 7091 zone_t *zone; 7092 int err = 0; 7093 7094 if ((zone = zone_find_by_id(zoneid)) == NULL) 7095 return (set_errno(EINVAL)); 7096 7097 mutex_enter(&zone->zone_lock); 7098 if ((zdl = zone_find_dl(zone, linkid)) == NULL) { 7099 err = ENXIO; 7100 } else { 7101 list_remove(&zone->zone_dl_list, zdl); 7102 nvlist_free(zdl->zdl_net); 7103 kmem_free(zdl, sizeof (zone_dl_t)); 7104 } 7105 mutex_exit(&zone->zone_lock); 7106 zone_rele(zone); 7107 return (err == 0 ? 0 : set_errno(err)); 7108 } 7109 7110 /* 7111 * Using the zoneidp as ALL_ZONES, we can lookup which zone has been assigned 7112 * the linkid. Otherwise we just check if the specified zoneidp has been 7113 * assigned the supplied linkid. 7114 */ 7115 int 7116 zone_check_datalink(zoneid_t *zoneidp, datalink_id_t linkid) 7117 { 7118 zone_t *zone; 7119 int err = ENXIO; 7120 7121 if (*zoneidp != ALL_ZONES) { 7122 if ((zone = zone_find_by_id(*zoneidp)) != NULL) { 7123 if (zone_dl_exists(zone, linkid)) 7124 err = 0; 7125 zone_rele(zone); 7126 } 7127 return (err); 7128 } 7129 7130 mutex_enter(&zonehash_lock); 7131 for (zone = list_head(&zone_active); zone != NULL; 7132 zone = list_next(&zone_active, zone)) { 7133 if (zone_dl_exists(zone, linkid)) { 7134 *zoneidp = zone->zone_id; 7135 err = 0; 7136 break; 7137 } 7138 } 7139 mutex_exit(&zonehash_lock); 7140 return (err); 7141 } 7142 7143 /* 7144 * Get the list of datalink IDs assigned to a zone. 7145 * 7146 * On input, *nump is the number of datalink IDs that can fit in the supplied 7147 * idarray. Upon return, *nump is either set to the number of datalink IDs 7148 * that were placed in the array if the array was large enough, or to the 7149 * number of datalink IDs that the function needs to place in the array if the 7150 * array is too small. 7151 */ 7152 static int 7153 zone_list_datalink(zoneid_t zoneid, int *nump, datalink_id_t *idarray) 7154 { 7155 uint_t num, dlcount; 7156 zone_t *zone; 7157 zone_dl_t *zdl; 7158 datalink_id_t *idptr = idarray; 7159 7160 if (copyin(nump, &dlcount, sizeof (dlcount)) != 0) 7161 return (set_errno(EFAULT)); 7162 if ((zone = zone_find_by_id(zoneid)) == NULL) 7163 return (set_errno(ENXIO)); 7164 7165 num = 0; 7166 mutex_enter(&zone->zone_lock); 7167 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL; 7168 zdl = list_next(&zone->zone_dl_list, zdl)) { 7169 /* 7170 * If the list is bigger than what the caller supplied, just 7171 * count, don't do copyout. 7172 */ 7173 if (++num > dlcount) 7174 continue; 7175 if (copyout(&zdl->zdl_id, idptr, sizeof (*idptr)) != 0) { 7176 mutex_exit(&zone->zone_lock); 7177 zone_rele(zone); 7178 return (set_errno(EFAULT)); 7179 } 7180 idptr++; 7181 } 7182 mutex_exit(&zone->zone_lock); 7183 zone_rele(zone); 7184 7185 /* Increased or decreased, caller should be notified. */ 7186 if (num != dlcount) { 7187 if (copyout(&num, nump, sizeof (num)) != 0) 7188 return (set_errno(EFAULT)); 7189 } 7190 return (0); 7191 } 7192 7193 /* 7194 * Public interface for looking up a zone by zoneid. It's a customized version 7195 * for netstack_zone_create(). It can only be called from the zsd create 7196 * callbacks, since it doesn't have reference on the zone structure hence if 7197 * it is called elsewhere the zone could disappear after the zonehash_lock 7198 * is dropped. 7199 * 7200 * Furthermore it 7201 * 1. Doesn't check the status of the zone. 7202 * 2. It will be called even before zone_init is called, in that case the 7203 * address of zone0 is returned directly, and netstack_zone_create() 7204 * will only assign a value to zone0.zone_netstack, won't break anything. 7205 * 3. Returns without the zone being held. 7206 */ 7207 zone_t * 7208 zone_find_by_id_nolock(zoneid_t zoneid) 7209 { 7210 zone_t *zone; 7211 7212 mutex_enter(&zonehash_lock); 7213 if (zonehashbyid == NULL) 7214 zone = &zone0; 7215 else 7216 zone = zone_find_all_by_id(zoneid); 7217 mutex_exit(&zonehash_lock); 7218 return (zone); 7219 } 7220 7221 /* 7222 * Walk the datalinks for a given zone 7223 */ 7224 int 7225 zone_datalink_walk(zoneid_t zoneid, int (*cb)(datalink_id_t, void *), 7226 void *data) 7227 { 7228 zone_t *zone; 7229 zone_dl_t *zdl; 7230 datalink_id_t *idarray; 7231 uint_t idcount = 0; 7232 int i, ret = 0; 7233 7234 if ((zone = zone_find_by_id(zoneid)) == NULL) 7235 return (ENOENT); 7236 7237 /* 7238 * We first build an array of linkid's so that we can walk these and 7239 * execute the callback with the zone_lock dropped. 7240 */ 7241 mutex_enter(&zone->zone_lock); 7242 for (zdl = list_head(&zone->zone_dl_list); zdl != NULL; 7243 zdl = list_next(&zone->zone_dl_list, zdl)) { 7244 idcount++; 7245 } 7246 7247 if (idcount == 0) { 7248 mutex_exit(&zone->zone_lock); 7249 zone_rele(zone); 7250 return (0); 7251 } 7252 7253 idarray = kmem_alloc(sizeof (datalink_id_t) * idcount, KM_NOSLEEP); 7254 if (idarray == NULL) { 7255 mutex_exit(&zone->zone_lock); 7256 zone_rele(zone); 7257 return (ENOMEM); 7258 } 7259 7260 for (i = 0, zdl = list_head(&zone->zone_dl_list); zdl != NULL; 7261 i++, zdl = list_next(&zone->zone_dl_list, zdl)) { 7262 idarray[i] = zdl->zdl_id; 7263 } 7264 7265 mutex_exit(&zone->zone_lock); 7266 7267 for (i = 0; i < idcount && ret == 0; i++) { 7268 if ((ret = (*cb)(idarray[i], data)) != 0) 7269 break; 7270 } 7271 7272 zone_rele(zone); 7273 kmem_free(idarray, sizeof (datalink_id_t) * idcount); 7274 return (ret); 7275 } 7276 7277 static char * 7278 zone_net_type2name(int type) 7279 { 7280 switch (type) { 7281 case ZONE_NETWORK_ADDRESS: 7282 return (ZONE_NET_ADDRNAME); 7283 case ZONE_NETWORK_DEFROUTER: 7284 return (ZONE_NET_RTRNAME); 7285 default: 7286 return (NULL); 7287 } 7288 } 7289 7290 static int 7291 zone_set_network(zoneid_t zoneid, zone_net_data_t *znbuf) 7292 { 7293 zone_t *zone; 7294 zone_dl_t *zdl; 7295 nvlist_t *nvl; 7296 int err = 0; 7297 uint8_t *new = NULL; 7298 char *nvname; 7299 int bufsize; 7300 datalink_id_t linkid = znbuf->zn_linkid; 7301 7302 if (secpolicy_zone_config(CRED()) != 0) 7303 return (set_errno(EPERM)); 7304 7305 if (zoneid == GLOBAL_ZONEID) 7306 return (set_errno(EINVAL)); 7307 7308 nvname = zone_net_type2name(znbuf->zn_type); 7309 bufsize = znbuf->zn_len; 7310 new = znbuf->zn_val; 7311 if (nvname == NULL) 7312 return (set_errno(EINVAL)); 7313 7314 if ((zone = zone_find_by_id(zoneid)) == NULL) { 7315 return (set_errno(EINVAL)); 7316 } 7317 7318 mutex_enter(&zone->zone_lock); 7319 if ((zdl = zone_find_dl(zone, linkid)) == NULL) { 7320 err = ENXIO; 7321 goto done; 7322 } 7323 if ((nvl = zdl->zdl_net) == NULL) { 7324 if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP)) { 7325 err = ENOMEM; 7326 goto done; 7327 } else { 7328 zdl->zdl_net = nvl; 7329 } 7330 } 7331 if (nvlist_exists(nvl, nvname)) { 7332 err = EINVAL; 7333 goto done; 7334 } 7335 err = nvlist_add_uint8_array(nvl, nvname, new, bufsize); 7336 ASSERT(err == 0); 7337 done: 7338 mutex_exit(&zone->zone_lock); 7339 zone_rele(zone); 7340 if (err != 0) 7341 return (set_errno(err)); 7342 else 7343 return (0); 7344 } 7345 7346 static int 7347 zone_get_network(zoneid_t zoneid, zone_net_data_t *znbuf) 7348 { 7349 zone_t *zone; 7350 zone_dl_t *zdl; 7351 nvlist_t *nvl; 7352 uint8_t *ptr; 7353 uint_t psize; 7354 int err = 0; 7355 char *nvname; 7356 int bufsize; 7357 void *buf; 7358 datalink_id_t linkid = znbuf->zn_linkid; 7359 7360 if (zoneid == GLOBAL_ZONEID) 7361 return (set_errno(EINVAL)); 7362 7363 nvname = zone_net_type2name(znbuf->zn_type); 7364 bufsize = znbuf->zn_len; 7365 buf = znbuf->zn_val; 7366 7367 if (nvname == NULL) 7368 return (set_errno(EINVAL)); 7369 if ((zone = zone_find_by_id(zoneid)) == NULL) 7370 return (set_errno(EINVAL)); 7371 7372 mutex_enter(&zone->zone_lock); 7373 if ((zdl = zone_find_dl(zone, linkid)) == NULL) { 7374 err = ENXIO; 7375 goto done; 7376 } 7377 if ((nvl = zdl->zdl_net) == NULL || !nvlist_exists(nvl, nvname)) { 7378 err = ENOENT; 7379 goto done; 7380 } 7381 err = nvlist_lookup_uint8_array(nvl, nvname, &ptr, &psize); 7382 ASSERT(err == 0); 7383 7384 if (psize > bufsize) { 7385 err = ENOBUFS; 7386 goto done; 7387 } 7388 znbuf->zn_len = psize; 7389 bcopy(ptr, buf, psize); 7390 done: 7391 mutex_exit(&zone->zone_lock); 7392 zone_rele(zone); 7393 if (err != 0) 7394 return (set_errno(err)); 7395 else 7396 return (0); 7397 } 7398