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 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #ifndef _SYS_ZONE_H 27 #define _SYS_ZONE_H 28 29 #pragma ident "%Z%%M% %I% %E% SMI" 30 31 #include <sys/types.h> 32 #include <sys/mutex.h> 33 #include <sys/param.h> 34 #include <sys/rctl.h> 35 #include <sys/pset.h> 36 #include <sys/tsol/label.h> 37 38 #ifdef __cplusplus 39 extern "C" { 40 #endif 41 42 /* 43 * NOTE 44 * 45 * The contents of this file are private to the implementation of 46 * Solaris and are subject to change at any time without notice. 47 * Applications and drivers using these interfaces may fail to 48 * run on future releases. 49 */ 50 51 /* Available both in kernel and for user space */ 52 53 /* zone id restrictions and special ids */ 54 #define MAX_ZONEID 9999 55 #define MIN_USERZONEID 1 /* lowest user-creatable zone ID */ 56 #define MIN_ZONEID 0 /* minimum zone ID on system */ 57 #define GLOBAL_ZONEID 0 58 #define ZONEID_WIDTH 4 /* for printf */ 59 60 /* 61 * Special zoneid_t token to refer to all zones. 62 */ 63 #define ALL_ZONES (-1) 64 65 /* system call subcodes */ 66 #define ZONE_CREATE 0 67 #define ZONE_DESTROY 1 68 #define ZONE_GETATTR 2 69 #define ZONE_ENTER 3 70 #define ZONE_LIST 4 71 #define ZONE_SHUTDOWN 5 72 #define ZONE_LOOKUP 6 73 #define ZONE_BOOT 7 74 #define ZONE_VERSION 8 75 76 /* zone attributes */ 77 #define ZONE_ATTR_ROOT 1 78 #define ZONE_ATTR_NAME 2 79 #define ZONE_ATTR_STATUS 3 80 #define ZONE_ATTR_PRIVSET 4 81 #define ZONE_ATTR_UNIQID 5 82 #define ZONE_ATTR_POOLID 6 83 #define ZONE_ATTR_INITPID 7 84 #define ZONE_ATTR_SLBL 8 85 86 #define ZONE_EVENT_CHANNEL "com.sun:zones:status" 87 #define ZONE_EVENT_STATUS_CLASS "status" 88 #define ZONE_EVENT_STATUS_SUBCLASS "change" 89 90 #define ZONE_EVENT_UNINITIALIZED "uninitialized" 91 #define ZONE_EVENT_READY "ready" 92 #define ZONE_EVENT_RUNNING "running" 93 #define ZONE_EVENT_SHUTTING_DOWN "shutting_down" 94 95 #define ZONE_CB_NAME "zonename" 96 #define ZONE_CB_NEWSTATE "newstate" 97 #define ZONE_CB_OLDSTATE "oldstate" 98 #define ZONE_CB_TIMESTAMP "when" 99 #define ZONE_CB_ZONEID "zoneid" 100 101 #ifdef _SYSCALL32 102 typedef struct { 103 caddr32_t zone_name; 104 caddr32_t zone_root; 105 caddr32_t zone_privs; 106 size32_t zone_privssz; 107 caddr32_t rctlbuf; 108 size32_t rctlbufsz; 109 caddr32_t extended_error; 110 caddr32_t zfsbuf; 111 size32_t zfsbufsz; 112 int match; /* match level */ 113 uint32_t doi; /* DOI for label */ 114 caddr32_t label; /* label associated with zone */ 115 } zone_def32; 116 #endif 117 typedef struct { 118 const char *zone_name; 119 const char *zone_root; 120 const struct priv_set *zone_privs; 121 size_t zone_privssz; 122 const char *rctlbuf; 123 size_t rctlbufsz; 124 int *extended_error; 125 const char *zfsbuf; 126 size_t zfsbufsz; 127 int match; /* match level */ 128 uint32_t doi; /* DOI for label */ 129 const bslabel_t *label; /* label associated with zone */ 130 } zone_def; 131 132 /* extended error information */ 133 #define ZE_UNKNOWN 0 /* No extended error info */ 134 #define ZE_CHROOTED 1 /* tried to zone_create from chroot */ 135 #define ZE_AREMOUNTS 2 /* there are mounts within the zone */ 136 137 /* zone_status */ 138 typedef enum { 139 ZONE_IS_UNINITIALIZED = 0, 140 ZONE_IS_READY, 141 ZONE_IS_BOOTING, 142 ZONE_IS_RUNNING, 143 ZONE_IS_SHUTTING_DOWN, 144 ZONE_IS_EMPTY, 145 ZONE_IS_DOWN, 146 ZONE_IS_DYING, 147 ZONE_IS_DEAD 148 } zone_status_t; 149 #define ZONE_MIN_STATE ZONE_IS_UNINITIALIZED 150 #define ZONE_MAX_STATE ZONE_IS_DEAD 151 152 /* 153 * Valid commands which may be issued by zoneadm to zoneadmd. The kernel also 154 * communicates with zoneadmd, but only uses Z_REBOOT and Z_HALT. 155 */ 156 typedef enum zone_cmd { 157 Z_READY, Z_BOOT, Z_REBOOT, Z_HALT, Z_NOTE_UNINSTALLING, 158 Z_MOUNT, Z_UNMOUNT 159 } zone_cmd_t; 160 161 #define ZONEBOOTARGS_MAX 257 /* uadmin()'s buffer is 257 bytes. */ 162 163 /* 164 * The structure of a request to zoneadmd. 165 */ 166 typedef struct zone_cmd_arg { 167 uint64_t uniqid; /* unique "generation number" */ 168 zone_cmd_t cmd; /* requested action */ 169 char locale[MAXPATHLEN]; /* locale in which to render messages */ 170 char bootbuf[ZONEBOOTARGS_MAX]; /* arguments passed to zone_boot() */ 171 } zone_cmd_arg_t; 172 173 /* 174 * Structure of zoneadmd's response to a request. A NULL return value means 175 * the caller should attempt to restart zoneadmd and retry. 176 */ 177 typedef struct zone_cmd_rval { 178 int rval; /* return value of request */ 179 char errbuf[1]; /* variable-sized buffer containing error messages */ 180 } zone_cmd_rval_t; 181 182 /* 183 * The zone support infrastructure uses the zone name as a component 184 * of unix domain (AF_UNIX) sockets, which are limited to 108 characters 185 * in length, so ZONENAME_MAX is limited by that. 186 */ 187 #define ZONENAME_MAX 64 188 189 #define GLOBAL_ZONENAME "global" 190 191 /* 192 * Extended Regular expression (see regex(5)) which matches all valid zone 193 * names. 194 */ 195 #define ZONENAME_REGEXP "[a-zA-Z0-9][-_.a-zA-Z0-9]{0,62}" 196 197 /* 198 * Where the zones support infrastructure places temporary files. 199 */ 200 #define ZONES_TMPDIR "/var/run/zones" 201 202 /* 203 * The path to the door used by clients to communicate with zoneadmd. 204 */ 205 #define ZONE_DOOR_PATH ZONES_TMPDIR "/%s.zoneadmd_door" 206 207 #ifdef _KERNEL 208 /* 209 * We need to protect the definition of 'list_t' from userland applications and 210 * libraries which may be defining ther own versions. 211 */ 212 #include <sys/list.h> 213 214 #define GLOBAL_ZONEUNIQID 0 /* uniqid of the global zone */ 215 216 /* zone_flags */ 217 #define ZF_DESTROYED 0x1 /* ZSD destructor callbacks run */ 218 #define ZF_HASHED_LABEL 0x2 /* zone has a unique label */ 219 #define ZF_IS_SCRATCH 0x4 /* scratch zone */ 220 221 struct pool; 222 223 /* 224 * Structure to record list of ZFS datasets exported to a zone. 225 */ 226 typedef struct zone_dataset { 227 char *zd_dataset; 228 list_node_t zd_linkage; 229 } zone_dataset_t; 230 231 typedef struct zone { 232 /* 233 * zone_name is never modified once set. 234 */ 235 char *zone_name; /* zone's configuration name */ 236 /* 237 * zone_nodename and zone_domain are never freed once allocated. 238 */ 239 char *zone_nodename; /* utsname.nodename equivalent */ 240 char *zone_domain; /* srpc_domain equivalent */ 241 /* 242 * zone_lock protects the following fields of a zone_t: 243 * zone_ref 244 * zone_cred_ref 245 * zone_ntasks 246 * zone_flags 247 * zone_zsd 248 */ 249 kmutex_t zone_lock; 250 /* 251 * zone_linkage is the zone's linkage into the active or 252 * death-row list. The field is protected by zonehash_lock. 253 */ 254 list_node_t zone_linkage; 255 zoneid_t zone_id; /* ID of zone */ 256 uint_t zone_ref; /* count of zone_hold()s on zone */ 257 uint_t zone_cred_ref; /* count of zone_hold_cred()s on zone */ 258 /* 259 * zone_rootvp and zone_rootpath can never be modified once set. 260 */ 261 struct vnode *zone_rootvp; /* zone's root vnode */ 262 char *zone_rootpath; /* Path to zone's root + '/' */ 263 ushort_t zone_flags; /* misc flags */ 264 zone_status_t zone_status; /* protected by zone_status_lock */ 265 uint_t zone_ntasks; /* number of tasks executing in zone */ 266 kmutex_t zone_nlwps_lock; /* protects zone_nlwps, and *_nlwps */ 267 /* counters in projects and tasks */ 268 /* that are within the zone */ 269 rctl_qty_t zone_nlwps; /* number of lwps in zone */ 270 rctl_qty_t zone_nlwps_ctl; /* protected by zone_rctls->rcs_lock */ 271 272 uint_t zone_rootpathlen; /* strlen(zone_rootpath) + 1 */ 273 uint32_t zone_shares; /* FSS shares allocated to zone */ 274 rctl_set_t *zone_rctls; /* zone-wide (zone.*) rctls */ 275 list_t zone_zsd; /* list of Zone-Specific Data values */ 276 kcondvar_t zone_cv; /* used to signal state changes */ 277 struct proc *zone_zsched; /* Dummy kernel "zsched" process */ 278 pid_t zone_proc_initpid; /* pid of "init" for this zone */ 279 int zone_boot_err; /* for zone_boot() if boot fails */ 280 char *zone_bootargs; /* arguments passed via zone_boot() */ 281 /* 282 * zone_kthreads is protected by zone_status_lock. 283 */ 284 kthread_t *zone_kthreads; /* kernel threads in zone */ 285 struct priv_set *zone_privset; /* limit set for zone */ 286 /* 287 * zone_vfslist is protected by vfs_list_lock(). 288 */ 289 struct vfs *zone_vfslist; /* list of FS's mounted in zone */ 290 uint64_t zone_uniqid; /* unique zone generation number */ 291 struct cred *zone_kcred; /* kcred-like, zone-limited cred */ 292 /* 293 * zone_pool is protected by pool_lock(). 294 */ 295 struct pool *zone_pool; /* pool the zone is bound to */ 296 hrtime_t zone_pool_mod; /* last pool bind modification time */ 297 /* zone_psetid is protected by cpu_lock */ 298 psetid_t zone_psetid; /* pset the zone is bound to */ 299 /* 300 * The following two can be read without holding any locks. They are 301 * updated under cpu_lock. 302 */ 303 int zone_ncpus; /* zone's idea of ncpus */ 304 int zone_ncpus_online; /* zone's idea of ncpus_online */ 305 /* 306 * List of ZFS datasets exported to this zone. 307 */ 308 list_t zone_datasets; /* list of datasets */ 309 310 ts_label_t *zone_slabel; /* zone sensitivity label */ 311 int zone_match; /* require label match for packets */ 312 tsol_mlp_list_t zone_mlps; /* MLPs on zone-private addresses */ 313 } zone_t; 314 315 /* 316 * Special value of zone_psetid to indicate that pools are disabled. 317 */ 318 #define ZONE_PS_INVAL PS_MYID 319 320 extern zone_t zone0; 321 extern zone_t *global_zone; 322 extern uint_t maxzones; 323 extern rctl_hndl_t rc_zone_nlwps; 324 325 extern const char * const zone_initname; 326 327 extern long zone(int, void *, void *, void *, void *); 328 extern void zone_zsd_init(void); 329 extern void zone_init(void); 330 extern void zone_hold(zone_t *); 331 extern void zone_rele(zone_t *); 332 extern void zone_cred_hold(zone_t *); 333 extern void zone_cred_rele(zone_t *); 334 extern void zone_task_hold(zone_t *); 335 extern void zone_task_rele(zone_t *); 336 extern zone_t *zone_find_by_id(zoneid_t); 337 extern zone_t *zone_find_by_label(const ts_label_t *); 338 extern zone_t *zone_find_by_name(char *); 339 extern zone_t *zone_find_by_any_path(const char *, boolean_t); 340 extern zone_t *zone_find_by_path(const char *); 341 extern zoneid_t getzoneid(void); 342 343 /* 344 * Zone-specific data (ZSD) APIs 345 */ 346 /* 347 * The following is what code should be initializing its zone_key_t to if it 348 * calls zone_getspecific() without necessarily knowing that zone_key_create() 349 * has been called on the key. 350 */ 351 #define ZONE_KEY_UNINITIALIZED 0 352 353 typedef uint_t zone_key_t; 354 355 extern void zone_key_create(zone_key_t *, void *(*)(zoneid_t), 356 void (*)(zoneid_t, void *), void (*)(zoneid_t, void *)); 357 extern int zone_key_delete(zone_key_t); 358 extern void *zone_getspecific(zone_key_t, zone_t *); 359 extern int zone_setspecific(zone_key_t, zone_t *, const void *); 360 361 /* 362 * The definition of a zsd_entry is truly private to zone.c and is only 363 * placed here so it can be shared with mdb. 364 */ 365 struct zsd_entry { 366 zone_key_t zsd_key; /* Key used to lookup value */ 367 void *zsd_data; /* Caller-managed value */ 368 /* 369 * Callbacks to be executed when a zone is created, shutdown, and 370 * destroyed, respectively. 371 */ 372 void *(*zsd_create)(zoneid_t); 373 void (*zsd_shutdown)(zoneid_t, void *); 374 void (*zsd_destroy)(zoneid_t, void *); 375 list_node_t zsd_linkage; 376 }; 377 378 /* 379 * Macros to help with zone visibility restrictions. 380 */ 381 382 /* 383 * Is process in the global zone? 384 */ 385 #define INGLOBALZONE(p) \ 386 ((p)->p_zone == global_zone) 387 388 /* 389 * Can process view objects in given zone? 390 */ 391 #define HASZONEACCESS(p, zoneid) \ 392 ((p)->p_zone->zone_id == (zoneid) || INGLOBALZONE(p)) 393 394 /* 395 * Convenience macro to see if a resolved path is visible from within a 396 * given zone. 397 * 398 * The basic idea is that the first (zone_rootpathlen - 1) bytes of the 399 * two strings must be equal. Since the rootpathlen has a trailing '/', 400 * we want to skip everything in the path up to (but not including) the 401 * trailing '/'. 402 */ 403 #define ZONE_PATH_VISIBLE(path, zone) \ 404 (strncmp((path), (zone)->zone_rootpath, \ 405 (zone)->zone_rootpathlen - 1) == 0) 406 407 /* 408 * Convenience macro to go from the global view of a path to that seen 409 * from within said zone. It is the responsibility of the caller to 410 * ensure that the path is a resolved one (ie, no '..'s or '.'s), and is 411 * in fact visible from within the zone. 412 */ 413 #define ZONE_PATH_TRANSLATE(path, zone) \ 414 (ASSERT(ZONE_PATH_VISIBLE(path, zone)), \ 415 (path) + (zone)->zone_rootpathlen - 2) 416 417 /* 418 * Special processes visible in all zones. 419 */ 420 #define ZONE_SPECIALPID(x) ((x) == 0 || (x) == 1) 421 422 /* 423 * Zone-safe version of thread_create() to be used when the caller wants to 424 * create a kernel thread to run within the current zone's context. 425 */ 426 extern kthread_t *zthread_create(caddr_t, size_t, void (*)(), void *, size_t, 427 pri_t); 428 extern void zthread_exit(void); 429 430 /* 431 * Functions for an external observer to register interest in a zone's status 432 * change. Observers will be woken up when the zone status equals the status 433 * argument passed in (in the case of zone_status_timedwait, the function may 434 * also return because of a timeout; zone_status_wait_sig may return early due 435 * to a signal being delivered; zone_status_timedwait_sig may return for any of 436 * the above reasons). 437 * 438 * Otherwise these behave identically to cv_timedwait(), cv_wait(), and 439 * cv_wait_sig() respectively. 440 */ 441 extern clock_t zone_status_timedwait(zone_t *, clock_t, zone_status_t); 442 extern clock_t zone_status_timedwait_sig(zone_t *, clock_t, zone_status_t); 443 extern void zone_status_wait(zone_t *, zone_status_t); 444 extern int zone_status_wait_sig(zone_t *, zone_status_t); 445 446 /* 447 * Get the status of the zone (at the time it was called). The state may 448 * have progressed by the time it is returned. 449 */ 450 extern zone_status_t zone_status_get(zone_t *); 451 452 /* 453 * Get the "kcred" credentials corresponding to the given zone. 454 */ 455 extern struct cred *zone_get_kcred(zoneid_t); 456 457 /* 458 * Get/set the pool the zone is currently bound to. 459 */ 460 extern struct pool *zone_pool_get(zone_t *); 461 extern void zone_pool_set(zone_t *, struct pool *); 462 463 /* 464 * Get/set the pset the zone is currently using. 465 */ 466 extern psetid_t zone_pset_get(zone_t *); 467 extern void zone_pset_set(zone_t *, psetid_t); 468 469 /* 470 * Get the number of cpus/online-cpus visible from the given zone. 471 */ 472 extern int zone_ncpus_get(zone_t *); 473 extern int zone_ncpus_online_get(zone_t *); 474 475 /* 476 * Returns true if the named pool/dataset is visible in the current zone. 477 */ 478 extern int zone_dataset_visible(const char *, int *); 479 480 /* 481 * zone version of uadmin() 482 */ 483 extern int zone_uadmin(int, int, struct cred *); 484 extern void zone_shutdown_global(void); 485 486 extern void mount_in_progress(void); 487 extern void mount_completed(void); 488 489 extern int zone_walk(int (*)(zone_t *, void *), void *); 490 491 #endif /* _KERNEL */ 492 493 #ifdef __cplusplus 494 } 495 #endif 496 497 #endif /* _SYS_ZONE_H */ 498