/*

 * CDDL HEADER START

 *

 * The contents of this file are subject to the terms of the

 * Common Development and Distribution License (the "License").

 * You may not use this file except in compliance with the License.

 *

 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE

 * or http://www.opensolaris.org/os/licensing.

 * See the License for the specific language governing permissions

 * and limitations under the License.

 *

 * When distributing Covered Code, include this CDDL HEADER in each

 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.

 * If applicable, add the following below this CDDL HEADER, with the

 * fields enclosed by brackets "[]" replaced with your own identifying

 * information: Portions Copyright [yyyy] [name of copyright owner]

 *

 * CDDL HEADER END

 */

/*

 * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.

 * Use is subject to license terms.

 */

#ifndef _SYS_ZONE_H

#define	_SYS_ZONE_H

#pragma ident	"%Z%%M%	%I%	%E% SMI"

#include <sys/types.h>

#include <sys/mutex.h>

#include <sys/param.h>

#include <sys/rctl.h>

#include <sys/ipc_rctl.h>

#include <sys/pset.h>

#include <sys/tsol/label.h>

#include <sys/cred.h>

#include <sys/netstack.h>

#include <sys/uadmin.h>

#ifdef	__cplusplus

extern "C" {

#endif

/*

 * NOTE

 *

 * The contents of this file are private to the implementation of

 * Solaris and are subject to change at any time without notice.

 * Applications and drivers using these interfaces may fail to

 * run on future releases.

 */

/* Available both in kernel and for user space */

/* zone id restrictions and special ids */

#define	MAX_ZONEID	9999

#define	MIN_USERZONEID	1	/* lowest user-creatable zone ID */

#define	MIN_ZONEID	0	/* minimum zone ID on system */

#define	GLOBAL_ZONEID	0

#define	ZONEID_WIDTH	4	/* for printf */

/*

 * Special zoneid_t token to refer to all zones.

 */

#define	ALL_ZONES	(-1)

/* system call subcodes */

#define	ZONE_CREATE		0

#define	ZONE_DESTROY		1

#define	ZONE_GETATTR		2

#define	ZONE_ENTER		3

#define	ZONE_LIST		4

#define	ZONE_SHUTDOWN		5

#define	ZONE_LOOKUP		6

#define	ZONE_BOOT		7

#define	ZONE_VERSION		8

#define	ZONE_SETATTR		9

#define	ZONE_ADD_DATALINK	10

#define	ZONE_DEL_DATALINK	11

#define	ZONE_CHECK_DATALINK	12

#define	ZONE_LIST_DATALINK	13

/* zone attributes */

#define	ZONE_ATTR_ROOT		1

#define	ZONE_ATTR_NAME		2

#define	ZONE_ATTR_STATUS	3

#define	ZONE_ATTR_PRIVSET	4

#define	ZONE_ATTR_UNIQID	5

#define	ZONE_ATTR_POOLID	6

#define	ZONE_ATTR_INITPID	7

#define	ZONE_ATTR_SLBL		8

#define	ZONE_ATTR_INITNAME	9

#define	ZONE_ATTR_BOOTARGS	10

#define	ZONE_ATTR_BRAND		11

#define	ZONE_ATTR_PHYS_MCAP	12

#define	ZONE_ATTR_SCHED_CLASS	13

#define	ZONE_ATTR_FLAGS		14

/* Start of the brand-specific attribute namespace */

#define	ZONE_ATTR_BRAND_ATTRS	32768

#define	ZONE_EVENT_CHANNEL	"com.sun:zones:status"

#define	ZONE_EVENT_STATUS_CLASS	"status"

#define	ZONE_EVENT_STATUS_SUBCLASS	"change"

#define	ZONE_EVENT_UNINITIALIZED	"uninitialized"

#define	ZONE_EVENT_READY		"ready"

#define	ZONE_EVENT_RUNNING		"running"

#define	ZONE_EVENT_SHUTTING_DOWN	"shutting_down"

#define	ZONE_CB_NAME		"zonename"

#define	ZONE_CB_NEWSTATE	"newstate"

#define	ZONE_CB_OLDSTATE	"oldstate"

#define	ZONE_CB_TIMESTAMP	"when"

#define	ZONE_CB_ZONEID		"zoneid"

/*

 * Exit values that may be returned by scripts or programs invoked by various

 * zone commands.

 *

 * These are defined as:

 *

 *	ZONE_SUBPROC_OK

 *	===============

 *	The subprocess completed successfully.

 *

 *	ZONE_SUBPROC_USAGE

 *	==================

 *	The subprocess failed with a usage message, or a usage message should

 *	be output in its behalf.

 *

 *	ZONE_SUBPROC_NOTCOMPLETE

 *	========================

 *	The subprocess did not complete, but the actions performed by the

 *	subprocess require no recovery actions by the user.

 *

 *	For example, if the subprocess were called by "zoneadm install," the

 *	installation of the zone did not succeed but the user need not perform

 *	a "zoneadm uninstall" before attempting another install.

 *

 *	ZONE_SUBPROC_FATAL

 *	==================

 *	The subprocess failed in a fatal manner, usually one that will require

 *	some type of recovery action by the user.

 *

 *	For example, if the subprocess were called by "zoneadm install," the

 *	installation of the zone did not succeed and the user will need to

 *	perform a "zoneadm uninstall" before another install attempt is

 *	possible.

 *

 *	The non-success exit values are large to avoid accidental collision

 *	with values used internally by some commands (e.g. "Z_ERR" and

 *	"Z_USAGE" as used by zoneadm.)

 */

#define	ZONE_SUBPROC_OK			0

#define	ZONE_SUBPROC_USAGE		253

#define	ZONE_SUBPROC_NOTCOMPLETE	254

#define	ZONE_SUBPROC_FATAL		255

#ifdef _SYSCALL32

typedef struct {

	caddr32_t zone_name;

	caddr32_t zone_root;

	caddr32_t zone_privs;

	size32_t zone_privssz;

	caddr32_t rctlbuf;

	size32_t rctlbufsz;

	caddr32_t extended_error;

	caddr32_t zfsbuf;

	size32_t  zfsbufsz;

	int match;			/* match level */

	uint32_t doi;			/* DOI for label */

	caddr32_t label;		/* label associated with zone */

	int flags;

} zone_def32;

#endif

typedef struct {

	const char *zone_name;

	const char *zone_root;

	const struct priv_set *zone_privs;

	size_t zone_privssz;

	const char *rctlbuf;

	size_t rctlbufsz;

	int *extended_error;

	const char *zfsbuf;

	size_t zfsbufsz;

	int match;			/* match level */

	uint32_t doi;			/* DOI for label */

	const bslabel_t *label;		/* label associated with zone */

	int flags;

} zone_def;

/* extended error information */

#define	ZE_UNKNOWN	0	/* No extended error info */

#define	ZE_CHROOTED	1	/* tried to zone_create from chroot */

#define	ZE_AREMOUNTS	2	/* there are mounts within the zone */

/* zone_status */

typedef enum {

	ZONE_IS_UNINITIALIZED = 0,

	ZONE_IS_READY,

	ZONE_IS_BOOTING,

	ZONE_IS_RUNNING,

	ZONE_IS_SHUTTING_DOWN,

	ZONE_IS_EMPTY,

	ZONE_IS_DOWN,

	ZONE_IS_DYING,

	ZONE_IS_DEAD

} zone_status_t;

#define	ZONE_MIN_STATE		ZONE_IS_UNINITIALIZED

#define	ZONE_MAX_STATE		ZONE_IS_DEAD

/*

 * Valid commands which may be issued by zoneadm to zoneadmd.  The kernel also

 * communicates with zoneadmd, but only uses Z_REBOOT and Z_HALT.

 */

typedef enum zone_cmd {

	Z_READY, Z_BOOT, Z_FORCEBOOT, Z_REBOOT, Z_HALT, Z_NOTE_UNINSTALLING,

	Z_MOUNT, Z_FORCEMOUNT, Z_UNMOUNT

} zone_cmd_t;

/*

 * The structure of a request to zoneadmd.

 */

typedef struct zone_cmd_arg {

	uint64_t	uniqid;		/* unique "generation number" */

	zone_cmd_t	cmd;		/* requested action */

	uint32_t	_pad;		/* need consistent 32/64 bit alignmt */

	char locale[MAXPATHLEN];	/* locale in which to render messages */

	char bootbuf[BOOTARGS_MAX];	/* arguments passed to zone_boot() */

} zone_cmd_arg_t;

/*

 * Structure of zoneadmd's response to a request.  A NULL return value means

 * the caller should attempt to restart zoneadmd and retry.

 */

typedef struct zone_cmd_rval {

	int rval;			/* return value of request */

	char errbuf[1];	/* variable-sized buffer containing error messages */

} zone_cmd_rval_t;

/*

 * The zone support infrastructure uses the zone name as a component

 * of unix domain (AF_UNIX) sockets, which are limited to 108 characters

 * in length, so ZONENAME_MAX is limited by that.

 */

#define	ZONENAME_MAX		64

#define	GLOBAL_ZONENAME		"global"

/*

 * Extended Regular expression (see regex(5)) which matches all valid zone

 * names.

 */

#define	ZONENAME_REGEXP		"[a-zA-Z0-9][-_.a-zA-Z0-9]{0,62}"

/*

 * Where the zones support infrastructure places temporary files.

 */

#define	ZONES_TMPDIR		"/var/run/zones"

/*

 * The path to the door used by clients to communicate with zoneadmd.

 */

#define	ZONE_DOOR_PATH		ZONES_TMPDIR "/%s.zoneadmd_door"

/* zone_flags */

#define	ZF_DESTROYED		0x1	/* ZSD destructor callbacks run */

#define	ZF_HASHED_LABEL		0x2	/* zone has a unique label */

#define	ZF_IS_SCRATCH		0x4	/* scratch zone */

#define	ZF_NET_EXCL		0x8	/* Zone has an exclusive IP stack */

/* zone_create flags */

#define	ZCF_NET_EXCL		0x1	/* Create a zone with exclusive IP */

#ifdef _KERNEL

/*

 * We need to protect the definition of 'list_t' from userland applications and

 * libraries which may be defining ther own versions.

 */

#include <sys/list.h>

#define	GLOBAL_ZONEUNIQID	0	/* uniqid of the global zone */

struct pool;

struct brand;

struct dlnamelist;

/*

 * Structure to record list of ZFS datasets exported to a zone.

 */

typedef struct zone_dataset {

	char		*zd_dataset;

	list_node_t	zd_linkage;

} zone_dataset_t;

/*

 * structure for zone kstats

 */

typedef struct zone_kstat {

	kstat_named_t zk_zonename;

	kstat_named_t zk_usage;

	kstat_named_t zk_value;

} zone_kstat_t;

typedef struct zone {

	/*

	 * zone_name is never modified once set.

	 */

	char		*zone_name;	/* zone's configuration name */

	/*

	 * zone_nodename and zone_domain are never freed once allocated.

	 */

	char		*zone_nodename;	/* utsname.nodename equivalent */

	char		*zone_domain;	/* srpc_domain equivalent */

	/*

	 * zone_lock protects the following fields of a zone_t:

	 * 	zone_ref

	 * 	zone_cred_ref

	 * 	zone_ntasks

	 * 	zone_flags

	 * 	zone_zsd

	 */

	kmutex_t	zone_lock;

	/*

	 * zone_linkage is the zone's linkage into the active or

	 * death-row list.  The field is protected by zonehash_lock.

	 */

	list_node_t	zone_linkage;

	zoneid_t	zone_id;	/* ID of zone */

	uint_t		zone_ref;	/* count of zone_hold()s on zone */

	uint_t		zone_cred_ref;	/* count of zone_hold_cred()s on zone */

	/*

	 * zone_rootvp and zone_rootpath can never be modified once set.

	 */

	struct vnode	*zone_rootvp;	/* zone's root vnode */

	char		*zone_rootpath;	/* Path to zone's root + '/' */

	ushort_t	zone_flags;	/* misc flags */

	zone_status_t	zone_status;	/* protected by zone_status_lock */

	uint_t		zone_ntasks;	/* number of tasks executing in zone */

	kmutex_t	zone_nlwps_lock; /* protects zone_nlwps, and *_nlwps */

					/* counters in projects and tasks */

					/* that are within the zone */

	rctl_qty_t	zone_nlwps;	/* number of lwps in zone */

	rctl_qty_t	zone_nlwps_ctl; /* protected by zone_rctls->rcs_lock */

	rctl_qty_t	zone_shmmax;	/* System V shared memory usage */

	ipc_rqty_t	zone_ipc;	/* System V IPC id resource usage */

	uint_t		zone_rootpathlen; /* strlen(zone_rootpath) + 1 */

	uint32_t	zone_shares;	/* FSS shares allocated to zone */

	rctl_set_t	*zone_rctls;	/* zone-wide (zone.*) rctls */

	kmutex_t	zone_mem_lock;	/* protects zone_locked_mem and */

					/* kpd_locked_mem for all */

					/* projects in zone. */

					/* Also protects zone_max_swap */

					/* grab after p_lock, before rcs_lock */

	rctl_qty_t	zone_locked_mem;	/* bytes of locked memory in */

						/* zone */

	rctl_qty_t	zone_locked_mem_ctl;	/* Current locked memory */

						/* limit.  Protected by */

						/* zone_rctls->rcs_lock */

	rctl_qty_t	zone_max_swap; /* bytes of swap reserved by zone */

	rctl_qty_t	zone_max_swap_ctl;	/* current swap limit. */

						/* Protected by */

						/* zone_rctls->rcs_lock */

	list_t		zone_zsd;	/* list of Zone-Specific Data values */

	kcondvar_t	zone_cv;	/* used to signal state changes */

	struct proc	*zone_zsched;	/* Dummy kernel "zsched" process */

	pid_t		zone_proc_initpid; /* pid of "init" for this zone */

	char		*zone_initname;	/* fs path to 'init' */

	int		zone_boot_err;  /* for zone_boot() if boot fails */

	char		*zone_bootargs;	/* arguments passed via zone_boot() */

	uint64_t	zone_phys_mcap;	/* physical memory cap */

	/*

	 * zone_kthreads is protected by zone_status_lock.

	 */

	kthread_t	*zone_kthreads;	/* kernel threads in zone */

	struct priv_set	*zone_privset;	/* limit set for zone */

	/*

	 * zone_vfslist is protected by vfs_list_lock().

	 */

	struct vfs	*zone_vfslist;	/* list of FS's mounted in zone */

	uint64_t	zone_uniqid;	/* unique zone generation number */

	struct cred	*zone_kcred;	/* kcred-like, zone-limited cred */

	/*

	 * zone_pool is protected by pool_lock().

	 */

	struct pool	*zone_pool;	/* pool the zone is bound to */

	hrtime_t	zone_pool_mod;	/* last pool bind modification time */

	/* zone_psetid is protected by cpu_lock */

	psetid_t	zone_psetid;	/* pset the zone is bound to */

	/*

	 * The following two can be read without holding any locks.  They are

	 * updated under cpu_lock.

	 */

	int		zone_ncpus;  /* zone's idea of ncpus */

	int		zone_ncpus_online; /* zone's idea of ncpus_online */

	/*

	 * List of ZFS datasets exported to this zone.

	 */

	list_t		zone_datasets;	/* list of datasets */

	ts_label_t	*zone_slabel;	/* zone sensitivity label */

	int		zone_match;	/* require label match for packets */

	tsol_mlp_list_t zone_mlps;	/* MLPs on zone-private addresses */

	boolean_t	zone_restart_init;	/* Restart init if it dies? */

	struct brand	*zone_brand;		/* zone's brand */

	id_t		zone_defaultcid;	/* dflt scheduling class id */

	kstat_t		*zone_swapresv_kstat;

	kstat_t		*zone_lockedmem_kstat;

	/*

	 * zone_dl_list is protected by zone_lock

	 */

	struct dlnamelist *zone_dl_list;

	netstack_t	*zone_netstack;

} zone_t;

/*

 * Special value of zone_psetid to indicate that pools are disabled.

 */

#define	ZONE_PS_INVAL	PS_MYID

extern zone_t zone0;

extern zone_t *global_zone;

extern uint_t maxzones;

extern rctl_hndl_t rc_zone_nlwps;

extern long zone(int, void *, void *, void *, void *);

extern void zone_zsd_init(void);

extern void zone_init(void);

extern void zone_hold(zone_t *);

extern void zone_rele(zone_t *);

extern void zone_cred_hold(zone_t *);

extern void zone_cred_rele(zone_t *);

extern void zone_task_hold(zone_t *);

extern void zone_task_rele(zone_t *);

extern zone_t *zone_find_by_id(zoneid_t);

extern zone_t *zone_find_by_label(const ts_label_t *);

extern zone_t *zone_find_by_name(char *);

extern zone_t *zone_find_by_any_path(const char *, boolean_t);

extern zone_t *zone_find_by_path(const char *);

extern zoneid_t getzoneid(void);

extern zone_t *zone_find_by_id_nolock(zoneid_t);

/*

 * Zone-specific data (ZSD) APIs

 */

/*

 * The following is what code should be initializing its zone_key_t to if it

 * calls zone_getspecific() without necessarily knowing that zone_key_create()

 * has been called on the key.

 */

#define	ZONE_KEY_UNINITIALIZED	0

typedef uint_t zone_key_t;

extern void	zone_key_create(zone_key_t *, void *(*)(zoneid_t),

    void (*)(zoneid_t, void *), void (*)(zoneid_t, void *));

extern int 	zone_key_delete(zone_key_t);

extern void	*zone_getspecific(zone_key_t, zone_t *);

extern int	zone_setspecific(zone_key_t, zone_t *, const void *);

/*

 * The definition of a zsd_entry is truly private to zone.c and is only

 * placed here so it can be shared with mdb.

 */

struct zsd_entry {

	zone_key_t		zsd_key;	/* Key used to lookup value */

	void			*zsd_data;	/* Caller-managed value */

	/*

	 * Callbacks to be executed when a zone is created, shutdown, and

	 * destroyed, respectively.

	 */

	void			*(*zsd_create)(zoneid_t);

	void			(*zsd_shutdown)(zoneid_t, void *);

	void			(*zsd_destroy)(zoneid_t, void *);

	list_node_t		zsd_linkage;

};

/*

 * Macros to help with zone visibility restrictions.

 */

/*

 * Is process in the global zone?

 */

#define	INGLOBALZONE(p) \

	((p)->p_zone == global_zone)

/*

 * Can process view objects in given zone?

 */

#define	HASZONEACCESS(p, zoneid) \

	((p)->p_zone->zone_id == (zoneid) || INGLOBALZONE(p))

/*

 * Convenience macro to see if a resolved path is visible from within a

 * given zone.

 *

 * The basic idea is that the first (zone_rootpathlen - 1) bytes of the

 * two strings must be equal.  Since the rootpathlen has a trailing '/',

 * we want to skip everything in the path up to (but not including) the

 * trailing '/'.