#include <stdbool.h>
#include <sys/types.h>
#include <sched.h>
+#include <netinet/in.h>
#ifndef IS_DOXYGEN
#if defined(__GNUC__)
#define VC_CAP_LEASE 28
#define VC_CAP_AUDIT_WRITE 29
#define VC_CAP_AUDIT_CONTROL 30
+#define VC_CAP_SETFCAP 31
+#define VC_CAP_MAC_OVERRIDE 32
+#define VC_CAP_MAC_ADMIN 33
#define VC_IMMUTABLE_FILE_FL 0x0000010lu
#define VC_IMMUTABLE_LINK_FL 0x0008000lu
#define VC_IATTR_ADMIN 0x00000001u
#define VC_IATTR_WATCH 0x00000002u
#define VC_IATTR_HIDE 0x00000004u
-#define VC_IATTR_FLAGS 0x00000007u
+#define VC_IATTR_WRITE 0x00000008u
+#define VC_IATTR_FLAGS 0x0000000fu
#define VC_IATTR_BARRIER 0x00010000u
#define VC_IATTR_IUNLINK 0x00020000u
#define VC_IATTR_IMMUTABLE 0x00040000u
+#define VC_IATTR_COW 0x00080000u
// the flags
#define VC_VXF_PROLIFIC (1ULL<<49)
#define VC_VXF_IGNEG_NICE (1ULL<<52)
+#define VC_VXF_IGNEG_IONICE (1ULL<<53)
// the ccapabilities
#define VC_VXC_SET_UTSNAME 0x00000001ull
#define VC_VXC_SET_RLIMIT 0x00000002ull
+#define VC_VXC_FS_SECURITY 0x00000004ull
+#define VC_VXC_TIOCSTI 0x00000010ull
#define VC_VXC_RAW_ICMP 0x00000100ull
#define VC_VXC_SYSLOG 0x00001000ull
+#define VC_VXC_OOM_ADJUST 0x00002000ull
+#define VC_VXC_AUDIT_CONTROL 0x00004000ull
#define VC_VXC_SECURE_MOUNT 0x00010000ull
#define VC_VXC_SECURE_REMOUNT 0x00020000ull
#define VC_VXC_ADMIN_MAPPER 0x00200000ull
#define VC_VXC_ADMIN_CLOOP 0x00400000ull
+#define VC_VXC_KTHREAD 0x01000000ull
+#define VC_VXC_NAMESPACE 0x02000000ull
+
// the scheduler flags
#define VC_VXSM_FILL_RATE 0x0001
#define VC_VXSM_IDLE_TIME 0x0200
#define VC_VXSM_FORCE 0x0400
+#define VC_VXSM_MSEC 0x4000
#define VC_VXSM_V3_MASK 0x0173
#define VC_NXF_INFO_PRIVATE 0x00000008ull
#define VC_NXF_SINGLE_IP 0x00000100ull
+#define VC_NXF_LBACK_REMAP 0x00000200ull
+#define VC_NXF_LBACK_ALLOW 0x00000400ull
#define VC_NXF_HIDE_NETIF 0x02000000ull
+#define VC_NXF_HIDE_LBACK 0x04000000ull
#define VC_NXF_STATE_SETUP (1ULL<<32)
#define VC_NXF_STATE_ADMIN (1ULL<<34)
#define VC_NXF_PERSISTENT (1ULL<<38)
+// the network capabilities
+#define VC_NXC_TUN_CREATE 0x00000001ull
+
+#define VC_NXC_RAW_ICMP 0x00000100ull
+
+
// the vserver specific limits
#define VC_VLIMIT_NSOCK 16
#define VC_VLIMIT_OPENFD 17
// the VCI bit values
#define VC_VCI_NO_DYNAMIC (1 << 0)
+#define VC_VCI_PROC_SECURE (1 << 4)
+#define VC_VCI_HARDCPU (1 << 5)
+#define VC_VCI_IDLELIMIT (1 << 6)
+#define VC_VCI_IDLETIME (1 << 7)
+#define VC_VCI_COWBL (1 << 8)
+#define VC_VCI_FULLCOWBL (1 << 9)
#define VC_VCI_SPACES (1 << 10)
+#define VC_VCI_NETV2 (1 << 11)
+#define VC_VCI_MEMCG (1 << 12)
+#define VC_VCI_DEBUG (1 << 16)
+#define VC_VCI_HISTORY (1 << 20)
+#define VC_VCI_TAGGED (1 << 24)
+#define VC_VCI_PPTAG (1 << 28)
// the device mapping flags
#define VC_DATTR_REMAP 0x00000010
+// the process context migration flags
+#define VC_VXM_SET_INIT 0x00000001
+#define VC_VXM_SET_REAPER 0x00000002
+
+
+// the network address flags
+#define VC_NXA_TYPE_IPV4 0x0001
+#define VC_NXA_TYPE_IPV6 0x0002
+
+#define VC_NXA_TYPE_NONE 0x0000
+#define VC_NXA_TYPE_ANY 0x00FF
+
+#define VC_NXA_TYPE_ADDR 0x0010
+#define VC_NXA_TYPE_MASK 0x0020
+#define VC_NXA_TYPE_RANGE 0x0040
+
+#define VC_NXA_MOD_BCAST 0x0100
+#define VC_NXA_MOD_LBACK 0x0200
+
+
#ifndef CLONE_NEWNS
# define CLONE_NEWNS 0x00020000
#endif
#ifndef CLONE_NEWIPC
# define CLONE_NEWIPC 0x08000000
#endif
-
+#ifndef CLONE_NEWUSER
+# define CLONE_NEWUSER 0x10000000
+#endif
+#ifndef CLONE_NEWPID
+# define CLONE_NEWPID 0x20000000
+#endif
+#ifndef CLONE_NEWNET
+# define CLONE_NEWNET 0x40000000
+#endif
#define VC_BAD_PERSONALITY ((uint_least32_t)(-1))
#ifdef IS_DOXYGEN
typedef an_unsigned_integer_type xid_t;
typedef an_unsigned_integer_type nid_t;
+typedef an_unsigned_integer_type tag_t;
#endif
#ifdef __cplusplus
*/
int vc_get_version();
+ typedef uint64_t vc_vci_t;
/** \brief Returns the kernel configuration bits
* \ingroup syscalls
* \returns The kernel configuration bits
*/
- int vc_get_vci();
+ vc_vci_t vc_get_vci();
+
+ int vc_get_kernel();
/** \brief Moves current process into a context
* \ingroup syscalls
*
* \returns the xid of the created context, or VC_NOCTX on errors. \c errno
* will be set appropriately. */
- xid_t vc_ctx_create(xid_t xid);
+ xid_t vc_ctx_create(xid_t xid, struct vc_ctx_flags *flags);
/** \brief Moves the current process into the specified context.
* \ingroup syscalls
*
- * \param xid The new context
+ * \param xid The new context
+ * \param flags The flags, see VC_VXM_*
* \returns 0 on success, -1 on errors */
- int vc_ctx_migrate(xid_t xid);
+ int vc_ctx_migrate(xid_t xid, uint_least64_t flags);
/** \brief Statistics about a context */
struct vc_ctx_stat {
/** \brief Contains further statistics about a context. */
struct vc_virt_stat {
uint_least64_t offset;
- uint_least32_t uptime;
+ uint_least64_t uptime;
uint_least32_t nr_threads;
uint_least32_t nr_running;
uint_least32_t nr_uninterruptible;
nid_t vc_get_task_nid(pid_t pid);
int vc_get_nx_info(nid_t nid, struct vc_nx_info *) VC_ATTR_NONNULL((2));
- typedef enum { vcNET_IPV4=1, vcNET_IPV6=2,
- vcNET_IPV4B=0x101, vcNET_IPV6B=0x102,
- vcNET_ANY=~0 } vc_net_nx_type;
-
- struct vc_net_nx {
- vc_net_nx_type type;
- size_t count;
- uint32_t ip[4];
- uint32_t mask[4];
+ struct vc_net_addr {
+ uint16_t vna_type;
+ uint16_t vna_flags;
+ uint16_t vna_prefix;
+ uint16_t vna_parent;
+ struct {
+ union {
+ struct in_addr v4;
+ struct in6_addr v6;
+ } ip;
+ union {
+ struct in_addr v4;
+ struct in6_addr v6;
+ } ip2;
+ union {
+ struct in_addr v4;
+ struct in6_addr v6;
+ } mask;
+ } s;
+#define vna_v4_ip s.ip.v4
+#define vna_v4_ip2 s.ip2.v4
+#define vna_v4_mask s.mask.v4
+#define vna_v6_ip s.ip.v6
+#define vna_v6_ip2 s.ip2.v6
+#define vna_v6_mask s.mask.v6
};
struct vc_net_flags {
nid_t vc_net_create(nid_t nid);
int vc_net_migrate(nid_t nid);
- int vc_net_add(nid_t nid, struct vc_net_nx const *info);
- int vc_net_remove(nid_t nid, struct vc_net_nx const *info);
+ int vc_net_add(nid_t nid, struct vc_net_addr const *info);
+ int vc_net_remove(nid_t nid, struct vc_net_addr const *info);
int vc_get_nflags(nid_t, struct vc_net_flags *);
int vc_set_nflags(nid_t, struct vc_net_flags const *);
int vc_set_iattr(char const *filename, xid_t xid,
uint_least32_t flags, uint_least32_t mask) VC_ATTR_NONNULL((1));
+ int vc_fset_iattr(int fd, xid_t xid,
+ uint_least32_t flags, uint_least32_t mask);
+
/** \brief Returns information about attributes and assigned context of a file.
* \ingroup syscalls
*
int vc_get_iattr(char const *filename, xid_t * /*@null@*/ xid,
uint_least32_t * /*@null@*/ flags,
uint_least32_t * /*@null@*/ mask) VC_ATTR_NONNULL((1));
+
+ int vc_fget_iattr(int fd, xid_t * /*@null@*/ xid,
+ uint_least32_t * /*@null@*/ flags,
+ uint_least32_t * /*@null@*/ mask) VC_ATTR_NONNULL((4));
/** \brief Returns the context of \c filename
* \ingroup syscalls
char *val, size_t len) VC_ATTR_NONNULL((3));
/* namespace related functions */
- int vc_enter_namespace(xid_t xid, uint_least64_t mask);
- int vc_set_namespace(xid_t xid, uint_least64_t mask);
- int vc_cleanup_namespace();
- uint_least64_t vc_get_space_mask();
+ int vc_enter_namespace(xid_t xid, uint_least64_t mask, uint32_t index);
+ int vc_set_namespace(xid_t xid, uint_least64_t mask, uint32_t index);
+ int vc_cleanup_namespace(void);
+ uint_least64_t vc_get_space_mask(void);
+ uint_least64_t vc_get_space_default(void);
/* disk limit related things */
/** Get a disk limit. */
int vc_get_dlimit(char const *filename, xid_t xid,
uint_least32_t flags,
- struct vc_ctx_dlimit *limits) VC_ATTR_NONNULL((1,4));
+ struct vc_ctx_dlimit *limits) VC_ATTR_NONNULL((1));
- /* misc. syscalls */
+ /** Get the filesystem tag for a process. */
+ tag_t vc_get_task_tag(pid_t pid);
+
+ /** Create a new filesystem tag space. */
+ int vc_tag_create(tag_t tag);
+
+ /** Migrate to an existing filesystem tag space. */
+ int vc_tag_migrate(tag_t tag);
+
+ /* scheduler related syscalls */
struct vc_set_sched {
uint_least32_t set_mask;
int_least32_t fill_rate;
};
int vc_set_sched(xid_t xid, struct vc_set_sched const *) VC_ATTR_NONNULL((2));
+ int vc_get_sched(xid_t xid, struct vc_set_sched *) VC_ATTR_NONNULL((2));
+
+ struct vc_sched_info {
+ int_least32_t cpu_id;
+ int_least32_t bucket_id;
+ uint_least64_t user_msec;
+ uint_least64_t sys_msec;
+ uint_least64_t hold_msec;
+ uint_least32_t token_usec;
+ int_least32_t vavavoom;
+ };
+ int vc_sched_info(xid_t xid, struct vc_sched_info *info) VC_ATTR_NONNULL((2));
+
+ /* misc. syscalls */
int vc_set_mapping(xid_t xid, const char *device, const char *target, uint32_t flags);
+ int vc_unset_mapping(xid_t xid, const char *device, const char *target, uint32_t flags);
+ int vc_get_badness(xid_t xid, int64_t *badness);
+ int vc_set_badness(xid_t xid, int64_t badness);
+
+
+ /** \brief Namespaces allowed to unshare */
+ struct vc_umask {
+ uint_least64_t umask;
+ uint_least64_t mask;
+ };
+
+ int vc_get_umask(xid_t xid, struct vc_umask *umask);
+ int vc_set_umask(xid_t xid, struct vc_umask const *umask);
/** \brief Information about parsing errors
* \ingroup helper
int vc_list2ccap(char const *, size_t len,
struct vc_err_listparser *err,
struct vc_ctx_caps *);
-
+ char const * vc_loumask2text(uint_least64_t *);
+ int vc_list2umask(char const *, size_t len,
+ struct vc_err_listparser *err,
+ struct vc_umask *);
int vc_list2cflag(char const *, size_t len,
struct vc_err_listparser *err,
struct vc_ctx_flags *flags);
uint_least64_t vc_get_insecurebcaps() VC_ATTR_CONST;
inline static uint_least64_t vc_get_insecureccaps() {
- return ~(VC_VXC_SET_UTSNAME|VC_VXC_RAW_ICMP);
+ return ~(VC_VXC_SET_UTSNAME|VC_VXC_RAW_ICMP|VC_VXC_AUDIT_CONTROL);
}
inline static int vc_setfilecontext(char const *filename, xid_t xid) {
vcFEATURE_COMPAT, vcFEATURE_MIGRATE, vcFEATURE_NAMESPACE,
vcFEATURE_SCHED, vcFEATURE_VINFO, vcFEATURE_VHI,
vcFEATURE_VSHELPER0, vcFEATURE_VSHELPER, vcFEATURE_VWAIT,
- vcFEATURE_VNET }
+ vcFEATURE_VNET, vcFEATURE_VSTAT, vcFEATURE_PPTAG,
+ vcFEATURE_PIDSPACE, vcFEATURE_SPACES, vcFEATURE_PERSISTENT,
+ vcFEATURE_PIVOT_ROOT, vcFEATURE_MEMCG, vcFEATURE_DYNAMIC,
+ vcFEATURE_BME,
+ }
vcFeatureSet;
bool vc_isSupported(vcFeatureSet) VC_ATTR_CONST;
xid_t vc_xidopt2xid(char const *, bool honor_static, char const **err_info);
/** Maps a nid given at '--nid' options to a nid_t */
nid_t vc_nidopt2nid(char const *, bool honor_static, char const **err_info);
+ /** Maps a tag given at '--tag' options to a tag_t */
+ tag_t vc_tagopt2tag(char const *, bool honor_static, char const **err_info);
vcCfgStyle vc_getVserverCfgStyle(char const *id);
* allocated and must be freed by the caller. */
char * vc_getVserverVdir(char const *id, vcCfgStyle style, bool physical);
+ typedef enum { vcCTX_XID = 1,
+ vcCTX_NID,
+ vcCTX_TAG,
+ } vcCtxType;
+
/** Returns the ctx of the given vserver. When vserver is not running and
* 'honor_static' is false, VC_NOCTX will be returned. Else, when
* 'honor_static' is true and a static assignment exists, those value will
* When 'is_running' is not null, the status of the vserver will be
* assigned to this variable. */
xid_t vc_getVserverCtx(char const *id, vcCfgStyle style,
- bool honor_static, bool /*@null@*/ *is_running);
+ bool honor_static, bool /*@null@*/ *is_running,
+ vcCtxType type);
/** Resolves the cfg-path of the vserver owning the given ctx. 'revdir' will
be used as the directory holding the mapping-links; when NULL, the
int vc_compareVserverById(char const *lhs, vcCfgStyle lhs_style,
char const *rhs, vcCfgStyle rhs_style);
+
+ void vc_exitLikeProcess(int pid, int ret) VC_ATTR_NORETURN;
#define vcSKEL_INTERFACES 1u
#define vcSKEL_PKGMGMT 2u