3 * Copyright (C) 2003 Enrico Scholz <enrico.scholz@informatik.tu-chemnitz.de>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2, or (at your option)
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
21 * \brief The public interface of the the libvserver library.
24 #ifndef H_VSERVER_SYSCALL_H
25 #define H_VSERVER_SYSCALL_H
30 #include <sys/types.h>
34 # define VC_ATTR_UNUSED __attribute__((__unused__))
35 # define VC_ATTR_NORETURN __attribute__((__noreturn__))
36 # define VC_ATTR_CONST __attribute__((__const__))
37 # define VC_ATTR_DEPRECATED __attribute__((__deprecated__))
38 # if __GNUC__*0x10000 + __GNUC_MINOR__*0x100 + __GNUC_PATCHLEVEL__ >= 0x30300
39 # define VC_ATTR_NONNULL(ARGS) __attribute__((__nonnull__ ARGS))
40 # define VC_ATTR_ALWAYSINLINE __attribute__((__always_inline__))
42 # define VC_ATTR_NONNULL(ARGS)
43 # define VC_ATTR_ALWAYSINLINE
45 # if __GNUC__*0x10000 + __GNUC_MINOR__*0x100 + __GNUC_PATCHLEVEL__ >= 0x30303
46 # define VC_ATTR_PURE __attribute__((__pure__))
51 # define VC_ATTR_NONNULL(ARGS)
52 # define VC_ATTR_UNUSED
53 # define VC_ATTR_NORETURN
54 # define VC_ATTR_ALWAYSINLINE
55 # define VC_ATTR_DEPRECATED
57 # define VC_ATTR_CONST
61 /** the value which is returned in error-case (no ctx found) */
62 #define VC_NOCTX ((xid_t)(-1))
63 #define VC_NOXID ((xid_t)(-1))
64 /** the value which means a random (the next free) ctx */
65 #define VC_DYNAMIC_XID ((xid_t)(-1))
66 /** the value which means the current ctx */
67 #define VC_SAMECTX ((xid_t)(-2))
69 #define VC_NONID ((nid_t)(-1))
70 #define VC_DYNAMIC_NID ((nid_t)(-1))
72 #define VC_LIM_INFINITY (~0ULL)
73 #define VC_LIM_KEEP (~1ULL)
76 #ifndef S_CTX_INFO_LOCK
77 # define S_CTX_INFO_LOCK 1
80 #ifndef S_CTX_INFO_SCHED
81 # define S_CTX_INFO_SCHED 2
84 #ifndef S_CTX_INFO_NPROC
85 # define S_CTX_INFO_NPROC 4
88 #ifndef S_CTX_INFO_PRIVATE
89 # define S_CTX_INFO_PRIVATE 8
92 #ifndef S_CTX_INFO_INIT
93 # define S_CTX_INFO_INIT 16
96 #ifndef S_CTX_INFO_HIDEINFO
97 # define S_CTX_INFO_HIDEINFO 32
100 #ifndef S_CTX_INFO_ULIMIT
101 # define S_CTX_INFO_ULIMIT 64
104 #ifndef S_CTX_INFO_NAMESPACE
105 # define S_CTX_INFO_NAMESPACE 128
108 #define VC_CAP_CHOWN 0
109 #define VC_CAP_DAC_OVERRIDE 1
110 #define VC_CAP_DAC_READ_SEARCH 2
111 #define VC_CAP_FOWNER 3
112 #define VC_CAP_FSETID 4
113 #define VC_CAP_KILL 5
114 #define VC_CAP_SETGID 6
115 #define VC_CAP_SETUID 7
116 #define VC_CAP_SETPCAP 8
117 #define VC_CAP_LINUX_IMMUTABLE 9
118 #define VC_CAP_NET_BIND_SERVICE 10
119 #define VC_CAP_NET_BROADCAST 11
120 #define VC_CAP_NET_ADMIN 12
121 #define VC_CAP_NET_RAW 13
122 #define VC_CAP_IPC_LOCK 14
123 #define VC_CAP_IPC_OWNER 15
124 #define VC_CAP_SYS_MODULE 16
125 #define VC_CAP_SYS_RAWIO 17
126 #define VC_CAP_SYS_CHROOT 18
127 #define VC_CAP_SYS_PTRACE 19
128 #define VC_CAP_SYS_PACCT 20
129 #define VC_CAP_SYS_ADMIN 21
130 #define VC_CAP_SYS_BOOT 22
131 #define VC_CAP_SYS_NICE 23
132 #define VC_CAP_SYS_RESOURCE 24
133 #define VC_CAP_SYS_TIME 25
134 #define VC_CAP_SYS_TTY_CONFIG 26
135 #define VC_CAP_MKNOD 27
136 #define VC_CAP_LEASE 28
137 #define VC_CAP_QUOTACTL 29
139 #define VC_IMMUTABLE_FILE_FL 0x0000010lu
140 #define VC_IMMUTABLE_LINK_FL 0x0008000lu
141 #define VC_IMMUTABLE_ALL (VC_IMMUTABLE_LINK_FL|VC_IMMUTABLE_FILE_FL)
143 #define VC_IATTR_XID 0x01000000u
145 #define VC_IATTR_ADMIN 0x00000001u
146 #define VC_IATTR_WATCH 0x00000002u
147 #define VC_IATTR_HIDE 0x00000004u
148 #define VC_IATTR_FLAGS 0x00000007u
150 #define VC_IATTR_BARRIER 0x00010000u
151 #define VC_IATTR_IUNLINK 0x00020000u
152 #define VC_IATTR_IMMUTABLE 0x00040000u
156 #define VC_VXF_INFO_LOCK 0x00000001ull
157 #define VC_VXF_INFO_NPROC 0x00000004ull
158 #define VC_VXF_INFO_PRIVATE 0x00000008ull
159 #define VC_VXF_INFO_INIT 0x00000010ull
161 #define VC_VXF_INFO_HIDEINFO 0x00000020ull
162 #define VC_VXF_INFO_ULIMIT 0x00000040ull
163 #define VC_VXF_INFO_NAMESPACE 0x00000080ull
165 #define VC_VXF_SCHED_HARD 0x00000100ull
166 #define VC_VXF_SCHED_PRIO 0x00000200ull
167 #define VC_VXF_SCHED_PAUSE 0x00000400ull
169 #define VC_VXF_VIRT_MEM 0x00010000ull
170 #define VC_VXF_VIRT_UPTIME 0x00020000ull
171 #define VC_VXF_VIRT_CPU 0x00040000ull
173 #define VC_VXF_HIDE_MOUNT 0x01000000ull
174 #define VC_VXF_HIDE_NETIF 0x02000000ull
176 #define VC_VXF_STATE_SETUP (1ULL<<32)
177 #define VC_VXF_STATE_INIT (1ULL<<33)
180 #define VC_VXC_SET_UTSNAME 0x00000001ull
181 #define VC_VXC_SET_RLIMIT 0x00000002ull
183 #define VC_VXC_ICMP_PING 0x00000100ull
185 #define VC_VXC_SECURE_MOUNT 0x00010000ull
188 /** \defgroup syscalls Syscall wrappers
189 * Functions which are calling the vserver syscall directly. */
191 /** \defgroup helper Helper functions
192 * Functions which are doing general helper tasks like parameter parsing. */
194 /** \typedef an_unsigned_integer_type xid_t
195 * The identifier of a context. */
198 typedef an_unsigned_integer_type xid_t;
199 typedef an_unsigned_integer_type nid_t;
206 struct vc_ip_mask_pair {
211 /** \brief The generic vserver syscall
214 * This function executes the generic vserver syscall. It uses the
215 * correct syscallnumber (which may differ between the different
218 * \params cmd the command to be executed
219 * \params xid the xid on which the cmd shall be applied
220 * \params data additional arguments; depends on \c cmd
221 * \returns depends on \c cmd; usually, -1 stands for an error
223 int vc_syscall(uint32_t cmd, xid_t xid, void *data);
225 /** \brief Returns the version of the current kernel API.
227 * \returns The versionnumber of the kernel API
229 int vc_get_version();
231 /** \brief Moves current process into a context
234 * Puts current process into context \a ctx, removes the capabilities
235 * given in \a remove_cap and sets \a flags.
237 * \param ctx The new context; special values for are
238 * - VC_SAMECTX which means the current context (just for changing caps and flags)
239 * - VC_DYNAMIC_XID which means the next free context; this value can be used by
240 * ordinary users also
241 * \param remove_cap The linux capabilities which will be \b removed.
242 * \param flags Special flags which will be set.
244 * \returns The new context-id, or VC_NOCTX on errors; errno
245 * will be set appropriately
247 * See http://vserver.13thfloor.at/Stuff/Logic.txt for details */
248 xid_t vc_new_s_context(xid_t ctx, unsigned int remove_cap, unsigned int flags);
250 /** \brief Sets the ipv4root information.
252 * \pre \a nb < NB_IPV4ROOT && \a ips != 0 */
253 int vc_set_ipv4root(uint32_t bcast, size_t nb,
254 struct vc_ip_mask_pair const *ips) VC_ATTR_NONNULL((3));
256 /** \brief Returns the value of NB_IPV4ROOT.
259 * This function returns the value of NB_IPV4ROOT which was used when the
260 * library was built, but \b not the value which is used by the currently
262 size_t vc_get_nb_ipv4root() VC_ATTR_CONST VC_ATTR_PURE;
264 /** \brief Creates a context without starting it.
267 * This functions initializes a new context. When already in a freshly
268 * created context, this old context will be discarded.
270 * \param xid The new context; special values are:
271 * - VC_DYNAMIC_XID which means to create a dynamic context
273 * \returns the xid of the created context, or VC_NOCTX on errors. errno
274 * will be set appropriately. */
275 xid_t vc_ctx_create(xid_t xid);
277 /** \brief Moves the current process into the specified context.
280 * \param xid The new context
281 * \returns 0 on success, -1 on errors */
282 int vc_ctx_migrate(xid_t xid);
284 /* rlimit related functions */
286 /** \brief The type which is used for a single limit value.
289 * - VC_LIM_INFINITY ... which is the infinite value
290 * - VC_LIM_KEEP ... which is used to mark values which shall not be
291 * modified by the vc_set_rlimit() operation.
293 * Else, the interpretation of the value depends on the corresponding
294 * resource; it might be bytes, pages, seconds or litres of beer. */
295 typedef uint_least64_t vc_limit_t;
297 /** \brief The limits of a resources.
299 * This is a triple consisting of a minimum, soft and hardlimit. */
301 vc_limit_t min; ///< the guaranted minimum of a resources
302 vc_limit_t soft; ///< the softlimit of a resource
303 vc_limit_t hard; ///< the absolute hardlimit of a resource
306 /** \brief Masks describing the supported limits. */
307 struct vc_rlimit_mask {
308 uint_least32_t min; ///< masks the resources supporting a minimum limit
309 uint_least32_t soft; ///< masks the resources supporting a soft limit
310 uint_least32_t hard; ///< masks the resources supporting a hard limit
313 /** \brief Returns the limits of \a resource.
316 * \param xid The id of the context
317 * \param resource The resource which will be queried
318 * \param lim The result which will be filled with the limits
320 * \returns 0 on success, and -1 on errors. */
321 int vc_get_rlimit(xid_t xid, int resource,
322 struct vc_rlimit /*@out@*/ *lim) VC_ATTR_NONNULL((3));
323 /** \brief Sets the limits of \a resource.
326 * \param xid The id of the context
327 * \param resource The resource which will be queried
328 * \param lim The new limits
330 * \returns 0 on success, and -1 on errors. */
331 int vc_set_rlimit(xid_t xid, int resource,
332 struct vc_rlimit const /*@in@*/ *lim) VC_ATTR_NONNULL((3));
333 int vc_get_rlimit_mask(xid_t xid,
334 struct vc_rlimit_mask *lim) VC_ATTR_NONNULL((2));
335 /** \brief Parses a string describing a limit
338 * This function parses \a str and interprets special words like \p "inf"
339 * or suffixes. Valid suffixes are
345 * \param str The string which shall be parsed
346 * \param res Will be filled with the interpreted value; in errorcase,
347 * this value is undefined.
349 * \returns \a true, iff the string \a str could be parsed. \a res will
350 * be filled with the interpreted value in this case.
352 * \pre \a str!=0 && \a res!=0
354 bool vc_parseLimit(char const /*@in@*/ *str, vc_limit_t /*@out@*/ *res) VC_ATTR_NONNULL((1,2));
357 /** \brief Sends a signal to a context/pid
360 * Special values for \a pid are:
361 * - -1 which means every process in ctx except the init-process
362 * - 0 which means every process in ctx inclusive the init-process */
363 int vc_ctx_kill(xid_t ctx, pid_t pid, int sig);
370 nid_t vc_get_task_nid(pid_t pid);
371 int vc_get_nx_info(nid_t nid, struct vc_nx_info *) VC_ATTR_NONNULL((2));
373 typedef enum { vcNET_IPV4, vcNET_IPV6, vcNET_IPV4R, vcNET_IPV6R } vc_net_nx_type;
382 nid_t vc_net_create(nid_t nid);
383 int vc_net_migrate(nid_t nid);
385 int vc_net_add(nid_t nid, struct vc_net_nx const *info);
386 int vc_net_remove(nid_t nid, struct vc_net_nx const *info);
388 struct vc_net_flags {
389 uint_least64_t flagword;
393 int vc_get_nflags(nid_t, struct vc_net_flags *);
394 int vc_set_nflags(nid_t, struct vc_net_flags const *);
398 uint_least64_t ncaps;
399 uint_least64_t cmask;
402 int vc_get_ncaps(nid_t, struct vc_net_caps *);
403 int vc_set_ncaps(nid_t, struct vc_net_caps const *);
408 int vc_set_iattr(char const *filename, xid_t xid,
409 uint_least32_t flags, uint_least32_t mask) VC_ATTR_NONNULL((1));
410 int vc_get_iattr(char const *filename, xid_t * /*@null@*/ xid,
411 uint_least32_t * /*@null@*/ flags,
412 uint_least32_t * /*@null@*/ mask) VC_ATTR_NONNULL((1));
419 /** \brief Returns the context of the given process.
421 * \params pid the process-id whose xid shall be determined;
422 * pid==0 means the current process.
423 * \returns the xid of process \c pid or -1 on errors
425 xid_t vc_get_task_xid(pid_t pid);
426 int vc_get_vx_info(xid_t xid, struct vc_vx_info *info) VC_ATTR_NONNULL((2));
429 typedef enum { vcVHI_CONTEXT, vcVHI_SYSNAME, vcVHI_NODENAME,
430 vcVHI_RELEASE, vcVHI_VERSION, vcVHI_MACHINE,
431 vcVHI_DOMAINNAME } vc_uts_type;
433 int vc_set_vhi_name(xid_t xid, vc_uts_type type,
434 char const *val, size_t len) VC_ATTR_NONNULL((3));
435 int vc_get_vhi_name(xid_t xid, vc_uts_type type,
436 char *val, size_t len) VC_ATTR_NONNULL((3));
438 /** Returns true iff \a xid is a dynamic xid */
439 bool vc_is_dynamic_xid(xid_t xid);
441 int vc_enter_namespace(xid_t xid);
442 int vc_set_namespace();
443 int vc_cleanup_namespace();
446 /** \brief Flags of process-contexts
448 struct vc_ctx_flags {
449 /** \brief Mask of set context flags */
450 uint_least64_t flagword;
451 /** \brief Mask of set and unset context flags when used by set
452 * operations, or modifiable flags when used by get
457 /** \brief Capabilities of process-contexts */
459 /** \brief Mask of set common system capabilities */
460 uint_least64_t bcaps;
461 /** \brief Mask of set and unset common system capabilities when used by
462 * set operations, or the modifiable capabilities when used by
464 uint_least64_t bmask;
465 /** \brief Mask of set process context capabilities */
466 uint_least64_t ccaps;
467 /** \brief Mask of set and unset process context capabilities when used
468 * by set operations, or the modifiable capabilities when used
469 * by get operations */
470 uint_least64_t cmask;
473 /** \brief Information about parsing errors
476 struct vc_err_listparser {
477 char const *ptr; ///< Pointer to the first character of an erroneous string
478 size_t len; ///< Length of the erroneous string
481 int vc_get_cflags(xid_t xid, struct vc_ctx_flags *) VC_ATTR_NONNULL((2));
482 int vc_set_cflags(xid_t xid, struct vc_ctx_flags const *) VC_ATTR_NONNULL((2));
484 int vc_get_ccaps(xid_t xid, struct vc_ctx_caps *);
485 int vc_set_ccaps(xid_t xid, struct vc_ctx_caps const *);
487 /** \brief Converts a single string into bcapability
490 * \param str The string to be parsed;
491 * both "CAP_xxx" and "xxx" will be accepted
492 * \param len The length of the string, or \c 0 for automatic detection
494 * \returns 0 on error; a bitmask on success
497 uint_least64_t vc_text2bcap(char const *str, size_t len);
499 /** \brief Converts the lowest bit of a bcapability or the entire value
500 * (when possible) to a textual representation
503 * \param val The string to be converted; on success, the detected bit(s)
504 * will be unset, in errorcase only the lowest set bit
506 * \returns A textual representation of \a val resp. of its lowest set bit;
507 * or \c NULL in errorcase.
509 * \post \a *val<sub>old</sub> \c != 0 \c <-->
510 * \a *val<sub>old</sub> > \a *val<sub>new</sub>
511 * \post \a *val<sub>old</sub> \c == 0 \c ---> \a result == 0
513 char const * vc_lobcap2text(uint_least64_t *val) VC_ATTR_NONNULL((1));
515 /** \brief Converts a string into a bcapability-bitmask
518 * Syntax of \a str: \verbinclude list2xxx.syntax
520 * When the \c `~' prefix is used, the bits will be unset and a `~' after
521 * another `~' will cancel both ones. The \c `^' prefix specifies a
522 * bitnumber instead of a bitmask.
524 * "literal name" is everything which will be accepted by the
525 * vc_text2bcap() function. The special values for \c NAME will be
526 * recognized case insensitively
528 * \param str The string to be parsed
529 * \param len The length of the string, or \c 0 for automatic detection
530 * \param err Pointer to a structure for error-information, or \c NULL.
531 * \param cap Pointer to a vc_ctx_caps structure holding the results;
532 * only the \a bcaps and \a bmask fields will be changed and
533 * already set values will not be honored. When an error
534 * occured, \a cap will have the value of all processed valid
537 * \returns 0 on success, -1 on error. In error case, \a err will hold
538 * position and length of the first not understood BCAP part
539 * \pre \a str != 0 && \a cap != 0;
540 * \a cap->bcaps and \a cap->bmask must be initialized
542 int vc_list2bcap(char const *str, size_t len,
543 struct vc_err_listparser *err,
544 struct vc_ctx_caps *cap) VC_ATTR_NONNULL((1,4));
546 uint_least64_t vc_text2ccap(char const *, size_t len);
547 char const * vc_loccap2text(uint_least64_t *);
548 int vc_list2ccap(char const *, size_t len,
549 struct vc_err_listparser *err,
550 struct vc_ctx_caps *);
552 int vc_list2cflag(char const *, size_t len,
553 struct vc_err_listparser *err,
554 struct vc_ctx_flags *flags);
555 uint_least64_t vc_text2cflag(char const *, size_t len);
556 char const * vc_locflag2text(uint_least64_t *);
558 uint_least32_t vc_list2cflag_compat(char const *, size_t len,
559 struct vc_err_listparser *err);
560 uint_least32_t vc_text2cflag_compat(char const *, size_t len);
561 char const * vc_hicflag2text_compat(uint_least32_t);
563 int vc_text2cap(char const *);
564 char const * vc_cap2text(unsigned int);
567 int vc_list2nflag(char const *, size_t len,
568 struct vc_err_listparser *err,
569 struct vc_net_flags *flags);
570 uint_least64_t vc_text2nflag(char const *, size_t len);
571 char const * vc_lonflag2text(uint_least64_t *);
573 uint_least64_t vc_text2ncap(char const *, size_t len);
574 char const * vc_loncap2text(uint_least64_t *);
575 int vc_list2ncap(char const *, size_t len,
576 struct vc_err_listparser *err,
577 struct vc_net_caps *);
579 uint_least64_t vc_get_insecurebcaps() VC_ATTR_CONST;
580 inline static uint_least64_t vc_get_insecureccaps() {
581 return ~(VC_VXC_SET_UTSNAME|VC_VXC_ICMP_PING);
584 inline static int vc_setfilecontext(char const *filename, xid_t xid) {
585 return vc_set_iattr(filename, xid, 0, VC_IATTR_XID);
588 inline static xid_t vc_getfilecontext(char const *filename) {
590 if (vc_get_iattr(filename, &res, 0,0)==-1) return VC_NOCTX;
595 struct vc_set_sched {
604 int vc_set_sched(xid_t xid, struct vc_set_sched const *);
607 typedef enum { vcFEATURE_VKILL, vcFEATURE_IATTR, vcFEATURE_RLIMIT,
608 vcFEATURE_COMPAT, vcFEATURE_MIGRATE, vcFEATURE_NAMESPACE,
609 vcFEATURE_SCHED, vcFEATURE_VINFO, vcFEATURE_VHI,
610 vcFEATURE_VSHELPER0, vcFEATURE_VSHELPER }
613 bool vc_isSupported(vcFeatureSet) VC_ATTR_CONST;
614 bool vc_isSupportedString(char const *);
616 /* The management part */
618 #define VC_LIMIT_VSERVER_NAME_LEN 1024
620 typedef enum { vcCFG_NONE, vcCFG_AUTO,
623 vcCFG_RECENT_FULL } vcCfgStyle;
626 /** Maps an xid given at '--xid' options to an xid_t */
627 xid_t vc_xidopt2xid(char const *, bool honor_static, char const **err_info);
629 vcCfgStyle vc_getVserverCfgStyle(char const *id);
631 /** Resolves the name of the vserver. The result will be allocated and must
632 be freed by the caller. */
633 char * vc_getVserverName(char const *id, vcCfgStyle style);
635 /** Returns the path of the vserver configuration directory. When the given
636 * vserver does not exist, or when it does not have such a directory, NULL
637 * will be returned. Else, the result will be allocated and must be freed
639 char * vc_getVserverCfgDir(char const *id, vcCfgStyle style);
641 /** Returns the path of the configuration directory for the given
642 * application. The result will be allocated and must be freed by the
644 char * vc_getVserverAppDir(char const *id, vcCfgStyle style, char const *app);
646 /** Returns the path to the vserver root-directory. The result will be
647 * allocated and must be freed by the caller. */
648 char * vc_getVserverVdir(char const *id, vcCfgStyle style, bool physical);
650 /** Returns the ctx of the given vserver. When vserver is not running and
651 * 'honor_static' is false, VC_NOCTX will be returned. Else, when
652 * 'honor_static' is true and a static assignment exists, those value will
653 * be returned. Else, the result will be VC_NOCTX.
655 * When 'is_running' is not null, the status of the vserver will be
656 * assigned to this variable. */
657 xid_t vc_getVserverCtx(char const *id, vcCfgStyle style,
658 bool honor_static, bool /*@null@*/ *is_running);
660 /** Resolves the cfg-path of the vserver owning the given ctx. 'revdir' will
661 be used as the directory holding the mapping-links; when NULL, the
662 default value will be assumed. The result will be allocated and must be
663 freed by the caller. */
664 char * vc_getVserverByCtx(xid_t ctx, /*@null@*/vcCfgStyle *style,
665 /*@null@*/char const *revdir);
667 #define vcSKEL_INTERFACES 1u
668 #define vcSKEL_PKGMGMT 2u
669 #define vcSKEL_FILESYSTEM 4u
671 /** Create a basic configuration skeleton for a vserver plus toplevel
672 * directories for pkgmanagemt and filesystem (when requested). */
673 int vc_createSkeleton(char const *id, vcCfgStyle style, int flags);
680 #undef VC_ATTR_ALWAYSINLINE
681 #undef VC_ATTR_NORETURN
682 #undef VC_ATTR_UNUSED
683 #undef VC_ATTR_NONNULL