1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 #ifndef __LXC_PROCESS_UTILS_H
4 #define __LXC_PROCESS_UTILS_H
8 #include <linux/sched.h>
14 #include <sys/syscall.h>
18 #include "syscall_numbers.h"
21 #define CSIGNAL 0x000000ff /* signal mask to be sent at exit */
25 #define CLONE_VM 0x00000100 /* set if VM shared between processes */
29 #define CLONE_FS 0x00000200 /* set if fs info shared between processes */
33 #define CLONE_FILES 0x00000400 /* set if open files shared between processes */
37 #define CLONE_SIGHAND 0x00000800 /* set if signal handlers and blocked signals shared */
41 #define CLONE_PIDFD 0x00001000 /* set if a pidfd should be placed in parent */
45 #define CLONE_PTRACE 0x00002000 /* set if we want to let tracing continue on the child too */
49 #define CLONE_VFORK 0x00004000 /* set if the parent wants the child to wake it up on mm_release */
53 #define CLONE_PARENT 0x00008000 /* set if we want to have the same parent as the cloner */
57 #define CLONE_THREAD 0x00010000 /* Same thread group? */
61 #define CLONE_NEWNS 0x00020000 /* New mount namespace group */
65 #define CLONE_SYSVSEM 0x00040000 /* share system V SEM_UNDO semantics */
69 #define CLONE_SETTLS 0x00080000 /* create a new TLS for the child */
72 #ifndef CLONE_PARENT_SETTID
73 #define CLONE_PARENT_SETTID 0x00100000 /* set the TID in the parent */
76 #ifndef CLONE_CHILD_CLEARTID
77 #define CLONE_CHILD_CLEARTID 0x00200000 /* clear the TID in the child */
80 #ifndef CLONE_DETACHED
81 #define CLONE_DETACHED 0x00400000 /* Unused, ignored */
84 #ifndef CLONE_UNTRACED
85 #define CLONE_UNTRACED 0x00800000 /* set if the tracing process can't force CLONE_PTRACE on this clone */
88 #ifndef CLONE_CHILD_SETTID
89 #define CLONE_CHILD_SETTID 0x01000000 /* set the TID in the child */
92 #ifndef CLONE_NEWCGROUP
93 #define CLONE_NEWCGROUP 0x02000000 /* New cgroup namespace */
97 #define CLONE_NEWUTS 0x04000000 /* New utsname namespace */
101 #define CLONE_NEWIPC 0x08000000 /* New ipc namespace */
104 #ifndef CLONE_NEWUSER
105 #define CLONE_NEWUSER 0x10000000 /* New user namespace */
109 #define CLONE_NEWPID 0x20000000 /* New pid namespace */
113 #define CLONE_NEWNET 0x40000000 /* New network namespace */
117 #define CLONE_IO 0x80000000 /* Clone io context */
120 /* Flags for the clone3() syscall. */
121 #ifndef CLONE_CLEAR_SIGHAND
122 #define CLONE_CLEAR_SIGHAND 0x100000000ULL /* Clear any signal handler and reset to SIG_DFL. */
125 #ifndef CLONE_INTO_CGROUP
126 #define CLONE_INTO_CGROUP 0x200000000ULL /* Clone into a specific cgroup given the right permissions. */
130 * cloning flags intersect with CSIGNAL so can be used with unshare and clone3
133 #ifndef CLONE_NEWTIME
134 #define CLONE_NEWTIME 0x00000080 /* New time namespace */
142 #ifndef CLONE_ARGS_SIZE_VER0
143 #define CLONE_ARGS_SIZE_VER0 64 /* sizeof first published struct */
146 #ifndef CLONE_ARGS_SIZE_VER1
147 #define CLONE_ARGS_SIZE_VER1 80 /* sizeof second published struct */
150 #ifndef CLONE_ARGS_SIZE_VER2
151 #define CLONE_ARGS_SIZE_VER2 88 /* sizeof third published struct */
155 #define ptr_to_u64(ptr) ((__u64)((uintptr_t)(ptr)))
158 #define u64_to_ptr(x) ((void *)(uintptr_t)x)
161 struct lxc_clone_args
{
164 __aligned_u64 child_tid
;
165 __aligned_u64 parent_tid
;
166 __aligned_u64 exit_signal
;
168 __aligned_u64 stack_size
;
170 __aligned_u64 set_tid
;
171 __aligned_u64 set_tid_size
;
172 __aligned_u64 cgroup
;
175 __returns_twice
static inline pid_t
lxc_clone3(struct lxc_clone_args
*args
, size_t size
)
177 return syscall(__NR_clone3
, args
, size
);
180 #if defined(__ia64__)
181 int __clone2(int (*__fn
)(void *__arg
), void *__child_stack_base
,
182 size_t __child_stack_size
, int __flags
, void *__arg
, ...);
184 int clone(int (*fn
)(void *), void *child_stack
, int flags
, void *arg
, ...
185 /* pid_t *ptid, struct user_desc *tls, pid_t *ctid */);
189 * lxc_clone() - create a new process
192 * This function allocates a new stack the size of page and passes it to the
195 * - support all CLONE_*flags:
196 * This function supports all CLONE_* flags. If in doubt or not sufficiently
197 * familiar with process creation in the kernel and interactions with libcs
198 * this function should be used.
200 * - pthread_atfork() handlers depending on libc:
201 * Whether this function runs pthread_atfork() handlers depends on the
202 * corresponding libc wrapper. glibc currently does not run pthread_atfork()
203 * handlers but does not guarantee that they are not. Other libcs might or
204 * might not run pthread_atfork() handlers. If you require guarantees please
205 * refer to the lxc_raw_clone*() functions in process_utils.{c,h}.
207 * - should call lxc_raw_getpid():
208 * The child should use lxc_raw_getpid() to retrieve its pid.
210 __hidden
extern pid_t
lxc_clone(int (*fn
)(void *), void *arg
, int flags
, int *pidfd
);
214 * lxc_raw_clone() - create a new process
217 * This function returns 0 in the child and > 0 in the parent.
220 * This function does not allocate a new stack and relies on copy-on-write
223 * - supports subset of ClONE_* flags:
224 * lxc_raw_clone() intentionally only supports a subset of the flags available
225 * to the actual system call. Please refer to the implementation what flags
226 * cannot be used. Also, please don't assume that just because a flag isn't
227 * explicitly checked for as being unsupported that it is supported. If in
228 * doubt or not sufficiently familiar with process creation in the kernel and
229 * interactions with libcs this function should be used.
231 * - no pthread_atfork() handlers:
232 * This function circumvents - as much as this this is possible - any libc
233 * wrappers and thus does not run any pthread_atfork() handlers. Make sure
234 * that this is safe to do in the context you are trying to call this
237 * - must call lxc_raw_getpid():
238 * The child must use lxc_raw_getpid() to retrieve its pid.
240 __hidden
extern pid_t
lxc_raw_clone(unsigned long flags
, int *pidfd
);
241 __hidden
extern pid_t
lxc_raw_legacy_clone(unsigned long flags
, int *pidfd
);
244 * lxc_raw_clone_cb() - create a new process
246 * - non-fork() behavior:
247 * Function does return pid of the child or -1 on error. Pass in a callback
248 * function via the "fn" argument that gets executed in the child process.
249 * The "args" argument is passed to "fn".
251 * All other comments that apply to lxc_raw_clone() apply to lxc_raw_clone_cb()
254 __hidden
extern pid_t
lxc_raw_clone_cb(int (*fn
)(void *), void *args
, unsigned long flags
,
258 static inline int execveat(int dirfd
, const char *pathname
, char *const argv
[],
259 char *const envp
[], int flags
)
261 return syscall(__NR_execveat
, dirfd
, pathname
, argv
, envp
, flags
);
264 extern int execveat(int dirfd
, const char *pathname
, char *const argv
[],
265 char *const envp
[], int flags
);
269 * Because of older glibc's pid cache (up to 2.25) whenever clone() is called
270 * the child must must retrieve it's own pid via lxc_raw_getpid().
272 static inline pid_t
lxc_raw_getpid(void)
274 return (pid_t
)syscall(SYS_getpid
);
277 static inline pid_t
lxc_raw_gettid(void)
280 return syscall(__NR_gettid
);
282 return lxc_raw_getpid();
286 __hidden
extern int lxc_raw_pidfd_send_signal(int pidfd
, int sig
, siginfo_t
*info
,
289 #endif /* __LXC_PROCESS_UTILS_H */