src/lxc/process_utils.h

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #ifndef __LXC_PROCESS_UTILS_H
   4 #define __LXC_PROCESS_UTILS_H
   5
   6 #ifndef _GNU_SOURCE
   7 #define _GNU_SOURCE 1
   8 #endif
   9 #include <sched.h>
  10 #include <signal.h>
  11 #include <stdbool.h>
  12 #include <stdio.h>
  13 #include <stdlib.h>
  14 #include <sys/syscall.h>
  15 #include <unistd.h>
  16
  17 #ifndef CSIGNAL
  18 #define CSIGNAL 0x000000ff /* signal mask to be sent at exit */
  19 #endif
  20
  21 #ifndef CLONE_VM
  22 #define CLONE_VM 0x00000100 /* set if VM shared between processes */
  23 #endif
  24
  25 #ifndef CLONE_FS
  26 #define CLONE_FS 0x00000200 /* set if fs info shared between processes */
  27 #endif
  28
  29 #ifndef CLONE_FILES
  30 #define CLONE_FILES 0x00000400 /* set if open files shared between processes */
  31 #endif
  32
  33 #ifndef CLONE_SIGHAND
  34 #define CLONE_SIGHAND 0x00000800 /* set if signal handlers and blocked signals shared */
  35 #endif
  36
  37 #ifndef CLONE_PIDFD
  38 #define CLONE_PIDFD 0x00001000 /* set if a pidfd should be placed in parent */
  39 #endif
  40
  41 #ifndef CLONE_PTRACE
  42 #define CLONE_PTRACE 0x00002000 /* set if we want to let tracing continue on the child too */
  43 #endif
  44
  45 #ifndef CLONE_VFORK
  46 #define CLONE_VFORK 0x00004000 /* set if the parent wants the child to wake it up on mm_release */
  47 #endif
  48
  49 #ifndef CLONE_PARENT
  50 #define CLONE_PARENT 0x00008000 /* set if we want to have the same parent as the cloner */
  51 #endif
  52
  53 #ifndef CLONE_THREAD
  54 #define CLONE_THREAD 0x00010000 /* Same thread group? */
  55 #endif
  56
  57 #ifndef CLONE_NEWNS
  58 #define CLONE_NEWNS 0x00020000 /* New mount namespace group */
  59 #endif
  60
  61 #ifndef CLONE_SYSVSEM
  62 #define CLONE_SYSVSEM 0x00040000 /* share system V SEM_UNDO semantics */
  63 #endif
  64
  65 #ifndef CLONE_SETTLS
  66 #define CLONE_SETTLS 0x00080000 /* create a new TLS for the child */
  67 #endif
  68
  69 #ifndef CLONE_PARENT_SETTID
  70 #define CLONE_PARENT_SETTID 0x00100000 /* set the TID in the parent */
  71 #endif
  72
  73 #ifndef CLONE_CHILD_CLEARTID
  74 #define CLONE_CHILD_CLEARTID 0x00200000 /* clear the TID in the child */
  75 #endif
  76
  77 #ifndef CLONE_DETACHED
  78 #define CLONE_DETACHED 0x00400000 /* Unused, ignored */
  79 #endif
  80
  81 #ifndef CLONE_UNTRACED
  82 #define CLONE_UNTRACED 0x00800000 /* set if the tracing process can't force CLONE_PTRACE on this clone */
  83 #endif
  84
  85 #ifndef CLONE_CHILD_SETTID
  86 #define CLONE_CHILD_SETTID 0x01000000 /* set the TID in the child */
  87 #endif
  88
  89 #ifndef CLONE_NEWCGROUP
  90 #define CLONE_NEWCGROUP 0x02000000 /* New cgroup namespace */
  91 #endif
  92
  93 #ifndef CLONE_NEWUTS
  94 #define CLONE_NEWUTS 0x04000000 /* New utsname namespace */
  95 #endif
  96
  97 #ifndef CLONE_NEWIPC
  98 #define CLONE_NEWIPC 0x08000000 /* New ipc namespace */
  99 #endif
 100
 101 #ifndef CLONE_NEWUSER
 102 #define CLONE_NEWUSER 0x10000000 /* New user namespace */
 103 #endif
 104
 105 #ifndef CLONE_NEWPID
 106 #define CLONE_NEWPID 0x20000000 /* New pid namespace */
 107 #endif
 108
 109 #ifndef CLONE_NEWNET
 110 #define CLONE_NEWNET 0x40000000 /* New network namespace */
 111 #endif
 112
 113 #ifndef CLONE_IO
 114 #define CLONE_IO 0x80000000 /* Clone io context */
 115 #endif
 116
 117 /* Flags for the clone3() syscall. */
 118 #ifndef CLONE_CLEAR_SIGHAND
 119 #define CLONE_CLEAR_SIGHAND 0x100000000ULL /* Clear any signal handler and reset to SIG_DFL. */
 120 #endif
 121
 122 #ifndef CLONE_INTO_CGROUP
 123 #define CLONE_INTO_CGROUP 0x200000000ULL /* Clone into a specific cgroup given the right permissions. */
 124 #endif
 125
 126 /*
 127  * cloning flags intersect with CSIGNAL so can be used with unshare and clone3
 128  * syscalls only:
 129  */
 130 #ifndef CLONE_NEWTIME
 131 #define CLONE_NEWTIME 0x00000080 /* New time namespace */
 132 #endif
 133
 134 /* waitid */
 135 #ifndef P_PIDFD
 136 #define P_PIDFD 3
 137 #endif
 138
 139
 140 #if defined(__ia64__)
 141 int __clone2(int (*__fn)(void *__arg), void *__child_stack_base,
 142              size_t __child_stack_size, int __flags, void *__arg, ...);
 143 #else
 144 int clone(int (*fn)(void *), void *child_stack, int flags, void *arg, ...
 145           /* pid_t *ptid, struct user_desc *tls, pid_t *ctid */);
 146 #endif
 147
 148 /**
 149  * lxc_clone() - create a new process
 150  *
 151  * - allocate stack:
 152  *   This function allocates a new stack the size of page and passes it to the
 153  *   kernel.
 154  *
 155  * - support all CLONE_*flags:
 156  *   This function supports all CLONE_* flags. If in doubt or not sufficiently
 157  *   familiar with process creation in the kernel and interactions with libcs
 158  *   this function should be used.
 159  *
 160  * - pthread_atfork() handlers depending on libc:
 161  *   Whether this function runs pthread_atfork() handlers depends on the
 162  *   corresponding libc wrapper. glibc currently does not run pthread_atfork()
 163  *   handlers but does not guarantee that they are not. Other libcs might or
 164  *   might not run pthread_atfork() handlers. If you require guarantees please
 165  *   refer to the lxc_raw_clone*() functions in process_utils.{c,h}.
 166  *
 167  * - should call lxc_raw_getpid():
 168  *   The child should use lxc_raw_getpid() to retrieve its pid.
 169  */
 170 extern pid_t lxc_clone(int (*fn)(void *), void *arg, int flags, int *pidfd);
 171
 172
 173 /*
 174  * lxc_raw_clone() - create a new process
 175  *
 176  * - fork() behavior:
 177  *   This function returns 0 in the child and > 0 in the parent.
 178  *
 179  * - copy-on-write:
 180  *   This function does not allocate a new stack and relies on copy-on-write
 181  *   semantics.
 182  *
 183  * - supports subset of ClONE_* flags:
 184  *   lxc_raw_clone() intentionally only supports a subset of the flags available
 185  *   to the actual system call. Please refer to the implementation what flags
 186  *   cannot be used. Also, please don't assume that just because a flag isn't
 187  *   explicitly checked for as being unsupported that it is supported. If in
 188  *   doubt or not sufficiently familiar with process creation in the kernel and
 189  *   interactions with libcs this function should be used.
 190  *
 191  * - no pthread_atfork() handlers:
 192  *   This function circumvents - as much as this this is possible - any libc
 193  *   wrappers and thus does not run any pthread_atfork() handlers. Make sure
 194  *   that this is safe to do in the context you are trying to call this
 195  *   function.
 196  *
 197  * - must call lxc_raw_getpid():
 198  *   The child must use lxc_raw_getpid() to retrieve its pid.
 199  */
 200 extern pid_t lxc_raw_clone(unsigned long flags, int *pidfd);
 201
 202 /*
 203  * lxc_raw_clone_cb() - create a new process
 204  *
 205  * - non-fork() behavior:
 206  *   Function does return pid of the child or -1 on error. Pass in a callback
 207  *   function via the "fn" argument that gets executed in the child process.
 208  *   The "args" argument is passed to "fn".
 209  *
 210  * All other comments that apply to lxc_raw_clone() apply to lxc_raw_clone_cb()
 211  * as well.
 212  */
 213 extern pid_t lxc_raw_clone_cb(int (*fn)(void *), void *args,
 214                               unsigned long flags, int *pidfd);
 215
 216 #ifndef HAVE_EXECVEAT
 217 static inline int execveat(int dirfd, const char *pathname, char *const argv[],
 218                            char *const envp[], int flags)
 219 {
 220         return syscall(__NR_execveat, dirfd, pathname, argv, envp, flags);
 221 }
 222 #else
 223 extern int execveat(int dirfd, const char *pathname, char *const argv[],
 224                     char *const envp[], int flags);
 225 #endif
 226
 227 /*
 228  * Because of older glibc's pid cache (up to 2.25) whenever clone() is called
 229  * the child must must retrieve it's own pid via lxc_raw_getpid().
 230  */
 231 static inline pid_t lxc_raw_getpid(void)
 232 {
 233         return (pid_t)syscall(SYS_getpid);
 234 }
 235
 236 static inline pid_t lxc_raw_gettid(void)
 237 {
 238 #if __NR_gettid > 0
 239         return syscall(__NR_gettid);
 240 #else
 241         return lxc_raw_getpid();
 242 #endif
 243 }
 244
 245 extern int lxc_raw_pidfd_send_signal(int pidfd, int sig, siginfo_t *info,
 246                                      unsigned int flags);
 247
 248 #endif /* __LXC_PROCESS_UTILS_H */