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 #include "config.h"
   7
   8 #include <linux/sched.h>
   9 #include <sched.h>
  10 #include <signal.h>
  11 #include <stdbool.h>
  12 #include <stdio.h>
  13 #include <stdlib.h>
  14 #include <string.h>
  15 #include <sys/syscall.h>
  16 #include <unistd.h>
  17
  18 #include "compiler.h"
  19 #include "syscall_numbers.h"
  20
  21 #ifndef CSIGNAL
  22 #define CSIGNAL 0x000000ff /* signal mask to be sent at exit */
  23 #endif
  24
  25 #ifndef CLONE_VM
  26 #define CLONE_VM 0x00000100 /* set if VM shared between processes */
  27 #endif
  28
  29 #ifndef CLONE_FS
  30 #define CLONE_FS 0x00000200 /* set if fs info shared between processes */
  31 #endif
  32
  33 #ifndef CLONE_FILES
  34 #define CLONE_FILES 0x00000400 /* set if open files shared between processes */
  35 #endif
  36
  37 #ifndef CLONE_SIGHAND
  38 #define CLONE_SIGHAND 0x00000800 /* set if signal handlers and blocked signals shared */
  39 #endif
  40
  41 #ifndef CLONE_PIDFD
  42 #define CLONE_PIDFD 0x00001000 /* set if a pidfd should be placed in parent */
  43 #endif
  44
  45 #ifndef CLONE_PTRACE
  46 #define CLONE_PTRACE 0x00002000 /* set if we want to let tracing continue on the child too */
  47 #endif
  48
  49 #ifndef CLONE_VFORK
  50 #define CLONE_VFORK 0x00004000 /* set if the parent wants the child to wake it up on mm_release */
  51 #endif
  52
  53 #ifndef CLONE_PARENT
  54 #define CLONE_PARENT 0x00008000 /* set if we want to have the same parent as the cloner */
  55 #endif
  56
  57 #ifndef CLONE_THREAD
  58 #define CLONE_THREAD 0x00010000 /* Same thread group? */
  59 #endif
  60
  61 #ifndef CLONE_NEWNS
  62 #define CLONE_NEWNS 0x00020000 /* New mount namespace group */
  63 #endif
  64
  65 #ifndef CLONE_SYSVSEM
  66 #define CLONE_SYSVSEM 0x00040000 /* share system V SEM_UNDO semantics */
  67 #endif
  68
  69 #ifndef CLONE_SETTLS
  70 #define CLONE_SETTLS 0x00080000 /* create a new TLS for the child */
  71 #endif
  72
  73 #ifndef CLONE_PARENT_SETTID
  74 #define CLONE_PARENT_SETTID 0x00100000 /* set the TID in the parent */
  75 #endif
  76
  77 #ifndef CLONE_CHILD_CLEARTID
  78 #define CLONE_CHILD_CLEARTID 0x00200000 /* clear the TID in the child */
  79 #endif
  80
  81 #ifndef CLONE_DETACHED
  82 #define CLONE_DETACHED 0x00400000 /* Unused, ignored */
  83 #endif
  84
  85 #ifndef CLONE_UNTRACED
  86 #define CLONE_UNTRACED 0x00800000 /* set if the tracing process can't force CLONE_PTRACE on this clone */
  87 #endif
  88
  89 #ifndef CLONE_CHILD_SETTID
  90 #define CLONE_CHILD_SETTID 0x01000000 /* set the TID in the child */
  91 #endif
  92
  93 #ifndef CLONE_NEWCGROUP
  94 #define CLONE_NEWCGROUP 0x02000000 /* New cgroup namespace */
  95 #endif
  96
  97 #ifndef CLONE_NEWUTS
  98 #define CLONE_NEWUTS 0x04000000 /* New utsname namespace */
  99 #endif
 100
 101 #ifndef CLONE_NEWIPC
 102 #define CLONE_NEWIPC 0x08000000 /* New ipc namespace */
 103 #endif
 104
 105 #ifndef CLONE_NEWUSER
 106 #define CLONE_NEWUSER 0x10000000 /* New user namespace */
 107 #endif
 108
 109 #ifndef CLONE_NEWPID
 110 #define CLONE_NEWPID 0x20000000 /* New pid namespace */
 111 #endif
 112
 113 #ifndef CLONE_NEWNET
 114 #define CLONE_NEWNET 0x40000000 /* New network namespace */
 115 #endif
 116
 117 #ifndef CLONE_IO
 118 #define CLONE_IO 0x80000000 /* Clone io context */
 119 #endif
 120
 121 /* Flags for the clone3() syscall. */
 122 #ifndef CLONE_CLEAR_SIGHAND
 123 #define CLONE_CLEAR_SIGHAND 0x100000000ULL /* Clear any signal handler and reset to SIG_DFL. */
 124 #endif
 125
 126 #ifndef CLONE_INTO_CGROUP
 127 #define CLONE_INTO_CGROUP 0x200000000ULL /* Clone into a specific cgroup given the right permissions. */
 128 #endif
 129
 130 /*
 131  * cloning flags intersect with CSIGNAL so can be used with unshare and clone3
 132  * syscalls only:
 133  */
 134 #ifndef CLONE_NEWTIME
 135 #define CLONE_NEWTIME 0x00000080 /* New time namespace */
 136 #endif
 137
 138 /* waitid */
 139 #ifndef P_PIDFD
 140 #define P_PIDFD 3
 141 #endif
 142
 143 #ifndef CLONE_ARGS_SIZE_VER0
 144 #define CLONE_ARGS_SIZE_VER0 64 /* sizeof first published struct */
 145 #endif
 146
 147 #ifndef CLONE_ARGS_SIZE_VER1
 148 #define CLONE_ARGS_SIZE_VER1 80 /* sizeof second published struct */
 149 #endif
 150
 151 #ifndef CLONE_ARGS_SIZE_VER2
 152 #define CLONE_ARGS_SIZE_VER2 88 /* sizeof third published struct */
 153 #endif
 154
 155 #ifndef ptr_to_u64
 156 #define ptr_to_u64(ptr) ((__u64)((uintptr_t)(ptr)))
 157 #endif
 158 #ifndef u64_to_ptr
 159 #define u64_to_ptr(x) ((void *)(uintptr_t)x)
 160 #endif
 161
 162 struct lxc_clone_args {
 163         __aligned_u64 flags;
 164         __aligned_u64 pidfd;
 165         __aligned_u64 child_tid;
 166         __aligned_u64 parent_tid;
 167         __aligned_u64 exit_signal;
 168         __aligned_u64 stack;
 169         __aligned_u64 stack_size;
 170         __aligned_u64 tls;
 171         __aligned_u64 set_tid;
 172         __aligned_u64 set_tid_size;
 173         __aligned_u64 cgroup;
 174 };
 175
 176 __returns_twice static inline pid_t lxc_clone3(struct lxc_clone_args *args, size_t size)
 177 {
 178         return syscall(__NR_clone3, args, size);
 179 }
 180
 181 #if defined(__ia64__)
 182 int __clone2(int (*__fn)(void *__arg), void *__child_stack_base,
 183              size_t __child_stack_size, int __flags, void *__arg, ...);
 184 #else
 185 int clone(int (*fn)(void *), void *child_stack, int flags, void *arg, ...
 186           /* pid_t *ptid, struct user_desc *tls, pid_t *ctid */);
 187 #endif
 188
 189 /**
 190  * lxc_clone() - create a new process
 191  *
 192  * - allocate stack:
 193  *   This function allocates a new stack the size of page and passes it to the
 194  *   kernel.
 195  *
 196  * - support all CLONE_*flags:
 197  *   This function supports all CLONE_* flags. If in doubt or not sufficiently
 198  *   familiar with process creation in the kernel and interactions with libcs
 199  *   this function should be used.
 200  *
 201  * - pthread_atfork() handlers depending on libc:
 202  *   Whether this function runs pthread_atfork() handlers depends on the
 203  *   corresponding libc wrapper. glibc currently does not run pthread_atfork()
 204  *   handlers but does not guarantee that they are not. Other libcs might or
 205  *   might not run pthread_atfork() handlers. If you require guarantees please
 206  *   refer to the lxc_raw_clone*() functions in process_utils.{c,h}.
 207  *
 208  * - should call lxc_raw_getpid():
 209  *   The child should use lxc_raw_getpid() to retrieve its pid.
 210  */
 211 __hidden extern pid_t lxc_clone(int (*fn)(void *), void *arg, int flags, int *pidfd);
 212
 213
 214 /*
 215  * lxc_raw_clone() - create a new process
 216  *
 217  * - fork() behavior:
 218  *   This function returns 0 in the child and > 0 in the parent.
 219  *
 220  * - copy-on-write:
 221  *   This function does not allocate a new stack and relies on copy-on-write
 222  *   semantics.
 223  *
 224  * - supports subset of ClONE_* flags:
 225  *   lxc_raw_clone() intentionally only supports a subset of the flags available
 226  *   to the actual system call. Please refer to the implementation what flags
 227  *   cannot be used. Also, please don't assume that just because a flag isn't
 228  *   explicitly checked for as being unsupported that it is supported. If in
 229  *   doubt or not sufficiently familiar with process creation in the kernel and
 230  *   interactions with libcs this function should be used.
 231  *
 232  * - no pthread_atfork() handlers:
 233  *   This function circumvents - as much as this this is possible - any libc
 234  *   wrappers and thus does not run any pthread_atfork() handlers. Make sure
 235  *   that this is safe to do in the context you are trying to call this
 236  *   function.
 237  *
 238  * - must call lxc_raw_getpid():
 239  *   The child must use lxc_raw_getpid() to retrieve its pid.
 240  */
 241 __hidden extern pid_t lxc_raw_clone(unsigned long flags, int *pidfd);
 242 __hidden extern pid_t lxc_raw_legacy_clone(unsigned long flags, int *pidfd);
 243
 244 /*
 245  * lxc_raw_clone_cb() - create a new process
 246  *
 247  * - non-fork() behavior:
 248  *   Function does return pid of the child or -1 on error. Pass in a callback
 249  *   function via the "fn" argument that gets executed in the child process.
 250  *   The "args" argument is passed to "fn".
 251  *
 252  * All other comments that apply to lxc_raw_clone() apply to lxc_raw_clone_cb()
 253  * as well.
 254  */
 255 __hidden extern pid_t lxc_raw_clone_cb(int (*fn)(void *), void *args, unsigned long flags,
 256                                        int *pidfd);
 257
 258 #if !HAVE_EXECVEAT
 259 static inline int execveat(int dirfd, const char *pathname, char *const argv[],
 260                            char *const envp[], int flags)
 261 {
 262         return syscall(__NR_execveat, dirfd, pathname, argv, envp, flags);
 263 }
 264 #else
 265 extern int execveat(int dirfd, const char *pathname, char *const argv[],
 266                     char *const envp[], int flags);
 267 #endif
 268
 269 /*
 270  * Because of older glibc's pid cache (up to 2.25) whenever clone() is called
 271  * the child must must retrieve it's own pid via lxc_raw_getpid().
 272  */
 273 static inline pid_t lxc_raw_getpid(void)
 274 {
 275         return (pid_t)syscall(SYS_getpid);
 276 }
 277
 278 static inline pid_t lxc_raw_gettid(void)
 279 {
 280 #if __NR_gettid > 0
 281         return syscall(__NR_gettid);
 282 #else
 283         return lxc_raw_getpid();
 284 #endif
 285 }
 286
 287 __hidden extern int lxc_raw_pidfd_send_signal(int pidfd, int sig, siginfo_t *info,
 288                                               unsigned int flags);
 289
 290 static inline const char *signal_name(int sig)
 291 {
 292         const char *s;
 293
 294 #if HAVE_SIGDESCR_NP
 295         s = sigdescr_np(sig);
 296 #else
 297         s = "UNSUPPORTED";
 298 #endif
 299         return s ?: "INVALID_SIGNAL_NUMBER";
 300 }
 301
 302 #endif /* __LXC_PROCESS_UTILS_H */