src/lxc/raw_syscalls.c

   1 #ifndef _GNU_SOURCE
   2 #define _GNU_SOURCE 1
   3 #endif
   4 #include <errno.h>
   5 #include <sched.h>
   6 #include <signal.h>
   7 #include <stdio.h>
   8 #include <stdlib.h>
   9 #include <sys/syscall.h>
  10 #include <unistd.h>
  11
  12 #include "config.h"
  13 #include "macro.h"
  14 #include "raw_syscalls.h"
  15
  16 int lxc_raw_execveat(int dirfd, const char *pathname, char *const argv[],
  17                      char *const envp[], int flags)
  18 {
  19 #ifdef __NR_execveat
  20         syscall(__NR_execveat, dirfd, pathname, argv, envp, flags);
  21 #else
  22         errno = ENOSYS;
  23         return -1;
  24 #endif
  25 }
  26
  27 /*
  28  * This is based on raw_clone in systemd but adapted to our needs. This uses
  29  * copy on write semantics and doesn't pass a stack. CLONE_VM is tricky and
  30  * doesn't really matter to us so disallow it.
  31  *
  32  * The nice thing about this is that we get fork() behavior. That is
  33  * lxc_raw_clone() returns 0 in the child and the child pid in the parent.
  34  */
  35 pid_t lxc_raw_clone(unsigned long flags)
  36 {
  37
  38         /*
  39          * These flags don't interest at all so we don't jump through any hoopes
  40          * of retrieving them and passing them to the kernel.
  41          */
  42         errno = EINVAL;
  43         if ((flags & (CLONE_VM | CLONE_PARENT_SETTID | CLONE_CHILD_SETTID |
  44                       CLONE_CHILD_CLEARTID | CLONE_SETTLS)))
  45                 return -EINVAL;
  46
  47 #if defined(__s390x__) || defined(__s390__) || defined(__CRIS__)
  48         /* On s390/s390x and cris the order of the first and second arguments
  49          * of the system call is reversed.
  50          */
  51         return (int)syscall(__NR_clone, NULL, flags | SIGCHLD);
  52 #elif defined(__sparc__) && defined(__arch64__)
  53         {
  54                 /*
  55                  * sparc64 always returns the other process id in %o0, and a
  56                  * boolean flag whether this is the child or the parent in %o1.
  57                  * Inline assembly is needed to get the flag returned in %o1.
  58                  */
  59                 int in_child;
  60                 int child_pid;
  61                 asm volatile("mov %2, %%g1\n\t"
  62                              "mov %3, %%o0\n\t"
  63                              "mov 0 , %%o1\n\t"
  64                              "t 0x6d\n\t"
  65                              "mov %%o1, %0\n\t"
  66                              "mov %%o0, %1"
  67                              : "=r"(in_child), "=r"(child_pid)
  68                              : "i"(__NR_clone), "r"(flags | SIGCHLD)
  69                              : "%o1", "%o0", "%g1");
  70
  71                 if (in_child)
  72                         return 0;
  73                 else
  74                         return child_pid;
  75         }
  76 #elif defined(__ia64__)
  77         /* On ia64 the stack and stack size are passed as separate arguments. */
  78         return (int)syscall(__NR_clone, flags | SIGCHLD, NULL, prctl_arg(0));
  79 #else
  80         return (int)syscall(__NR_clone, flags | SIGCHLD, NULL);
  81 #endif
  82 }
  83
  84 pid_t lxc_raw_clone_cb(int (*fn)(void *), void *args, unsigned long flags)
  85 {
  86         pid_t pid;
  87
  88         pid = lxc_raw_clone(flags);
  89         if (pid < 0)
  90                 return -1;
  91
  92         /*
  93          * exit() is not thread-safe and might mess with the parent's signal
  94          * handlers and other stuff when exec() fails.
  95          */
  96         if (pid == 0)
  97                 _exit(fn(args));
  98
  99         return pid;
 100 }