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 "compiler.h"
  13 #include "config.h"
  14 #include "macro.h"
  15 #include "raw_syscalls.h"
  16
  17 int lxc_raw_execveat(int dirfd, const char *pathname, char *const argv[],
  18                      char *const envp[], int flags)
  19 {
  20 #ifdef __NR_execveat
  21         syscall(__NR_execveat, dirfd, pathname, argv, envp, flags);
  22 #else
  23         errno = ENOSYS;
  24 #endif
  25         return -1;
  26 }
  27
  28 /*
  29  * This is based on raw_clone in systemd but adapted to our needs. This uses
  30  * copy on write semantics and doesn't pass a stack. CLONE_VM is tricky and
  31  * doesn't really matter to us so disallow it.
  32  *
  33  * The nice thing about this is that we get fork() behavior. That is
  34  * lxc_raw_clone() returns 0 in the child and the child pid in the parent.
  35  */
  36 __returns_twice pid_t lxc_raw_clone(unsigned long flags, int *pidfd)
  37 {
  38         /*
  39          * These flags don't interest at all so we don't jump through any hoops
  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 syscall(__NR_clone, NULL, flags | SIGCHLD, pidfd);
  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                 register long g1 asm("g1") = __NR_clone;
  60                 register long o0 asm("o0") = flags | SIGCHLD;
  61                 register long o1 asm("o1") = 0; /* is parent/child indicator */
  62                 register long o2 asm("o2") = (unsigned long)pidfd;
  63                 long is_error, retval, in_child;
  64                 pid_t child_pid;
  65
  66                 asm volatile(
  67 #if defined(__arch64__)
  68                     "t 0x6d\n\t" /* 64-bit trap */
  69 #else
  70                     "t 0x10\n\t" /* 32-bit trap */
  71 #endif
  72                     /*
  73                      * catch errors: On sparc, the carry bit (csr) in the
  74                      * processor status register (psr) is used instead of a
  75                      * full register.
  76                      */
  77                     "addx %%g0, 0, %g1"
  78                     : "=r"(g1), "=r"(o0), "=r"(o1), "=r"(o2) /* outputs */
  79                     : "r"(g1), "r"(o0), "r"(o1), "r"(o2)     /* inputs */
  80                     : "%cc");                                /* clobbers */
  81
  82                 is_error = g1;
  83                 retval = o0;
  84                 in_child = o1;
  85
  86                 if (is_error) {
  87                         errno = retval;
  88                         return -1;
  89                 }
  90
  91                 if (in_child)
  92                         return 0;
  93
  94                 child_pid = retval;
  95                 return child_pid;
  96         }
  97 #elif defined(__ia64__)
  98         /* On ia64 the stack and stack size are passed as separate arguments. */
  99         return syscall(__NR_clone, flags | SIGCHLD, NULL, prctl_arg(0), pidfd);
 100 #else
 101         return syscall(__NR_clone, flags | SIGCHLD, NULL, pidfd);
 102 #endif
 103 }
 104
 105 pid_t lxc_raw_clone_cb(int (*fn)(void *), void *args, unsigned long flags,
 106                        int *pidfd)
 107 {
 108         pid_t pid;
 109
 110         pid = lxc_raw_clone(flags, pidfd);
 111         if (pid < 0)
 112                 return -1;
 113
 114         /*
 115          * exit() is not thread-safe and might mess with the parent's signal
 116          * handlers and other stuff when exec() fails.
 117          */
 118         if (pid == 0)
 119                 _exit(fn(args));
 120
 121         return pid;
 122 }
 123
 124 int lxc_raw_pidfd_send_signal(int pidfd, int sig, siginfo_t *info,
 125                               unsigned int flags)
 126 {
 127 #ifdef __NR_pidfd_send_signal
 128         syscall(__NR_pidfd_send_signal, pidfd, sig, info, flags);
 129 #else
 130         errno = ENOSYS;
 131 #endif
 132         return -1;
 133 }