src/nspawn/nspawn-stub-pid1.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <sys/reboot.h>
   4 #include <sys/wait.h>
   5 #include <sys/prctl.h>
   6 #include <unistd.h>
   7
   8 #include "def.h"
   9 #include "exit-status.h"
  10 #include "fd-util.h"
  11 #include "log.h"
  12 #include "missing.h"
  13 #include "nspawn-stub-pid1.h"
  14 #include "process-util.h"
  15 #include "signal-util.h"
  16 #include "time-util.h"
  17
  18 static int reset_environ(const char *new_environment, size_t length) {
  19         unsigned long start, end;
  20
  21         start = (unsigned long) new_environment;
  22         end = start + length;
  23
  24         if (prctl(PR_SET_MM, PR_SET_MM_ENV_START, start, 0, 0) < 0)
  25                 return -errno;
  26
  27         if (prctl(PR_SET_MM, PR_SET_MM_ENV_END, end, 0, 0) < 0)
  28                 return -errno;
  29
  30         return 0;
  31 }
  32
  33 int stub_pid1(sd_id128_t uuid) {
  34         enum {
  35                 STATE_RUNNING,
  36                 STATE_REBOOT,
  37                 STATE_POWEROFF,
  38         } state = STATE_RUNNING;
  39
  40         sigset_t fullmask, oldmask, waitmask;
  41         usec_t quit_usec = USEC_INFINITY;
  42         pid_t pid;
  43         int r;
  44
  45         /* The new environment we set up, on the stack. */
  46         char new_environment[] =
  47                 "container=systemd-nspawn\0"
  48                 "container_uuid=XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX";
  49
  50         /* Implements a stub PID 1, that reaps all processes and processes a couple of standard signals. This is useful
  51          * for allowing arbitrary processes run in a container, and still have all zombies reaped. */
  52
  53         assert_se(sigfillset(&fullmask) >= 0);
  54         assert_se(sigprocmask(SIG_BLOCK, &fullmask, &oldmask) >= 0);
  55
  56         pid = fork();
  57         if (pid < 0)
  58                 return log_error_errno(errno, "Failed to fork child pid: %m");
  59
  60         if (pid == 0) {
  61                 /* Return in the child */
  62                 assert_se(sigprocmask(SIG_SETMASK, &oldmask, NULL) >= 0);
  63                 setsid();
  64                 return 0;
  65         }
  66
  67         reset_all_signal_handlers();
  68
  69         log_close();
  70         close_all_fds(NULL, 0);
  71         log_open();
  72
  73         /* Flush out /proc/self/environ, so that we don't leak the environment from the host into the container. Also,
  74          * set $container= and $container_uuid= so that clients in the container that query it from /proc/1/environ
  75          * find them set. */
  76         sd_id128_to_string(uuid, new_environment + sizeof(new_environment) - SD_ID128_STRING_MAX);
  77         reset_environ(new_environment, sizeof(new_environment));
  78
  79         (void) rename_process("(sd-stubinit)");
  80
  81         assert_se(sigemptyset(&waitmask) >= 0);
  82         assert_se(sigset_add_many(&waitmask,
  83                                   SIGCHLD,          /* posix: process died */
  84                                   SIGINT,           /* sysv: ctrl-alt-del */
  85                                   SIGRTMIN+3,       /* systemd: halt */
  86                                   SIGRTMIN+4,       /* systemd: poweroff */
  87                                   SIGRTMIN+5,       /* systemd: reboot */
  88                                   SIGRTMIN+6,       /* systemd: kexec */
  89                                   SIGRTMIN+13,      /* systemd: halt */
  90                                   SIGRTMIN+14,      /* systemd: poweroff */
  91                                   SIGRTMIN+15,      /* systemd: reboot */
  92                                   SIGRTMIN+16,      /* systemd: kexec */
  93                                   -1) >= 0);
  94
  95         /* Note that we ignore SIGTERM (sysv's reexec), SIGHUP (reload), and all other signals here, since we don't
  96          * support reexec/reloading in this stub process. */
  97
  98         for (;;) {
  99                 siginfo_t si;
 100                 usec_t current_usec;
 101
 102                 si.si_pid = 0;
 103                 r = waitid(P_ALL, 0, &si, WEXITED|WNOHANG);
 104                 if (r < 0) {
 105                         r = log_error_errno(errno, "Failed to reap children: %m");
 106                         goto finish;
 107                 }
 108
 109                 current_usec = now(CLOCK_MONOTONIC);
 110
 111                 if (si.si_pid == pid || current_usec >= quit_usec) {
 112
 113                         /* The child we started ourselves died or we reached a timeout. */
 114
 115                         if (state == STATE_REBOOT) { /* dispatch a queued reboot */
 116                                 (void) reboot(RB_AUTOBOOT);
 117                                 r = log_error_errno(errno, "Failed to reboot: %m");
 118                                 goto finish;
 119
 120                         } else if (state == STATE_POWEROFF)
 121                                 (void) reboot(RB_POWER_OFF); /* if this fails, fall back to normal exit. */
 122
 123                         if (si.si_pid == pid && si.si_code == CLD_EXITED)
 124                                 r = si.si_status; /* pass on exit code */
 125                         else
 126                                 r = EXIT_EXCEPTION; /* signal, coredump, timeout, … */
 127
 128                         goto finish;
 129                 }
 130                 if (si.si_pid != 0)
 131                         /* We reaped something. Retry until there's nothing more to reap. */
 132                         continue;
 133
 134                 if (quit_usec == USEC_INFINITY)
 135                         r = sigwaitinfo(&waitmask, &si);
 136                 else {
 137                         struct timespec ts;
 138                         r = sigtimedwait(&waitmask, &si, timespec_store(&ts, quit_usec - current_usec));
 139                 }
 140                 if (r < 0) {
 141                         if (errno == EINTR) /* strace -p attach can result in EINTR, let's handle this nicely. */
 142                                 continue;
 143                         if (errno == EAGAIN) /* timeout reached */
 144                                 continue;
 145
 146                         r = log_error_errno(errno, "Failed to wait for signal: %m");
 147                         goto finish;
 148                 }
 149
 150                 if (si.si_signo == SIGCHLD)
 151                         continue; /* Let's reap this */
 152
 153                 if (state != STATE_RUNNING)
 154                         continue;
 155
 156                 /* Would love to use a switch() statement here, but SIGRTMIN is actually a function call, not a
 157                  * constant… */
 158
 159                 if (si.si_signo == SIGRTMIN+3 ||
 160                     si.si_signo == SIGRTMIN+4 ||
 161                     si.si_signo == SIGRTMIN+13 ||
 162                     si.si_signo == SIGRTMIN+14)
 163
 164                         state = STATE_POWEROFF;
 165
 166                 else if (si.si_signo == SIGINT ||
 167                          si.si_signo == SIGRTMIN+5 ||
 168                          si.si_signo == SIGRTMIN+6 ||
 169                          si.si_signo == SIGRTMIN+15 ||
 170                          si.si_signo == SIGRTMIN+16)
 171
 172                         state = STATE_REBOOT;
 173                 else
 174                         assert_not_reached("Got unexpected signal");
 175
 176                 r = kill_and_sigcont(pid, SIGTERM);
 177
 178                 /* Let's send a SIGHUP after the SIGTERM, as shells tend to ignore SIGTERM but do react to SIGHUP. We
 179                  * do it strictly in this order, so that the SIGTERM is dispatched first, and SIGHUP second for those
 180                  * processes which handle both. That's because services tend to bind configuration reload or something
 181                  * else to SIGHUP. */
 182
 183                 if (r != -ESRCH)
 184                         (void) kill(pid, SIGHUP);
 185
 186                 quit_usec = now(CLOCK_MONOTONIC) + DEFAULT_TIMEOUT_USEC;
 187         }
 188
 189 finish:
 190         _exit(r < 0 ? EXIT_FAILURE : r);
 191 }