src/nspawn/nspawn-stub-pid1.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2 /***
   3   This file is part of systemd.
   4
   5   Copyright 2016 Lennart Poettering
   6 ***/
   7
   8 #include <sys/reboot.h>
   9 #include <sys/wait.h>
  10 #include <sys/prctl.h>
  11 #include <unistd.h>
  12
  13 #include "fd-util.h"
  14 #include "log.h"
  15 #include "missing.h"
  16 #include "nspawn-stub-pid1.h"
  17 #include "process-util.h"
  18 #include "signal-util.h"
  19 #include "time-util.h"
  20 #include "def.h"
  21
  22 static int reset_environ(const char *new_environment, size_t length) {
  23         unsigned long start, end;
  24
  25         start = (unsigned long) new_environment;
  26         end = start + length;
  27
  28         if (prctl(PR_SET_MM, PR_SET_MM_ENV_START, start, 0, 0) < 0)
  29                 return -errno;
  30
  31         if (prctl(PR_SET_MM, PR_SET_MM_ENV_END, end, 0, 0) < 0)
  32                 return -errno;
  33
  34         return 0;
  35 }
  36
  37 int stub_pid1(sd_id128_t uuid) {
  38         enum {
  39                 STATE_RUNNING,
  40                 STATE_REBOOT,
  41                 STATE_POWEROFF,
  42         } state = STATE_RUNNING;
  43
  44         sigset_t fullmask, oldmask, waitmask;
  45         usec_t quit_usec = USEC_INFINITY;
  46         pid_t pid;
  47         int r;
  48
  49         /* The new environment we set up, on the stack. */
  50         char new_environment[] =
  51                 "container=systemd-nspawn\0"
  52                 "container_uuid=XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX";
  53
  54         /* Implements a stub PID 1, that reaps all processes and processes a couple of standard signals. This is useful
  55          * for allowing arbitrary processes run in a container, and still have all zombies reaped. */
  56
  57         assert_se(sigfillset(&fullmask) >= 0);
  58         assert_se(sigprocmask(SIG_BLOCK, &fullmask, &oldmask) >= 0);
  59
  60         pid = fork();
  61         if (pid < 0)
  62                 return log_error_errno(errno, "Failed to fork child pid: %m");
  63
  64         if (pid == 0) {
  65                 /* Return in the child */
  66                 assert_se(sigprocmask(SIG_SETMASK, &oldmask, NULL) >= 0);
  67                 setsid();
  68                 return 0;
  69         }
  70
  71         reset_all_signal_handlers();
  72
  73         log_close();
  74         close_all_fds(NULL, 0);
  75         log_open();
  76
  77         /* Flush out /proc/self/environ, so that we don't leak the environment from the host into the container. Also,
  78          * set $container= and $container_uuid= so that clients in the container that query it from /proc/1/environ
  79          * find them set. */
  80         sd_id128_to_string(uuid, new_environment + sizeof(new_environment) - SD_ID128_STRING_MAX);
  81         reset_environ(new_environment, sizeof(new_environment));
  82
  83         (void) rename_process("(sd-stubinit)");
  84
  85         assert_se(sigemptyset(&waitmask) >= 0);
  86         assert_se(sigset_add_many(&waitmask,
  87                                   SIGCHLD,          /* posix: process died */
  88                                   SIGINT,           /* sysv: ctrl-alt-del */
  89                                   SIGRTMIN+3,       /* systemd: halt */
  90                                   SIGRTMIN+4,       /* systemd: poweroff */
  91                                   SIGRTMIN+5,       /* systemd: reboot */
  92                                   SIGRTMIN+6,       /* systemd: kexec */
  93                                   SIGRTMIN+13,      /* systemd: halt */
  94                                   SIGRTMIN+14,      /* systemd: poweroff */
  95                                   SIGRTMIN+15,      /* systemd: reboot */
  96                                   SIGRTMIN+16,      /* systemd: kexec */
  97                                   -1) >= 0);
  98
  99         /* Note that we ignore SIGTERM (sysv's reexec), SIGHUP (reload), and all other signals here, since we don't
 100          * support reexec/reloading in this stub process. */
 101
 102         for (;;) {
 103                 siginfo_t si;
 104                 usec_t current_usec;
 105
 106                 si.si_pid = 0;
 107                 r = waitid(P_ALL, 0, &si, WEXITED|WNOHANG);
 108                 if (r < 0) {
 109                         r = log_error_errno(errno, "Failed to reap children: %m");
 110                         goto finish;
 111                 }
 112
 113                 current_usec = now(CLOCK_MONOTONIC);
 114
 115                 if (si.si_pid == pid || current_usec >= quit_usec) {
 116
 117                         /* The child we started ourselves died or we reached a timeout. */
 118
 119                         if (state == STATE_REBOOT) { /* dispatch a queued reboot */
 120                                 (void) reboot(RB_AUTOBOOT);
 121                                 r = log_error_errno(errno, "Failed to reboot: %m");
 122                                 goto finish;
 123
 124                         } else if (state == STATE_POWEROFF)
 125                                 (void) reboot(RB_POWER_OFF); /* if this fails, fall back to normal exit. */
 126
 127                         if (si.si_pid == pid && si.si_code == CLD_EXITED)
 128                                 r = si.si_status; /* pass on exit code */
 129                         else
 130                                 r = 255; /* signal, coredump, timeout, … */
 131
 132                         goto finish;
 133                 }
 134                 if (si.si_pid != 0)
 135                         /* We reaped something. Retry until there's nothing more to reap. */
 136                         continue;
 137
 138                 if (quit_usec == USEC_INFINITY)
 139                         r = sigwaitinfo(&waitmask, &si);
 140                 else {
 141                         struct timespec ts;
 142                         r = sigtimedwait(&waitmask, &si, timespec_store(&ts, quit_usec - current_usec));
 143                 }
 144                 if (r < 0) {
 145                         if (errno == EINTR) /* strace -p attach can result in EINTR, let's handle this nicely. */
 146                                 continue;
 147                         if (errno == EAGAIN) /* timeout reached */
 148                                 continue;
 149
 150                         r = log_error_errno(errno, "Failed to wait for signal: %m");
 151                         goto finish;
 152                 }
 153
 154                 if (si.si_signo == SIGCHLD)
 155                         continue; /* Let's reap this */
 156
 157                 if (state != STATE_RUNNING)
 158                         continue;
 159
 160                 /* Would love to use a switch() statement here, but SIGRTMIN is actually a function call, not a
 161                  * constant… */
 162
 163                 if (si.si_signo == SIGRTMIN+3 ||
 164                     si.si_signo == SIGRTMIN+4 ||
 165                     si.si_signo == SIGRTMIN+13 ||
 166                     si.si_signo == SIGRTMIN+14)
 167
 168                         state = STATE_POWEROFF;
 169
 170                 else if (si.si_signo == SIGINT ||
 171                          si.si_signo == SIGRTMIN+5 ||
 172                          si.si_signo == SIGRTMIN+6 ||
 173                          si.si_signo == SIGRTMIN+15 ||
 174                          si.si_signo == SIGRTMIN+16)
 175
 176                         state = STATE_REBOOT;
 177                 else
 178                         assert_not_reached("Got unexpected signal");
 179
 180                 r = kill_and_sigcont(pid, SIGTERM);
 181
 182                 /* Let's send a SIGHUP after the SIGTERM, as shells tend to ignore SIGTERM but do react to SIGHUP. We
 183                  * do it strictly in this order, so that the SIGTERM is dispatched first, and SIGHUP second for those
 184                  * processes which handle both. That's because services tend to bind configuration reload or something
 185                  * else to SIGHUP. */
 186
 187                 if (r != -ESRCH)
 188                         (void) kill(pid, SIGHUP);
 189
 190                 quit_usec = now(CLOCK_MONOTONIC) + DEFAULT_TIMEOUT_USEC;
 191         }
 192
 193 finish:
 194         _exit(r < 0 ? EXIT_FAILURE : r);
 195 }