src/nspawn/nspawn-stub-pid1.c

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