TarPull* tar_pull_unref(TarPull *i) {
if (!i)
return NULL;
if (i->tar_pid > 1) {
(void) kill_and_sigcont(i->tar_pid, SIGKILL);
(void) wait_for_terminate(i->tar_pid, NULL);
}
pull_job_unref(i->tar_job);
pull_job_unref(i->settings_job);
pull_job_unref(i->checksum_job);
pull_job_unref(i->signature_job);
curl_glue_unref(i->glue);
sd_event_unref(i->event);
if (i->temp_path) {
(void) rm_rf(i->temp_path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
free(i->temp_path);
}
if (i->settings_temp_path) {
(void) unlink(i->settings_temp_path);
free(i->settings_temp_path);
}
free(i->final_path);
free(i->settings_path);
free(i->image_root);
free(i->local);
return mfree(i);
}
TarImport* tar_import_unref(TarImport *i) {
if (!i)
return NULL;
sd_event_source_unref(i->input_event_source);
if (i->tar_pid > 1) {
(void) kill_and_sigcont(i->tar_pid, SIGKILL);
(void) wait_for_terminate(i->tar_pid, NULL);
}
if (i->temp_path) {
(void) rm_rf(i->temp_path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
free(i->temp_path);
}
import_compress_free(&i->compress);
sd_event_unref(i->event);
safe_close(i->tar_fd);
free(i->final_path);
free(i->image_root);
free(i->local);
free(i);
return NULL;
}
TarExport *tar_export_unref(TarExport *e) {
if (!e)
return NULL;
sd_event_source_unref(e->output_event_source);
if (e->tar_pid > 1) {
(void) kill_and_sigcont(e->tar_pid, SIGKILL);
(void) wait_for_terminate(e->tar_pid, NULL);
}
if (e->temp_path) {
(void) btrfs_subvol_remove(e->temp_path, BTRFS_REMOVE_QUOTA);
free(e->temp_path);
}
import_compress_free(&e->compress);
sd_event_unref(e->event);
safe_close(e->tar_fd);
free(e->buffer);
free(e->path);
free(e);
return NULL;
}
TarImport* tar_import_unref(TarImport *i) {
if (!i)
return NULL;
if (i->tar_pid > 1) {
(void) kill_and_sigcont(i->tar_pid, SIGKILL);
(void) wait_for_terminate(i->tar_pid, NULL);
}
import_job_unref(i->tar_job);
import_job_unref(i->checksum_job);
import_job_unref(i->signature_job);
curl_glue_unref(i->glue);
sd_event_unref(i->event);
if (i->temp_path) {
(void) btrfs_subvol_remove(i->temp_path);
(void) rm_rf_dangerous(i->temp_path, false, true, false);
free(i->temp_path);
}
free(i->final_path);
free(i->image_root);
free(i->local);
free(i);
return NULL;
}
DkrPull* dkr_pull_unref(DkrPull *i) {
if (!i)
return NULL;
if (i->tar_pid > 1) {
(void) kill_and_sigcont(i->tar_pid, SIGKILL);
(void) wait_for_terminate(i->tar_pid, NULL);
}
pull_job_unref(i->images_job);
pull_job_unref(i->tags_job);
pull_job_unref(i->ancestry_job);
pull_job_unref(i->json_job);
pull_job_unref(i->layer_job);
curl_glue_unref(i->glue);
sd_event_unref(i->event);
if (i->temp_path) {
(void) rm_rf(i->temp_path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
free(i->temp_path);
}
free(i->name);
free(i->reference);
free(i->id);
free(i->response_token);
free(i->response_registries);
strv_free(i->ancestry);
free(i->final_path);
free(i->index_url);
free(i->image_root);
free(i->local);
free(i);
return NULL;
}
static int setup_machine_raw(uint64_t size, sd_bus_error *error) {
_cleanup_free_ char *tmp = NULL;
_cleanup_close_ int fd = -1;
struct statvfs ss;
pid_t pid = 0;
siginfo_t si;
int r;
/* We want to be able to make use of btrfs-specific file
* system features, in particular subvolumes, reflinks and
* quota. Hence, if we detect that /var/lib/machines.raw is
* not located on btrfs, let's create a loopback file, place a
* btrfs file system into it, and mount it to
* /var/lib/machines. */
fd = open("/var/lib/machines.raw", O_RDWR|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
if (fd >= 0) {
r = fd;
fd = -1;
return r;
}
if (errno != ENOENT)
return sd_bus_error_set_errnof(error, errno, "Failed to open /var/lib/machines.raw: %m");
r = tempfn_xxxxxx("/var/lib/machines.raw", NULL, &tmp);
if (r < 0)
return r;
(void) mkdir_p_label("/var/lib", 0755);
fd = open(tmp, O_RDWR|O_CREAT|O_EXCL|O_NOCTTY|O_CLOEXEC, 0600);
if (fd < 0)
return sd_bus_error_set_errnof(error, errno, "Failed to create /var/lib/machines.raw: %m");
if (fstatvfs(fd, &ss) < 0) {
r = sd_bus_error_set_errnof(error, errno, "Failed to determine free space on /var/lib/machines.raw: %m");
goto fail;
}
if (ss.f_bsize * ss.f_bavail < VAR_LIB_MACHINES_FREE_MIN) {
r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "Not enough free disk space to set up /var/lib/machines.");
goto fail;
}
if (ftruncate(fd, size) < 0) {
r = sd_bus_error_set_errnof(error, errno, "Failed to enlarge /var/lib/machines.raw: %m");
goto fail;
}
pid = fork();
if (pid < 0) {
r = sd_bus_error_set_errnof(error, errno, "Failed to fork mkfs.btrfs: %m");
goto fail;
}
if (pid == 0) {
/* Child */
(void) reset_all_signal_handlers();
(void) reset_signal_mask();
assert_se(prctl(PR_SET_PDEATHSIG, SIGTERM) == 0);
fd = safe_close(fd);
execlp("mkfs.btrfs", "-Lvar-lib-machines", tmp, NULL);
if (errno == ENOENT)
return 99;
_exit(EXIT_FAILURE);
}
r = wait_for_terminate(pid, &si);
if (r < 0) {
sd_bus_error_set_errnof(error, r, "Failed to wait for mkfs.btrfs: %m");
goto fail;
}
pid = 0;
if (si.si_code != CLD_EXITED) {
r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "mkfs.btrfs died abnormally.");
goto fail;
}
if (si.si_status == 99) {
r = sd_bus_error_set_errnof(error, ENOENT, "Cannot set up /var/lib/machines, mkfs.btrfs is missing");
goto fail;
}
if (si.si_status != 0) {
r = sd_bus_error_setf(error, SD_BUS_ERROR_FAILED, "mkfs.btrfs failed with error code %i", si.si_status);
goto fail;
}
r = rename_noreplace(AT_FDCWD, tmp, AT_FDCWD, "/var/lib/machines.raw");
if (r < 0) {
sd_bus_error_set_errnof(error, r, "Failed to move /var/lib/machines.raw into place: %m");
goto fail;
}
r = fd;
fd = -1;
return r;
fail:
unlink_noerrno(tmp);
if (pid > 1)
kill_and_sigcont(pid, SIGKILL);
return r;
}
int stub_pid1(sd_id128_t uuid) {
enum {
STATE_RUNNING,
STATE_REBOOT,
STATE_POWEROFF,
} state = STATE_RUNNING;
sigset_t fullmask, oldmask, waitmask;
usec_t quit_usec = USEC_INFINITY;
pid_t pid;
int r;
/* The new environment we set up, on the stack. */
char new_environment[] =
"container=systemd-nspawn\0"
"container_uuid=XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX";
/* Implements a stub PID 1, that reaps all processes and processes a couple of standard signals. This is useful
* for allowing arbitrary processes run in a container, and still have all zombies reaped. */
assert_se(sigfillset(&fullmask) >= 0);
assert_se(sigprocmask(SIG_BLOCK, &fullmask, &oldmask) >= 0);
pid = fork();
if (pid < 0)
return log_error_errno(errno, "Failed to fork child pid: %m");
if (pid == 0) {
/* Return in the child */
assert_se(sigprocmask(SIG_SETMASK, &oldmask, NULL) >= 0);
setsid();
return 0;
}
reset_all_signal_handlers();
log_close();
close_all_fds(NULL, 0);
log_open();
/* Flush out /proc/self/environ, so that we don't leak the environment from the host into the container. Also,
* set $container= and $container_uuid= so that clients in the container that query it from /proc/1/environ
* find them set. */
sd_id128_to_string(uuid, new_environment + sizeof(new_environment) - SD_ID128_STRING_MAX);
reset_environ(new_environment, sizeof(new_environment));
rename_process("STUBINIT");
assert_se(sigemptyset(&waitmask) >= 0);
assert_se(sigset_add_many(&waitmask,
SIGCHLD, /* posix: process died */
SIGINT, /* sysv: ctrl-alt-del */
SIGRTMIN+3, /* systemd: halt */
SIGRTMIN+4, /* systemd: poweroff */
SIGRTMIN+5, /* systemd: reboot */
SIGRTMIN+6, /* systemd: kexec */
SIGRTMIN+13, /* systemd: halt */
SIGRTMIN+14, /* systemd: poweroff */
SIGRTMIN+15, /* systemd: reboot */
SIGRTMIN+16, /* systemd: kexec */
-1) >= 0);
/* Note that we ignore SIGTERM (sysv's reexec), SIGHUP (reload), and all other signals here, since we don't
* support reexec/reloading in this stub process. */
for (;;) {
siginfo_t si;
usec_t current_usec;
si.si_pid = 0;
r = waitid(P_ALL, 0, &si, WEXITED|WNOHANG);
if (r < 0) {
r = log_error_errno(errno, "Failed to reap children: %m");
goto finish;
}
current_usec = now(CLOCK_MONOTONIC);
if (si.si_pid == pid || current_usec >= quit_usec) {
/* The child we started ourselves died or we reached a timeout. */
if (state == STATE_REBOOT) { /* dispatch a queued reboot */
(void) reboot(RB_AUTOBOOT);
r = log_error_errno(errno, "Failed to reboot: %m");
goto finish;
} else if (state == STATE_POWEROFF)
(void) reboot(RB_POWER_OFF); /* if this fails, fall back to normal exit. */
if (si.si_pid == pid && si.si_code == CLD_EXITED)
r = si.si_status; /* pass on exit code */
else
r = 255; /* signal, coredump, timeout, … */
goto finish;
}
if (si.si_pid != 0)
/* We reaped something. Retry until there's nothing more to reap. */
continue;
if (quit_usec == USEC_INFINITY)
r = sigwaitinfo(&waitmask, &si);
else {
struct timespec ts;
r = sigtimedwait(&waitmask, &si, timespec_store(&ts, quit_usec - current_usec));
}
if (r < 0) {
if (errno == EINTR) /* strace -p attach can result in EINTR, let's handle this nicely. */
continue;
if (errno == EAGAIN) /* timeout reached */
continue;
r = log_error_errno(errno, "Failed to wait for signal: %m");
goto finish;
}
if (si.si_signo == SIGCHLD)
continue; /* Let's reap this */
if (state != STATE_RUNNING)
continue;
/* Would love to use a switch() statement here, but SIGRTMIN is actually a function call, not a
* constant… */
if (si.si_signo == SIGRTMIN+3 ||
si.si_signo == SIGRTMIN+4 ||
si.si_signo == SIGRTMIN+13 ||
si.si_signo == SIGRTMIN+14)
state = STATE_POWEROFF;
else if (si.si_signo == SIGINT ||
si.si_signo == SIGRTMIN+5 ||
si.si_signo == SIGRTMIN+6 ||
si.si_signo == SIGRTMIN+15 ||
si.si_signo == SIGRTMIN+16)
state = STATE_REBOOT;
else
assert_not_reached("Got unexpected signal");
r = kill_and_sigcont(pid, SIGTERM);
/* Let's send a SIGHUP after the SIGTERM, as shells tend to ignore SIGTERM but do react to SIGHUP. We
* do it strictly in this order, so that the SIGTERM is dispatched first, and SIGHUP second for those
* processes which handle both. That's because services tend to bind configuration reload or something
* else to SIGHUP. */
if (r != -ESRCH)
(void) kill(pid, SIGHUP);
quit_usec = now(CLOCK_MONOTONIC) + DEFAULT_TIMEOUT_USEC;
}
finish:
_exit(r < 0 ? EXIT_FAILURE : r);
}
