#include <signal.h>
#include <sched.h>
#include <unistd.h>
-#include <attr/xattr.h>
#include <sys/types.h>
#include <sys/syscall.h>
#include <sys/mount.h>
#include <sys/signalfd.h>
#include <grp.h>
#include <linux/fs.h>
+#include <linux/netlink.h>
#include <sys/un.h>
#include <sys/socket.h>
#include <systemd/sd-daemon.h>
+#include <systemd/sd-bus.h>
#include "log.h"
#include "util.h"
#include "fdset.h"
#include "build.h"
#include "fileio.h"
+#include "bus-internal.h"
+#include "bus-message.h"
#ifndef TTY_GID
#define TTY_GID 5
static char *arg_directory = NULL;
static char *arg_user = NULL;
-static char **arg_controllers = NULL;
-static char *arg_uuid = NULL;
+static sd_id128_t arg_uuid = {};
static char *arg_machine = NULL;
+static const char *arg_slice = NULL;
static bool arg_private_network = false;
static bool arg_read_only = false;
static bool arg_boot = false;
" -D --directory=NAME Root directory for the container\n"
" -b --boot Boot up full system (i.e. invoke init)\n"
" -u --user=USER Run the command under specified user or uid\n"
- " -C --controllers=LIST Put the container in specified comma-separated\n"
- " cgroup hierarchies\n"
" --uuid=UUID Set a specific machine UUID for the container\n"
" -M --machine=NAME Set the machine name for the container\n"
+ " -S --slice=SLICE Place the container in the specified slice\n"
" --private-network Disable network in container\n"
" --read-only Mount the root directory read-only\n"
" --capability=CAP In addition to the default, retain specified\n"
{ "version", no_argument, NULL, ARG_VERSION },
{ "directory", required_argument, NULL, 'D' },
{ "user", required_argument, NULL, 'u' },
- { "controllers", required_argument, NULL, 'C' },
{ "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
{ "boot", no_argument, NULL, 'b' },
{ "uuid", required_argument, NULL, ARG_UUID },
{ "bind", required_argument, NULL, ARG_BIND },
{ "bind-ro", required_argument, NULL, ARG_BIND_RO },
{ "machine", required_argument, NULL, 'M' },
+ { "slice", required_argument, NULL, 'S' },
{ NULL, 0, NULL, 0 }
};
- int c;
+ int c, r;
assert(argc >= 0);
assert(argv);
- while ((c = getopt_long(argc, argv, "+hD:u:C:bM:j", options, NULL)) >= 0) {
+ while ((c = getopt_long(argc, argv, "+hD:u:bM:jS:", options, NULL)) >= 0) {
switch (c) {
break;
- case 'C':
- strv_free(arg_controllers);
- arg_controllers = strv_split(optarg, ",");
- if (!arg_controllers)
- return log_oom();
-
- cg_shorten_controllers(arg_controllers);
- break;
-
case ARG_PRIVATE_NETWORK:
arg_private_network = true;
break;
break;
case ARG_UUID:
- if (!id128_is_valid(optarg)) {
+ r = sd_id128_from_string(optarg, &arg_uuid);
+ if (r < 0) {
log_error("Invalid UUID: %s", optarg);
- return -EINVAL;
+ return r;
}
+ break;
- arg_uuid = optarg;
+ case 'S':
+ arg_slice = strdup(optarg);
break;
case 'M':
_cleanup_free_ char *a = NULL, *b = NULL;
char *e;
char ***x;
- int r;
x = c == ARG_BIND ? &arg_bind : &arg_bind_ro;
return 0;
}
-static int setup_cgroup(const char *path) {
- char **c;
- int r;
-
- r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, path, 1);
- if (r < 0) {
- log_error("Failed to create cgroup: %s", strerror(-r));
- return r;
- }
-
- STRV_FOREACH(c, arg_controllers) {
- r = cg_create_and_attach(*c, path, 1);
- if (r < 0)
- log_warning("Failed to create cgroup in controller %s: %s", *c, strerror(-r));
- }
-
- return 0;
-}
-
-static int save_attributes(const char *cgroup, pid_t pid, const char *uuid, const char *directory) {
- _cleanup_free_ char *path = NULL;
- char buf[DECIMAL_STR_MAX(pid_t)];
- int r = 0, k;
-
- assert(cgroup);
- assert(pid >= 0);
- assert(arg_directory);
-
-#ifdef HAVE_XATTR
- assert_se(snprintf(buf, sizeof(buf), "%lu", (unsigned long) pid) < (int) sizeof(buf));
-
- r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, cgroup, NULL, &path);
- if (r < 0) {
- log_error("Failed to get path: %s", strerror(-r));
- return r;
- }
-
- r = setxattr(path, "trusted.init_pid", buf, strlen(buf), XATTR_CREATE);
- if (r < 0)
- log_warning("Failed to set %s attribute on %s: %m", "trusted.init_pid", path);
-
- if (uuid) {
- k = setxattr(path, "trusted.machine_id", uuid, strlen(uuid), XATTR_CREATE);
- if (k < 0) {
- log_warning("Failed to set %s attribute on %s: %m", "trusted.machine_id", path);
- if (r == 0)
- r = k;
- }
- }
-
- k = setxattr(path, "trusted.root_directory", directory, strlen(directory), XATTR_CREATE);
- if (k < 0) {
- log_warning("Failed to set %s attribute on %s: %m", "trusted.root_directory", path);
- if (r == 0)
- r = k;
- }
-#endif
- return r;
-}
-
static int drop_capabilities(void) {
return capability_bounding_set_drop(~arg_retain, false);
}
return r;
}
+static int register_machine(void) {
+ _cleanup_bus_error_free_ sd_bus_error error = SD_BUS_ERROR_NULL;
+ _cleanup_bus_unref_ sd_bus *bus = NULL;
+ int r;
+
+ r = sd_bus_open_system(&bus);
+ if (r < 0) {
+ log_error("Failed to open system bus: %s", strerror(-r));
+ return r;
+ }
+
+ r = sd_bus_call_method(
+ bus,
+ "org.freedesktop.machine1",
+ "/org/freedesktop/machine1",
+ "org.freedesktop.machine1.Manager",
+ "CreateMachine",
+ &error,
+ NULL,
+ "sayssusa(sv)",
+ arg_machine,
+ SD_BUS_APPEND_ID128(arg_uuid),
+ "nspawn",
+ "container",
+ (uint32_t) 0,
+ strempty(arg_directory),
+ 1, "Slice", "s", strempty(arg_slice));
+ if (r < 0) {
+ log_error("Failed to register machine: %s", error.message ? error.message : strerror(-r));
+ return r;
+ }
+
+ return 0;
+}
+
+static bool audit_enabled(void) {
+ int fd;
+
+ fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_AUDIT);
+ if (fd >= 0) {
+ close_nointr_nofail(fd);
+ return true;
+ }
+ return false;
+}
+
int main(int argc, char *argv[]) {
pid_t pid = 0;
int r = EXIT_FAILURE, k;
- _cleanup_free_ char *newcg = NULL;
_cleanup_close_ int master = -1;
int n_fd_passed;
const char *console = NULL;
goto finish;
}
- hostname_cleanup(arg_machine);
+ hostname_cleanup(arg_machine, false);
if (isempty(arg_machine)) {
log_error("Failed to determine machine name automatically, please use -M.");
goto finish;
goto finish;
}
+ if (arg_boot && audit_enabled()) {
+ log_warning("The kernel auditing subsystem is known to be incompatible with containers.\n"
+ "Please make sure to turn off auditing with 'audit=0' on the kernel command\n"
+ "line before using systemd-nspawn. Sleeping for 5s...\n");
+ sleep(5);
+ }
+
if (path_equal(arg_directory, "/")) {
log_error("Spawning container on root directory not supported.");
goto finish;
}
if (path_is_os_tree(arg_directory) <= 0) {
- log_error("Directory %s doesn't look like an OS root directory. Refusing.", arg_directory);
+ log_error("Directory %s doesn't look like an OS root directory (/etc/os-release is missing). Refusing.", arg_directory);
goto finish;
}
fdset_close_others(fds);
log_open();
- k = cg_get_machine_path(arg_machine, &newcg);
- if (k < 0) {
- log_error("Failed to determine machine cgroup path: %s", strerror(-k));
- goto finish;
- }
-
- k = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, newcg, true);
- if (k <= 0 && k != -ENOENT) {
- log_error("Container already running.");
-
- free(newcg);
- newcg = NULL;
-
- goto finish;
- }
-
master = posix_openpt(O_RDWR|O_NOCTTY|O_CLOEXEC|O_NDELAY);
if (master < 0) {
log_error("Failed to acquire pseudo tty: %m");
goto child_fail;
}
- if (setup_cgroup(newcg) < 0)
- goto child_fail;
-
close_pipe(pipefd2);
+ r = register_machine();
+ if (r < 0)
+ goto finish;
+
/* Mark everything as slave, so that we still
* receive mounts from the real root, but don't
* propagate mounts to the real root. */
goto child_fail;
}
- if (arg_uuid) {
- if (asprintf((char**)(envp + n_env++), "container_uuid=%s", arg_uuid) < 0) {
+ if (!sd_id128_equal(arg_uuid, SD_ID128_NULL)) {
+ if (asprintf((char**)(envp + n_env++), "container_uuid=" SD_ID128_FORMAT_STR, SD_ID128_FORMAT_VAL(arg_uuid)) < 0) {
log_oom();
goto child_fail;
}
fd_wait_for_event(pipefd2[0], POLLHUP, -1);
close_nointr_nofail(pipefd2[0]);
- save_attributes(newcg, pid, arg_uuid, arg_directory);
-
fdset_free(fds);
fds = NULL;
}
if (status.si_code == CLD_EXITED) {
+ r = status.si_status;
if (status.si_status != 0) {
log_error("Container failed with error code %i.", status.si_status);
- r = status.si_status;
break;
}
close_pipe(kmsg_socket_pair);
- if (newcg)
- cg_kill_recursive_and_wait(SYSTEMD_CGROUP_CONTROLLER, newcg, true);
+ if (pid > 0)
+ kill(pid, SIGKILL);
free(arg_directory);
free(arg_machine);
- strv_free(arg_controllers);
fdset_free(fds);
~ianmdlvl / elogind.git / blobdiff