chiark / gitweb /
nspawn: add simple chroot(1) like tool to execute commands in a namespace container
authorLennart Poettering <lennart@poettering.net>
Mon, 14 Mar 2011 01:40:36 +0000 (02:40 +0100)
committerLennart Poettering <lennart@poettering.net>
Mon, 14 Mar 2011 02:12:25 +0000 (03:12 +0100)
.gitignore
Makefile.am
src/nspawn.c [new file with mode: 0644]

index ffc602a..d679f79 100644 (file)
@@ -1,3 +1,4 @@
+systemd-nspawn
 systemd-stdio-bridge
 systemd-machine-id-setup
 systemd-detect-virt
index 872bcc2..7d6cfd1 100644 (file)
@@ -115,6 +115,9 @@ bin_PROGRAMS = \
        systemd-cgls \
        systemd-stdio-bridge
 
+sbin_PROGRAMS = \
+        systemd-nspawn
+
 if HAVE_GTK
 bin_PROGRAMS += \
        systemadm \
@@ -983,6 +986,15 @@ systemd_cgls_CFLAGS = \
 systemd_cgls_LDADD = \
        libsystemd-basic.la
 
+systemd_nspawn_SOURCES = \
+       src/nspawn.c
+
+systemd_nspawn_CFLAGS = \
+       $(AM_CFLAGS)
+
+systemd_nspawn_LDADD = \
+       libsystemd-basic.la
+
 systemd_stdio_bridge_SOURCES = \
        src/bridge.c
 
diff --git a/src/nspawn.c b/src/nspawn.c
new file mode 100644 (file)
index 0000000..4e4d40e
--- /dev/null
@@ -0,0 +1,444 @@
+/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
+
+/***
+  This file is part of systemd.
+
+  Copyright 2010 Lennart Poettering
+
+  systemd is free software; you can redistribute it and/or modify it
+  under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  systemd is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with systemd; If not, see <http://www.gnu.org/licenses/>.
+***/
+
+#include <signal.h>
+#include <sched.h>
+#include <unistd.h>
+#include <sys/types.h>
+#include <sys/syscall.h>
+#include <sys/mount.h>
+#include <sys/wait.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include <errno.h>
+#include <sys/prctl.h>
+#include <sys/capability.h>
+#include <getopt.h>
+
+#include "log.h"
+#include "util.h"
+
+static char *arg_directory = NULL;
+
+static int help(void) {
+
+        printf("%s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
+               "Spawn a minimal namespace container for debugging, testing and building.\n\n"
+               "  -h --help            Show this help\n"
+               "  -D --directory=NAME  Root directory for the container\n",
+               program_invocation_short_name);
+
+        return 0;
+}
+
+static int parse_argv(int argc, char *argv[]) {
+
+        static const struct option options[] = {
+                { "help",      no_argument,       NULL, 'h' },
+                { "directory", required_argument, NULL, 'D' },
+                { NULL,        0,                 NULL, 0   }
+        };
+
+        int c;
+
+        assert(argc >= 0);
+        assert(argv);
+
+        while ((c = getopt_long(argc, argv, "+hD:", options, NULL)) >= 0) {
+
+                switch (c) {
+
+                case 'h':
+                        help();
+                        return 0;
+
+                case 'D':
+                        free(arg_directory);
+                        if (!(arg_directory = strdup(optarg))) {
+                                log_error("Failed to duplicate root directory.");
+                                return -ENOMEM;
+                        }
+
+                        break;
+
+                case '?':
+                        return -EINVAL;
+
+                default:
+                        log_error("Unknown option code %c", c);
+                        return -EINVAL;
+                }
+        }
+
+        return 1;
+}
+
+static int mount_all(const char *dest) {
+
+        typedef struct MountPoint {
+                const char *what;
+                const char *where;
+                const char *type;
+                const char *options;
+                unsigned long flags;
+        } MountPoint;
+
+        static const MountPoint mount_table[] = {
+                { "proc",      "/proc",     "proc",   NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV },
+                { "/proc/sys", "/proc/sys", "bind",   NULL,        MS_BIND },                      /* Bind mount first */
+                { "/proc/sys", "/proc/sys", "bind",   NULL,        MS_BIND|MS_RDONLY|MS_REMOUNT }, /* Then, make it r/o */
+                { "sysfs",     "/sys",      "sysfs",  NULL,        MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY },
+                { "tmpfs",     "/dev",      "tmpfs",  "mode=755",  MS_NOSUID },
+                { "/dev/pts",  "/dev/pts",  "bind",   NULL,        MS_BIND },
+                { "tmpfs",     "/dev/.run", "tmpfs",  "mode=755",  MS_NOSUID|MS_NOEXEC|MS_NODEV },
+        };
+
+        unsigned k;
+        int r = 0;
+
+        for (k = 0; k < ELEMENTSOF(mount_table); k++) {
+                char *where;
+                int t;
+
+                if (asprintf(&where, "%s/%s", dest, mount_table[k].where) < 0) {
+                        log_error("Out of memory");
+
+                        if (r == 0)
+                                r = -ENOMEM;
+
+                        break;
+                }
+
+                if ((t = path_is_mount_point(where)) < 0) {
+                        log_error("Failed to detect whether %s is a mount point: %s", where, strerror(-t));
+                        free(where);
+
+                        if (r == 0)
+                                r = t;
+
+                        continue;
+                }
+
+                mkdir_p(where, 0755);
+
+                if (mount(mount_table[k].what,
+                          where,
+                          mount_table[k].type,
+                          mount_table[k].flags,
+                          mount_table[k].options) < 0) {
+
+                        log_error("mount(%s) failed: %m", where);
+
+                        if (r == 0)
+                                r = -errno;
+                }
+
+                free(where);
+        }
+
+        return r;
+}
+
+static int copy_devnodes(const char *dest) {
+
+        static const char devnodes[] =
+                "null\0"
+                "zero\0"
+                "full\0"
+                "random\0"
+                "urandom\0"
+                "tty\0"
+                "ptmx\0"
+                "kmsg\0"
+                "rtc0\0";
+
+        const char *d;
+        int r = 0, k;
+        char *tty = NULL;
+        dev_t tty_devnum;
+
+        NULSTR_FOREACH(d, devnodes) {
+                char *from = NULL, *to = NULL;
+                struct stat st;
+
+                asprintf(&from, "/dev/%s", d);
+                asprintf(&to, "%s/dev/%s", dest, d);
+
+                if (!from || !to) {
+                        log_error("Failed to allocate devnode path");
+
+                        free(from);
+                        free(to);
+
+                        if (r == 0)
+                                r = -ENOMEM;
+
+                        break;
+                }
+
+                if (stat(from, &st) < 0) {
+
+                        if (errno != ENOENT) {
+                                log_error("Failed to stat %s: %m", from);
+
+                                if (r == 0)
+                                        r = -errno;
+                        }
+
+                } else {
+                        if (mknod(to, st.st_mode, st.st_rdev) < 0) {
+                                log_error("mknod(%s) failed: %m", dest);
+
+                                if (r == 0)
+                                        r = -errno;
+                        }
+                }
+
+                free(from);
+                free(to);
+        }
+
+        if ((k = get_ctty(&tty, &tty_devnum)) < 0) {
+                log_error("Failed to determine controlling tty: %s", strerror(-k));
+
+                if (r == 0)
+                        r = k;
+        } else {
+                char *from = NULL, *to = NULL;
+
+                asprintf(&from, "/dev/%s", tty);
+                asprintf(&to, "%s/dev/console", dest);
+
+                if (!from || !to) {
+                        log_error("Out of memory");
+
+                        if (r == 0)
+                                r = k;
+                } else {
+                        /* We need to bind mount our own tty on
+                         * /dev/console, since ptys cannot be used
+                         * unless on a devpts file system. But to bind
+                         * mount it we first have to create a device
+                         * node where we can bind mount it on. This is
+                         * kinda ugly since the TTY will very likely
+                         * be owned by a user/group that does not
+                         * exist in the container. */
+
+                        if (mknod(to, S_IFCHR|0600, tty_devnum) < 0) {
+                                log_error("mknod for /dev/console failed: %m");
+
+                                if (r == 0)
+                                        r = -errno;
+                        }
+
+                        if (mount(from, to, "bind", MS_BIND, NULL) < 0) {
+                                log_error("bind mount for /dev/console failed: %m");
+
+                                if (r == 0)
+                                        r = -errno;
+                        }
+                }
+
+                free(from);
+                free(to);
+        }
+
+        free(tty);
+
+        return r;
+}
+
+static int drop_capabilities(void) {
+        static const unsigned long retain[] = {
+                CAP_CHOWN,
+                CAP_DAC_OVERRIDE,
+                CAP_DAC_READ_SEARCH,
+                CAP_FOWNER,
+                CAP_FSETID,
+                CAP_IPC_OWNER,
+                CAP_KILL,
+                CAP_LEASE,
+                CAP_LINUX_IMMUTABLE,
+                CAP_NET_BIND_SERVICE,
+                CAP_NET_BROADCAST,
+                CAP_NET_RAW,
+                CAP_SETGID,
+                CAP_SETFCAP,
+                CAP_SETPCAP,
+                CAP_SETUID,
+                CAP_SYS_ADMIN,
+                CAP_SYS_CHROOT,
+                CAP_SYS_NICE,
+                CAP_SYS_PTRACE,
+                CAP_SYS_TTY_CONFIG
+        };
+
+        unsigned long l;
+
+        for (l = 0; l <= MAX(63LU, (unsigned long) CAP_LAST_CAP); l ++) {
+                unsigned i;
+
+                for (i = 0; i < ELEMENTSOF(retain); i++)
+                        if (retain[i] == l)
+                                break;
+
+                if (i < ELEMENTSOF(retain))
+                        continue;
+
+                if (prctl(PR_CAPBSET_DROP, l) < 0) {
+
+                        /* If this capability is not known, EINVAL
+                         * will be returned, let's ignore this. */
+                        if (errno == EINVAL)
+                                continue;
+
+                        log_error("PR_CAPBSET_DROP failed: %m");
+                        return -errno;
+                }
+        }
+
+        return 0;
+}
+
+static int is_os_tree(const char *path) {
+        int r;
+        char *p;
+        /* We use /bin/sh as flag file if something is an OS */
+
+        if (asprintf(&p, "%s/bin/sh", path) < 0)
+                return -ENOMEM;
+
+        r = access(p, F_OK);
+        free(p);
+
+        return r < 0 ? 0 : 1;
+}
+
+
+int main(int argc, char *argv[]) {
+        pid_t pid = 0;
+        int r = EXIT_FAILURE;
+
+        log_parse_environment();
+        log_open();
+
+        if ((r = parse_argv(argc, argv)) <= 0)
+                goto finish;
+
+        if (arg_directory) {
+                char *p;
+
+                p = path_make_absolute_cwd(arg_directory);
+                free(arg_directory);
+                arg_directory = p;
+        } else
+                arg_directory = get_current_dir_name();
+
+        if (!arg_directory) {
+                log_error("Failed to determine path");
+                goto finish;
+        }
+
+        path_kill_slashes(arg_directory);
+
+        if (geteuid() != 0) {
+                log_error("Need to be root.");
+                goto finish;
+        }
+
+        if (path_equal(arg_directory, "/")) {
+                log_error("Spawning constainer on root directory not supported.");
+                goto finish;
+        }
+
+        if (is_os_tree(arg_directory) <= 0) {
+                log_error("Directory %s doesn't look like an OS root directory. Refusing.", arg_directory);
+                goto finish;
+        }
+
+        log_info("Spawning namespace container on %s.", arg_directory);
+
+        if ((pid = syscall(__NR_clone, SIGCHLD|CLONE_NEWIPC|CLONE_NEWNS|CLONE_NEWPID|CLONE_NEWUTS|CLONE_NEWNET, NULL)) < 0) {
+                log_error("clone() failed: %m");
+                goto finish;
+        }
+
+        if (pid == 0) {
+                const char *hn;
+
+                /* child */
+
+                if (mount_all(arg_directory) < 0)
+                        goto child_fail;
+
+                if (copy_devnodes(arg_directory) < 0)
+                        goto child_fail;
+
+                if (chdir(arg_directory) < 0) {
+                        log_error("chdir(%s) failed: %m", arg_directory);
+                        goto child_fail;
+                }
+                if (mount(arg_directory, "/", "bind", MS_BIND|MS_MOVE, NULL) < 0) {
+                        log_error("mount(MS_MOVE) failed: %m");
+                        goto child_fail;
+                }
+
+                if (chroot(".") < 0) {
+                        log_error("chroot() failed: %m");
+                        goto child_fail;
+                }
+
+                if (chdir("/") < 0) {
+                        log_error("chdir() failed: %m");
+                        goto child_fail;
+                }
+
+                if (drop_capabilities() < 0)
+                        goto child_fail;
+
+                if ((hn = file_name_from_path(arg_directory)))
+                        sethostname(hn, strlen(hn));
+
+                if (argc > optind)
+                        execvp(argv[optind], argv + optind);
+                else
+                        execl("/bin/bash", "/bin/bash", NULL);
+
+                log_error("execv() failed: %m");
+
+        child_fail:
+                _exit(EXIT_FAILURE);
+        }
+
+        r = wait_for_terminate_and_warn("container", pid);
+
+        if (r < 0)
+                r = EXIT_FAILURE;
+
+finish:
+        free(arg_directory);
+
+        if (pid > 0)
+                kill(pid, SIGTERM);
+
+        return r;
+}