chiark / gitweb /
util: split-out conf-file.[ch]
[elogind.git] / src / nspawn / nspawn.c
index ce727b6e1d8b3d663855a19a69fac4e52c8e2d2b..2a1f37bffcd50dca1a40bc3763e4f0827dc230e7 100644 (file)
 static char *arg_directory = NULL;
 static char *arg_user = NULL;
 static char **arg_controllers = NULL;
+static char *arg_uuid = NULL;
 static bool arg_private_network = false;
+static bool arg_read_only = false;
+static bool arg_boot = false;
 
 static int help(void) {
 
@@ -63,9 +66,12 @@ static int help(void) {
                "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"
+               "  -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 cgroup hierarchies\n"
-               "     --private-network  Disable network in container\n",
+               "     --uuid=UUID        Set a specific machine UUID for the container\n"
+               "     --private-network  Disable network in container\n"
+               "     --read-only        Mount the root directory read-only\n",
                program_invocation_short_name);
 
         return 0;
@@ -74,7 +80,9 @@ static int help(void) {
 static int parse_argv(int argc, char *argv[]) {
 
         enum {
-                ARG_PRIVATE_NETWORK = 0x100
+                ARG_PRIVATE_NETWORK = 0x100,
+                ARG_UUID,
+                ARG_READ_ONLY
         };
 
         static const struct option options[] = {
@@ -83,6 +91,9 @@ static int parse_argv(int argc, char *argv[]) {
                 { "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            },
+                { "read-only",       no_argument,       NULL, ARG_READ_ONLY       },
                 { NULL,              0,                 NULL, 0                   }
         };
 
@@ -91,7 +102,7 @@ static int parse_argv(int argc, char *argv[]) {
         assert(argc >= 0);
         assert(argv);
 
-        while ((c = getopt_long(argc, argv, "+hD:u:C:", options, NULL)) >= 0) {
+        while ((c = getopt_long(argc, argv, "+hD:u:C:b", options, NULL)) >= 0) {
 
                 switch (c) {
 
@@ -101,8 +112,9 @@ static int parse_argv(int argc, char *argv[]) {
 
                 case 'D':
                         free(arg_directory);
-                        if (!(arg_directory = strdup(optarg))) {
-                                log_error("Failed to duplicate root directory.");
+                        arg_directory = canonicalize_file_name(optarg);
+                        if (!arg_directory) {
+                                log_error("Failed to canonicalize root directory.");
                                 return -ENOMEM;
                         }
 
@@ -132,6 +144,18 @@ static int parse_argv(int argc, char *argv[]) {
                         arg_private_network = true;
                         break;
 
+                case 'b':
+                        arg_boot = true;
+                        break;
+
+                case ARG_UUID:
+                        arg_uuid = optarg;
+                        break;
+
+                case ARG_READ_ONLY:
+                        arg_read_only = true;
+                        break;
+
                 case '?':
                         return -EINVAL;
 
@@ -247,6 +271,28 @@ static int setup_timezone(const char *dest) {
         return 0;
 }
 
+static int setup_resolv_conf(const char *dest) {
+        char *where;
+
+        assert(dest);
+
+        if (arg_private_network)
+                return 0;
+
+        /* Fix resolv.conf, if possible */
+        if (asprintf(&where, "%s/etc/resolv.conf", dest) < 0) {
+                log_error("Out of memory");
+                return -ENOMEM;
+        }
+
+        if (mount("/etc/resolv.conf", where, "bind", MS_BIND, NULL) >= 0)
+                mount("/etc/resolv.conf", where, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
+
+        free(where);
+
+        return 0;
+}
+
 static int copy_devnodes(const char *dest) {
 
         static const char devnodes[] =
@@ -394,6 +440,13 @@ static int setup_kmsg(const char *dest, int kmsg_socket) {
 
         u = umask(0000);
 
+        /* We create the kmsg FIFO as /dev/kmsg, but immediately
+         * delete it after bind mounting it to /proc/kmsg. While FIFOs
+         * on the reading side behave very similar to /proc/kmsg,
+         * their writing side behaves differently from /dev/kmsg in
+         * that writing blocks when nothing is reading. In order to
+         * avoid any problems with containers deadlocking due to this
+         * we simply make /dev/kmsg unavailable to the container. */
         if (asprintf(&from, "%s/dev/kmsg", dest) < 0) {
                 log_error("Out of memory");
                 r = -ENOMEM;
@@ -456,6 +509,9 @@ static int setup_kmsg(const char *dest, int kmsg_socket) {
                 goto finish;
         }
 
+        /* And now make the FIFO unavailable as /dev/kmsg... */
+        unlink(from);
+
 finish:
         free(from);
         free(to);
@@ -464,6 +520,28 @@ finish:
         return r;
 }
 
+static int setup_hostname(void) {
+        char *hn;
+        int r = 0;
+
+        hn = file_name_from_path(arg_directory);
+        if (hn) {
+                hn = strdup(hn);
+                if (!hn)
+                        return -ENOMEM;
+
+                hostname_cleanup(hn);
+
+                if (!isempty(hn))
+                        if (sethostname(hn, strlen(hn)) < 0)
+                                r = -errno;
+
+                free(hn);
+        }
+
+        return r;
+}
+
 static int drop_capabilities(void) {
         static const unsigned long retain[] = {
                 CAP_CHOWN,
@@ -862,7 +940,6 @@ int main(int argc, char *argv[]) {
         if (pid == 0) {
                 /* child */
 
-                const char *hn;
                 const char *home = NULL;
                 uid_t uid = (uid_t) -1;
                 gid_t gid = (gid_t) -1;
@@ -873,6 +950,7 @@ int main(int argc, char *argv[]) {
                         NULL, /* HOME */
                         NULL, /* USER */
                         NULL, /* LOGNAME */
+                        NULL, /* container_uuid */
                         NULL
                 };
 
@@ -901,6 +979,18 @@ int main(int argc, char *argv[]) {
                 if (mount(NULL, "/", NULL, MS_PRIVATE|MS_REC, NULL) < 0)
                         goto child_fail;
 
+                /* Turn directory into bind mount */
+                if (mount(arg_directory, arg_directory, "bind", MS_BIND, NULL) < 0) {
+                        log_error("Failed to make bind mount.");
+                        goto child_fail;
+                }
+
+                if (arg_read_only)
+                        if (mount(arg_directory, arg_directory, "bind", MS_BIND|MS_REMOUNT|MS_RDONLY, NULL) < 0) {
+                                log_error("Failed to make read-only.");
+                                goto child_fail;
+                        }
+
                 if (mount_all(arg_directory) < 0)
                         goto child_fail;
 
@@ -918,6 +1008,9 @@ int main(int argc, char *argv[]) {
                 if (setup_timezone(arg_directory) < 0)
                         goto child_fail;
 
+                if (setup_resolv_conf(arg_directory) < 0)
+                        goto child_fail;
+
                 if (chdir(arg_directory) < 0) {
                         log_error("chdir(%s) failed: %m", arg_directory);
                         goto child_fail;
@@ -928,8 +1021,8 @@ int main(int argc, char *argv[]) {
                     dup2(STDIN_FILENO, STDERR_FILENO) != STDERR_FILENO)
                         goto child_fail;
 
-                if (mount(arg_directory, "/", "bind", MS_BIND|MS_MOVE, NULL) < 0) {
-                        log_error("mount(MS_MOVE) failed: %m");
+                if (mount(arg_directory, "/", "bind", MS_MOVE, NULL) < 0) {
+                        log_error("mount(MS_BIND) failed: %m");
                         goto child_fail;
                 }
 
@@ -983,17 +1076,41 @@ int main(int argc, char *argv[]) {
                         }
                 }
 
-                if ((asprintf((char**)(envp + 3), "HOME=%s", home? home: "/root") < 0) ||
-                    (asprintf((char**)(envp + 4), "USER=%s", arg_user? arg_user : "root") < 0) ||
-                    (asprintf((char**)(envp + 5), "LOGNAME=%s", arg_user? arg_user : "root") < 0)) {
+                if ((asprintf((char**)(envp + 3), "HOME=%s", home ? home: "/root") < 0) ||
+                    (asprintf((char**)(envp + 4), "USER=%s", arg_user ? arg_user : "root") < 0) ||
+                    (asprintf((char**)(envp + 5), "LOGNAME=%s", arg_user ? arg_user : "root") < 0)) {
                     log_error("Out of memory");
                     goto child_fail;
                 }
 
-                if ((hn = file_name_from_path(arg_directory)))
-                        sethostname(hn, strlen(hn));
+                if (arg_uuid) {
+                        if (asprintf((char**)(envp + 6), "container_uuid=%s", arg_uuid) < 0) {
+                                log_error("Out of memory");
+                                goto child_fail;
+                        }
+                }
+
+                setup_hostname();
+
+                if (arg_boot) {
+                        char **a;
+                        size_t l;
+
+                        /* Automatically search for the init system */
+
+                        l = 1 + argc - optind;
+                        a = newa(char*, l + 1);
+                        memcpy(a + 1, argv + optind, l * sizeof(char*));
+
+                        a[0] = (char*) "/usr/lib/systemd/systemd";
+                        execve(a[0], a, (char**) envp);
+
+                        a[0] = (char*) "/lib/systemd/systemd";
+                        execve(a[0], a, (char**) envp);
 
-                if (argc > optind)
+                        a[0] = (char*) "/sbin/init";
+                        execve(a[0], a, (char**) envp);
+                } else if (argc > optind)
                         execvpe(argv[optind], argv + optind, (char**) envp);
                 else {
                         chdir(home ? home : "/root");