machinectl: implement "bind" command to create additional bind mounts from host to...

[elogind.git] / src / nspawn / nspawn.c

diff --git a/src/nspawn/nspawn.c b/src/nspawn/nspawn.c
index f6f20abdaf23b1db88865e4db0897f06d43a34d2..72f7d66782ecd81e5ffea76d0dd52596d8829ee8 100644 (file)
--- a/src/nspawn/nspawn.c
+++ b/src/nspawn/nspawn.c
@@ -2082,6 +2082,27 @@ finish:
 
 }
 
+static int setup_propagate(const char *root) {
+        const char *p, *q;
+
+        (void) mkdir_p("/run/systemd/nspawn/", 0755);
+        (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
+        p = strappenda("/run/systemd/nspawn/propagate/", arg_machine);
+        (void) mkdir_p(p, 0600);
+
+        q = strappenda(root, "/run/systemd/nspawn/incoming");
+        mkdir_parents(q, 0755);
+        mkdir_p(q, 0600);
+
+        if (mount(p, q, NULL, MS_BIND, NULL) < 0)
+                return log_error_errno(errno, "Failed to install propagation bind mount.");
+
+        if (mount(NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL) < 0)
+                return log_error_errno(errno, "Failed to make propagation mount read-only");
+
+        return 0;
+}
+
 static int setup_image(char **device_path, int *loop_nr) {
         struct loop_info64 info = {
                 .lo_flags = LO_FLAGS_AUTOCLEAR|LO_FLAGS_PARTSCAN
@@ -2875,7 +2896,11 @@ static int determine_names(void) {
         }
 
         if (!arg_machine) {
-                arg_machine = strdup(basename(arg_image ?: arg_directory));
+                if (arg_directory && path_equal(arg_directory, "/"))
+                        arg_machine = gethostname_malloc();
+                else
+                        arg_machine = strdup(basename(arg_image ?: arg_directory));
+
                 if (!arg_machine)
                         return log_oom();
 
@@ -2884,6 +2909,21 @@ static int determine_names(void) {
                         log_error("Failed to determine machine name automatically, please use -M.");
                         return -EINVAL;
                 }
+
+                if (arg_ephemeral) {
+                        char *b;
+
+                        /* Add a random suffix when this is an
+                         * ephemeral machine, so that we can run many
+                         * instances at once without manually having
+                         * to specify -M each time. */
+
+                        if (asprintf(&b, "%s-%016" PRIx64, arg_machine, random_u64()) < 0)
+                                return log_oom();
+
+                        free(arg_machine);
+                        arg_machine = b;
+                }
         }
 
         return 0;
@@ -2942,8 +2982,8 @@ int main(int argc, char *argv[]) {
         if (arg_directory) {
                 assert(!arg_image);
 
-                if (path_equal(arg_directory, "/")) {
-                        log_error("Spawning container on root directory not supported.");
+                if (path_equal(arg_directory, "/") && !arg_ephemeral) {
+                        log_error("Spawning container on root directory is not supported. Consider using --ephemeral.");
                         r = -EINVAL;
                         goto finish;
                 }
@@ -2964,7 +3004,21 @@ int main(int argc, char *argv[]) {
                 } else if (arg_ephemeral) {
                         char *np;
 
-                        r = tempfn_random(arg_directory, &np);
+                        /* If the specified path is a mount point we
+                         * generate the new snapshot immediately
+                         * inside it under a random name. However if
+                         * the specified is not a mount point we
+                         * create the new snapshot in the parent
+                         * directory, just next to it. */
+                        r = path_is_mount_point(arg_directory, false);
+                        if (r < 0) {
+                                log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
+                                goto finish;
+                        }
+                        if (r > 0)
+                                r = tempfn_random_child(arg_directory, &np);
+                        else
+                                r = tempfn_random(arg_directory, &np);
                         if (r < 0) {
                                 log_error_errno(r, "Failed to generate name for snapshot: %m");
                                 goto finish;
@@ -3100,9 +3154,9 @@ int main(int argc, char *argv[]) {
                         goto finish;
                 }
 
-                pid = syscall(__NR_clone, SIGCHLD|CLONE_NEWNS|
-                                          (arg_share_system ? 0 : CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS)|
-                                          (arg_private_network ? CLONE_NEWNET : 0), NULL);
+                pid = raw_clone(SIGCHLD|CLONE_NEWNS|
+                                (arg_share_system ? 0 : CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS)|
+                                (arg_private_network ? CLONE_NEWNET : 0), NULL);
                 if (pid < 0) {
                         if (errno == EINVAL)
                                 r = log_error_errno(errno, "clone() failed, do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in): %m");
@@ -3227,6 +3281,9 @@ int main(int argc, char *argv[]) {
 
                         dev_setup(arg_directory);
 
+                        if (setup_propagate(arg_directory) < 0)
+                                _exit(EXIT_FAILURE);
+
                         if (setup_seccomp() < 0)
                                 _exit(EXIT_FAILURE);
 
@@ -3538,6 +3595,13 @@ finish:
                         log_warning_errno(k, "Cannot remove subvolume '%s', ignoring: %m", arg_directory);
         }
 
+        if (arg_machine) {
+                const char *p;
+
+                p = strappenda("/run/systemd/nspawn/propagate", arg_machine);
+                (void) rm_rf(p, false, true, false);
+        }
+
         free(arg_directory);
         free(arg_template);
         free(arg_image);