Specially handle block device rootfs

It is not possible to mount a block device from a non-init user namespace.
Therefore if root on the host is starting a container with a uid
mapping, and the rootfs is a block device, then mount the rootfs before
we spawn the container init task.

This addresses #221

Signed-off-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Stéphane Graber <stgraber@ubuntu.com>

src/lxc/bdev.c

@@ -2745,11 +2745,9 @@ struct bdev *bdev_get(const char *type)
 	return bdev;
 }
 
-struct bdev *bdev_init(struct lxc_conf *conf, const char *src, const char *dst, const char *mntopts)
+static const struct bdev_type *bdev_query(const char *src)
 {
 	int i;
-	struct bdev *bdev;
-
 	for (i=0; i<numbdevs; i++) {
 		int r;
 		r = bdevs[i].ops->detect(src);
@@ -2759,12 +2757,24 @@ struct bdev *bdev_init(struct lxc_conf *conf, const char *src, const char *dst,
 
 	if (i == numbdevs)
 		return NULL;
+	return &bdevs[i];
+}
+
+struct bdev *bdev_init(struct lxc_conf *conf, const char *src, const char *dst, const char *mntopts)
+{
+	struct bdev *bdev;
+	const struct bdev_type *q;
+
+	q = bdev_query(src);
+	if (!q)
+		return NULL;
+
 	bdev = malloc(sizeof(struct bdev));
 	if (!bdev)
 		return NULL;
 	memset(bdev, 0, sizeof(struct bdev));
-	bdev->ops = bdevs[i].ops;
-	bdev->type = bdevs[i].name;
+	bdev->ops = q->ops;
+	bdev->type = q->name;
 	if (mntopts)
 		bdev->mntopts = strdup(mntopts);
 	if (src)
@@ -3087,3 +3097,22 @@ char *overlay_getlower(char *p)
 		*p1 = '\0';
 	return p;
 }
+
+bool rootfs_is_blockdev(struct lxc_conf *conf)
+{
+	const struct bdev_type *q;
+	struct stat st;
+	int ret;
+
+	ret = stat(conf->rootfs.path, &st);
+	if (ret == 0 && S_ISBLK(st.st_mode))
+		return true;
+	q = bdev_query(conf->rootfs.path);
+	if (!q)
+		return false;
+	if (strcmp(q->name, "lvm") == 0 ||
+		strcmp(q->name, "loop") == 0 ||
+		strcmp(q->name, "nbd") == 0)
+		return true;
+	return false;
+}

src/lxc/bdev.h

@@ -102,6 +102,8 @@ void bdev_put(struct bdev *bdev);
 bool attach_block_device(struct lxc_conf *conf);
 void detach_block_device(struct lxc_conf *conf);
 
+bool rootfs_is_blockdev(struct lxc_conf *conf);
+
 /* define constants if the kernel/glibc headers don't define them */
 #ifndef MS_DIRSYNC
 #define MS_DIRSYNC  128

src/lxc/conf.c

@@ -3811,22 +3811,59 @@ static void remount_all_slave(void)
 		free(line);
 }
 
-int lxc_setup(struct lxc_handler *handler)
+/*
+ * This does the work of remounting / if it is shared, calling the
+ * container pre-mount hooks, and mounting the rootfs.
+ */
+int do_rootfs_setup(struct lxc_conf *conf, const char *name, const char *lxcpath)
 {
-	const char *name = handler->name;
-	struct lxc_conf *lxc_conf = handler->conf;
-	const char *lxcpath = handler->lxcpath;
-	void *data = handler->data;
+	if (conf->rootfs_setup) {
+		/*
+		 * rootfs was set up in another namespace.  bind-mount it
+		 * to give us a mount in our own ns so we can pivot_root to it
+		 */
+		const char *path = conf->rootfs.mount;
+		if (mount(path, path, "rootfs", MS_BIND, NULL) < 0) {
+			ERROR("Failed to bind-mount container / onto itself");
+			return false;
+		}
+	}
 
 	if (detect_ramfs_rootfs()) {
-		if (chroot_into_slave(lxc_conf)) {
+		if (chroot_into_slave(conf)) {
 			ERROR("Failed to chroot into slave /");
 			return -1;
 		}
 	}
 
 	remount_all_slave();
 
+	if (run_lxc_hooks(name, "pre-mount", conf, lxcpath, NULL)) {
+		ERROR("failed to run pre-mount hooks for container '%s'.", name);
+		return -1;
+	}
+
+	if (setup_rootfs(conf)) {
+		ERROR("failed to setup rootfs for '%s'", name);
+		return -1;
+	}
+
+	conf->rootfs_setup = true;
+	return 0;
+}
+
+int lxc_setup(struct lxc_handler *handler)
+{
+	const char *name = handler->name;
+	struct lxc_conf *lxc_conf = handler->conf;
+	const char *lxcpath = handler->lxcpath;
+	void *data = handler->data;
+
+	if (do_rootfs_setup(lxc_conf, name, lxcpath) < 0) {
+		ERROR("Error setting up rootfs mount after spawn");
+		return -1;
+	}
+
 	if (lxc_conf->inherit_ns_fd[LXC_NS_UTS] == -1) {
 		if (setup_utsname(lxc_conf->utsname)) {
 			ERROR("failed to setup the utsname for '%s'", name);
@@ -3839,16 +3876,6 @@ int lxc_setup(struct lxc_handler *handler)
 		return -1;
 	}
 
-	if (run_lxc_hooks(name, "pre-mount", lxc_conf, lxcpath, NULL)) {
-		ERROR("failed to run pre-mount hooks for container '%s'.", name);
-		return -1;
-	}
-
-	if (setup_rootfs(lxc_conf)) {
-		ERROR("failed to setup rootfs for '%s'", name);
-		return -1;
-	}
-
 	if (lxc_conf->autodev < 0) {
 		lxc_conf->autodev = check_autodev(lxc_conf->rootfs.mount, data);
 	}

src/lxc/conf.h

@@ -335,6 +335,9 @@ struct lxc_conf {
 	int start_order;
 	struct lxc_list groups;
 	int nbd_idx;
+
+	/* set to true when rootfs has been setup */
+	bool rootfs_setup;
 };
 
 int run_lxc_hooks(const char *name, char *hook, struct lxc_conf *conf,
@@ -371,6 +374,9 @@ extern int lxc_clear_hooks(struct lxc_conf *c, const char *key);
 extern int lxc_clear_idmaps(struct lxc_conf *c);
 extern int lxc_clear_groups(struct lxc_conf *c);
 
+extern int do_rootfs_setup(struct lxc_conf *conf, const char *name,
+			   const char *lxcpath);
+
 /*
  * Configure the container from inside
  */

src/lxc/start.c

@@ -1060,6 +1060,21 @@ int __lxc_start(const char *name, struct lxc_conf *conf,
 		goto out_fini_nonet;
 	}
 
+	if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
+		/* if the backing store is a device, mount it here and now */
+		if (rootfs_is_blockdev(conf)) {
+			if (unshare(CLONE_NEWNS) < 0) {
+				ERROR("Error unsharing mounts");
+				goto out_fini_nonet;
+			}
+			if (do_rootfs_setup(conf, name, lxcpath) < 0) {
+				ERROR("Error setting up rootfs mount as root before spawn");
+				goto out_fini_nonet;
+			}
+			INFO("Set up container rootfs as host root");
+		}
+	}
+
 	err = lxc_spawn(handler);
 	if (err) {
 		ERROR("failed to spawn '%s'", name);