Merge pull request #752 from hallyn/2016-01-04/cgroup.use

2016 01 04/cgroup.use

doc/lxc.container.conf.sgml.in

@@ -1593,6 +1593,22 @@ mknod errno 0
           </listitem>
         </varlistentry>
       </variablelist>
+      <variablelist>
+        <varlistentry>
+          <term>
+            <option>LXC_CGNS_AWARE</option>
+          </term>
+          <listitem>
+            <para>
+	      If unset, then this version of lxc is not aware of cgroup
+	      namespaces.  If set, it will be set to 1, and lxc is aware
+	      of cgroup namespaces.  Note this does not guarantee that
+	      cgroup namespaces are enabled in the kernel.  This is used
+	      by the lxcfs mount hook.
+            </para>
+          </listitem>
+        </varlistentry>
+      </variablelist>
     </refsect2>
     <refsect2>
     <title>Logging</title>

src/lxc/attach.c

@@ -957,6 +957,13 @@ int lxc_attach(const char* name, const char* lxcpath, lxc_attach_exec_t exec_fun
 		WARN("could not change directory to '%s'", new_cwd);
 	free(cwd);
 
+	if (options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP && cgns_supported()) {
+		if (unshare(CLONE_NEWCGROUP) != 0) {
+			SYSERROR("cgroupns unshare: permission denied");
+			rexit(-1);
+		}
+	}
+
 	/* now create the real child process */
 	{
 		struct attach_clone_payload payload = {

src/lxc/cgmanager.c

@@ -210,11 +210,6 @@ static void check_supports_multiple_controllers(pid_t pid)
 	cgm_supports_multiple_controllers = false;
 	cgm_all_controllers_same = false;
 
-	if (api_version < CGM_SUPPORTS_MULT_CONTROLLERS) {
-		cgm_supports_multiple_controllers = false;
-		return;
-	}
-
 	cgm_supports_multiple_controllers = true;
 
 	if (pid == -1)
@@ -544,17 +539,13 @@ static void *cgm_init(const char *name)
 {
 	struct cgm_data *d;
 
-	if (!cgm_dbus_connect()) {
-		ERROR("Error connecting to cgroup manager");
-		return NULL;
-	}
-
-	check_supports_multiple_controllers(-1);
-
 	d = malloc(sizeof(*d));
-	if (!d) {
-		cgm_dbus_disconnect();
+	if (!d)
 		return NULL;
+
+	if (!cgm_dbus_connect()) {
+		ERROR("Error connecting to cgroup manager");
+		goto err1;
 	}
 
 	memset(d, 0, sizeof(*d));
@@ -1132,6 +1123,9 @@ static void cull_user_controllers(void)
 	}
 }
 
+/*
+ * return true if inword is in the comma-delimited list cgroup_use
+ */
 static bool in_comma_list(const char *inword, const char *cgroup_use)
 {
 	char *e;
@@ -1148,6 +1142,23 @@ static bool in_comma_list(const char *inword, const char *cgroup_use)
 	return false;
 }
 
+/*
+ * inlist is a comma-delimited list of cgroups;  so is checklist.  Return
+ * true if any member of inlist is in checklist.
+ */
+static bool any_in_comma_list(const char *inlist, const char *checklist)
+{
+	char *tmp = alloca(strlen(inlist) + 1), *tok, *saveptr = NULL;
+
+	strcpy(tmp, inlist);
+	for (tok = strtok_r(tmp, ",", &saveptr); tok; tok = strtok_r(NULL, ",", &saveptr)) {
+		if (in_comma_list(tok, checklist))
+			return true;
+	}
+
+	return false;
+}
+
 static bool in_subsystem_list(const char *c)
 {
 	int i;
@@ -1202,6 +1213,132 @@ static bool verify_and_prune(const char *cgroup_use)
 	return true;
 }
 
+static void drop_subsystem(int which)
+{
+	int i;
+
+	if (which < 0 || which >= nr_subsystems) {
+		ERROR("code error: dropping invalid subsystem index\n");
+		exit(1);
+	}
+
+	free(subsystems[which]);
+	/* note - we have nr_subsystems+1 entries, last one a NULL */
+	for (i = which; i < nr_subsystems; i++)
+		subsystems[i] = subsystems[i+1];
+	nr_subsystems -= 1;
+}
+
+/*
+ * Check whether we can create the cgroups we would want
+ */
+static bool subsys_is_writeable(const char *controller, const char *probe)
+{
+	int32_t existed;
+	bool ret = true;
+
+	if ( cgmanager_create_sync(NULL, cgroup_manager, controller,
+				       probe, &existed) != 0) {
+		NihError *nerr;
+		nerr = nih_error_get();
+		ERROR("call to cgmanager_create_sync failed: %s", nerr->message);
+		nih_free(nerr);
+		ERROR("Failed to create %s:%s", controller, probe);
+		ret = false;
+	}
+
+	return ret;
+}
+
+/*
+ * Return true if this is a subsystem which we cannot do
+ * without
+ */
+static bool is_crucial_subsys(const char *s)
+{
+	if (strcmp(s, "systemd") == 0)
+		return true;
+	if (strcmp(s, "name=systemd") == 0)
+		return true;
+	if (strcmp(s, "freezer") == 0)
+		return true;
+	return false;
+}
+
+static char *get_last_controller_in_list(char *list)
+{
+	char *p;
+
+	while ((p = strchr(list, ',')) != NULL)
+		list = p + 1;
+
+	return list;
+}
+
+/*
+ * Make sure that all the controllers are writeable.
+ * If any are not, then
+ *   - if they are listed in lxc.cgroup.use, refuse to start
+ *   - else if they are crucial subsystems, refuse to start
+ *   - else warn and do not use them
+ */
+static bool verify_final_subsystems(const char *cgroup_use)
+{
+	int i = 0;
+	bool dropped_any = false;
+	bool ret = false;
+	const char *cgroup_pattern;
+	char tmpnam[50], *probe;
+
+	if (!cgm_dbus_connect()) {
+		ERROR("Error connecting to cgroup manager");
+		return false;
+	}
+
+	cgroup_pattern = lxc_global_config_value("lxc.cgroup.pattern");
+	i = snprintf(tmpnam, 50, "lxcprobe-%d", getpid());
+	if (i < 0 || i >= 50) {
+		ERROR("Attack - format string modified?");
+		return false;
+	}
+	probe = lxc_string_replace("%n", tmpnam, cgroup_pattern);
+	if (!probe)
+		goto out;
+
+	while (i < nr_subsystems) {
+		char *p = get_last_controller_in_list(subsystems[i]);
+
+		if (!subsys_is_writeable(p, probe)) {
+			if (is_crucial_subsys(p)) {
+				ERROR("Cannot write to crucial subsystem %s\n",
+					subsystems[i]);
+				goto out;
+			}
+			if (cgroup_use && any_in_comma_list(subsystems[i], cgroup_use)) {
+				ERROR("Cannot write to subsystem %s which is requested in lxc.cgroup.use\n",
+					subsystems[i]);
+				goto out;
+			}
+			WARN("Cannot write to subsystem %s, continuing with out it\n",
+				subsystems[i]);
+			dropped_any = true;
+			drop_subsystem(i);
+		} else {
+			cgm_remove_cgroup(subsystems[i], probe);
+			i++;
+		}
+	}
+
+	if (dropped_any)
+		cgm_all_controllers_same = false;
+	ret = true;
+
+out:
+	free(probe);
+	cgm_dbus_disconnect();
+	return ret;
+}
+
 static bool collect_subsytems(void)
 {
 	char *line = NULL;
@@ -1285,7 +1422,7 @@ static bool collect_subsytems(void)
 	/* make sure that cgroup.use can be and is honored */
 	const char *cgroup_use = lxc_global_config_value("lxc.cgroup.use");
 	if (!cgroup_use && errno != 0)
-		goto out_good;
+		goto final_verify;
 	if (cgroup_use) {
 		if (!verify_and_prune(cgroup_use)) {
 			free_subsystems();
@@ -1295,8 +1432,8 @@ static bool collect_subsytems(void)
 		cgm_all_controllers_same = false;
 	}
 
-out_good:
-	return true;
+final_verify:
+	return verify_final_subsystems(cgroup_use);
 
 out_free:
 	free(line);
@@ -1313,23 +1450,20 @@ static bool collect_subsytems(void)
  */
 struct cgroup_ops *cgm_ops_init(void)
 {
+	check_supports_multiple_controllers(-1);
 	if (!collect_subsytems())
 		return NULL;
-	if (!cgm_dbus_connect())
-		goto err1;
 
-	// root;  try to escape to root cgroup
-	if (geteuid() == 0 && !cgm_escape())
-		goto err2;
-	cgm_dbus_disconnect();
+	if (api_version < CGM_SUPPORTS_MULT_CONTROLLERS)
+		cgm_supports_multiple_controllers = false;
 
-	return &cgmanager_ops;
+	// if root, try to escape to root cgroup
+	if (geteuid() == 0 && !cgm_escape()) {
+		free_subsystems();
+		return NULL;
+	}
 
-err2:
-	cgm_dbus_disconnect();
-err1:
-	free_subsystems();
-	return NULL;
+	return &cgmanager_ops;
 }
 
 /* unfreeze is called by the command api after killing a container.  */

src/lxc/namespace.h

@@ -34,6 +34,9 @@
 #ifndef CLONE_NEWNS
 #  define CLONE_NEWNS             0x00020000
 #endif
+#ifndef CLONE_NEWCGROUP
+#  define CLONE_NEWCGROUP         0x02000000
+#endif
 #ifndef CLONE_NEWUTS
 #  define CLONE_NEWUTS            0x04000000
 #endif

src/lxc/start.c

@@ -451,6 +451,9 @@ struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char
 	if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) {
 		SYSERROR("failed to set environment variable for console log");
 	}
+	if (setenv("LXC_CGNS_AWARE", "1", 1)) {
+		SYSERROR("failed to set LXC_CGNS_AWARE environment variable");
+	}
 	/* End of environment variable setup for hooks */
 
 	if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
@@ -842,6 +845,11 @@ static int do_start(void *data)
 	if (handler->backgrounded && null_stdfds() < 0)
 		goto out_warn_father;
 
+	if (cgns_supported() && unshare(CLONE_NEWCGROUP) != 0) {
+		SYSERROR("Failed to unshare cgroup namespace");
+		goto out_warn_father;
+	}
+
 	/* after this call, we are in error because this
 	 * ops should not return as it execs */
 	handler->ops->start(handler, handler->data);

src/lxc/utils.c

@@ -1185,6 +1185,11 @@ bool file_exists(const char *f)
 	return stat(f, &statbuf) == 0;
 }
 
+bool cgns_supported(void)
+{
+	return file_exists("/proc/self/ns/cgroup");
+}
+
 /* historically lxc-init has been under /usr/lib/lxc and under
  * /usr/lib/$ARCH/lxc.  It now lives as $prefix/sbin/init.lxc.
  */

src/lxc/utils.h

@@ -273,6 +273,7 @@ int detect_shared_rootfs(void);
 int detect_ramfs_rootfs(void);
 char *on_path(char *cmd, const char *rootfs);
 bool file_exists(const char *f);
+bool cgns_supported(void);
 char *choose_init(const char *rootfs);
 int print_to_file(const char *file, const char *content);
 bool switch_to_ns(pid_t pid, const char *ns);