From: Christian Brauner <christian.brauner@ubuntu.com>
Date: Thu, 26 Mar 2020 18:27:07 +0000 (+0100)
Subject: cgroups: fix attaching to the unified cgroup
X-Git-Tag: lxc-5.0.0~491^2
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=refs%2Fpull%2F3329%2Fhead;p=thirdparty%2Flxc.git

cgroups: fix attaching to the unified cgroup

Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
---

diff --git a/src/lxc/attach.c b/src/lxc/attach.c
index 083f709bb..ca1fd2bd6 100644
--- a/src/lxc/attach.c
+++ b/src/lxc/attach.c
@@ -1174,7 +1174,7 @@ int lxc_attach(struct lxc_container *container, lxc_attach_exec_t exec_function,
 			 * If this is the unified hierarchy cgroup_attach() is
 			 * enough.
 			 */
-			ret = cgroup_attach(name, lxcpath, pid);
+			ret = cgroup_attach(conf, name, lxcpath, pid);
 			if (ret) {
 				call_cleaner(cgroup_exit) struct cgroup_ops *cgroup_ops = NULL;
 
@@ -1182,7 +1182,7 @@ int lxc_attach(struct lxc_container *container, lxc_attach_exec_t exec_function,
 				if (!cgroup_ops)
 					goto on_error;
 
-				if (!cgroup_ops->attach(cgroup_ops, name, lxcpath, pid))
+				if (!cgroup_ops->attach(cgroup_ops, conf, name, lxcpath, pid))
 					goto on_error;
 			}
 			TRACE("Moved intermediate process %d into container's cgroups", pid);
diff --git a/src/lxc/cgroups/cgfsng.c b/src/lxc/cgroups/cgfsng.c
index 440a19efa..ad37291e9 100644
--- a/src/lxc/cgroups/cgfsng.c
+++ b/src/lxc/cgroups/cgfsng.c
@@ -2057,12 +2057,11 @@ static inline char *build_full_cgpath_from_monitorpath(struct hierarchy *h,
 	return must_make_path(h->mountpoint, inpath, filename, NULL);
 }
 
-static int cgroup_attach_leaf(int unified_fd, int64_t pid)
+static int cgroup_attach_leaf(const struct lxc_conf *conf, int unified_fd, pid_t pid)
 {
 	int idx = 1;
 	int ret;
 	char pidstr[INTTYPE_TO_STRLEN(int64_t) + 1];
-	char attach_cgroup[STRLITERALLEN("lxc-1000/cgroup.procs") + 1];
 	size_t pidstr_len;
 
 	/* Create leaf cgroup. */
@@ -2070,7 +2069,7 @@ static int cgroup_attach_leaf(int unified_fd, int64_t pid)
 	if (ret < 0 && errno != EEXIST)
 		return log_error_errno(-1, errno, "Failed to create leaf cgroup \"lxc\"");
 
-	pidstr_len = sprintf(pidstr, INT64_FMT, pid);
+	pidstr_len = sprintf(pidstr, INT64_FMT, (int64_t)pid);
 	ret = lxc_writeat(unified_fd, "lxc/cgroup.procs", pidstr, pidstr_len);
 	if (ret < 0)
 		ret = lxc_writeat(unified_fd, "cgroup.procs", pidstr, pidstr_len);
@@ -2082,6 +2081,8 @@ static int cgroup_attach_leaf(int unified_fd, int64_t pid)
 		return log_error_errno(-1, errno, "Failed to attach to unified cgroup");
 
 	do {
+		bool rm = false;
+		char attach_cgroup[STRLITERALLEN("lxc-1000/cgroup.procs") + 1];
 		char *slash;
 
 		sprintf(attach_cgroup, "lxc-%d/cgroup.procs", idx);
@@ -2091,6 +2092,8 @@ static int cgroup_attach_leaf(int unified_fd, int64_t pid)
 		ret = mkdirat(unified_fd, attach_cgroup, 0755);
 		if (ret < 0 && errno != EEXIST)
 			return log_error_errno(-1, errno, "Failed to create cgroup %s", attach_cgroup);
+		if (ret == 0)
+			rm = true;
 
 		*slash = '/';
 
@@ -2098,6 +2101,9 @@ static int cgroup_attach_leaf(int unified_fd, int64_t pid)
 		if (ret == 0)
 			return 0;
 
+		if (rm && unlinkat(unified_fd, attach_cgroup, AT_REMOVEDIR))
+			SYSERROR("Failed to remove cgroup \"%d(%s)\"", unified_fd, attach_cgroup);
+
 		/* this is a non-leaf node */
 		if (errno != EBUSY)
 			return log_error_errno(-1, errno, "Failed to attach to unified cgroup");
@@ -2108,15 +2114,66 @@ static int cgroup_attach_leaf(int unified_fd, int64_t pid)
 	return log_error_errno(-1, errno, "Failed to attach to unified cgroup");
 }
 
-int cgroup_attach(const char *name, const char *lxcpath, int64_t pid)
+struct userns_exec_unified_attach_data {
+	const struct lxc_conf *conf;
+	int unified_fd;
+	pid_t pid;
+	uid_t origuid;
+};
+
+static int cgroup_unified_attach_wrapper(void *data)
+{
+	struct userns_exec_unified_attach_data *args = data;
+	uid_t nsuid = (args->conf->root_nsuid_map != NULL) ? 0 : args->conf->init_uid;
+	gid_t nsgid = (args->conf->root_nsgid_map != NULL) ? 0 : args->conf->init_gid;
+	int ret;
+
+	if (!args->conf || args->unified_fd < 0 || args->pid <= 0)
+		return ret_errno(EINVAL);
+
+	if (!lxc_setgroups(0, NULL) && errno != EPERM)
+		return log_error_errno(-1, errno, "Failed to setgroups(0, NULL)");
+
+	ret = setresgid(nsgid, nsgid, nsgid);
+	if (ret < 0)
+		return log_error_errno(-1, errno, "Failed to setresgid(%d, %d, %d)",
+				       (int)nsgid, (int)nsgid, (int)nsgid);
+
+	ret = setresuid(nsuid, nsuid, nsuid);
+	if (ret < 0)
+		return log_error_errno(-1, errno, "Failed to setresuid(%d, %d, %d)",
+				       (int)nsuid, (int)nsuid, (int)nsuid);
+
+	return cgroup_attach_leaf(args->conf, args->unified_fd, args->pid);
+}
+
+int cgroup_attach(const struct lxc_conf *conf, const char *name,
+		  const char *lxcpath, pid_t pid)
 {
 	__do_close int unified_fd = -EBADF;
+	int ret;
+
+	if (!conf || !name || !lxcpath || pid <= 0)
+		return ret_errno(EINVAL);
 
 	unified_fd = lxc_cmd_get_cgroup2_fd(name, lxcpath);
 	if (unified_fd < 0)
-		return -1;
+		return ret_errno(EBADF);
+
+	if (!lxc_list_empty(&conf->id_map)) {
+		struct userns_exec_unified_attach_data args = {
+			.conf		= conf,
+			.unified_fd	= unified_fd,
+			.pid		= pid,
+		};
+
+		ret = userns_exec_1(conf, cgroup_unified_attach_wrapper, &args,
+				    "cgroup_unified_attach_wrapper");
+	} else {
+		ret = cgroup_attach_leaf(conf, unified_fd, pid);
+	}
 
-	return cgroup_attach_leaf(unified_fd, pid);
+	return ret;
 }
 
 /* Technically, we're always at a delegation boundary here (This is especially
@@ -2128,14 +2185,18 @@ int cgroup_attach(const char *name, const char *lxcpath, int64_t pid)
  * created when we started the container in the latter case we create our own
  * cgroup for the attaching process.
  */
-static int __cg_unified_attach(const struct hierarchy *h, const char *name,
+static int __cg_unified_attach(const struct hierarchy *h,
+			       const struct lxc_conf *conf, const char *name,
 			       const char *lxcpath, pid_t pid,
 			       const char *controller)
 {
 	__do_close int unified_fd = -EBADF;
 	int ret;
 
-	ret = cgroup_attach(name, lxcpath, pid);
+	if (!conf || !name || !lxcpath || pid <= 0)
+		return ret_errno(EINVAL);
+
+	ret = cgroup_attach(conf, name, lxcpath, pid);
 	if (ret < 0) {
 		__do_free char *path = NULL, *cgroup = NULL;
 
@@ -2148,13 +2209,28 @@ static int __cg_unified_attach(const struct hierarchy *h, const char *name,
 		unified_fd = open(path, O_DIRECTORY | O_RDONLY | O_CLOEXEC);
 	}
 	if (unified_fd < 0)
-		return -1;
+		return ret_errno(EBADF);
 
-	return cgroup_attach_leaf(unified_fd, pid);
+	if (!lxc_list_empty(&conf->id_map)) {
+		struct userns_exec_unified_attach_data args = {
+			.conf		= conf,
+			.unified_fd	= unified_fd,
+			.pid		= pid,
+		};
+
+		ret = userns_exec_1(conf, cgroup_unified_attach_wrapper, &args,
+				    "cgroup_unified_attach_wrapper");
+	} else {
+		ret = cgroup_attach_leaf(conf, unified_fd, pid);
+	}
+
+	return ret;
 }
 
-__cgfsng_ops static bool cgfsng_attach(struct cgroup_ops *ops, const char *name,
-					 const char *lxcpath, pid_t pid)
+__cgfsng_ops static bool cgfsng_attach(struct cgroup_ops *ops,
+				       const struct lxc_conf *conf,
+				       const char *name, const char *lxcpath,
+				       pid_t pid)
 {
 	int len, ret;
 	char pidstr[INTTYPE_TO_STRLEN(pid_t)];
@@ -2174,7 +2250,7 @@ __cgfsng_ops static bool cgfsng_attach(struct cgroup_ops *ops, const char *name,
 		struct hierarchy *h = ops->hierarchies[i];
 
 		if (h->version == CGROUP2_SUPER_MAGIC) {
-			ret = __cg_unified_attach(h, name, lxcpath, pid,
+			ret = __cg_unified_attach(h, conf, name, lxcpath, pid,
 						  h->controllers[0]);
 			if (ret < 0)
 				return false;
diff --git a/src/lxc/cgroups/cgroup.h b/src/lxc/cgroups/cgroup.h
index 9f44ad51b..1e08a017a 100644
--- a/src/lxc/cgroups/cgroup.h
+++ b/src/lxc/cgroups/cgroup.h
@@ -160,8 +160,8 @@ struct cgroup_ops {
 				    struct lxc_conf *conf, bool with_devices);
 	bool (*setup_limits)(struct cgroup_ops *ops, struct lxc_handler *handler);
 	bool (*chown)(struct cgroup_ops *ops, struct lxc_conf *conf);
-	bool (*attach)(struct cgroup_ops *ops, const char *name,
-		       const char *lxcpath, pid_t pid);
+	bool (*attach)(struct cgroup_ops *ops, const struct lxc_conf *conf,
+		       const char *name, const char *lxcpath, pid_t pid);
 	bool (*mount)(struct cgroup_ops *ops, struct lxc_handler *handler,
 		      const char *root, int type);
 	bool (*devices_activate)(struct cgroup_ops *ops,
@@ -178,7 +178,8 @@ define_cleanup_function(struct cgroup_ops *, cgroup_exit);
 
 extern void prune_init_scope(char *cg);
 
-extern int cgroup_attach(const char *name, const char *lxcpath, int64_t pid);
+extern int cgroup_attach(const struct lxc_conf *conf, const char *name,
+			 const char *lxcpath, pid_t pid);
 
 static inline bool pure_unified_layout(const struct cgroup_ops *ops)
 {
diff --git a/src/lxc/conf.c b/src/lxc/conf.c
index 4bab3ee67..8cd06abf9 100644
--- a/src/lxc/conf.c
+++ b/src/lxc/conf.c
@@ -1440,7 +1440,7 @@ static int lxc_setup_rootfs_switch_root(const struct lxc_rootfs *rootfs)
 	return lxc_pivot_root(rootfs->mount);
 }
 
-static const struct id_map *find_mapped_nsid_entry(struct lxc_conf *conf,
+static const struct id_map *find_mapped_nsid_entry(const struct lxc_conf *conf,
 						   unsigned id,
 						   enum idtype idtype)
 {
@@ -2845,7 +2845,7 @@ int mapped_hostid(unsigned id, const struct lxc_conf *conf, enum idtype idtype)
 	return -1;
 }
 
-int find_unmapped_nsid(struct lxc_conf *conf, enum idtype idtype)
+int find_unmapped_nsid(const struct lxc_conf *conf, enum idtype idtype)
 {
 	struct id_map *map;
 	struct lxc_list *it;
@@ -3869,7 +3869,7 @@ static int run_userns_fn(void *data)
 	return d->fn(d->arg);
 }
 
-static struct id_map *mapped_nsid_add(struct lxc_conf *conf, unsigned id,
+static struct id_map *mapped_nsid_add(const struct lxc_conf *conf, unsigned id,
 				      enum idtype idtype)
 {
 	const struct id_map *map;
@@ -3887,7 +3887,7 @@ static struct id_map *mapped_nsid_add(struct lxc_conf *conf, unsigned id,
 	return retmap;
 }
 
-static struct id_map *find_mapped_hostid_entry(struct lxc_conf *conf,
+static struct id_map *find_mapped_hostid_entry(const struct lxc_conf *conf,
 					       unsigned id, enum idtype idtype)
 {
 	struct id_map *map;
@@ -3911,7 +3911,7 @@ static struct id_map *find_mapped_hostid_entry(struct lxc_conf *conf,
 /* Allocate a new {g,u}id mapping for the given {g,u}id. Re-use an already
  * existing one or establish a new one.
  */
-static struct id_map *mapped_hostid_add(struct lxc_conf *conf, uid_t id,
+static struct id_map *mapped_hostid_add(const struct lxc_conf *conf, uid_t id,
 					enum idtype type)
 {
 	__do_free struct id_map *entry = NULL;
@@ -3940,7 +3940,7 @@ static struct id_map *mapped_hostid_add(struct lxc_conf *conf, uid_t id,
 	return move_ptr(entry);
 }
 
-struct lxc_list *get_minimal_idmap(struct lxc_conf *conf)
+static struct lxc_list *get_minimal_idmap(const struct lxc_conf *conf)
 {
 	__do_free struct id_map *container_root_uid = NULL,
 				*container_root_gid = NULL,
@@ -4044,7 +4044,7 @@ struct lxc_list *get_minimal_idmap(struct lxc_conf *conf)
  * retrieve from the container's configured {g,u}id mappings as it must have been
  * there to start the container in the first place.
  */
-int userns_exec_1(struct lxc_conf *conf, int (*fn)(void *), void *data,
+int userns_exec_1(const struct lxc_conf *conf, int (*fn)(void *), void *data,
 		  const char *fn_name)
 {
 	pid_t pid;
diff --git a/src/lxc/conf.h b/src/lxc/conf.h
index 7f91c9fb9..f37e09e4a 100644
--- a/src/lxc/conf.h
+++ b/src/lxc/conf.h
@@ -436,12 +436,12 @@ extern int lxc_setup_rootfs_prepare_root(struct lxc_conf *conf,
 extern int lxc_setup(struct lxc_handler *handler);
 extern int lxc_setup_parent(struct lxc_handler *handler);
 extern int setup_resource_limits(struct lxc_list *limits, pid_t pid);
-extern int find_unmapped_nsid(struct lxc_conf *conf, enum idtype idtype);
+extern int find_unmapped_nsid(const struct lxc_conf *conf, enum idtype idtype);
 extern int mapped_hostid(unsigned id, const struct lxc_conf *conf,
 			 enum idtype idtype);
 extern int chown_mapped_root(const char *path, const struct lxc_conf *conf);
-extern int userns_exec_1(struct lxc_conf *conf, int (*fn)(void *), void *data,
-			 const char *fn_name);
+extern int userns_exec_1(const struct lxc_conf *conf, int (*fn)(void *),
+			 void *data, const char *fn_name);
 extern int userns_exec_full(struct lxc_conf *conf, int (*fn)(void *),
 			    void *data, const char *fn_name);
 extern int parse_mntopts(const char *mntopts, unsigned long *mntflags,