vfs-6.11.pidfs

-----BEGIN PGP SIGNATURE-----
 
 iHUEABYKAB0WIQRAhzRXHqcMeLMyaSiRxhvAZXjcogUCZpEHIgAKCRCRxhvAZXjc
 ovTvAQDvxpq1CIJz4arkf6lkI1VX1PcSfyV1+aIsXkrGF01tfwD+PekJH0xJ7RqU
 ysuMo1uG3i1OO2xIdrdwCXJDng4QggE=
 =LtRf
 -----END PGP SIGNATURE-----

Merge tag 'vfs-6.11.pidfs' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs

Pull pidfs updates from Christian Brauner:
 "This contains work to make it possible to derive namespace file
  descriptors from pidfd file descriptors.

  Right now it is already possible to use a pidfd with setns() to
  atomically change multiple namespaces at the same time. In other
  words, it is possible to switch to the namespace context of a process
  using a pidfd. There is no need to first open namespace file
  descriptors via procfs.

  The work included here is an extension of these abilities by allowing
  to open namespace file descriptors using a pidfd. This means it is now
  possible to interact with namespaces without ever touching procfs.

  To this end a new set of ioctls() on pidfds is introduced covering all
  supported namespace types"

* tag 'vfs-6.11.pidfs' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs:
  pidfs: allow retrieval of namespace file descriptors
  nsfs: add open_namespace()
  nsproxy: add helper to go from arbitrary namespace to ns_common
  nsproxy: add a cleanup helper for nsproxy
  file: add take_fd() cleanup helper

fs/internal.h

@@ -17,6 +17,7 @@ struct fs_context;
 struct pipe_inode_info;
 struct iov_iter;
 struct mnt_idmap;
+struct ns_common;
 
 /*
  * block/bdev.c
@@ -239,6 +240,7 @@ extern void mnt_pin_kill(struct mount *m);
  * fs/nsfs.c
  */
 extern const struct dentry_operations ns_dentry_operations;
+int open_namespace(struct ns_common *ns);
 
 /*
  * fs/stat.c:

fs/nsfs.c

@@ -84,40 +84,47 @@ int ns_get_path(struct path *path, struct task_struct *task,
 	return ns_get_path_cb(path, ns_get_path_task, &args);
 }
 
-int open_related_ns(struct ns_common *ns,
+/**
-		   struct ns_common *(*get_ns)(struct ns_common *ns))
+ * open_namespace - open a namespace
+ * @ns: the namespace to open
+ *
+ * This will consume a reference to @ns indendent of success or failure.
+ *
+ * Return: A file descriptor on success or a negative error code on failure.
+ */
+int open_namespace(struct ns_common *ns)
 {
-	struct path path = {};
+	struct path path __free(path_put) = {};
-	struct ns_common *relative;
 	struct file *f;
 	int err;
-	int fd;
 
-	fd = get_unused_fd_flags(O_CLOEXEC);
+	/* call first to consume reference */
+	err = path_from_stashed(&ns->stashed, nsfs_mnt, ns, &path);
+	if (err < 0)
+		return err;
+
+	CLASS(get_unused_fd, fd)(O_CLOEXEC);
 	if (fd < 0)
 		return fd;
 
-	relative = get_ns(ns);
-	if (IS_ERR(relative)) {
-		put_unused_fd(fd);
-		return PTR_ERR(relative);
-	}
-
-	err = path_from_stashed(&relative->stashed, nsfs_mnt, relative, &path);
-	if (err < 0) {
-		put_unused_fd(fd);
-		return err;
-	}
-
 	f = dentry_open(&path, O_RDONLY, current_cred());
-	path_put(&path);
+	if (IS_ERR(f))
-	if (IS_ERR(f)) {
+		return PTR_ERR(f);
-		put_unused_fd(fd);
-		fd = PTR_ERR(f);
-	} else
-		fd_install(fd, f);
 
-	return fd;
+	fd_install(fd, f);
+	return take_fd(fd);
+}
+
+int open_related_ns(struct ns_common *ns,
+		   struct ns_common *(*get_ns)(struct ns_common *ns))
+{
+	struct ns_common *relative;
+
+	relative = get_ns(ns);
+	if (IS_ERR(relative))
+		return PTR_ERR(relative);
+
+	return open_namespace(relative);
 }
 EXPORT_SYMBOL_GPL(open_related_ns);
 

fs/pidfs.c

@@ -11,10 +11,16 @@
 #include <linux/proc_fs.h>
 #include <linux/proc_ns.h>
 #include <linux/pseudo_fs.h>
+#include <linux/ptrace.h>
 #include <linux/seq_file.h>
 #include <uapi/linux/pidfd.h>
+#include <linux/ipc_namespace.h>
+#include <linux/time_namespace.h>
+#include <linux/utsname.h>
+#include <net/net_namespace.h>
 
 #include "internal.h"
+#include "mount.h"
 
 #ifdef CONFIG_PROC_FS
 /**
@@ -108,11 +114,95 @@ static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
 	return poll_flags;
 }
 
+static long pidfd_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+	struct task_struct *task __free(put_task) = NULL;
+	struct nsproxy *nsp __free(put_nsproxy) = NULL;
+	struct pid *pid = pidfd_pid(file);
+	struct ns_common *ns_common;
+
+	if (arg)
+		return -EINVAL;
+
+	task = get_pid_task(pid, PIDTYPE_PID);
+	if (!task)
+		return -ESRCH;
+
+	scoped_guard(task_lock, task) {
+		nsp = task->nsproxy;
+		if (nsp)
+			get_nsproxy(nsp);
+	}
+	if (!nsp)
+		return -ESRCH; /* just pretend it didn't exist */
+
+	/*
+	 * We're trying to open a file descriptor to the namespace so perform a
+	 * filesystem cred ptrace check. Also, we mirror nsfs behavior.
+	 */
+	if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
+		return -EACCES;
+
+	switch (cmd) {
+	/* Namespaces that hang of nsproxy. */
+	case PIDFD_GET_CGROUP_NAMESPACE:
+		get_cgroup_ns(nsp->cgroup_ns);
+		ns_common = to_ns_common(nsp->cgroup_ns);
+		break;
+	case PIDFD_GET_IPC_NAMESPACE:
+		get_ipc_ns(nsp->ipc_ns);
+		ns_common = to_ns_common(nsp->ipc_ns);
+		break;
+	case PIDFD_GET_MNT_NAMESPACE:
+		get_mnt_ns(nsp->mnt_ns);
+		ns_common = to_ns_common(nsp->mnt_ns);
+		break;
+	case PIDFD_GET_NET_NAMESPACE:
+		ns_common = to_ns_common(nsp->net_ns);
+		get_net_ns(ns_common);
+		break;
+	case PIDFD_GET_PID_FOR_CHILDREN_NAMESPACE:
+		get_pid_ns(nsp->pid_ns_for_children);
+		ns_common = to_ns_common(nsp->pid_ns_for_children);
+		break;
+	case PIDFD_GET_TIME_NAMESPACE:
+		get_time_ns(nsp->time_ns);
+		ns_common = to_ns_common(nsp->time_ns);
+		break;
+	case PIDFD_GET_TIME_FOR_CHILDREN_NAMESPACE:
+		get_time_ns(nsp->time_ns_for_children);
+		ns_common = to_ns_common(nsp->time_ns_for_children);
+		break;
+	case PIDFD_GET_UTS_NAMESPACE:
+		get_uts_ns(nsp->uts_ns);
+		ns_common = to_ns_common(nsp->uts_ns);
+		break;
+	/* Namespaces that don't hang of nsproxy. */
+	case PIDFD_GET_USER_NAMESPACE:
+		rcu_read_lock();
+		ns_common = to_ns_common(get_user_ns(task_cred_xxx(task, user_ns)));
+		rcu_read_unlock();
+		break;
+	case PIDFD_GET_PID_NAMESPACE:
+		rcu_read_lock();
+		ns_common = to_ns_common(get_pid_ns(task_active_pid_ns(task)));
+		rcu_read_unlock();
+		break;
+	default:
+		return -ENOIOCTLCMD;
+	}
+
+	/* open_namespace() unconditionally consumes the reference */
+	return open_namespace(ns_common);
+}
+
 static const struct file_operations pidfs_file_operations = {
 	.poll		= pidfd_poll,
 #ifdef CONFIG_PROC_FS
 	.show_fdinfo	= pidfd_show_fdinfo,
 #endif
+	.unlocked_ioctl	= pidfd_ioctl,
+	.compat_ioctl   = compat_ptr_ioctl,
 };
 
 struct pid *pidfd_pid(const struct file *file)

include/linux/cleanup.h

@@ -63,17 +63,20 @@
 
 #define __free(_name)	__cleanup(__free_##_name)
 
-#define __get_and_null_ptr(p) \
+#define __get_and_null(p, nullvalue)   \
-	({ __auto_type __ptr = &(p); \
+	({                                  \
+		__auto_type __ptr = &(p);   \
 		__auto_type __val = *__ptr; \
-	   *__ptr = NULL;  __val; })
+		*__ptr = nullvalue;         \
+		__val;                      \
+	})
 
 static inline __must_check
 const volatile void * __must_check_fn(const volatile void *val)
 { return val; }
 
 #define no_free_ptr(p) \
-	((typeof(p)) __must_check_fn(__get_and_null_ptr(p)))
+	((typeof(p)) __must_check_fn(__get_and_null(p, NULL)))
 
 #define return_ptr(p)	return no_free_ptr(p)
 

include/linux/file.h

@@ -97,6 +97,26 @@ extern void put_unused_fd(unsigned int fd);
 DEFINE_CLASS(get_unused_fd, int, if (_T >= 0) put_unused_fd(_T),
 	     get_unused_fd_flags(flags), unsigned flags)
 
+/*
+ * take_fd() will take care to set @fd to -EBADF ensuring that
+ * CLASS(get_unused_fd) won't call put_unused_fd(). This makes it
+ * easier to rely on CLASS(get_unused_fd):
+ *
+ * struct file *f;
+ *
+ * CLASS(get_unused_fd, fd)(O_CLOEXEC);
+ * if (fd < 0)
+ *         return fd;
+ *
+ * f = dentry_open(&path, O_RDONLY, current_cred());
+ * if (IS_ERR(f))
+ *         return PTR_ERR(fd);
+ *
+ * fd_install(fd, f);
+ * return take_fd(fd);
+ */
+#define take_fd(fd) __get_and_null(fd, -EBADF)
+
 extern void fd_install(unsigned int fd, struct file *file);
 
 int receive_fd(struct file *file, int __user *ufd, unsigned int o_flags);

include/linux/nsproxy.h

@@ -42,6 +42,17 @@ struct nsproxy {
 };
 extern struct nsproxy init_nsproxy;
 
+#define to_ns_common(__ns)                              \
+	_Generic((__ns),                                \
+		struct cgroup_namespace *: &(__ns->ns), \
+		struct ipc_namespace *:    &(__ns->ns), \
+		struct net *:              &(__ns->ns), \
+		struct pid_namespace *:    &(__ns->ns), \
+		struct mnt_namespace *:    &(__ns->ns), \
+		struct time_namespace *:   &(__ns->ns), \
+		struct user_namespace *:   &(__ns->ns), \
+		struct uts_namespace *:    &(__ns->ns))
+
 /*
  * A structure to encompass all bits needed to install
  * a partial or complete new set of namespaces.
@@ -112,4 +123,6 @@ static inline void get_nsproxy(struct nsproxy *ns)
 	refcount_inc(&ns->count);
 }
 
+DEFINE_FREE(put_nsproxy, struct nsproxy *, if (_T) put_nsproxy(_T))
+
 #endif

include/uapi/linux/pidfd.h

@@ -5,6 +5,7 @@
 
 #include <linux/types.h>
 #include <linux/fcntl.h>
+#include <linux/ioctl.h>
 
 /* Flags for pidfd_open().  */
 #define PIDFD_NONBLOCK	O_NONBLOCK
@@ -15,4 +16,17 @@
 #define PIDFD_SIGNAL_THREAD_GROUP	(1UL << 1)
 #define PIDFD_SIGNAL_PROCESS_GROUP	(1UL << 2)
 
+#define PIDFS_IOCTL_MAGIC 0xFF
+
+#define PIDFD_GET_CGROUP_NAMESPACE            _IO(PIDFS_IOCTL_MAGIC, 1)
+#define PIDFD_GET_IPC_NAMESPACE               _IO(PIDFS_IOCTL_MAGIC, 2)
+#define PIDFD_GET_MNT_NAMESPACE               _IO(PIDFS_IOCTL_MAGIC, 3)
+#define PIDFD_GET_NET_NAMESPACE               _IO(PIDFS_IOCTL_MAGIC, 4)
+#define PIDFD_GET_PID_NAMESPACE               _IO(PIDFS_IOCTL_MAGIC, 5)
+#define PIDFD_GET_PID_FOR_CHILDREN_NAMESPACE  _IO(PIDFS_IOCTL_MAGIC, 6)
+#define PIDFD_GET_TIME_NAMESPACE              _IO(PIDFS_IOCTL_MAGIC, 7)
+#define PIDFD_GET_TIME_FOR_CHILDREN_NAMESPACE _IO(PIDFS_IOCTL_MAGIC, 8)
+#define PIDFD_GET_USER_NAMESPACE              _IO(PIDFS_IOCTL_MAGIC, 9)
+#define PIDFD_GET_UTS_NAMESPACE               _IO(PIDFS_IOCTL_MAGIC, 10)
+
 #endif /* _UAPI_LINUX_PIDFD_H */