seccomp: introduce writer locking

Normally, task_struct.seccomp.filter is only ever read or modified by
the task that owns it (current). This property aids in fast access
during system call filtering as read access is lockless.

Updating the pointer from another task, however, opens up race
conditions. To allow cross-thread filter pointer updates, writes to the
seccomp fields are now protected by the sighand spinlock (which is shared
by all threads in the thread group). Read access remains lockless because
pointer updates themselves are atomic. However, writes (or cloning)
often entail additional checking (like maximum instruction counts)
which require locking to perform safely.

In the case of cloning threads, the child is invisible to the system
until it enters the task list. To make sure a child can't be cloned from
a thread and left in a prior state, seccomp duplication is additionally
moved under the sighand lock. Then parent and child are certain have
the same seccomp state when they exit the lock.

Based on patches by Will Drewry and David Drysdale.

Signed-off-by: Kees Cook <[e-mail]>
Reviewed-by: Oleg Nesterov <[e-mail]>
Reviewed-by: Andy Lutomirski <[e-mail]>

include/linux/seccomp.h

···
  *
  * @mode:  indicates one of the valid values above for controlled
  *         system calls available to a process.
- * @filter: The metadata and ruleset for determining what system calls
- *          are allowed for a task.
+ * @filter: must always point to a valid seccomp-filter or NULL as it is
+ *          accessed without locking during system call entry.
  *
  *          @filter must only be accessed from the context of current as there
- *          is no locking.
+ *          is no read locking.
  */
 struct seccomp {
 	int mode;

kernel/fork.c

···
 		goto free_ti;
 
 	tsk->stack = ti;
+#ifdef CONFIG_SECCOMP
+	/*
+	 * We must handle setting up seccomp filters once we're under
+	 * the sighand lock in case orig has changed between now and
+	 * then. Until then, filter must be NULL to avoid messing up
+	 * the usage counts on the error path calling free_task.
+	 */
+	tsk->seccomp.filter = NULL;
+#endif
 
 	setup_thread_stack(tsk, orig);
 	clear_user_return_notifier(tsk);
···
 	return 0;
 }
 
+static void copy_seccomp(struct task_struct *p)
+{
+#ifdef CONFIG_SECCOMP
+	/*
+	 * Must be called with sighand->lock held, which is common to
+	 * all threads in the group. Holding cred_guard_mutex is not
+	 * needed because this new task is not yet running and cannot
+	 * be racing exec.
+	 */
+	BUG_ON(!spin_is_locked(&current->sighand->siglock));
+
+	/* Ref-count the new filter user, and assign it. */
+	get_seccomp_filter(current);
+	p->seccomp = current->seccomp;
+
+	/*
+	 * Explicitly enable no_new_privs here in case it got set
+	 * between the task_struct being duplicated and holding the
+	 * sighand lock. The seccomp state and nnp must be in sync.
+	 */
+	if (task_no_new_privs(current))
+		task_set_no_new_privs(p);
+
+	/*
+	 * If the parent gained a seccomp mode after copying thread
+	 * flags and between before we held the sighand lock, we have
+	 * to manually enable the seccomp thread flag here.
+	 */
+	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
+		set_tsk_thread_flag(p, TIF_SECCOMP);
+#endif
+}
+
 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
 {
 	current->clear_child_tid = tidptr;
···
 		goto fork_out;
 
 	ftrace_graph_init_task(p);
-	get_seccomp_filter(p);
 
 	rt_mutex_init_task(p);
 
···
 	}
 
 	spin_lock(&current->sighand->siglock);
+
+	/*
+	 * Copy seccomp details explicitly here, in case they were changed
+	 * before holding sighand lock.
+	 */
+	copy_seccomp(p);
 
 	/*
 	 * Process group and session signals need to be delivered to just the

kernel/seccomp.c

···
 
 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
 {
+	BUG_ON(!spin_is_locked(&current->sighand->siglock));
+
 	if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
 		return false;
 
···
 
 static inline void seccomp_assign_mode(unsigned long seccomp_mode)
 {
+	BUG_ON(!spin_is_locked(&current->sighand->siglock));
+
 	current->seccomp.mode = seccomp_mode;
 	set_tsk_thread_flag(current, TIF_SECCOMP);
 }
···
  * @flags:  flags to change filter behavior
  * @filter: seccomp filter to add to the current process
  *
+ * Caller must be holding current->sighand->siglock lock.
+ *
  * Returns 0 on success, -ve on error.
  */
 static long seccomp_attach_filter(unsigned int flags,
···
 {
 	unsigned long total_insns;
 	struct seccomp_filter *walker;
+
+	BUG_ON(!spin_is_locked(&current->sighand->siglock));
 
 	/* Validate resulting filter length. */
 	total_insns = filter->prog->len;
···
 	const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
 	long ret = -EINVAL;
 
+	spin_lock_irq(&current->sighand->siglock);
+
 	if (!seccomp_may_assign_mode(seccomp_mode))
 		goto out;
 
···
 	ret = 0;
 
 out:
+	spin_unlock_irq(&current->sighand->siglock);
 
 	return ret;
 }
···
 
 	/* Validate flags. */
 	if (flags != 0)
-		goto out;
+		return -EINVAL;
 
 	/* Prepare the new filter before holding any locks. */
 	prepared = seccomp_prepare_user_filter(filter);
 	if (IS_ERR(prepared))
 		return PTR_ERR(prepared);
+
+	spin_lock_irq(&current->sighand->siglock);
 
 	if (!seccomp_may_assign_mode(seccomp_mode))
 		goto out;
···
 
 	seccomp_assign_mode(seccomp_mode);
 out:
+	spin_unlock_irq(&current->sighand->siglock);
 	seccomp_filter_free(prepared);
 	return ret;
 }