eventfd: track eventfd_signal() recursion depth

commit b5e683d5 upstream.

eventfd use cases from aio and io_uring can deadlock due to circular
or resursive calling, when eventfd_signal() tries to grab the waitqueue
lock. On top of that, it's also possible to construct notification
chains that are deep enough that we could blow the stack.

Add a percpu counter that tracks the percpu recursion depth, warn if we
exceed it. The counter is also exposed so that users of eventfd_signal()
can do the right thing if it's non-zero in the context where it is
called.

Cc: stable@vger.kernel.org # 4.19+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

fs/eventfd.c

@@ -22,6 +22,8 @@
 #include <linux/proc_fs.h>
 #include <linux/seq_file.h>
 
+DEFINE_PER_CPU(int, eventfd_wake_count);
+
 struct eventfd_ctx {
 	struct kref kref;
 	wait_queue_head_t wqh;
@@ -55,12 +57,25 @@ __u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n)
 {
 	unsigned long flags;
 
+	/*
+	 * Deadlock or stack overflow issues can happen if we recurse here
+	 * through waitqueue wakeup handlers. If the caller users potentially
+	 * nested waitqueues with custom wakeup handlers, then it should
+	 * check eventfd_signal_count() before calling this function. If
+	 * it returns true, the eventfd_signal() call should be deferred to a
+	 * safe context.
+	 */
+	if (WARN_ON_ONCE(this_cpu_read(eventfd_wake_count)))
+		return 0;
+
 	spin_lock_irqsave(&ctx->wqh.lock, flags);
+	this_cpu_inc(eventfd_wake_count);
 	if (ULLONG_MAX - ctx->count < n)
 		n = ULLONG_MAX - ctx->count;
 	ctx->count += n;
 	if (waitqueue_active(&ctx->wqh))
 		wake_up_locked_poll(&ctx->wqh, EPOLLIN);
+	this_cpu_dec(eventfd_wake_count);
 	spin_unlock_irqrestore(&ctx->wqh.lock, flags);
 
 	return n;

include/linux/eventfd.h

@@ -12,6 +12,8 @@
 #include <linux/fcntl.h>
 #include <linux/wait.h>
 #include <linux/err.h>
+#include <linux/percpu-defs.h>
+#include <linux/percpu.h>
 
 /*
  * CAREFUL: Check include/uapi/asm-generic/fcntl.h when defining
@@ -40,6 +42,13 @@ __u64 eventfd_signal(struct eventfd_ctx *ctx, __u64 n);
 int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx, wait_queue_entry_t *wait,
 				  __u64 *cnt);
 
+DECLARE_PER_CPU(int, eventfd_wake_count);
+
+static inline bool eventfd_signal_count(void)
+{
+	return this_cpu_read(eventfd_wake_count);
+}
+
 #else /* CONFIG_EVENTFD */
 
 /*
@@ -68,6 +77,11 @@ static inline int eventfd_ctx_remove_wait_queue(struct eventfd_ctx *ctx,
 	return -ENOSYS;
 }
 
+static inline bool eventfd_signal_count(void)
+{
+	return false;
+}
+
 #endif
 
 #endif /* _LINUX_EVENTFD_H */