fs/pipe: use spinlock in pipe_read() only if there is a watch_queue

mainline inclusion
from mainline-v6.7-rc1
category: performance

If there is no watch_queue, holding the pipe mutex is enough to
prevent concurrent writes, and we can avoid the spinlock.

O_NOTIFICATION_QUEUE is an exotic and rarely used feature, and of all
the pipes that exist at any given time, only very few actually have a
watch_queue, therefore it appears worthwile to optimize the common
case.

This patch does not optimize pipe_resize_ring() where the spinlocks
could be avoided as well; that does not seem like a worthwile
optimization because this function is not called often.

Related commits:

- commit 8df4412 ("pipe: Check for ring full inside of the
  spinlock in pipe_write()")
- commit b667b86 ("pipe: Advance tail pointer inside of wait
  spinlock in pipe_read()")
- commit 189b0dd ("pipe: Fix missing lock in pipe_resize_ring()")

Signed-off-by: Max Kellermann <max.kellermann@ionos.com>
Message-Id: <20230921075755.1378787-4-max.kellermann@ionos.com>
Reviewed-by: David Howells <dhowells@redhat.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
(cherry picked from commit 478dbf1)
Signed-off-by: Wentao Guan <guanwentao@uniontech.com>
Change-Id: Icdeb15fb11c0c6e6d07a3adb17b46457795396a0

Author: MaxKellermann

fs/pipe.c

@@ -227,6 +227,36 @@ static inline bool pipe_readable(const struct pipe_inode_info *pipe)
 	return !pipe_empty(head, tail) || !writers;
 }
 
+static inline unsigned int pipe_update_tail(struct pipe_inode_info *pipe,
+					    struct pipe_buffer *buf,
+					    unsigned int tail)
+{
+	pipe_buf_release(pipe, buf);
+
+	/*
+	 * If the pipe has a watch_queue, we need additional protection
+	 * by the spinlock because notifications get posted with only
+	 * this spinlock, no mutex
+	 */
+	if (pipe_has_watch_queue(pipe)) {
+		spin_lock_irq(&pipe->rd_wait.lock);
+#ifdef CONFIG_WATCH_QUEUE
+		if (buf->flags & PIPE_BUF_FLAG_LOSS)
+			pipe->note_loss = true;
+#endif
+		pipe->tail = ++tail;
+		spin_unlock_irq(&pipe->rd_wait.lock);
+		return tail;
+	}
+
+	/*
+	 * Without a watch_queue, we can simply increment the tail
+	 * without the spinlock - the mutex is enough.
+	 */
+	pipe->tail = ++tail;
+	return tail;
+}
+
 static ssize_t
 pipe_read(struct kiocb *iocb, struct iov_iter *to)
 {
@@ -320,17 +350,8 @@ pipe_read(struct kiocb *iocb, struct iov_iter *to)
 				buf->len = 0;
 			}
 
-			if (!buf->len) {
-				pipe_buf_release(pipe, buf);
-				spin_lock_irq(&pipe->rd_wait.lock);
-#ifdef CONFIG_WATCH_QUEUE
-				if (buf->flags & PIPE_BUF_FLAG_LOSS)
-					pipe->note_loss = true;
-#endif
-				tail++;
-				pipe->tail = tail;
-				spin_unlock_irq(&pipe->rd_wait.lock);
-			}
+			if (!buf->len)
+				tail = pipe_update_tail(pipe, buf, tail);
 			total_len -= chars;
 			if (!total_len)
 				break;	/* common path: read succeeded */