xfs: reduce i_rwsem contention on buffered randrw workload

Warm up page cache before buffered read to minimize time spent under
shared ilock.

Truncate and hole punch page cache invalidation is protected by MMAPLOCK,
but pNFS leases require IOLOCK protection of buffered reads, so skip
page cache warm up when pNFS block export is enabled.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>

fs/xfs/xfs_file.c

@@ -264,9 +264,29 @@ xfs_file_buffered_read(
 
 	trace_xfs_file_buffered_read(iocb, to);
 
-	ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
-	if (ret)
-		return ret;
+	if (iocb->ki_flags & IOCB_NOWAIT) {
+		if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
+			return -EAGAIN;
+	} else {
+#ifndef CONFIG_EXPORTFS_BLOCK_OPS
+		/*
+		 * Warm up page cache to minimize time spent under
+		 * shared ilock. truncate and hole punch page invalidation
+		 * is protected by MMAPLOCK, but pNFS leases require IOLOCK
+		 * protection of buffered reads.
+		 */
+		struct iov_iter	iter;
+		loff_t pos = iocb->ki_pos;
+
+		iov_iter_discard(&iter, READ, iov_iter_count(to));
+		ret = generic_file_read_iter(iocb, &iter);
+		if (ret <= 0)
+			return ret;
+
+		iocb->ki_pos = pos;
+#endif
+		xfs_ilock(ip, XFS_IOLOCK_SHARED);
+	}
 	ret = generic_file_read_iter(iocb, to);
 	xfs_iunlock(ip, XFS_IOLOCK_SHARED);