iio: buffer: make sure O_NONBLOCK is respected

For output buffers, there's no guarantee that the buffer won't be full
in the first iteration of the loop in which case we would block
independently of userspace passing O_NONBLOCK or not. Fix it by always
checking the flag before going to sleep.

While at it (and as it's a bit related), refactored the loop so that the
stop condition is 'written != n', i.e, run the loop until all data has
been copied into the IIO buffers. This makes the code a bit simpler.

Fixes: 9eeee3b0 ("iio: Add output buffer support")
Signed-off-by: Nuno Sá <nuno.sa@analog.com>
Reviewed-by: Lars-Peter Clausen <lars@metafoo.de>
Link: https://lore.kernel.org/r/20230216101452.591805-3-nuno.sa@analog.com
Cc: <Stable@vger.kernel.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>

drivers/iio/industrialio-buffer.c

@@ -203,21 +203,24 @@ static ssize_t iio_buffer_write(struct file *filp, const char __user *buf,
 				break;
 			}
 
+			if (filp->f_flags & O_NONBLOCK) {
+				if (!written)
+					ret = -EAGAIN;
+				break;
+			}
+
 			wait_woken(&wait, TASK_INTERRUPTIBLE,
 					MAX_SCHEDULE_TIMEOUT);
 			continue;
 		}
 
 		ret = rb->access->write(rb, n - written, buf + written);
-		if (ret == 0 && (filp->f_flags & O_NONBLOCK))
-			ret = -EAGAIN;
+		if (ret < 0)
+			break;
 
-		if (ret > 0) {
-			written += ret;
-			if (written != n && !(filp->f_flags & O_NONBLOCK))
-				continue;
-		}
-	} while (ret == 0);
+		written += ret;
+
+	} while (written != n);
 	remove_wait_queue(&rb->pollq, &wait);
 
 	return ret < 0 ? ret : written;