From 96d3832f025452af33fda17e1a09eeb78455e7fa Mon Sep 17 00:00:00 2001
From: Stanislav Baiduzhyi <stanislav.baiduzhyi@kape.com>
Date: Wed, 19 Jun 2024 15:11:23 +0100
Subject: [PATCH] CVPN-956: handling of Full error on tun io-uring send

Treating tun io-uring Full error as OK scenario.
It is effectively the same scenario as a buffer in a network switch/router filling up so dropping the traffic is appropriate, higher level protocols (e.g. TCP) running over the tunnel will use their congestion control algorithms to adjust their send rate.
---
 Cargo.lock                        |  1 +
 lightway-app-utils/Cargo.toml     |  1 +
 lightway-app-utils/src/iouring.rs | 16 +++++++++++++---
 lightway-app-utils/src/lib.rs     |  1 +
 lightway-app-utils/src/metrics.rs |  8 ++++++++
 5 files changed, 24 insertions(+), 3 deletions(-)
 create mode 100644 lightway-app-utils/src/metrics.rs

diff --git a/Cargo.lock b/Cargo.lock
index 9e1b11c..7d57021 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -992,6 +992,7 @@ dependencies = [
  "humantime",
  "io-uring",
  "lightway-core",
+ "metrics",
  "pnet",
  "serde",
  "serde_with",
diff --git a/lightway-app-utils/Cargo.toml b/lightway-app-utils/Cargo.toml
index c095e51..01cadcc 100644
--- a/lightway-app-utils/Cargo.toml
+++ b/lightway-app-utils/Cargo.toml
@@ -32,6 +32,7 @@ tokio-eventfd = "0.2.1"
 dashmap = "5.5.3"
 thiserror = "1.0.57"
 tokio-tun = "0.11.2"
+metrics = "0.23.0"
 
 [[example]]
 name = "udprelay"
diff --git a/lightway-app-utils/src/iouring.rs b/lightway-app-utils/src/iouring.rs
index 48e9fb3..4c9a743 100644
--- a/lightway-app-utils/src/iouring.rs
+++ b/lightway-app-utils/src/iouring.rs
@@ -18,6 +18,7 @@ use std::{
 };
 use tokio::{io::AsyncReadExt, sync::Semaphore};
 use tokio_eventfd::EventFd;
+use crate::metrics::inside_io_send_failed;
 
 const REGISTERED_FD_INDEX: u32 = 0;
 const IOURING_SQPOLL_IDLE_TIME: u32 = 100;
@@ -190,11 +191,20 @@ impl<T: AsRawFd> IOUring<T> {
 
     /// Try Send packet to Tun device
     pub fn try_send(&self, buf: BytesMut) -> IOUringResult<()> {
-        self.inner
+        let try_send_res = self.inner
             .send_q
             .tx
-            .try_send(buf)
-            .map_err(IOUringError::SendError)
+            .try_send(buf);
+        match try_send_res {
+            Ok(()) => Ok(()),
+            Err(e) if e.is_full() => {
+                // it is effectively the same scenario as a buffer in a network
+                // switch/router filling up so dropping the traffic is appropriate
+                inside_io_send_failed();
+                Ok(())
+            },
+            Err(e) => Err(IOUringError::SendError(e))
+        }
     }
 }
 
diff --git a/lightway-app-utils/src/lib.rs b/lightway-app-utils/src/lib.rs
index a892ec9..24cb1c1 100644
--- a/lightway-app-utils/src/lib.rs
+++ b/lightway-app-utils/src/lib.rs
@@ -26,5 +26,6 @@ pub use iouring::IOUring;
 
 pub use tun::Tun;
 
+mod metrics;
 mod utils;
 pub use utils::is_file_path_valid;
diff --git a/lightway-app-utils/src/metrics.rs b/lightway-app-utils/src/metrics.rs
new file mode 100644
index 0000000..682bd20
--- /dev/null
+++ b/lightway-app-utils/src/metrics.rs
@@ -0,0 +1,8 @@
+use metrics::counter;
+
+const METRIC_TUN_IOURING_PACKET_DROPPED: &str = "tun_iouring_packet_dropped";
+
+/// Receiving [`crate::iouring::IOUringError::SendError`] means that the
+pub(crate) fn inside_io_send_failed() {
+    counter!(METRIC_TUN_IOURING_PACKET_DROPPED).increment(1);
+}