From ba8c043079b38748e57adf1f80e3d86a4060efc5 Mon Sep 17 00:00:00 2001
From: Ben Kochie <superq@gmail.com>
Date: Mon, 24 Oct 2022 09:30:32 +0100
Subject: [PATCH] Add procfs fallback to netdev collector (#2509)

Some systems have broken netlink messages due to patched kernels. Since
these messages can not be parsed, add a flag to fall back to parsing
from `/proc/net/dev`.

Fixes: https://github.com/prometheus/node_exporter/issues/2502

Signed-off-by: Ben Kochie <superq@gmail.com>

Signed-off-by: Ben Kochie <superq@gmail.com>
---
 collector/netdev_linux.go      | 66 +++++++++++++++++++++++++++++++---
 collector/netdev_linux_test.go | 10 +++---
 2 files changed, 67 insertions(+), 9 deletions(-)

diff --git a/collector/netdev_linux.go b/collector/netdev_linux.go
index 325d10b7..527d33e4 100644
--- a/collector/netdev_linux.go
+++ b/collector/netdev_linux.go
@@ -17,28 +17,42 @@
 package collector
 
 import (
+	"fmt"
+
 	"github.com/go-kit/log"
 	"github.com/go-kit/log/level"
-
 	"github.com/jsimonetti/rtnetlink"
+	"github.com/prometheus/procfs"
+	"gopkg.in/alecthomas/kingpin.v2"
+)
+
+var (
+	netDevNetlink = kingpin.Flag("collector.netdev.netlink", "Use netlink to gather stats instead of /proc/net/dev.").Default("true").Bool()
 )
 
 func getNetDevStats(filter *deviceFilter, logger log.Logger) (netDevStats, error) {
+	if *netDevNetlink {
+		return netlinkStats(filter, logger)
+	}
+	return procNetDevStats(filter, logger)
+}
+
+func netlinkStats(filter *deviceFilter, logger log.Logger) (netDevStats, error) {
 	conn, err := rtnetlink.Dial(nil)
 	if err != nil {
 		return nil, err
 	}
-	defer conn.Close()
 
+	defer conn.Close()
 	links, err := conn.Link.List()
 	if err != nil {
 		return nil, err
 	}
 
-	return netlinkStats(links, filter, logger), nil
+	return parseNetlinkStats(links, filter, logger), nil
 }
 
-func netlinkStats(links []rtnetlink.LinkMessage, filter *deviceFilter, logger log.Logger) netDevStats {
+func parseNetlinkStats(links []rtnetlink.LinkMessage, filter *deviceFilter, logger log.Logger) netDevStats {
 	metrics := netDevStats{}
 
 	for _, msg := range links {
@@ -87,3 +101,47 @@ func netlinkStats(links []rtnetlink.LinkMessage, filter *deviceFilter, logger lo
 
 	return metrics
 }
+
+func procNetDevStats(filter *deviceFilter, logger log.Logger) (netDevStats, error) {
+	metrics := netDevStats{}
+
+	fs, err := procfs.NewFS(*procPath)
+	if err != nil {
+		return metrics, fmt.Errorf("failed to open procfs: %w", err)
+	}
+
+	netDev, err := fs.NetDev()
+	if err != nil {
+		return metrics, fmt.Errorf("failed to parse /proc/net/dev: %w", err)
+	}
+
+	for _, stats := range netDev {
+		name := stats.Name
+
+		if filter.ignored(name) {
+			level.Debug(logger).Log("msg", "Ignoring device", "device", name)
+			continue
+		}
+
+		metrics[name] = map[string]uint64{
+			"receive_bytes":       stats.RxBytes,
+			"receive_packets":     stats.RxPackets,
+			"receive_errors":      stats.RxErrors,
+			"receive_dropped":     stats.RxDropped,
+			"receive_fifo":        stats.RxFIFO,
+			"receive_frame":       stats.RxFrame,
+			"receive_compressed":  stats.RxCompressed,
+			"receive_multicast":   stats.RxMulticast,
+			"transmit_bytes":      stats.TxBytes,
+			"transmit_packets":    stats.TxPackets,
+			"transmit_errors":     stats.TxErrors,
+			"transmit_dropped":    stats.TxDropped,
+			"transmit_fifo":       stats.TxFIFO,
+			"transmit_colls":      stats.TxCollisions,
+			"transmit_carrier":    stats.TxCarrier,
+			"transmit_compressed": stats.TxCompressed,
+		}
+	}
+
+	return metrics, nil
+}
diff --git a/collector/netdev_linux_test.go b/collector/netdev_linux_test.go
index 32e3d16b..ff5e1c22 100644
--- a/collector/netdev_linux_test.go
+++ b/collector/netdev_linux_test.go
@@ -163,7 +163,7 @@ var links = []rtnetlink.LinkMessage{
 func TestNetDevStatsIgnore(t *testing.T) {
 	filter := newDeviceFilter("^veth", "")
 
-	netStats := netlinkStats(links, &filter, log.NewNopLogger())
+	netStats := parseNetlinkStats(links, &filter, log.NewNopLogger())
 
 	if want, got := uint64(10437182923), netStats["wlan0"]["receive_bytes"]; want != got {
 		t.Errorf("want netstat wlan0 bytes %v, got %v", want, got)
@@ -196,7 +196,7 @@ func TestNetDevStatsIgnore(t *testing.T) {
 
 func TestNetDevStatsAccept(t *testing.T) {
 	filter := newDeviceFilter("", "^💩0$")
-	netStats := netlinkStats(links, &filter, log.NewNopLogger())
+	netStats := parseNetlinkStats(links, &filter, log.NewNopLogger())
 
 	if want, got := 1, len(netStats); want != got {
 		t.Errorf("want count of devices to be %d, got %d", want, got)
@@ -227,7 +227,7 @@ func TestNetDevLegacyMetricNames(t *testing.T) {
 	}
 
 	filter := newDeviceFilter("", "")
-	netStats := netlinkStats(links, &filter, log.NewNopLogger())
+	netStats := parseNetlinkStats(links, &filter, log.NewNopLogger())
 
 	for dev, devStats := range netStats {
 		legacy(devStats)
@@ -260,7 +260,7 @@ func TestNetDevLegacyMetricValues(t *testing.T) {
 	}
 
 	filter := newDeviceFilter("", "^enp0s0f0$")
-	netStats := netlinkStats(links, &filter, log.NewNopLogger())
+	netStats := parseNetlinkStats(links, &filter, log.NewNopLogger())
 	metrics, ok := netStats["enp0s0f0"]
 	if !ok {
 		t.Error("expected stats for interface enp0s0f0")
@@ -282,7 +282,7 @@ func TestNetDevLegacyMetricValues(t *testing.T) {
 
 func TestNetDevMetricValues(t *testing.T) {
 	filter := newDeviceFilter("", "")
-	netStats := netlinkStats(links, &filter, log.NewNopLogger())
+	netStats := parseNetlinkStats(links, &filter, log.NewNopLogger())
 
 	for _, msg := range links {
 		device := msg.Attributes.Name