From dbb9c935981cba8e37b57ec344606f2218d48bc8 Mon Sep 17 00:00:00 2001
From: sidneychang <2190206983@qq.com>
Date: Mon, 26 Jan 2026 18:36:53 -0500
Subject: [PATCH] fix(network): add cleanupOrphanTaps() to remove orphan TAPs
 by carrier state

- Call new function cleanupOrphanTaps() at the start of
  DynamicNetwork.NetworkSetup().
- Add cleanupOrphanTaps(): scan netns for interfaces matching
  ^tap.*_urunc$ and use kernel carrier/operational state as the
  sole criterion:
  - NO-CARRIER => delete orphan (remove TC/qdisc, then delete link)
  - LOWER_UP / operational up / FlagRunning => treat as in-use and abort
- Do not scan /proc or check /dev/net/tun; do not attempt to reuse TAPs.
- Skip cleanup when no container interface (e.g. no eth0) is present.
- Remove PID/FD based checks and netns flock; document the single
  unikernel-per-netns assumption.
- Preserve networkSetup() create-only semantics and ensure TC/qdisc
  cleanup before link deletion.

This resolves an issue on Kubernetes where restarting urunc left
orphan TAP devices in the pod network namespace and prevented
subsequent network setup.

Signed-off-by: sidneychang <2190206983@qq.com>
---
 pkg/network/network.go         | 60 ++++++++++++++++++++++++++++++++++
 pkg/network/network_dynamic.go |  5 +++
 2 files changed, 65 insertions(+)

diff --git a/pkg/network/network.go b/pkg/network/network.go
index 6aa55e22..ceeb2b94 100644
--- a/pkg/network/network.go
+++ b/pkg/network/network.go
@@ -18,6 +18,7 @@ import (
 	"errors"
 	"fmt"
 	"net"
+	"regexp"
 	"strings"
 
 	"github.com/jackpal/gateway"
@@ -389,3 +390,62 @@ func deleteTapDevice(device netlink.Link) error {
 	}
 	return nil
 }
+
+func cleanupOrphanTaps() error {
+	netlog.Debug("running cleanupOrphanTaps (carrier-state based)")
+
+	// If there is no container interface (e.g. no eth0), do not attempt to create/delete taps.
+	// This avoids touching taps in netns that aren't ready or belong to other runtimes (ctr).
+	if _, err := netlink.LinkByName(DefaultInterface); err != nil {
+		netlog.Debug("no container interface found in namespace; skipping orphan TAP cleanup")
+		return nil
+	}
+
+	// Per design: assume at-most-one unikernel per netns. No inter-process netns lock is used.
+
+	handle, err := netlink.NewHandle()
+	if err != nil {
+		return fmt.Errorf("failed to get netlink handle: %w", err)
+	}
+	defer handle.Close()
+
+	links, err := handle.LinkList()
+	if err != nil {
+		return fmt.Errorf("failed to list links: %w", err)
+	}
+
+	tapRe := regexp.MustCompile(`^tap.*_urunc$`)
+	for _, link := range links {
+		attrs := link.Attrs()
+		if attrs == nil {
+			continue
+		}
+		name := attrs.Name
+		if !tapRe.MatchString(name) {
+			continue
+		}
+
+		// The device is in a 'Zombie' state: Administrative status is UP, but
+		// Operational status is DOWN with NO-CARRIER.
+		// In the Linux TUN/TAP driver model, NO-CARRIER on an UP interface
+		// definitively proves that no userspace process holds the file descriptor
+		// for this device.
+		if (attrs.Flags&net.FlagRunning) != 0 || attrs.OperState == netlink.OperUp {
+			return fmt.Errorf("found tap %s with carrier/oper state UP: aborting cleanup (unikernel may be running)", name)
+		}
+
+		netlog.Debugf("deleting orphan tap %s (no carrier)", name)
+		if err := deleteAllTCFilters(link); err != nil {
+			return fmt.Errorf("failed to delete tc filters for %s: %w", name, err)
+		}
+		if err := deleteAllQDiscs(link); err != nil {
+			return fmt.Errorf("failed to delete qdiscs for %s: %w", name, err)
+		}
+		if err := deleteTapDevice(link); err != nil {
+			return fmt.Errorf("failed to delete tap %s: %w", name, err)
+		}
+		netlog.Debugf("deleted orphan tap %s", name)
+	}
+
+	return nil
+}
diff --git a/pkg/network/network_dynamic.go b/pkg/network/network_dynamic.go
index f4264bbe..58990549 100644
--- a/pkg/network/network_dynamic.go
+++ b/pkg/network/network_dynamic.go
@@ -35,6 +35,11 @@ type DynamicNetwork struct {
 // for multiple unikernels in the same pod/network namespace.
 // See: https://github.com/urunc-dev/urunc/issues/13
 func (n DynamicNetwork) NetworkSetup(uid uint32, gid uint32) (*UnikernelNetworkInfo, error) {
+	// Attempt to clean up orphan TAPs created by urunc in this netns
+	if err := cleanupOrphanTaps(); err != nil {
+		return nil, fmt.Errorf("cleanupOrphanTaps failed: %w", err)
+	}
+
 	tapIndex, err := getTapIndex()
 	if err != nil {
 		return nil, fmt.Errorf("getTapIndex failed: %w", err)