pin tun reader threads to CPUs so per-flow packets keep wire order

Each listenIn goroutine locks its OS thread and pins it to one CPU
(sched_setaffinity), so every UDP send from that goroutine leaves
through the same XPS-selected NIC TX ring instead of being sprayed
across rings and reordered. On by default via tun.pin_threads; queue i
pins to the i-th entry of the process's allowed CPU set (respecting
cpuset/taskset masks, whose IDs are often not 0..NumCPU-1), or to an
explicit tun.cpu_affinity list, validated against that same allowed
set. Linux only; pinning is a no-op elsewhere.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
JackDoan
2026-07-16 12:25:23 -05:00
parent 913a37cfee
commit 10e9514e44
6 changed files with 242 additions and 2 deletions
+67
View File
@@ -7,6 +7,7 @@ import (
"net"
"net/netip"
"runtime/debug"
"slices"
"strings"
"time"
@@ -231,6 +232,8 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
relayManager: NewRelayManager(ctx, l, hostMap, c),
punchy: punchy,
ConntrackCacheTimeout: conntrackCacheTimeout,
CpuAffinity: parseCpuAffinity(c, l, routines),
PinThreads: c.GetBool("tun.pin_threads", true),
l: l,
}
@@ -282,6 +285,70 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
}, nil
}
// parseCpuAffinity reads `tun.cpu_affinity` from the config — a list of
// integer CPU IDs, one per TUN reader goroutine. Empty / unset returns nil
// (listenIn falls back to spreading queues across the allowed CPU set).
// Length mismatch with `routines` is a warning, not an error: shorter lists
// are modulo-cycled across queues, longer lists' tail is ignored. Invalid
// entries (non-integer, or a CPU ID we're not allowed to run on) are also a
// warning and disable the override entirely so we don't silently pin to the
// wrong CPU. Entries are validated against the process's current affinity
// mask (util.AllowedCPUs) rather than 0..NumCPU-1: under a cgroup cpuset or
// taskset the runnable IDs are frequently not that contiguous range, and
// pinning to an unrunnable ID always fails. If the allowed set can't be
// determined we fall back to a plain non-negative check.
func parseCpuAffinity(c *config.C, l *slog.Logger, routines int) []int {
raw := c.Get("tun.cpu_affinity")
if raw == nil {
return nil
}
rv, ok := raw.([]any)
if !ok {
l.Warn("tun.cpu_affinity must be a list of integers; ignoring", "value", raw)
return nil
}
// allowed is the set of CPU IDs we're actually permitted to run on. A nil
// slice (unsupported platform or lookup error) means "can't tell", so we
// only apply the weaker non-negative check in that case.
allowed, err := util.AllowedCPUs()
if err != nil {
l.Warn("could not determine allowed CPUs; validating tun.cpu_affinity against non-negative only", "error", err)
allowed = nil
}
cpus := make([]int, 0, len(rv))
for i, e := range rv {
var cpu int
switch v := e.(type) {
case int:
cpu = v
case int64:
cpu = int(v)
case float64:
cpu = int(v)
default:
l.Warn("tun.cpu_affinity entry not an integer; ignoring affinity",
"index", i, "value", e)
return nil
}
if cpu < 0 {
l.Warn("tun.cpu_affinity entry out of range; ignoring affinity",
"index", i, "cpu", cpu)
return nil
}
if len(allowed) > 0 && !slices.Contains(allowed, cpu) {
l.Warn("tun.cpu_affinity entry not in allowed CPU set; ignoring affinity",
"index", i, "cpu", cpu, "allowed", allowed)
return nil
}
cpus = append(cpus, cpu)
}
if len(cpus) != routines {
l.Warn("tun.cpu_affinity length doesn't match routines; queues will modulo-cycle through the list",
"affinity_len", len(cpus), "routines", routines)
}
return cpus
}
func moduleVersion() string {
info, ok := debug.ReadBuildInfo()
if !ok {