client, cmd/tailscale/cli, feature/relayserver, net/udprelay: implement tailscale debug peer-relay-sessions (#17239)

Fixes tailscale/corp#30035

Signed-off-by: Dylan Bargatze <dylan@tailscale.com>
Signed-off-by: Jordan Whited <jordan@tailscale.com>
Co-authored-by: Dylan Bargatze <dylan@tailscale.com>
This commit is contained in:
Jordan Whited
2025-09-23 15:26:10 -07:00
committed by GitHub
parent 87ccfbd250
commit 4657cbdb11
13 changed files with 341 additions and 7 deletions

View File

@@ -31,6 +31,7 @@ import (
"tailscale.com/net/sockopts"
"tailscale.com/net/stun"
"tailscale.com/net/udprelay/endpoint"
"tailscale.com/net/udprelay/status"
"tailscale.com/tstime"
"tailscale.com/types/key"
"tailscale.com/types/logger"
@@ -95,6 +96,8 @@ type serverEndpoint struct {
boundAddrPorts [2]netip.AddrPort // or zero value if a handshake has never completed for that relay leg
lastSeen [2]time.Time // TODO(jwhited): consider using mono.Time
challenge [2][disco.BindUDPRelayChallengeLen]byte
packetsRx [2]uint64 // num packets received from/sent by each client after they are bound
bytesRx [2]uint64 // num bytes received from/sent by each client after they are bound
lamportID uint64
vni uint32
@@ -223,9 +226,13 @@ func (e *serverEndpoint) handlePacket(from netip.AddrPort, gh packet.GeneveHeade
switch {
case from == e.boundAddrPorts[0]:
e.lastSeen[0] = time.Now()
e.packetsRx[0]++
e.bytesRx[0] += uint64(len(b))
return b, e.boundAddrPorts[1]
case from == e.boundAddrPorts[1]:
e.lastSeen[1] = time.Now()
e.packetsRx[1]++
e.bytesRx[1] += uint64(len(b))
return b, e.boundAddrPorts[0]
default:
// unrecognized source
@@ -782,3 +789,41 @@ func (s *Server) AllocateEndpoint(discoA, discoB key.DiscoPublic) (endpoint.Serv
SteadyStateLifetime: tstime.GoDuration{Duration: s.steadyStateLifetime},
}, nil
}
// extractClientInfo constructs a [status.ClientInfo] for one of the two peer
// relay clients involved in this session.
func extractClientInfo(idx int, ep *serverEndpoint) status.ClientInfo {
if idx != 0 && idx != 1 {
panic(fmt.Sprintf("idx passed to extractClientInfo() must be 0 or 1; got %d", idx))
}
return status.ClientInfo{
Endpoint: ep.boundAddrPorts[idx],
ShortDisco: ep.discoPubKeys.Get()[idx].ShortString(),
PacketsTx: ep.packetsRx[idx],
BytesTx: ep.bytesRx[idx],
}
}
// GetSessions returns a slice of peer relay session statuses, with each
// entry containing detailed info about the server and clients involved in
// each session. This information is intended for debugging/status UX, and
// should not be relied on for any purpose outside of that.
func (s *Server) GetSessions() []status.ServerSession {
s.mu.Lock()
defer s.mu.Unlock()
if s.closed {
return nil
}
var sessions = make([]status.ServerSession, 0, len(s.byDisco))
for _, se := range s.byDisco {
c1 := extractClientInfo(0, se)
c2 := extractClientInfo(1, se)
sessions = append(sessions, status.ServerSession{
VNI: se.vni,
Client1: c1,
Client2: c2,
})
}
return sessions
}

View File

@@ -0,0 +1,75 @@
// Copyright (c) Tailscale Inc & AUTHORS
// SPDX-License-Identifier: BSD-3-Clause
// Package status contains types relating to the status of peer relay sessions
// between peer relay client nodes via a peer relay server.
package status
import (
"net/netip"
)
// ServerStatus contains the listening UDP port and active sessions (if any) for
// this node's peer relay server at a point in time.
type ServerStatus struct {
// UDPPort is the UDP port number that the peer relay server forwards over,
// as configured by the user with 'tailscale set --relay-server-port=<PORT>'.
// If the port has not been configured, UDPPort will be nil.
UDPPort *int
// Sessions is a slice of detailed status information about each peer
// relay session that this node's peer relay server is involved with. It
// may be empty.
Sessions []ServerSession
}
// ClientInfo contains status-related information about a single peer relay
// client involved in a single peer relay session.
type ClientInfo struct {
// Endpoint is the [netip.AddrPort] of this peer relay client's underlay
// endpoint participating in the session, or a zero value if the client
// has not completed a handshake.
Endpoint netip.AddrPort
// ShortDisco is a string representation of this peer relay client's disco
// public key.
//
// TODO: disco keys are pretty meaningless to end users, and they are also
// ephemeral. We really need node keys (or translation to first ts addr),
// but those are not fully plumbed into the [udprelay.Server]. Disco keys
// can also be ambiguous to a node key, but we could add node key into a
// [disco.AllocateUDPRelayEndpointRequest] in similar fashion to
// [disco.Ping]. There's also the problem of netmap trimming, where we
// can't verify a node key maps to a disco key.
ShortDisco string
// PacketsTx is the number of packets this peer relay client has sent to
// the other client via the relay server after completing a handshake. This
// is identical to the number of packets that the peer relay server has
// received from this client.
PacketsTx uint64
// BytesTx is the total overlay bytes this peer relay client has sent to
// the other client via the relay server after completing a handshake. This
// is identical to the total overlay bytes that the peer relay server has
// received from this client.
BytesTx uint64
}
// ServerSession contains status information for a single session between two
// peer relay clients, which are relayed via one peer relay server. This is the
// status as seen by the peer relay server; each client node may have a
// different view of the session's current status based on connectivity and
// where the client is in the peer relay endpoint setup (allocation, binding,
// pinging, active).
type ServerSession struct {
// VNI is the Virtual Network Identifier for this peer relay session, which
// comes from the Geneve header and is unique to this session.
VNI uint32
// Client1 contains status information about one of the two peer relay
// clients involved in this session. Note that 'Client1' does NOT mean this
// was/wasn't the allocating client, or the first client to bind, etc; this
// is just one client of two.
Client1 ClientInfo
// Client2 contains status information about one of the two peer relay
// clients involved in this session. Note that 'Client2' does NOT mean this
// was/wasn't the allocating client, or the second client to bind, etc; this
// is just one client of two.
Client2 ClientInfo
}