Files
lneto/x/xnet/tcppool.go
T
Derek den Haas 2dda96b610 fix(xnet): install a retransmission timer on TCP connections
No connection created through x/xnet had one. Neither NewTCPPool nor the
dial path in StackGo set ConnConfig.LossRecovery and ConnConfig.Nanotime,
so tcp.Handler ran with loss recovery disabled: nothing noticed a lost
segment, and the connection simply stopped: the sender waiting for an
ACK that cannot arrive, the receiver for data nobody will resend.

TCPPoolConfig.NanoTime already documents itself as "passed to each
tcp.Conn for retransmission timing (RFC 6298)", including the
time.Now fallback when it is nil; this makes that true. Each connection
gets its own tcp.RTO, which shadows that connection's send sequence space
and so cannot be shared. That is 80 bytes per connection, allocated once
at pool construction, against buffers measured in kilobytes.

The test drops exactly one data segment and requires the byte to arrive
anyway. It depends on the FIN-WAIT-1 retransmission fix, since the server
closes after writing.
2026-09-07 14:12:28 -07:00

203 lines
5.7 KiB
Go

package xnet
import (
"context"
"log/slog"
"sync"
"time"
"github.com/soypat/lneto"
"github.com/soypat/lneto/tcp"
)
// TCPPool implements tcp.pool.
type TCPPool struct {
mu sync.Mutex
naqcuired int
conns []tcp.Conn
userData []any
acquiredAt []int64
closingAt []int64
abortedAt []int64
nextISS tcp.Value
_now func() int64
estbTimeout time.Duration
closingTimeout time.Duration
logger *slog.Logger
}
func _() {
var l tcp.Listener
l.Reset(0, &TCPPool{}) // compile time guarantee of interface implementation.
}
type TCPPoolConfig struct {
// PoolSize determines the maximum number of active incoming TCP connections to the pool.
PoolSize uint16
QueueSize int
TxBufSize int
RxBufSize int
Logger *slog.Logger
ConnLogger *slog.Logger
// NanoTime returns the current monotonic time in nanoseconds.
// Used for pool timeout tracking and passed to each [tcp.Conn] for
// retransmission timing (RFC 6298). If nil, defaults to time.Now().UnixNano().
NanoTime func() int64
// EstablishedTimeout sets the timeout for a TCP connection since it is acquired until it is established.
// If the connection does not establish in this time it will be closed by the pool.
EstablishedTimeout time.Duration
// ClosingTimeout sets the timeout for a TCP connection to close and be returned to Pool.
// If the connection is not closed in this time it will be aborted by the pool.
ClosingTimeout time.Duration
// NewUserData is used to create user data used for each individual TCP connection and returned on GetTCP.
NewUserData func() any
// NewBackoff returns the backoff to use for every newly configured TCP connection. Must be non-nil.
// This should always return a static(non-method) function unless you know what you are doing.
NewBackoff func() lneto.BackoffStrategy
}
func NewTCPPool(cfg TCPPoolConfig) (*TCPPool, error) {
if cfg.EstablishedTimeout <= 0 || cfg.ClosingTimeout <= 0 {
return nil, lneto.ErrInvalidConfig
} else if cfg.NewBackoff == nil {
return nil, lneto.ErrMissingHALConfig
}
n := int(cfg.PoolSize)
pool := &TCPPool{
acquiredAt: make([]int64, n),
closingAt: make([]int64, n),
abortedAt: make([]int64, n),
conns: make([]tcp.Conn, n),
userData: make([]any, n),
_now: cfg.NanoTime,
estbTimeout: cfg.EstablishedTimeout,
closingTimeout: cfg.ClosingTimeout,
logger: cfg.Logger,
}
allocPerConn := cfg.TxBufSize + cfg.RxBufSize
bufSpace := make([]byte, n*allocPerConn)
for i := range pool.conns {
bufoff := i * allocPerConn
txOff := bufoff + cfg.RxBufSize
conncfg := tcp.ConnConfig{
RxBuf: bufSpace[bufoff:txOff],
TxBuf: bufSpace[txOff : txOff+cfg.TxBufSize],
TxPacketQueueSize: cfg.QueueSize,
Logger: cfg.ConnLogger,
RWBackoff: cfg.NewBackoff(),
// Retransmission timing, as NanoTime's contract promises. One RTO per
// connection: it shadows that connection's send sequence space and so
// cannot be shared.
LossRecovery: new(tcp.RTO),
Nanotime: pool.now,
}
err := pool.conns[i].Configure(conncfg)
if err != nil {
return nil, err
}
if cfg.NewUserData != nil {
pool.userData[i] = cfg.NewUserData()
}
}
return pool, nil
}
func (p *TCPPool) NumberOfAcquired() int {
p.mu.Lock()
defer p.mu.Unlock()
return p.naqcuired
}
func (p *TCPPool) GetTCP() (conn *tcp.Conn, userData any, SuggestedISS tcp.Value) {
p.mu.Lock()
defer p.mu.Unlock()
p.debug("TCPPool:get")
for i := range p.conns {
if p.acquiredAt[i] == 0 {
p.acquiredAt[i] = p.now()
p.nextISS += 1000
p.naqcuired++
return &p.conns[i], p.userData[i], p.nextISS
}
}
return nil, nil, 0
}
func (p *TCPPool) PutTCP(conn *tcp.Conn) {
p.mu.Lock()
defer p.mu.Unlock()
p.debug("TCPPool:put", slog.Uint64("lport", uint64(conn.LocalPort())))
for i := range p.conns {
if &p.conns[i] == conn {
// p.mu.Lock()
p.conns[i].Abort()
p.acquiredAt[i] = 0
p.abortedAt[i] = 0
p.closingAt[i] = 0
p.naqcuired--
// p.mu.Unlock()
return
}
}
panic("conn does not belong to this pool")
}
func (p *TCPPool) CheckTimeouts() {
p.mu.Lock()
defer p.mu.Unlock()
p.debug("TCPPool:checktimeouts", slog.Int("acq", p.naqcuired))
for i := range p.conns {
conn := &p.conns[i]
st := conn.State()
if st == tcp.StateEstablished {
continue
}
// p.mu.Lock()
acq := p.acquiredAt[i]
// p.mu.Unlock()
if acq == 0 {
continue
} else if st.IsPreestablished() && p.since(acq) > p.estbTimeout {
// Was acquired and did not reach establishment state so we close.
// This is part of a syn-flood defense mechanism.
conn.Close()
} else if st.IsClosed() || st.IsClosing() {
// p.mu.Lock()
if p.closingAt[i] == 0 {
p.closingAt[i] = p.now()
} else if p.abortedAt[i] == 0 && p.since(p.closingAt[i]) > p.closingTimeout {
p.abortedAt[i] = p.now()
conn.Abort()
} else if p.abortedAt[i] != 0 && p.since(p.abortedAt[i]) > 10*time.Second {
println("connection aborted and still not returned to TCPPool")
println("source", conn.LocalPort(), "remote", conn.RemotePort(), "state", conn.State().String())
}
}
}
}
func (p *TCPPool) since(t int64) time.Duration {
return time.Duration(p.now() - t)
}
func (p *TCPPool) now() int64 {
if p._now == nil {
return time.Now().UnixNano()
}
return p._now()
}
func (p *TCPPool) trace(msg string, attrs ...slog.Attr) {
p.log(slog.LevelDebug-2, msg, attrs...)
}
func (p *TCPPool) debug(msg string, attrs ...slog.Attr) {
p.log(slog.LevelDebug, msg, attrs...)
}
func (p *TCPPool) log(lvl slog.Level, msg string, attrs ...slog.Attr) {
if p.logger != nil {
p.logger.LogAttrs(context.Background(), lvl, msg, attrs...)
}
}