Merge remote-tracking branch 'origin/stable' into v2

Add TCP_NOTSENT_LOWAT setting

.github/workflows/ci.yaml

@@ -122,7 +122,7 @@ jobs:
   artifacts:
     name: Build release artifacts
     runs-on: ubuntu-latest
-    timeout-minutes: 10
+    timeout-minutes: 20
     steps:
       - name: Checkout
         uses: actions/checkout@v6

Dockerfile

@@ -33,7 +33,7 @@ RUN go mod download
 COPY . /app
 
 RUN set -x \
-  && version="$(git describe --exact-match HEAD 2>/dev/null || git describe --tags --always)" \
+  && version="$(git describe --exact-match HEAD 2>/dev/null || git describe --tags --always 2>/dev/null || echo dev)" \
   && go build \
       -trimpath \
       -mod=readonly \

README.md

@@ -29,6 +29,8 @@ are the most notable:
 * [Official](https://github.com/TelegramMessenger/MTProxy)
 * [Python](https://github.com/alexbers/mtprotoproxy)
 * [Erlang](https://github.com/seriyps/mtproto_proxy)
+* [Teleproxy (C)](https://github.com/teleproxy/teleproxy)
+* [mtproto.zig (Zig)](https://github.com/sleep3r/mtproto.zig)
 * [Telemt (Rust)](https://github.com/telemt/telemt)
 
 You can use any of these. They work great and all implementations have

antireplay/stable_bloom_filter_test.go

@@ -12,7 +12,7 @@ type StableBloomFilterTestSuite struct {
 }
 
 func (suite *StableBloomFilterTestSuite) TestOp() {
-	filter := antireplay.NewStableBloomFilter(500, 0.001)
+	filter := antireplay.NewStableBloomFilter(100000, 0.001)
 
 	suite.False(filter.SeenBefore([]byte{1, 2, 3}))
 	suite.False(filter.SeenBefore([]byte{4, 5, 6}))

example.config.toml

@@ -204,14 +204,22 @@ proxies = [
 # define a global timeout on establishing of network connections. idle
 # means a timeout on pumping data between sockset when nothing is
 # happening.
-#
-# please be noticed that handshakes have no timeouts intentionally. You can
-# find a reasoning here:
-# https://www.ndss-symposium.org/wp-content/uploads/2020/02/23087-paper.pdf
 [network.timeout]
 tcp = "5s"
 http = "10s"
-idle = "1m"
+idle = "5m"
+handshake = "10s"
+
+# this defines a configuration for TCP keep alives. Default values are taken
+# from Golang default behavior.
+[network.keep-alive]
+disabled = false
+# idle means a time period after which we start sending TCP Keep Alive probes
+idle = "15s"
+# interval is a period between 2 consecutive probes
+interval = "15s"
+# if we miss that many probes, a connection will be considered as a dead one.
+count = 9
 
 # mtg has to mimic real websites. It does not mean domain fronting, it also
 # means that traffic characteristics should be similar to real world traffic.

go.mod

@@ -4,7 +4,7 @@ go 1.26
 
 require (
 	github.com/OneOfOne/xxhash v1.2.8
-	github.com/alecthomas/kong v1.14.0
+	github.com/alecthomas/kong v1.15.0
 	github.com/alecthomas/units v0.0.0-20240927000941-0f3dac36c52b
 	github.com/armon/go-socks5 v0.0.0-20160902184237-e75332964ef5
 	github.com/babolivier/go-doh-client v0.0.0-20201028162107-a76cff4cb8b6
@@ -28,7 +28,7 @@ require (
 
 require (
 	github.com/beevik/ntp v1.5.0
-	github.com/ncruces/go-dns v1.3.2
+	github.com/ncruces/go-dns v1.3.3
 	github.com/pelletier/go-toml/v2 v2.3.0
 	github.com/pires/go-proxyproto v0.11.0
 	github.com/things-go/go-socks5 v0.1.0

go.sum

@@ -2,8 +2,8 @@ github.com/OneOfOne/xxhash v1.2.8 h1:31czK/TI9sNkxIKfaUfGlU47BAxQ0ztGgd9vPyqimf8
 github.com/OneOfOne/xxhash v1.2.8/go.mod h1:eZbhyaAYD41SGSSsnmcpxVoRiQ/MPUTjUdIIOT9Um7Q=
 github.com/alecthomas/assert/v2 v2.11.0 h1:2Q9r3ki8+JYXvGsDyBXwH3LcJ+WK5D0gc5E8vS6K3D0=
 github.com/alecthomas/assert/v2 v2.11.0/go.mod h1:Bze95FyfUr7x34QZrjL+XP+0qgp/zg8yS+TtBj1WA3k=
-github.com/alecthomas/kong v1.14.0 h1:gFgEUZWu2ZmZ+UhyZ1bDhuutbKN1nTtJTwh19Wsn21s=
+github.com/alecthomas/kong v1.15.0 h1:BVJstKbpO73zKpmIu+m/aLRrNmWwxXPIGTNin9VmLVI=
-github.com/alecthomas/kong v1.14.0/go.mod h1:wrlbXem1CWqUV5Vbmss5ISYhsVPkBb1Yo7YKJghju2I=
+github.com/alecthomas/kong v1.15.0/go.mod h1:wrlbXem1CWqUV5Vbmss5ISYhsVPkBb1Yo7YKJghju2I=
 github.com/alecthomas/repr v0.5.2 h1:SU73FTI9D1P5UNtvseffFSGmdNci/O6RsqzeXJtP0Qs=
 github.com/alecthomas/repr v0.5.2/go.mod h1:Fr0507jx4eOXV7AlPV6AVZLYrLIuIeSOWtW57eE/O/4=
 github.com/alecthomas/units v0.0.0-20240927000941-0f3dac36c52b h1:mimo19zliBX/vSQ6PWWSL9lK8qwHozUj03+zLoEB8O0=
@@ -48,8 +48,8 @@ github.com/miekg/dns v1.1.51 h1:0+Xg7vObnhrz/4ZCZcZh7zPXlmU0aveS2HDBd0m0qSo=
 github.com/miekg/dns v1.1.51/go.mod h1:2Z9d3CP1LQWihRZUf29mQ19yDThaI4DAYzte2CaQW5c=
 github.com/munnerz/goautoneg v0.0.0-20191010083416-a7dc8b61c822 h1:C3w9PqII01/Oq1c1nUAm88MOHcQC9l5mIlSMApZMrHA=
 github.com/munnerz/goautoneg v0.0.0-20191010083416-a7dc8b61c822/go.mod h1:+n7T8mK8HuQTcFwEeznm/DIxMOiR9yIdICNftLE1DvQ=
-github.com/ncruces/go-dns v1.3.2 h1:kBLuUZBgkQ4qF4WDXZRQ4rG0Gk6sLVJQ5tESkWrxUa0=
+github.com/ncruces/go-dns v1.3.3 h1:59OV7XoJrTCoUMZjWRVs4GOjtntMTZqiQ5Mn+BT13hk=
-github.com/ncruces/go-dns v1.3.2/go.mod h1:tuzixNY8PY/M7yUzcvRbUaeLs3ifIdydpi5H2bfRU+s=
+github.com/ncruces/go-dns v1.3.3/go.mod h1:tuzixNY8PY/M7yUzcvRbUaeLs3ifIdydpi5H2bfRU+s=
 github.com/panjf2000/ants/v2 v2.12.0 h1:u9JhESo83i/GkZnhfTNuFMMWcNt7mnV1bGJ6FT4wXH8=
 github.com/panjf2000/ants/v2 v2.12.0/go.mod h1:tSQuaNQ6r6NRhPt+IZVUevvDyFMTs+eS4ztZc52uJTY=
 github.com/patrickmn/go-cache v2.1.0+incompatible h1:HRMgzkcYKYpi3C8ajMPV8OFXaaRUnok+kx1WdO15EQc=

internal/cli/doctor.go

@@ -97,6 +97,12 @@ func (d *Doctor) Run(cli *CLI, version string) error {
 		conf.Network.Timeout.TCP.Get(10*time.Second),
 		conf.Network.Timeout.HTTP.Get(0),
 		conf.Network.Timeout.Idle.Get(0),
+		net.KeepAliveConfig{
+			Enable:   !conf.Network.KeepAlive.Disabled.Get(false),
+			Idle:     conf.Network.KeepAlive.Idle.Get(0),
+			Interval: conf.Network.KeepAlive.Interval.Get(0),
+			Count:    int(conf.Network.KeepAlive.Count.Get(0)),
+		},
 	)
 
 	fmt.Println("Validate native network connectivity")

internal/cli/run_proxy.go

@@ -5,7 +5,6 @@ import (
 	"fmt"
 	"net"
 	"os"
-	"time"
 
 	"github.com/9seconds/mtg/v2/antireplay"
 	"github.com/9seconds/mtg/v2/events"
@@ -51,6 +50,12 @@ func makeNetwork(conf *config.Config, version string) (mtglib.Network, error) {
 		conf.Network.Timeout.TCP.Get(0),
 		conf.Network.Timeout.HTTP.Get(0),
 		conf.Network.Timeout.Idle.Get(0),
+		net.KeepAliveConfig{
+			Enable:   !conf.Network.KeepAlive.Disabled.Get(false),
+			Idle:     conf.Network.KeepAlive.Idle.Get(0),
+			Interval: conf.Network.KeepAlive.Interval.Get(0),
+			Count:    int(conf.Network.KeepAlive.Count.Get(0)),
+		},
 	)
 
 	proxyDialers := make([]mtglib.Network, len(conf.Network.Proxies))
@@ -263,7 +268,8 @@ func runProxy(conf *config.Config, version string) error { //nolint: funlen
 
 		AllowFallbackOnUnknownDC: conf.AllowFallbackOnUnknownDC.Get(false),
 		TolerateTimeSkewness:     conf.TolerateTimeSkewness.Value,
-		IdleTimeout:              conf.Network.Timeout.Idle.Get(time.Minute),
+		IdleTimeout:              conf.Network.Timeout.Idle.Get(mtglib.DefaultIdleTimeout),
+		HandshakeTimeout:         conf.Network.Timeout.Handshake.Get(mtglib.DefaultHandshakeTimeout),
 
 		DoppelGangerURLs:    doppelGangerURLs,
 		DoppelGangerPerRaid: conf.Defense.Doppelganger.Repeats.Get(mtglib.DoppelGangerPerRaid),

internal/config/config.go

@@ -52,18 +52,25 @@ type Config struct {
 		Blocklist    ListConfig `json:"blocklist"`
 		Allowlist    ListConfig `json:"allowlist"`
 		Doppelganger struct {
-			URLs            []TypeHttpsURL  `json:"urls"`
+			URLs       []TypeHttpsURL  `json:"urls"`
-			Repeats         TypeConcurrency `json:"repeats_per_raid"`
+			Repeats    TypeConcurrency `json:"repeats_per_raid"`
-			UpdateEach      TypeDuration    `json:"raid_each"`
+			UpdateEach TypeDuration    `json:"raid_each"`
-			DRS             TypeBool        `json:"drs"`
+			DRS        TypeBool        `json:"drs"`
 		} `json:"doppelganger"`
 	} `json:"defense"`
 	Network struct {
 		Timeout struct {
-			TCP  TypeDuration `json:"tcp"`
+			TCP       TypeDuration `json:"tcp"`
-			HTTP TypeDuration `json:"http"`
+			HTTP      TypeDuration `json:"http"`
-			Idle TypeDuration `json:"idle"`
+			Idle      TypeDuration `json:"idle"`
+			Handshake TypeDuration `json:"handshake"`
 		} `json:"timeout"`
+		KeepAlive struct {
+			Disabled TypeBool        `json:"disabled"`
+			Idle     TypeDuration    `json:"idle"`
+			Interval TypeDuration    `json:"interval"`
+			Count    TypeConcurrency `json:"count"`
+		} `json:"keepAlive"`
 		DOHIP   TypeIP         `json:"dohIp"`
 		DNS     TypeDNSURI     `json:"dns"`
 		Proxies []TypeProxyURL `json:"proxies"`

internal/config/parse.go

@@ -47,18 +47,25 @@ type tomlConfig struct {
 			UpdateEach          string   `toml:"update-each" json:"updateEach,omitempty"`
 		} `toml:"allowlist" json:"allowlist,omitempty"`
 		Doppelganger struct {
-			URLs            []string `toml:"urls" json:"urls,omitempty"`
+			URLs       []string `toml:"urls" json:"urls,omitempty"`
-			Repeats         uint     `toml:"repeats-per-raid" json:"repeats_per_raid,omitempty"`
+			Repeats    uint     `toml:"repeats-per-raid" json:"repeats_per_raid,omitempty"`
-			UpdateEach      string   `toml:"raid-each" json:"raid_each,omitempty"`
+			UpdateEach string   `toml:"raid-each" json:"raid_each,omitempty"`
-			DRS             bool     `toml:"drs" json:"drs,omitempty"`
+			DRS        bool     `toml:"drs" json:"drs,omitempty"`
 		} `toml:"doppelganger" json:"doppelganger,omitempty"`
 	} `toml:"defense" json:"defense,omitempty"`
 	Network struct {
 		Timeout struct {
-			TCP  string `toml:"tcp" json:"tcp,omitempty"`
+			TCP       string `toml:"tcp" json:"tcp,omitempty"`
-			HTTP string `toml:"http" json:"http,omitempty"`
+			HTTP      string `toml:"http" json:"http,omitempty"`
-			Idle string `toml:"idle" json:"idle,omitempty"`
+			Idle      string `toml:"idle" json:"idle,omitempty"`
+			Handshake string `toml:"handshake" json:"handshake,omitempty"`
 		} `toml:"timeout" json:"timeout,omitempty"`
+		KeepAlive struct {
+			Disabled bool   `toml:"disabled" json:"disabled,omitempty"`
+			Idle     string `toml:"idle" json:"idle,omitempty"`
+			Interval string `toml:"interval" json:"interval,omitempty"`
+			Count    uint   `toml:"count" json:"count,omitempty"`
+		} `toml:"keep-alive" json:"keepAlive,omitempty"`
 		DOHIP   string   `toml:"doh-ip" json:"dohIp,omitempty"`
 		DNS     string   `toml:"dns" json:"dns,omitempty"`
 		Proxies []string `toml:"proxies" json:"proxies,omitempty"`

mise.lock

@@ -82,40 +82,40 @@ checksum = "sha256:4932cfca5e75bf60fe1c576edf459e5e809e6644664a068185d64b84af3fa
 url = "https://github.com/golangci/golangci-lint/releases/download/v2.11.4/golangci-lint-2.11.4-windows-amd64.zip"
 
 [[tools.goreleaser]]
-version = "2.14.3"
+version = "2.15.2"
 backend = "aqua:goreleaser/goreleaser"
 
 [tools.goreleaser."platforms.linux-arm64"]
-checksum = "sha256:581a10e53c1176b3e81ee45cf531e02dbf899db0bc7b795669347df4276ce948"
+checksum = "sha256:5db66761a98f6693161e49e1a95d28d2673a892ba60cb4a5e16736cafd41c4c9"
-url = "https://github.com/goreleaser/goreleaser/releases/download/v2.14.3/goreleaser_Linux_arm64.tar.gz"
+url = "https://github.com/goreleaser/goreleaser/releases/download/v2.15.2/goreleaser_Linux_arm64.tar.gz"
 provenance = "cosign"
 
 [tools.goreleaser."platforms.linux-arm64-musl"]
-checksum = "sha256:581a10e53c1176b3e81ee45cf531e02dbf899db0bc7b795669347df4276ce948"
+checksum = "sha256:5db66761a98f6693161e49e1a95d28d2673a892ba60cb4a5e16736cafd41c4c9"
-url = "https://github.com/goreleaser/goreleaser/releases/download/v2.14.3/goreleaser_Linux_arm64.tar.gz"
+url = "https://github.com/goreleaser/goreleaser/releases/download/v2.15.2/goreleaser_Linux_arm64.tar.gz"
 provenance = "cosign"
 
 [tools.goreleaser."platforms.linux-x64"]
-checksum = "sha256:dc7faeeeb6da8bdfda788626263a4ae725892a8c7504b975c3234127d4a44579"
+checksum = "sha256:0ebdbf0353aba566b969dde746cc4e4806f96c27aa2f3971b229a9df7611fedc"
-url = "https://github.com/goreleaser/goreleaser/releases/download/v2.14.3/goreleaser_Linux_x86_64.tar.gz"
+url = "https://github.com/goreleaser/goreleaser/releases/download/v2.15.2/goreleaser_Linux_x86_64.tar.gz"
 provenance = "cosign"
 
 [tools.goreleaser."platforms.linux-x64-musl"]
-checksum = "sha256:dc7faeeeb6da8bdfda788626263a4ae725892a8c7504b975c3234127d4a44579"
+checksum = "sha256:0ebdbf0353aba566b969dde746cc4e4806f96c27aa2f3971b229a9df7611fedc"
-url = "https://github.com/goreleaser/goreleaser/releases/download/v2.14.3/goreleaser_Linux_x86_64.tar.gz"
+url = "https://github.com/goreleaser/goreleaser/releases/download/v2.15.2/goreleaser_Linux_x86_64.tar.gz"
 provenance = "cosign"
 
 [tools.goreleaser."platforms.macos-arm64"]
-checksum = "sha256:3507798489e107a78aff36b169de48148a335ac26eb3161608d905f3f3a957bd"
+checksum = "sha256:0e6bd67688ac949780bf1166813a91f89856898ef4c40d7d46c2c74ebaa4b9ee"
-url = "https://github.com/goreleaser/goreleaser/releases/download/v2.14.3/goreleaser_Darwin_all.tar.gz"
+url = "https://github.com/goreleaser/goreleaser/releases/download/v2.15.2/goreleaser_Darwin_all.tar.gz"
-provenance = "cosign"
+provenance = "github-attestations"
 
 [tools.goreleaser."platforms.macos-x64"]
-checksum = "sha256:3507798489e107a78aff36b169de48148a335ac26eb3161608d905f3f3a957bd"
+checksum = "sha256:0e6bd67688ac949780bf1166813a91f89856898ef4c40d7d46c2c74ebaa4b9ee"
-url = "https://github.com/goreleaser/goreleaser/releases/download/v2.14.3/goreleaser_Darwin_all.tar.gz"
+url = "https://github.com/goreleaser/goreleaser/releases/download/v2.15.2/goreleaser_Darwin_all.tar.gz"
 provenance = "cosign"
 
 [tools.goreleaser."platforms.windows-x64"]
-checksum = "sha256:3deea8ff471aa258a2d99f3e5302971d7028647ae8ddaf103257a8113e485a31"
+checksum = "sha256:7459832946dbe122c144f8d7f87484d8572ca005b779310aa6bb03346e8de17a"
-url = "https://github.com/goreleaser/goreleaser/releases/download/v2.14.3/goreleaser_Windows_x86_64.zip"
+url = "https://github.com/goreleaser/goreleaser/releases/download/v2.15.2/goreleaser_Windows_x86_64.zip"
 provenance = "cosign"

mtglib/conns.go

@@ -3,9 +3,11 @@ package mtglib
 import (
 	"bytes"
 	"context"
+	"errors"
 	"fmt"
 	"io"
 	"net"
+	"sync/atomic"
 	"time"
 
 	"github.com/9seconds/mtg/v2/essentials"
@@ -97,20 +99,63 @@ func newConnProxyProtocol(source, target essentials.Conn) *connProxyProtocol {
 	}
 }
 
+// idleTracker is a shared idle tracker for a pair of relay connections.
+// Both directions update the same timestamp so that activity in one direction
+// prevents the other (idle) direction from timing out.
+type idleTracker struct {
+	lastActive atomic.Pointer[time.Time]
+	timeout    time.Duration
+}
+
+func newIdleTracker(timeout time.Duration) *idleTracker {
+	t := &idleTracker{timeout: timeout}
+	t.touch()
+
+	return t
+}
+
+func (t *idleTracker) touch() {
+	stamp := time.Now()
+	t.lastActive.Store(&stamp)
+}
+
+func (t *idleTracker) isIdle() bool {
+	return time.Since(*t.lastActive.Load()) >= t.timeout
+}
+
 type connIdleTimeout struct {
 	essentials.Conn
 
-	timeout time.Duration
+	tracker *idleTracker
 }
 
 func (c connIdleTimeout) Read(b []byte) (int, error) {
-	c.SetReadDeadline(time.Now().Add(c.timeout)) //nolint: errcheck
+	var netErr net.Error
 
-	return c.Conn.Read(b) //nolint: wrapcheck
+	for {
+		c.SetReadDeadline(time.Now().Add(c.tracker.timeout)) //nolint: errcheck
+
+		n, err := c.Conn.Read(b)
+
+		switch {
+		case err == nil:
+			c.tracker.touch()
+			return n, nil
+		case errors.As(err, &netErr) && netErr.Timeout() && !c.tracker.isIdle():
+			continue
+		}
+
+		return n, err
+	}
 }
 
 func (c connIdleTimeout) Write(b []byte) (int, error) {
-	c.SetWriteDeadline(time.Now().Add(c.timeout)) //nolint: errcheck
+	c.SetWriteDeadline(time.Now().Add(c.tracker.timeout)) //nolint: errcheck
 
-	return c.Conn.Write(b) //nolint: wrapcheck
+	n, err := c.Conn.Write(b)
+	if n > 0 {
+		c.tracker.touch()
+	}
+
+	return n, err //nolint: wrapcheck
 }

mtglib/conns_internal_test.go

@@ -16,6 +16,12 @@ import (
 	"github.com/stretchr/testify/suite"
 )
 
+type netTimeoutError struct{}
+
+func (e netTimeoutError) Error() string   { return "i/o timeout" }
+func (e netTimeoutError) Timeout() bool   { return true }
+func (e netTimeoutError) Temporary() bool { return true }
+
 type ConnRewindBaseConn struct {
 	testlib.EssentialsConnMock
 
@@ -291,6 +297,141 @@ func (suite *ConnProxyProtocolTestSuite) TearDownTest() {
 	suite.targetConnMock.AssertExpectations(suite.T())
 }
 
+type IdleTrackerTestSuite struct {
+	suite.Suite
+}
+
+func (suite *IdleTrackerTestSuite) TestNewNotIdle() {
+	tracker := newIdleTracker(time.Second)
+	suite.False(tracker.isIdle())
+}
+
+func (suite *IdleTrackerTestSuite) TestIdleAfterTimeout() {
+	tracker := newIdleTracker(10 * time.Millisecond)
+	time.Sleep(20 * time.Millisecond)
+
+	suite.True(tracker.isIdle())
+}
+
+func (suite *IdleTrackerTestSuite) TestTouchResetsIdle() {
+	tracker := newIdleTracker(50 * time.Millisecond)
+	time.Sleep(30 * time.Millisecond)
+
+	tracker.touch()
+
+	suite.False(tracker.isIdle())
+}
+
+type ConnIdleTimeoutTestSuite struct {
+	suite.Suite
+
+	connMock *testlib.EssentialsConnMock
+	tracker  *idleTracker
+	conn     connIdleTimeout
+}
+
+func (suite *ConnIdleTimeoutTestSuite) SetupTest() {
+	suite.connMock = &testlib.EssentialsConnMock{}
+	suite.tracker = newIdleTracker(time.Second)
+	suite.conn = connIdleTimeout{
+		Conn:    suite.connMock,
+		tracker: suite.tracker,
+	}
+}
+
+func (suite *ConnIdleTimeoutTestSuite) TearDownTest() {
+	suite.connMock.AssertExpectations(suite.T())
+}
+
+func (suite *ConnIdleTimeoutTestSuite) TestReadOk() {
+	suite.connMock.On("SetReadDeadline", mock.Anything).Return(nil)
+	suite.connMock.On("Read", mock.Anything).Once().Return(5, nil)
+
+	n, err := suite.conn.Read(make([]byte, 10))
+	suite.NoError(err)
+	suite.Equal(5, n)
+}
+
+func (suite *ConnIdleTimeoutTestSuite) TestReadNonTimeoutErr() {
+	suite.connMock.On("SetReadDeadline", mock.Anything).Return(nil)
+	suite.connMock.On("Read", mock.Anything).Once().Return(0, io.EOF)
+
+	n, err := suite.conn.Read(make([]byte, 10))
+	suite.True(errors.Is(err, io.EOF))
+	suite.Equal(0, n)
+}
+
+func (suite *ConnIdleTimeoutTestSuite) TestReadTimeoutRetriesWhenNotIdle() {
+	suite.connMock.On("SetReadDeadline", mock.Anything).Return(nil)
+	suite.connMock.On("Read", mock.Anything).Once().Return(0, netTimeoutError{})
+	suite.connMock.On("Read", mock.Anything).Once().Return(5, nil)
+
+	n, err := suite.conn.Read(make([]byte, 10))
+	suite.NoError(err)
+	suite.Equal(5, n)
+}
+
+func (suite *ConnIdleTimeoutTestSuite) TestReadTimeoutClosesWhenIdle() {
+	suite.tracker = newIdleTracker(time.Millisecond)
+	suite.conn = connIdleTimeout{
+		Conn:    suite.connMock,
+		tracker: suite.tracker,
+	}
+
+	time.Sleep(5 * time.Millisecond)
+
+	suite.connMock.On("SetReadDeadline", mock.Anything).Return(nil)
+	suite.connMock.On("Read", mock.Anything).Once().Return(0, netTimeoutError{})
+
+	n, err := suite.conn.Read(make([]byte, 10))
+	suite.Equal(0, n)
+
+	netErr, ok := err.(net.Error) //nolint: errorlint
+	suite.True(ok)
+	suite.True(netErr.Timeout())
+}
+
+func (suite *ConnIdleTimeoutTestSuite) TestSharedTrackerPreventsFalseTimeout() {
+	connMock2 := &testlib.EssentialsConnMock{}
+	conn2 := connIdleTimeout{
+		Conn:    connMock2,
+		tracker: suite.tracker,
+	}
+
+	connMock2.On("SetWriteDeadline", mock.Anything).Return(nil)
+	connMock2.On("Write", mock.Anything).Once().Return(5, nil)
+
+	_, _ = conn2.Write(make([]byte, 5))
+
+	suite.connMock.On("SetReadDeadline", mock.Anything).Return(nil)
+	suite.connMock.On("Read", mock.Anything).Once().Return(0, netTimeoutError{})
+	suite.connMock.On("Read", mock.Anything).Once().Return(3, nil)
+
+	n, err := suite.conn.Read(make([]byte, 10))
+	suite.NoError(err)
+	suite.Equal(3, n)
+
+	connMock2.AssertExpectations(suite.T())
+}
+
+func (suite *ConnIdleTimeoutTestSuite) TestWriteOk() {
+	suite.connMock.On("SetWriteDeadline", mock.Anything).Return(nil)
+	suite.connMock.On("Write", mock.Anything).Once().Return(5, nil)
+
+	n, err := suite.conn.Write(make([]byte, 5))
+	suite.NoError(err)
+	suite.Equal(5, n)
+}
+
+func (suite *ConnIdleTimeoutTestSuite) TestWriteErr() {
+	suite.connMock.On("SetWriteDeadline", mock.Anything).Return(nil)
+	suite.connMock.On("Write", mock.Anything).Once().Return(0, io.EOF)
+
+	n, err := suite.conn.Write(make([]byte, 5))
+	suite.True(errors.Is(err, io.EOF))
+	suite.Equal(0, n)
+}
+
 func TestConnTraffic(t *testing.T) {
 	t.Parallel()
 	suite.Run(t, &ConnTrafficTestSuite{})
@@ -305,3 +446,13 @@ func TestConnProxyProtocol(t *testing.T) {
 	t.Parallel()
 	suite.Run(t, &ConnProxyProtocolTestSuite{})
 }
+
+func TestIdleTracker(t *testing.T) {
+	t.Parallel()
+	suite.Run(t, &IdleTrackerTestSuite{})
+}
+
+func TestConnIdleTimeout(t *testing.T) {
+	t.Parallel()
+	suite.Run(t, &ConnIdleTimeoutTestSuite{})
+}

mtglib/init.go

@@ -77,8 +77,13 @@ const (
 	// DefaultIdleTimeout is a default timeout for closing a connection in case of
 	// idling.
 	//
-	// Deprecated: no longer in use because of changed TCP relay algorithm.
+	// Set to 5 minutes to survive typical mobile sleep periods (2-5 min) and
-	DefaultIdleTimeout = time.Minute
+	// avoid racing with MTProto ping_delay_disconnect (~60s interval).
+	DefaultIdleTimeout = 5 * time.Minute
+
+	// DefaultHandshakeTimeout defines a time period during which the
+	// all handshake ceremonies must be completed.
+	DefaultHandshakeTimeout = 10 * time.Second
 
 	// DefaultTolerateTimeSkewness is a default timeout for time skewness on a
 	// faketls timeout verification.

mtglib/internal/dc/view.go

@@ -5,15 +5,19 @@ type dcView struct {
 }
 
 func (d dcView) getV4(dc int) []Addr {
-	addrs := d.publicConfigs.getV4(dc)
+	var addrs []Addr
+
 	addrs = append(addrs, defaultDCAddrSet.getV4(dc)...)
+	addrs = append(addrs, d.publicConfigs.getV4(dc)...)
 
 	return addrs
 }
 
 func (d dcView) getV6(dc int) []Addr {
-	addrs := d.publicConfigs.getV6(dc)
+	var addrs []Addr
+
 	addrs = append(addrs, defaultDCAddrSet.getV6(dc)...)
+	addrs = append(addrs, d.publicConfigs.getV6(dc)...)
 
 	return addrs
 }

mtglib/internal/doppel/scout.go

@@ -61,23 +61,29 @@ func (s Scout) learn(ctx context.Context, url string) (ScoutResult, error) {
 		client.CloseIdleConnections()
 	}
 
-	if err != nil || len(results.data) == 0 {
+	if err != nil {
 		return ScoutResult{}, err
 	}
 
+	data, writeIndex := results.Snapshot()
+
+	if len(data) == 0 {
+		return ScoutResult{}, nil
+	}
+
 	var result ScoutResult
 
 	// Compute inter-record durations (existing logic).
 	lastTimestamp := time.Time{}
 
-	for i, v := range results.data {
+	for i, v := range data {
 		if v.recordType != tls.TypeApplicationData {
 			continue
 		}
 
 		if lastTimestamp.IsZero() {
 			if i > 0 {
-				lastTimestamp = results.data[i-1].timestamp
+				lastTimestamp = data[i-1].timestamp
 			} else {
 				lastTimestamp = v.timestamp
 			}
@@ -90,12 +96,12 @@ func (s Scout) learn(ctx context.Context, url string) (ScoutResult, error) {
 	// Compute cert size: sum of ApplicationData payload between CCS and
 	// the first client Write (which marks the end of server handshake).
 	seenCCS := false
-	boundary := results.writeIndex
+	boundary := writeIndex
 	if boundary < 0 {
-		boundary = len(results.data)
+		boundary = len(data)
 	}
 
-	for i, v := range results.data {
+	for i, v := range data {
 		if i >= boundary {
 			break
 		}

mtglib/internal/doppel/scout_conn_collected.go

@@ -1,6 +1,10 @@
 package doppel
 
-import "time"
+import (
+	"slices"
+	"sync"
+	"time"
+)
 
 const (
 	ScoutConnCollectedPreallocSize = 100
@@ -13,23 +17,38 @@ type ScoutConnResult struct {
 }
 
 type ScoutConnCollected struct {
+	mu         sync.Mutex
 	data       []ScoutConnResult
 	writeIndex int // index at which client first wrote post-handshake data; -1 if not set
 }
 
 func (s *ScoutConnCollected) Add(record byte, payloadLen int) {
+	s.mu.Lock()
 	s.data = append(s.data, ScoutConnResult{
 		timestamp:  time.Now(),
 		recordType: record,
 		payloadLen: payloadLen,
 	})
+	s.mu.Unlock()
 }
 
 // MarkWrite records the current data length as the handshake boundary.
 func (s *ScoutConnCollected) MarkWrite() {
+	s.mu.Lock()
 	if s.writeIndex < 0 {
 		s.writeIndex = len(s.data)
 	}
+	s.mu.Unlock()
+}
+
+// Snapshot returns a copy of the collected data and the write index.
+func (s *ScoutConnCollected) Snapshot() ([]ScoutConnResult, int) {
+	s.mu.Lock()
+	snapshot := slices.Clone(s.data)
+	writeIndex := s.writeIndex
+	s.mu.Unlock()
+
+	return snapshot, writeIndex
 }
 
 func NewScoutConnCollected() *ScoutConnCollected {

mtglib/internal/doppel/scout_conn_collected_test.go

@@ -1,6 +1,7 @@
 package doppel
 
 import (
+	"sync"
 	"testing"
 	"time"
 
@@ -16,8 +17,10 @@ func (suite *ScoutConnCollectedTestSuite) TestAddSingle() {
 	collected := NewScoutConnCollected()
 	collected.Add(tls.TypeApplicationData, 100)
 
-	suite.Len(collected.data, 1)
+	data, _ := collected.Snapshot()
-	suite.Equal(byte(tls.TypeApplicationData), collected.data[0].recordType)
+
+	suite.Len(data, 1)
+	suite.Equal(byte(tls.TypeApplicationData), data[0].recordType)
 }
 
 func (suite *ScoutConnCollectedTestSuite) TestAddTimestampsAreMonotonic() {
@@ -31,11 +34,52 @@ func (suite *ScoutConnCollectedTestSuite) TestAddTimestampsAreMonotonic() {
 	time.Sleep(time.Microsecond)
 	collected.Add(tls.TypeApplicationData, 100)
 
-	for i := 1; i < len(collected.data); i++ {
+	data, _ := collected.Snapshot()
-		suite.True(collected.data[i].timestamp.After(collected.data[i-1].timestamp))
+
+	for i := 1; i < len(data); i++ {
+		suite.True(data[i].timestamp.After(data[i-1].timestamp))
 	}
 }
 
+func (suite *ScoutConnCollectedTestSuite) TestConcurrentAddSnapshot() {
+	collected := NewScoutConnCollected()
+
+	var wg sync.WaitGroup
+
+	wg.Add(3)
+
+	go func() {
+		defer wg.Done()
+
+		for i := 0; i < 1000; i++ {
+			collected.Add(tls.TypeApplicationData, i)
+		}
+	}()
+
+	go func() {
+		defer wg.Done()
+
+		for i := 0; i < 100; i++ {
+			collected.MarkWrite()
+		}
+	}()
+
+	go func() {
+		defer wg.Done()
+
+		for i := 0; i < 1000; i++ {
+			// call Snapshot concurrently to exercise the lock under -race
+			collected.Snapshot() //nolint:errcheck
+		}
+	}()
+
+	wg.Wait()
+
+	data, writeIndex := collected.Snapshot()
+	suite.Len(data, 1000)
+	suite.GreaterOrEqual(writeIndex, 0)
+}
+
 func TestScoutConnCollected(t *testing.T) {
 	t.Parallel()
 	suite.Run(t, &ScoutConnCollectedTestSuite{})

mtglib/internal/tls/fake/client_side.go

@@ -6,13 +6,12 @@ import (
 	"crypto/sha256"
 	"crypto/subtle"
 	"encoding/binary"
+	"errors"
 	"fmt"
 	"io"
 	"net"
 	"slices"
 	"time"
-
-	"github.com/9seconds/mtg/v2/mtglib/internal/tls"
 )
 
 const (
@@ -22,12 +21,19 @@ const (
 	// record_type(1) + version(2) + size(2) + handshake_type(1) + uint24_length(3) + client_version(2)
 	RandomOffset = 1 + 2 + 2 + 1 + 3 + 2
 
+	// https://datatracker.ietf.org/doc/html/rfc8701#name-grease-values
+	// https://medium.com/asecuritysite-when-bob-met-alice/in-cybersecurity-what-is-grease-9f8850558dea
+	GreaseMask      = 0x0f0f
+	GreaseValueType = 0x0a0a
+
 	sniDNSNamesListType = 0
 )
 
 var (
 	emptyRandom = [RandomLen]byte{}
 	extTypeSNI  = [2]byte{}
+
+	ErrCannotFindCipher = errors.New("cannot find a cipher")
 )
 
 type ClientHello struct {
@@ -42,11 +48,6 @@ func ReadClientHello(
 	hostname string,
 	tolerateTimeSkewness time.Duration,
 ) (*ClientHello, error) {
-	if err := conn.SetReadDeadline(time.Now().Add(ClientHelloReadTimeout)); err != nil {
-		return nil, fmt.Errorf("cannot set read deadline: %w", err)
-	}
-	defer conn.SetReadDeadline(resetDeadline) //nolint: errcheck
-
 	// This is how FakeTLS is organized:
 	//  1. We create sha256 HMAC with a given secret
 	//  2. We dump there a whole TLS frame except of the fact that random
@@ -56,25 +57,17 @@ func ReadClientHello(
 	//  4. New digest should be all 0 except of last 4 bytes
 	//  5. Last 4 bytes are little endian uint32 of UNIX timestamp when
 	//     this message was created.
-	handshakeCopyBuf := &bytes.Buffer{}
+	clientHelloCopy, handshakeReader, err := parseClientHello(conn)
-	reader := io.TeeReader(conn, handshakeCopyBuf)
-
-	reader, err := parseTLSHeader(reader)
 	if err != nil {
-		return nil, fmt.Errorf("cannot parse tls header: %w", err)
+		return nil, fmt.Errorf("cannot read client hello: %w", err)
 	}
 
-	reader, err = parseHandshakeHeader(reader)
+	hello, err := parseHandshake(handshakeReader)
-	if err != nil {
-		return nil, fmt.Errorf("cannot parse handshake header: %w", err)
-	}
-
-	hello, err := parseHandshake(reader)
 	if err != nil {
 		return nil, fmt.Errorf("cannot parse handshake: %w", err)
 	}
 
-	sniHostnames, err := parseSNI(reader)
+	sniHostnames, err := parseSNI(handshakeReader)
 	if err != nil {
 		return nil, fmt.Errorf("cannot parse SNI: %w", err)
 	}
@@ -85,10 +78,10 @@ func ReadClientHello(
 
 	digest := hmac.New(sha256.New, secret)
 	// we write a copy of the handshake with client random all nullified.
-	digest.Write(handshakeCopyBuf.Next(RandomOffset))
+	digest.Write(clientHelloCopy.Next(RandomOffset))
-	handshakeCopyBuf.Next(RandomLen)
+	clientHelloCopy.Next(RandomLen)
 	digest.Write(emptyRandom[:])
-	digest.Write(handshakeCopyBuf.Bytes())
+	digest.Write(clientHelloCopy.Bytes())
 
 	computed := digest.Sum(nil)
 
@@ -110,58 +103,6 @@ func ReadClientHello(
 	return hello, nil
 }
 
-func parseTLSHeader(r io.Reader) (io.Reader, error) {
-	// record_type(1) + version(2) + size(2)
-	//   16 - type is 0x16 (handshake record)
-	//   03 01 - protocol version is "3,1" (also known as TLS 1.0)
-	//   00 f8 - 0xF8 (248) bytes of handshake message follows
-	header := [1 + 2 + 2]byte{}
-
-	if _, err := io.ReadFull(r, header[:]); err != nil {
-		return nil, fmt.Errorf("cannot read record header: %w", err)
-	}
-
-	if header[0] != tls.TypeHandshake {
-		return nil, fmt.Errorf("unexpected record type %#x", header[0])
-	}
-
-	if header[1] != 3 || header[2] != 1 {
-		return nil, fmt.Errorf("unexpected protocol version %#x %#x", header[1], header[2])
-	}
-
-	length := int64(binary.BigEndian.Uint16(header[3:]))
-	buf := &bytes.Buffer{}
-
-	_, err := io.CopyN(buf, r, length)
-
-	return buf, err
-}
-
-func parseHandshakeHeader(r io.Reader) (io.Reader, error) {
-	// type(1) + size(3 / uint24)
-	// 01 - handshake message type 0x01 (client hello)
-	// 00 00 f4 - 0xF4 (244) bytes of client hello data follows
-	header := [1 + 3]byte{}
-
-	if _, err := io.ReadFull(r, header[:]); err != nil {
-		return nil, fmt.Errorf("cannot read handshake header: %w", err)
-	}
-
-	if header[0] != TypeHandshakeClient {
-		return nil, fmt.Errorf("incorrect handshake type: %#x", header[0])
-	}
-
-	// unfortunately there is not uint24 in golang, so we just reust header
-	header[0] = 0
-
-	length := int64(binary.BigEndian.Uint32(header[:]))
-	buf := &bytes.Buffer{}
-
-	_, err := io.CopyN(buf, r, length)
-
-	return buf, err
-}
-
 func parseHandshake(r io.Reader) (*ClientHello, error) {
 	//  A protocol version of "3,3" (meaning TLS 1.2) is given.
 	header := [2]byte{}
@@ -192,16 +133,27 @@ func parseHandshake(r io.Reader) (*ClientHello, error) {
 
 	cipherSuiteLen := int64(binary.BigEndian.Uint16(header[:]))
 
-	// we do not care about picking up any cipher. we pick the first one,
+	// Pick the first non-GREASE cipher suite from the list.
-	// so it is always should be present.
+	// Real TLS servers never select GREASE values (RFC 8701, pattern 0x?a?a),
-	if _, err := io.ReadFull(r, header[:]); err != nil {
+	// so echoing them back is a trivial DPI fingerprint.
-		return nil, fmt.Errorf("cannot read first cipher suite: %w", err)
+	// cipherSuiteLen is in bytes; each cipher suite is 2 bytes.
+	for range cipherSuiteLen / 2 {
+		if _, err := io.ReadFull(r, header[:]); err != nil {
+			return nil, fmt.Errorf("cannot read cipher suite: %w", err)
+		}
+
+		if hello.CipherSuite != 0 {
+			// do not forget we have to scan until the end
+			continue
+		}
+
+		if cs := binary.BigEndian.Uint16(header[:]); cs&GreaseMask != GreaseValueType {
+			hello.CipherSuite = cs
+		}
 	}
 
-	hello.CipherSuite = binary.BigEndian.Uint16(header[:])
+	if hello.CipherSuite == 0 {
-
+		return nil, ErrCannotFindCipher
-	if _, err := io.CopyN(io.Discard, r, cipherSuiteLen-2); err != nil {
-		return nil, fmt.Errorf("cannot skip remaining cipher suites: %w", err)
 	}
 
 	if _, err := io.ReadFull(r, header[:1]); err != nil {

mtglib/internal/tls/fake/client_side_snapshot_test.go

@@ -12,7 +12,6 @@ import (
 	"github.com/9seconds/mtg/v2/mtglib"
 	"github.com/9seconds/mtg/v2/mtglib/internal/tls/fake"
 	"github.com/stretchr/testify/assert"
-	"github.com/stretchr/testify/mock"
 	"github.com/stretchr/testify/require"
 	"github.com/stretchr/testify/suite"
 )
@@ -71,11 +70,6 @@ func (suite *ParseClientHelloSnapshotTestSuite) makeConn(data []byte) *parseClie
 		readBuf: readBuf,
 	}
 
-	connMock.
-		On("SetReadDeadline", mock.AnythingOfType("time.Time")).
-		Twice().
-		Return(nil)
-
 	return connMock
 }
 

mtglib/internal/tls/fake/client_side_test.go

@@ -2,9 +2,11 @@ package fake_test
 
 import (
 	"bytes"
+	cryptotls "crypto/tls"
 	"encoding/binary"
-	"errors"
+	"encoding/json"
 	"io"
+	"os"
 	"testing"
 	"time"
 
@@ -12,7 +14,6 @@ import (
 	"github.com/9seconds/mtg/v2/mtglib"
 	"github.com/9seconds/mtg/v2/mtglib/internal/tls"
 	"github.com/9seconds/mtg/v2/mtglib/internal/tls/fake"
-	"github.com/stretchr/testify/mock"
 	"github.com/stretchr/testify/require"
 	"github.com/stretchr/testify/suite"
 )
@@ -51,11 +52,6 @@ func (suite *ParseClientHelloTestSuite) SetupTest() {
 	suite.connMock = &parseClientHelloConnMock{
 		readBuf: suite.readBuf,
 	}
-
-	suite.connMock.
-		On("SetReadDeadline", mock.AnythingOfType("time.Time")).
-		Twice().
-		Return(nil)
 }
 
 func (suite *ParseClientHelloTestSuite) TearDownTest() {
@@ -67,23 +63,11 @@ type ParseClientHello_TLSHeaderTestSuite struct {
 }
 
 func (suite *ParseClientHello_TLSHeaderTestSuite) TestEmpty() {
-	suite.connMock.ExpectedCalls = []*mock.Call{}
-	suite.connMock.
-		On("SetReadDeadline", mock.AnythingOfType("time.Time")).
-		Once().
-		Return(errors.New("fail"))
-
 	_, err := fake.ReadClientHello(suite.connMock, suite.secret.Key[:], suite.secret.Host, TolerateTime)
-	suite.ErrorContains(err, "fail")
+	suite.ErrorContains(err, "cannot read client hello")
 }
 
 func (suite *ParseClientHello_TLSHeaderTestSuite) TestNothing() {
-	suite.connMock.ExpectedCalls = []*mock.Call{}
-	suite.connMock.
-		On("SetReadDeadline", mock.AnythingOfType("time.Time")).
-		Twice().
-		Return(nil)
-
 	_, err := fake.ReadClientHello(suite.connMock, suite.secret.Key[:], suite.secret.Host, TolerateTime)
 	suite.ErrorIs(err, io.EOF)
 }
@@ -232,12 +216,13 @@ func (suite *ParseClientHelloHandshakeBodyTestSuite) TestCannotReadCipherSuiteLe
 }
 
 func (suite *ParseClientHelloHandshakeBodyTestSuite) TestCannotReadFirstCipherSuite() {
-	body := make([]byte, 2+fake.RandomLen+1+2)
+	body := make([]byte, 2+fake.RandomLen+1+2+1) // cipherSuiteLen=2 but only 1 byte available
+	binary.BigEndian.PutUint16(body[2+fake.RandomLen+1:], 2)
 
 	suite.writeBody(body)
 
 	_, err := fake.ReadClientHello(suite.connMock, suite.secret.Key[:], suite.secret.Host, TolerateTime)
-	suite.ErrorContains(err, "cannot read first cipher suite")
+	suite.ErrorContains(err, "cannot read cipher suite")
 }
 
 func (suite *ParseClientHelloHandshakeBodyTestSuite) TestCannotSkipRemainingCipherSuites() {
@@ -247,12 +232,27 @@ func (suite *ParseClientHelloHandshakeBodyTestSuite) TestCannotSkipRemainingCiph
 	suite.writeBody(body)
 
 	_, err := fake.ReadClientHello(suite.connMock, suite.secret.Key[:], suite.secret.Host, TolerateTime)
-	suite.ErrorContains(err, "cannot skip remaining cipher suites")
+	suite.ErrorContains(err, "cannot read cipher suite")
+}
+
+func (suite *ParseClientHelloHandshakeBodyTestSuite) TestCannotFindCipher() {
+	// All cipher suites are GREASE values — must return ErrCannotFindCipher.
+	body := make([]byte, 2+fake.RandomLen+1+2+4+1)
+	binary.BigEndian.PutUint16(body[2+fake.RandomLen+1:], 4)
+	binary.BigEndian.PutUint16(body[2+fake.RandomLen+1+2:], 0x0a0a)
+	binary.BigEndian.PutUint16(body[2+fake.RandomLen+1+2+2:], 0x1a1a)
+	body[2+fake.RandomLen+1+2+4] = 1
+
+	suite.writeBody(body)
+
+	_, err := fake.ReadClientHello(suite.connMock, suite.secret.Key[:], suite.secret.Host, TolerateTime)
+	suite.ErrorIs(err, fake.ErrCannotFindCipher)
 }
 
 func (suite *ParseClientHelloHandshakeBodyTestSuite) TestCannotReadCompressionMethodsLength() {
 	body := make([]byte, 2+fake.RandomLen+1+2+2)
 	binary.BigEndian.PutUint16(body[2+fake.RandomLen+1:], 2)
+	binary.BigEndian.PutUint16(body[2+fake.RandomLen+1+2:], cryptotls.TLS_AES_128_GCM_SHA256)
 
 	suite.writeBody(body)
 
@@ -263,6 +263,7 @@ func (suite *ParseClientHelloHandshakeBodyTestSuite) TestCannotReadCompressionMe
 func (suite *ParseClientHelloHandshakeBodyTestSuite) TestCannotSkipCompressionMethods() {
 	body := make([]byte, 2+fake.RandomLen+1+2+2+1)
 	binary.BigEndian.PutUint16(body[2+fake.RandomLen+1:], 2)
+	binary.BigEndian.PutUint16(body[2+fake.RandomLen+1+2:], cryptotls.TLS_AES_128_GCM_SHA256)
 	body[2+fake.RandomLen+1+2+2] = 1
 
 	suite.writeBody(body)
@@ -298,6 +299,7 @@ func (suite *ParseClientHelloSNITestSuite) writeExtensions(extensions []byte) {
 	// cipherSuite(2) + compressionLen(1) + compression(1) = 41
 	body := make([]byte, 41)
 	binary.BigEndian.PutUint16(body[35:], 2)
+	binary.BigEndian.PutUint16(body[37:], cryptotls.TLS_AES_128_GCM_SHA256)
 	body[39] = 1
 
 	suite.readBuf.Write(body)
@@ -393,3 +395,268 @@ func TestParseClientHelloSNI(t *testing.T) {
 	t.Parallel()
 	suite.Run(t, &ParseClientHelloSNITestSuite{})
 }
+
+// fragmentTLSRecord splits a single TLS record into n TLS records by
+// dividing the payload into roughly equal parts. Each part gets its own
+// TLS record header with the same record type and version.
+func fragmentTLSRecord(t testing.TB, full []byte, n int) []byte {
+	t.Helper()
+
+	recordType := full[0]
+	version := full[1:3]
+	payload := full[tls.SizeHeader:]
+
+	chunkSize := len(payload) / n
+	result := &bytes.Buffer{}
+
+	for i := 0; i < n; i++ {
+		start := i * chunkSize
+		end := start + chunkSize
+
+		if i == n-1 {
+			end = len(payload)
+		}
+
+		chunk := payload[start:end]
+		result.WriteByte(recordType)
+		result.Write(version)
+		require.NoError(t, binary.Write(result, binary.BigEndian, uint16(len(chunk))))
+		result.Write(chunk)
+	}
+
+	return result.Bytes()
+}
+
+// splitPayloadAt creates two TLS records from a single record by splitting
+// the payload at the given byte position.
+func splitPayloadAt(t testing.TB, full []byte, pos int) []byte {
+	t.Helper()
+
+	payload := full[tls.SizeHeader:]
+	buf := &bytes.Buffer{}
+
+	buf.WriteByte(tls.TypeHandshake)
+	buf.Write(full[1:3])
+	require.NoError(t, binary.Write(buf, binary.BigEndian, uint16(pos)))
+	buf.Write(payload[:pos])
+
+	buf.WriteByte(tls.TypeHandshake)
+	buf.Write(full[1:3])
+	require.NoError(t, binary.Write(buf, binary.BigEndian, uint16(len(payload)-pos)))
+	buf.Write(payload[pos:])
+
+	return buf.Bytes()
+}
+
+type ParseClientHelloFragmentedTestSuite struct {
+	suite.Suite
+
+	secret   mtglib.Secret
+	snapshot *clientHelloSnapshot
+}
+
+func (s *ParseClientHelloFragmentedTestSuite) SetupSuite() {
+	parsed, err := mtglib.ParseSecret(
+		"ee367a189aee18fa31c190054efd4a8e9573746f726167652e676f6f676c65617069732e636f6d",
+	)
+	require.NoError(s.T(), err)
+
+	s.secret = parsed
+
+	fileData, err := os.ReadFile("testdata/client-hello-ok-19dfe38384b9884b.json")
+	require.NoError(s.T(), err)
+
+	s.snapshot = &clientHelloSnapshot{}
+	require.NoError(s.T(), json.Unmarshal(fileData, s.snapshot))
+}
+
+func (s *ParseClientHelloFragmentedTestSuite) makeConn(data []byte) *parseClientHelloConnMock {
+	readBuf := &bytes.Buffer{}
+	readBuf.Write(data)
+
+	connMock := &parseClientHelloConnMock{
+		readBuf: readBuf,
+	}
+
+	return connMock
+}
+
+func (s *ParseClientHelloFragmentedTestSuite) TestReassemblySuccess() {
+	full := s.snapshot.GetFull()
+
+	tests := []struct {
+		name string
+		data []byte
+	}{
+		{"two equal fragments", fragmentTLSRecord(s.T(), full, 2)},
+		{"three equal fragments", fragmentTLSRecord(s.T(), full, 3)},
+		{"single byte first fragment", splitPayloadAt(s.T(), full, 1)},
+		{"three byte first fragment", splitPayloadAt(s.T(), full, 3)},
+	}
+
+	for _, tt := range tests {
+		s.Run(tt.name, func() {
+			connMock := s.makeConn(tt.data)
+			defer connMock.AssertExpectations(s.T())
+
+			hello, err := fake.ReadClientHello(
+				connMock,
+				s.secret.Key[:],
+				s.secret.Host,
+				TolerateTime,
+			)
+			s.Require().NoError(err)
+
+			s.Equal(s.snapshot.GetRandom(), hello.Random[:])
+			s.Equal(s.snapshot.GetSessionID(), hello.SessionID)
+			s.Equal(uint16(s.snapshot.CipherSuite), hello.CipherSuite)
+		})
+	}
+}
+
+func (s *ParseClientHelloFragmentedTestSuite) TestReassemblyErrors() {
+	full := s.snapshot.GetFull()
+	payload := full[tls.SizeHeader:]
+
+	tests := []struct {
+		name      string
+		buildData func() []byte
+		errMsg    string
+	}{
+		{
+			name: "wrong continuation record type",
+			buildData: func() []byte {
+				buf := &bytes.Buffer{}
+				buf.WriteByte(tls.TypeHandshake)
+				buf.Write(full[1:3])
+				require.NoError(s.T(), binary.Write(buf, binary.BigEndian, uint16(10)))
+				buf.Write(payload[:10])
+				// Wrong type: application data instead of handshake
+				buf.WriteByte(tls.TypeApplicationData)
+				buf.Write(full[1:3])
+				require.NoError(s.T(), binary.Write(buf, binary.BigEndian, uint16(len(payload)-10)))
+				buf.Write(payload[10:])
+				return buf.Bytes()
+			},
+			errMsg: "unexpected record type",
+		},
+		{
+			name: "too many continuation records",
+			buildData: func() []byte {
+				// Handshake header claiming 256 bytes, but we only send 1 byte per continuation
+				handshakePayload := []byte{0x01, 0x00, 0x01, 0x00}
+				buf := &bytes.Buffer{}
+				buf.WriteByte(tls.TypeHandshake)
+				buf.Write([]byte{3, 1})
+				require.NoError(s.T(), binary.Write(buf, binary.BigEndian, uint16(len(handshakePayload))))
+				buf.Write(handshakePayload)
+				for range 11 {
+					buf.WriteByte(tls.TypeHandshake)
+					buf.Write([]byte{3, 1})
+					require.NoError(s.T(), binary.Write(buf, binary.BigEndian, uint16(1)))
+					buf.WriteByte(0xAB)
+				}
+				return buf.Bytes()
+			},
+			errMsg: "too many fragments",
+		},
+		{
+			name: "zero-length continuation record",
+			buildData: func() []byte {
+				buf := &bytes.Buffer{}
+				buf.WriteByte(tls.TypeHandshake)
+				buf.Write(full[1:3])
+				require.NoError(s.T(), binary.Write(buf, binary.BigEndian, uint16(10)))
+				buf.Write(payload[:10])
+				// Valid header but zero-length payload
+				buf.WriteByte(tls.TypeHandshake)
+				buf.Write(full[1:3])
+				require.NoError(s.T(), binary.Write(buf, binary.BigEndian, uint16(0)))
+				return buf.Bytes()
+			},
+			errMsg: "cannot read record header",
+		},
+		{
+			name: "wrong continuation record version",
+			buildData: func() []byte {
+				buf := &bytes.Buffer{}
+				buf.WriteByte(tls.TypeHandshake)
+				buf.Write(full[1:3])
+				require.NoError(s.T(), binary.Write(buf, binary.BigEndian, uint16(10)))
+				buf.Write(payload[:10])
+				// Wrong version: 3.3 instead of 3.1
+				buf.WriteByte(tls.TypeHandshake)
+				buf.Write([]byte{3, 3})
+				require.NoError(s.T(), binary.Write(buf, binary.BigEndian, uint16(len(payload)-10)))
+				buf.Write(payload[10:])
+				return buf.Bytes()
+			},
+			errMsg: "unexpected protocol version",
+		},
+		{
+			name: "handshake message too large",
+			buildData: func() []byte {
+				// Handshake header claiming 0x010000 (65536) bytes — exceeds 0xFFFF limit
+				handshakePayload := []byte{0x01, 0x01, 0x00, 0x00}
+				buf := &bytes.Buffer{}
+				buf.WriteByte(tls.TypeHandshake)
+				buf.Write([]byte{3, 1})
+				require.NoError(s.T(), binary.Write(buf, binary.BigEndian, uint16(len(handshakePayload))))
+				buf.Write(handshakePayload)
+				return buf.Bytes()
+			},
+			errMsg: "cannot read record header",
+		},
+		{
+			name: "truncated continuation record header",
+			buildData: func() []byte {
+				buf := &bytes.Buffer{}
+				buf.WriteByte(tls.TypeHandshake)
+				buf.Write(full[1:3])
+				require.NoError(s.T(), binary.Write(buf, binary.BigEndian, uint16(10)))
+				buf.Write(payload[:10])
+				// Connection ends mid-header (only 2 bytes)
+				buf.WriteByte(tls.TypeHandshake)
+				buf.WriteByte(3)
+				return buf.Bytes()
+			},
+			errMsg: "cannot read record header",
+		},
+		{
+			name: "truncated continuation record payload",
+			buildData: func() []byte {
+				buf := &bytes.Buffer{}
+				buf.WriteByte(tls.TypeHandshake)
+				buf.Write(full[1:3])
+				require.NoError(s.T(), binary.Write(buf, binary.BigEndian, uint16(10)))
+				buf.Write(payload[:10])
+				// Claims 100 bytes but no payload follows
+				buf.WriteByte(tls.TypeHandshake)
+				buf.Write(full[1:3])
+				require.NoError(s.T(), binary.Write(buf, binary.BigEndian, uint16(100)))
+				return buf.Bytes()
+			},
+			errMsg: "EOF",
+		},
+	}
+
+	for _, tt := range tests {
+		s.Run(tt.name, func() {
+			connMock := s.makeConn(tt.buildData())
+			defer connMock.AssertExpectations(s.T())
+
+			_, err := fake.ReadClientHello(
+				connMock,
+				s.secret.Key[:],
+				s.secret.Host,
+				TolerateTime,
+			)
+			s.ErrorContains(err, tt.errMsg)
+		})
+	}
+}
+
+func TestParseClientHelloFragmented(t *testing.T) {
+	t.Parallel()
+	suite.Run(t, &ParseClientHelloFragmentedTestSuite{})
+}

mtglib/internal/tls/fake/init.go

@@ -2,15 +2,6 @@ package fake
 
 import (
 	"errors"
-	"time"
 )
 
-const (
+var ErrBadDigest = errors.New("incorrect client random")
-	ClientHelloReadTimeout = 5 * time.Second
-)
-
-var (
-	resetDeadline time.Time
-
-	ErrBadDigest = errors.New("incorrect client random")
-)

mtglib/internal/tls/fake/server_side_test.go

@@ -58,7 +58,7 @@ func (suite *SendServerHelloTestSuite) TestRecordStructure() {
 	recordType, length, err := tls.ReadRecord(suite.buf, &rec)
 	suite.NoError(err)
 	suite.Equal(byte(tls.TypeApplicationData), recordType)
-	suite.Greater(length, int64(2500))
+	suite.GreaterOrEqual(length, int64(2500))
 
 	suite.Empty(suite.buf.Bytes())
 }

mtglib/internal/tls/fake/testdata/client-hello-ok-grease-first.json

@@ -0,0 +1,8 @@
+{
+  "time": 1617181365,
+  "random": "w4TaDfYg/aUKdx1oi68vxMKvHJczRNvtRRppLETzeNE=",
+  "sessionId": "St2BZ2uHMFn3B2trD1jfdtpjoJOOg6JBeLhFcyCMCq4=",
+  "host": "storage.googleapis.com",
+  "cipherSuite": 4867,
+  "full": "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"
+}

mtglib/internal/tls/fake/utils.go

@@ -0,0 +1,158 @@
+package fake
+
+import (
+	"bytes"
+	"encoding/binary"
+	"errors"
+	"fmt"
+	"io"
+
+	"github.com/9seconds/mtg/v2/mtglib/internal/tls"
+)
+
+const (
+	maxFragmentsCount = 10
+)
+
+var ErrTooManyFragments = errors.New("too many fragments")
+
+// https://datatracker.ietf.org/doc/html/rfc5246#section-6.2.1
+// client hello can be fragmented in a series of packets:
+//
+//	Bytes on the wire:
+//
+// 16 03 01 00 F8 01 00 00 F4 03 03 [32 bytes random] [session_id] [ciphers] [SNI...]
+// ├─────────────┤├──────────────────────────────────────────────────────────────────┤
+//
+//	TLS record       Payload (248 bytes)
+//	header (5B)
+//
+//	16    = Handshake
+//	03 01 = TLS 1.0 (record layer version)
+//	00 F8 = 248 bytes follow
+//
+//	01       = ClientHello (handshake type)
+//	00 00 F4 = 244 bytes of handshake body
+//	03 03    = TLS 1.2 (actual protocol version)
+//	...rest of ClientHello...
+//
+// Fragmented record look like:
+//
+//	Record 1:
+//
+// 16 03 01 00 03 01 00 00
+// ├─────────────┤├──────┤
+//
+//	TLS header    3 bytes of payload
+//
+//	16    = Handshake
+//	03 01 = TLS 1.0
+//	00 03 = only 3 bytes follow
+//
+//	01       = ClientHello type
+//	00 00    = first 2 bytes of the uint24 length (INCOMPLETE!)
+//
+// Record 2:
+// 16 03 01 00 F5 F4 03 03 [32 bytes random] [session_id] [ciphers] [SNI...]
+// ├─────────────┤├────────────────────────────────────────────────────────────┤
+//
+//	TLS header    remaining 245 bytes of payload
+//
+//	16    = Handshake
+//	03 01 = TLS 1.0
+//	00 F5 = 245 bytes follow
+//
+//	F4    = last byte of uint24 length (now complete: 00 00 F4 = 244)
+//	03 03 = TLS 1.2
+//	...rest of ClientHello continues...
+//
+// So it means that there could be a series of handshake packets of different
+// lengths. The goal of this function is to concatenate these fragments.
+type fragmentedHandshakeReader struct {
+	r             io.Reader
+	buf           bytes.Buffer
+	readFragments int
+}
+
+func (f *fragmentedHandshakeReader) Read(p []byte) (int, error) {
+	if n, err := f.buf.Read(p); err == nil {
+		return n, nil
+	}
+
+	f.buf.Reset()
+
+	for f.buf.Len() == 0 {
+		if f.readFragments > maxFragmentsCount {
+			return 0, ErrTooManyFragments
+		}
+
+		if err := f.parseNextFragment(); err != nil {
+			return 0, err
+		}
+
+		f.readFragments++
+	}
+
+	return f.buf.Read(p)
+}
+
+func (f *fragmentedHandshakeReader) parseNextFragment() error {
+	// record_type(1) + version(2) + size(2)
+	//   16 - type is 0x16 (handshake record)
+	//   03 01 - protocol version is "3,1" (also known as TLS 1.0)
+	//   00 f8 - 0xF8 (248) bytes of handshake message follows
+	header := [1 + 2 + 2]byte{}
+
+	if _, err := io.ReadFull(f.r, header[:]); err != nil {
+		return fmt.Errorf("cannot read record header: %w", err)
+	}
+
+	if header[0] != tls.TypeHandshake {
+		return fmt.Errorf("unexpected record type %#x", header[0])
+	}
+
+	if header[1] != 3 || header[2] != 1 {
+		return fmt.Errorf("unexpected protocol version %#x %#x", header[1], header[2])
+	}
+
+	length := int64(binary.BigEndian.Uint16(header[3:]))
+	_, err := io.CopyN(&f.buf, f.r, length)
+
+	return err
+}
+
+func parseClientHello(r io.Reader) (*bytes.Buffer, *bytes.Buffer, error) {
+	r = &fragmentedHandshakeReader{r: r}
+	header := [1 + 3]byte{}
+
+	if _, err := io.ReadFull(r, header[:]); err != nil {
+		return nil, nil, fmt.Errorf("cannot read handshake header: %w", err)
+	}
+
+	if header[0] != TypeHandshakeClient {
+		return nil, nil, fmt.Errorf("incorrect handshake type: %#x", header[0])
+	}
+
+	// unfortunately there is not uint24 in golang, so we just reuse header
+	header[0] = 0
+	length := int64(binary.BigEndian.Uint32(header[:]))
+
+	clientHelloCopy := &bytes.Buffer{}
+	clientHelloCopy.Write([]byte{tls.TypeHandshake, 3, 1})
+	binary.Write( //nolint: errcheck
+		clientHelloCopy,
+		binary.BigEndian,
+		// 1 for handshake type
+		// 3 for handshake length
+		uint16(1+3+length),
+	)
+	clientHelloCopy.WriteByte(TypeHandshakeClient)
+	clientHelloCopy.Write(header[1:])
+
+	handshakeCopy := &bytes.Buffer{}
+	writer := io.MultiWriter(clientHelloCopy, handshakeCopy)
+
+	_, err := io.CopyN(writer, r, length)
+
+	return clientHelloCopy, handshakeCopy, err
+}

mtglib/proxy.go

@@ -28,6 +28,7 @@ type Proxy struct {
 	allowFallbackOnUnknownDC    bool
 	tolerateTimeSkewness        time.Duration
 	idleTimeout                 time.Duration
+	handshakeTimeout            time.Duration
 	domainFrontingPort          int
 	domainFrontingIP            string
 	domainFrontingProxyProtocol bool
@@ -66,6 +67,11 @@ func (p *Proxy) ServeConn(conn essentials.Conn) {
 	ctx := newStreamContext(p.ctx, p.logger, conn)
 	defer ctx.Close()
 
+	if err := ctx.clientConn.SetDeadline(time.Now().Add(p.handshakeTimeout)); err != nil {
+		ctx.logger.WarningError("cannot set handshake timeout", err)
+		return
+	}
+
 	stop := context.AfterFunc(ctx, func() {
 		ctx.Close()
 	})
@@ -97,16 +103,23 @@ func (p *Proxy) ServeConn(conn essentials.Conn) {
 		return
 	}
 
+	if err := ctx.clientConn.SetDeadline(time.Time{}); err != nil {
+		ctx.logger.WarningError("cannot set deadline", err)
+		return
+	}
+
 	if err := p.doTelegramCall(ctx); err != nil {
 		ctx.logger.WarningError("cannot dial to telegram", err)
 		return
 	}
 
+	tracker := newIdleTracker(p.idleTimeout)
+
 	relay.Relay(
 		ctx,
 		ctx.logger.Named("relay"),
-		connIdleTimeout{Conn: ctx.telegramConn, timeout: p.idleTimeout},
+		connIdleTimeout{Conn: ctx.telegramConn, tracker: tracker},
-		connIdleTimeout{Conn: ctx.clientConn, timeout: p.idleTimeout},
+		connIdleTimeout{Conn: ctx.clientConn, tracker: tracker},
 	)
 }
 
@@ -305,11 +318,13 @@ func (p *Proxy) doDomainFronting(ctx *streamContext, conn *connRewind) {
 		stream:   p.eventStream,
 	}
 
+	tracker := newIdleTracker(p.idleTimeout)
+
 	relay.Relay(
 		ctx,
 		ctx.logger.Named("domain-fronting"),
-		connIdleTimeout{Conn: frontConn, timeout: p.idleTimeout},
+		connIdleTimeout{Conn: frontConn, tracker: tracker},
-		connIdleTimeout{Conn: conn, timeout: p.idleTimeout},
+		connIdleTimeout{Conn: conn, tracker: tracker},
 	)
 }
 
@@ -342,6 +357,7 @@ func NewProxy(opts ProxyOpts) (*Proxy, error) {
 		domainFrontingIP:         opts.DomainFrontingIP,
 		tolerateTimeSkewness:     opts.getTolerateTimeSkewness(),
 		idleTimeout:              opts.getIdleTimeout(),
+		handshakeTimeout:         opts.getHandshakeTimeout(),
 		allowFallbackOnUnknownDC: opts.AllowFallbackOnUnknownDC,
 		telegram:                 tg,
 		doppelGanger: doppel.NewGanger(

mtglib/proxy_opts.go

@@ -70,6 +70,12 @@ type ProxyOpts struct {
 	// This is an optional setting.
 	IdleTimeout time.Duration
 
+	// HandshakeTimeout is a timeout during which all handshake ceremonies must
+	// be completed, otherwise this process will be aborted
+	//
+	// This is an optional setting.
+	HandshakeTimeout time.Duration
+
 	// TolerateTimeSkewness is a time boundary that defines a time range where
 	// faketls timestamp is acceptable.
 	//
@@ -160,7 +166,6 @@ type ProxyOpts struct {
 
 	// DoppelGangerDRS defines if TLS Dynamic Record Sizing is active.
 	DoppelGangerDRS bool
-
 }
 
 func (p ProxyOpts) valid() error {
@@ -216,9 +221,17 @@ func (p ProxyOpts) getPreferIP() string {
 	return p.PreferIP
 }
 
+func (p ProxyOpts) getHandshakeTimeout() time.Duration {
+	if p.HandshakeTimeout == 0 {
+		return DefaultHandshakeTimeout
+	}
+
+	return p.HandshakeTimeout
+}
+
 func (p ProxyOpts) getIdleTimeout() time.Duration {
 	if p.IdleTimeout == 0 {
-		return time.Minute
+		return DefaultIdleTimeout
 	}
 
 	return p.IdleTimeout

mtglib/proxy_test.go

@@ -175,7 +175,7 @@ func (suite *ProxyTestSuite) TestHTTPSRequest() {
 	addr := fmt.Sprintf("https://%s/headers", suite.ProxyAddress())
 
 	resp, err := client.Get(addr) //nolint: noctx
-	suite.NoError(err)
+	suite.Require().NoError(err)
 
 	defer resp.Body.Close() //nolint: errcheck
 

network/init.go

@@ -36,8 +36,22 @@ const (
 
 	// DefaultTCPKeepAlivePeriod defines a time period between 2 consequitive
 	// probes.
+	//
+	// Deprecated: use DefaultKeepAliveIdle and DefaultKeepAliveInterval instead.
 	DefaultTCPKeepAlivePeriod = 10 * time.Second
 
+	// DefaultKeepAliveIdle is the time a connection must be idle before
+	// the first keepalive probe is sent.
+	DefaultKeepAliveIdle = 30 * time.Second
+
+	// DefaultKeepAliveInterval is the time between consecutive keepalive
+	// probes.
+	DefaultKeepAliveInterval = 10 * time.Second
+
+	// DefaultKeepAliveCount is the number of unacknowledged probes before
+	// the connection is considered dead.
+	DefaultKeepAliveCount = 3
+
 	// ProxyDialerOpenThreshold is used for load balancing SOCKS5 dialer only.
 	//
 	// This dialer uses circuit breaker with of 3 stages: OPEN, HALF_OPEN and

network/sockopts.go

@@ -20,8 +20,13 @@ func SetServerSocketOptions(conn net.Conn, bufferSize int) error {
 }
 
 func setCommonSocketOptions(conn *net.TCPConn) error {
-	if err := conn.SetKeepAlivePeriod(DefaultTCPKeepAlivePeriod); err != nil {
+	if err := conn.SetKeepAliveConfig(net.KeepAliveConfig{
-		return fmt.Errorf("cannot set time period of TCP keepalive probes: %w", err)
+		Enable:   true,
+		Idle:     DefaultKeepAliveIdle,
+		Interval: DefaultKeepAliveInterval,
+		Count:    DefaultKeepAliveCount,
+	}); err != nil {
+		return fmt.Errorf("cannot configure TCP keepalive: %w", err)
 	}
 
 	if err := conn.SetLinger(tcpLingerTimeout); err != nil {

network/sockopts_test.go

@@ -0,0 +1,93 @@
+//go:build linux || darwin
+// +build linux darwin
+
+package network_test
+
+import (
+	"net"
+	"runtime"
+	"syscall"
+	"testing"
+	"time"
+
+	"github.com/9seconds/mtg/v2/network"
+	"github.com/stretchr/testify/require"
+	"golang.org/x/sys/unix"
+)
+
+func tcpKeepIdleOption() int {
+	if runtime.GOOS == "darwin" {
+		return 0x10 // TCP_KEEPALIVE on macOS
+	}
+
+	return 0x4 // TCP_KEEPIDLE on Linux
+}
+
+func TestSetClientSocketOptionsKeepAlive(t *testing.T) {
+	t.Parallel()
+
+	listener, err := net.Listen("tcp", "127.0.0.1:0")
+	require.NoError(t, err)
+	defer func() {
+		err := listener.Close()
+		require.NoError(t, err)
+	}()
+
+	type dialResult struct {
+		conn net.Conn
+		err  error
+	}
+
+	dialDone := make(chan dialResult, 1)
+
+	go func() {
+		c, err := net.Dial("tcp", listener.Addr().String())
+		dialDone <- dialResult{conn: c, err: err}
+	}()
+
+	tcpListener, ok := listener.(*net.TCPListener)
+	require.True(t, ok, "listener must be a *net.TCPListener")
+
+	require.NoError(t, tcpListener.SetDeadline(time.Now().Add(5*time.Second)))
+
+	accepted, err := listener.Accept()
+	require.NoError(t, err)
+	defer func() {
+		err := accepted.Close()
+		require.NoError(t, err)
+	}()
+
+	dr := <-dialDone
+	require.NoError(t, dr.err)
+	defer func() {
+		err := dr.conn.Close()
+		require.NoError(t, err)
+	}()
+
+	err = network.SetClientSocketOptions(accepted, 0)
+	require.NoError(t, err)
+
+	tcpConn := accepted.(*net.TCPConn)
+
+	rawConn, err := tcpConn.SyscallConn()
+	require.NoError(t, err)
+
+	err = rawConn.Control(func(fd uintptr) {
+		val, err := unix.GetsockoptInt(int(fd), syscall.SOL_SOCKET, syscall.SO_KEEPALIVE)
+		require.NoError(t, err)
+		require.NotEqual(t, 0, val, "SO_KEEPALIVE should be enabled")
+
+		idle, err := unix.GetsockoptInt(int(fd), syscall.IPPROTO_TCP, tcpKeepIdleOption())
+		require.NoError(t, err)
+		require.Equal(t, int(network.DefaultKeepAliveIdle.Seconds()), idle)
+
+		interval, err := unix.GetsockoptInt(int(fd), syscall.IPPROTO_TCP, unix.TCP_KEEPINTVL)
+		require.NoError(t, err)
+		require.Equal(t, int(network.DefaultKeepAliveInterval.Seconds()), interval)
+
+		count, err := unix.GetsockoptInt(int(fd), syscall.IPPROTO_TCP, unix.TCP_KEEPCNT)
+		require.NoError(t, err)
+		require.Equal(t, network.DefaultKeepAliveCount, count)
+	})
+	require.NoError(t, err)
+}

network/v2/base_http_test.go

@@ -25,7 +25,7 @@ func (suite *BaseHTTPTestSuite) SetupSuite() {
 }
 
 func (suite *BaseHTTPTestSuite) SetupTest() {
-	suite.client = network.New(nil, "mtg/1", 0, 0, 0).MakeHTTPClient(nil)
+	suite.client = network.New(nil, "mtg/1", 0, 0, 0, network.DefaultKeepAliveConfig).MakeHTTPClient(nil)
 }
 
 func (suite *BaseHTTPTestSuite) TestGet() {

network/v2/base_network_test.go

@@ -19,7 +19,7 @@ type BaseNetworkTestSuite struct {
 func (suite *BaseNetworkTestSuite) SetupSuite() {
 	suite.EchoServerTestSuite.SetupSuite()
 
-	suite.net = network.New(nil, "agent", 0, 0, 0)
+	suite.net = network.New(nil, "agent", 0, 0, 0, network.DefaultKeepAliveConfig)
 }
 
 func (suite *BaseNetworkTestSuite) TestDialUnknownNetwork() {

network/v2/init.go

@@ -11,6 +11,7 @@ package network
 
 import (
 	"errors"
+	"net"
 	"time"
 )
 
@@ -26,14 +27,55 @@ const (
 
 	// DefaultTCPKeepAlivePeriod defines a time period between 2 consecuitive
 	// probes.
+	//
+	// Deprecated: use DefaultKeepAliveConfig
 	DefaultTCPKeepAlivePeriod = 10 * time.Second
 
+	// DefaultKeepAliveIdle is the time a connection must be idle before
+	// the first keepalive probe is sent.
+	//
+	// Deprecated: use DefaultKeepAliveConfig
+	DefaultKeepAliveIdle = 30 * time.Second
+
+	// DefaultKeepAliveInterval is the time between consecutive keepalive
+	// probes.
+	//
+	// Deprecated: use DefaultKeepAliveConfig
+	DefaultKeepAliveInterval = 10 * time.Second
+
+	// DefaultKeepAliveCount is the number of unacknowledged probes before
+	// the connection is considered dead.
+	//
+	// Deprecated: use DefaultKeepAliveConfig
+	DefaultKeepAliveCount = 3
+
 	// User Agent to use in HTTP client.
 	UserAgent = "curl/8.5.0"
 
 	// tcpLingerTimeout defines a number of seconds to wait for sending
 	// unacknowledged data.
 	tcpLingerTimeout = 1
+
+	// tcpNotSentLowat limits the amount of unsent data queued in the
+	// kernel write buffer per socket. When the unsent data drops below
+	// this threshold, the socket becomes writable again. This reduces
+	// per-connection memory usage and bufferbloat by applying
+	// back-pressure to the relay loop instead of piling up data in
+	// kernel buffers.
+	tcpNotSentLowat = 128 * 1024
 )
 
-var ErrCannotDial = errors.New("cannot dial to any address")
+var (
+	ErrCannotDial = errors.New("cannot dial to any address")
+
+	// DefaultKeepAliveConfig defines a default configuration for
+	// keep alive settings. As per official documentation, if keep alive
+	// is enabled, then:
+	//
+	//  Idle = 15 * time.Second
+	//  Interval = 15 * time.Second
+	//  Count = 9
+	DefaultKeepAliveConfig = net.KeepAliveConfig{
+		Enable: true,
+	}
+)

network/v2/network.go

@@ -14,9 +14,10 @@ import (
 type network struct {
 	net.Dialer
 
-	httpTimeout time.Duration
+	keepAliveConfig net.KeepAliveConfig
-	idleTimeout time.Duration
+	httpTimeout     time.Duration
-	userAgent   string
+	idleTimeout     time.Duration
+	userAgent       string
 }
 
 func (n *network) Dial(network, address string) (essentials.Conn, error) {
@@ -37,7 +38,7 @@ func (n *network) DialContext(ctx context.Context, network, address string) (ess
 
 	tcpConn := conn.(*net.TCPConn)
 
-	return tcpConn, setCommonSocketOptions(tcpConn)
+	return tcpConn, setCommonSocketOptions(tcpConn, n.keepAliveConfig)
 }
 
 func (n *network) MakeHTTPClient(
@@ -71,6 +72,7 @@ func New(
 	tcpTimeout,
 	httpTimeout,
 	idleTimeout time.Duration,
+	keepAliveConfig net.KeepAliveConfig,
 ) mtglib.Network {
 	if dnsResolver == nil {
 		dnsResolver = net.DefaultResolver
@@ -86,8 +88,9 @@ func New(
 			Resolver:      dnsResolver,
 			FallbackDelay: -1,
 		},
-		userAgent:   userAgent,
+		userAgent:       userAgent,
-		idleTimeout: idleTimeout,
+		idleTimeout:     idleTimeout,
-		httpTimeout: httpTimeout,
+		httpTimeout:     httpTimeout,
+		keepAliveConfig: keepAliveConfig,
 	}
 }

network/v2/sockopts.go

@@ -5,9 +5,9 @@ import (
 	"net"
 )
 
-func setCommonSocketOptions(conn *net.TCPConn) error {
+func setCommonSocketOptions(conn *net.TCPConn, keepAliveConfig net.KeepAliveConfig) error {
-	if err := conn.SetKeepAlivePeriod(DefaultTCPKeepAlivePeriod); err != nil {
+	if err := conn.SetKeepAliveConfig(keepAliveConfig); err != nil {
-		return fmt.Errorf("cannot set time period of TCP keepalive probes: %w", err)
+		return fmt.Errorf("cannot configure TCP keepalive: %w", err)
 	}
 
 	if err := conn.SetLinger(tcpLingerTimeout); err != nil {
@@ -23,5 +23,9 @@ func setCommonSocketOptions(conn *net.TCPConn) error {
 		return fmt.Errorf("cannot setup SO_REUSEADDR/PORT: %w", err)
 	}
 
+	setCongestionControl(rawConn)
+	setTCPUserTimeout(rawConn, keepAliveConfig)
+	setNotSentLowat(rawConn)
+
 	return nil
 }

network/v2/sockopts_congestion.go

@@ -0,0 +1,20 @@
+//go:build linux
+
+package network
+
+import (
+	"syscall"
+
+	"golang.org/x/sys/unix"
+)
+
+// setCongestionControl sets BBR as the TCP congestion control algorithm.
+// BBR provides better throughput over lossy and high-latency links compared
+// to the default cubic, which is especially beneficial for mobile and
+// home internet clients. This is best-effort: silently ignored if the
+// kernel does not have tcp_bbr available.
+func setCongestionControl(conn syscall.RawConn) {
+	conn.Control(func(fd uintptr) { //nolint: errcheck
+		unix.SetsockoptString(int(fd), unix.IPPROTO_TCP, unix.TCP_CONGESTION, "bbr") //nolint: errcheck
+	})
+}

network/v2/sockopts_congestion_stub.go

@@ -0,0 +1,7 @@
+//go:build !linux
+
+package network
+
+import "syscall"
+
+func setCongestionControl(conn syscall.RawConn) {}

network/v2/sockopts_lowat.go

@@ -0,0 +1,20 @@
+//go:build linux || darwin
+
+package network
+
+import (
+	"syscall"
+
+	"golang.org/x/sys/unix"
+)
+
+// setNotSentLowat sets TCP_NOTSENT_LOWAT which limits the amount of
+// unsent data queued in the kernel write buffer. Once unsent data drops
+// below this threshold the socket becomes writable again, applying
+// back-pressure to the relay loop instead of piling up data in kernel
+// buffers. This reduces per-connection memory and bufferbloat.
+func setNotSentLowat(conn syscall.RawConn) {
+	conn.Control(func(fd uintptr) { //nolint: errcheck
+		unix.SetsockoptInt(int(fd), unix.IPPROTO_TCP, unix.TCP_NOTSENT_LOWAT, tcpNotSentLowat) //nolint: errcheck
+	})
+}

network/v2/sockopts_lowat_stub.go

@@ -0,0 +1,7 @@
+//go:build !linux && !darwin
+
+package network
+
+import "syscall"
+
+func setNotSentLowat(conn syscall.RawConn) {}

network/v2/sockopts_reuseaddr.go

@@ -1,5 +1,4 @@
 //go:build !windows
-// +build !windows
 
 package network
 

network/v2/sockopts_reuseaddr_stub.go

@@ -1,5 +1,4 @@
 //go:build windows
-// +build windows
 
 package network
 

network/v2/sockopts_test.go

@@ -0,0 +1,92 @@
+//go:build linux || darwin
+// +build linux darwin
+
+package network
+
+import (
+	"net"
+	"runtime"
+	"syscall"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/require"
+	"golang.org/x/sys/unix"
+)
+
+func tcpKeepIdleOption() int {
+	if runtime.GOOS == "darwin" {
+		return 0x10 // TCP_KEEPALIVE on macOS
+	}
+
+	return 0x4 // TCP_KEEPIDLE on Linux
+}
+
+func TestSetCommonSocketOptionsKeepAlive(t *testing.T) {
+	t.Parallel()
+
+	listener, err := net.Listen("tcp", "127.0.0.1:0")
+	require.NoError(t, err)
+	defer func() {
+		err := listener.Close()
+		require.NoError(t, err)
+	}()
+
+	type dialResult struct {
+		conn net.Conn
+		err  error
+	}
+
+	dialDone := make(chan dialResult, 1)
+
+	go func() {
+		c, err := net.Dial("tcp", listener.Addr().String())
+		dialDone <- dialResult{conn: c, err: err}
+	}()
+
+	tcpListener, ok := listener.(*net.TCPListener)
+	require.True(t, ok, "listener must be a *net.TCPListener")
+
+	require.NoError(t, tcpListener.SetDeadline(time.Now().Add(5*time.Second)))
+
+	accepted, err := listener.Accept()
+	require.NoError(t, err)
+	defer func() {
+		err := accepted.Close()
+		require.NoError(t, err)
+	}()
+
+	dr := <-dialDone
+	require.NoError(t, dr.err)
+	defer func() {
+		err := dr.conn.Close()
+		require.NoError(t, err)
+	}()
+
+	tcpConn := accepted.(*net.TCPConn)
+
+	err = setCommonSocketOptions(tcpConn, DefaultKeepAliveConfig)
+	require.NoError(t, err)
+
+	rawConn, err := tcpConn.SyscallConn()
+	require.NoError(t, err)
+
+	err = rawConn.Control(func(fd uintptr) {
+		val, err := unix.GetsockoptInt(int(fd), syscall.SOL_SOCKET, syscall.SO_KEEPALIVE)
+		require.NoError(t, err)
+		require.NotEqual(t, 0, val, "SO_KEEPALIVE should be enabled")
+
+		idle, err := unix.GetsockoptInt(int(fd), syscall.IPPROTO_TCP, tcpKeepIdleOption())
+		require.NoError(t, err)
+		require.Equal(t, 15, idle, "keepalive idle should match DefaultKeepAliveIdle")
+
+		interval, err := unix.GetsockoptInt(int(fd), syscall.IPPROTO_TCP, unix.TCP_KEEPINTVL)
+		require.NoError(t, err)
+		require.Equal(t, 15, interval, "keepalive interval should match DefaultKeepAliveInterval")
+
+		count, err := unix.GetsockoptInt(int(fd), syscall.IPPROTO_TCP, unix.TCP_KEEPCNT)
+		require.NoError(t, err)
+		require.Equal(t, 9, count, "keepalive count should match DefaultKeepAliveCount")
+	})
+	require.NoError(t, err)
+}

network/v2/sockopts_usertimeout.go

@@ -0,0 +1,48 @@
+//go:build linux
+
+package network
+
+import (
+	"net"
+	"syscall"
+	"time"
+
+	"golang.org/x/sys/unix"
+)
+
+// Go runtime defaults for KeepAliveConfig when fields are zero.
+const (
+	goDefaultKeepAliveIdle     = 15 * time.Second
+	goDefaultKeepAliveInterval = 15 * time.Second
+	goDefaultKeepAliveCount    = 9
+)
+
+// setTCPUserTimeout sets TCP_USER_TIMEOUT on a socket. If transmitted
+// data remains unacknowledged for this long, the kernel closes the
+// connection. As recommended by Cloudflare
+// (https://blog.cloudflare.com/when-tcp-sockets-refuse-to-die/),
+// the value is computed as: keepidle + keepintvl * keepcnt. This
+// ensures TCP_USER_TIMEOUT and keepalives agree on when to give up.
+// Best-effort: silently ignored if unsupported.
+func setTCPUserTimeout(conn syscall.RawConn, cfg net.KeepAliveConfig) {
+	idle := cfg.Idle
+	if idle == 0 {
+		idle = goDefaultKeepAliveIdle
+	}
+
+	interval := cfg.Interval
+	if interval == 0 {
+		interval = goDefaultKeepAliveInterval
+	}
+
+	count := cfg.Count
+	if count == 0 {
+		count = goDefaultKeepAliveCount
+	}
+
+	timeout := idle + interval*time.Duration(count)
+
+	conn.Control(func(fd uintptr) { //nolint: errcheck
+		unix.SetsockoptInt(int(fd), unix.IPPROTO_TCP, unix.TCP_USER_TIMEOUT, int(timeout.Milliseconds())) //nolint: errcheck
+	})
+}

network/v2/sockopts_usertimeout_stub.go

@@ -0,0 +1,10 @@
+//go:build !linux
+
+package network
+
+import (
+	"net"
+	"syscall"
+)
+
+func setTCPUserTimeout(conn syscall.RawConn, cfg net.KeepAliveConfig) {}

network/v2/socks_proxy_test.go

@@ -66,7 +66,7 @@ func (suite *SocksProxyTestSuite) SetupSuite() {
 	require.NoError(suite.T(), err)
 	suite.authURL = parsed
 
-	suite.baseNetwork = network.New(nil, "mtg", 0, 0, 0)
+	suite.baseNetwork = network.New(nil, "mtg", 0, 0, 0, network.DefaultKeepAliveConfig)
 }
 
 func (suite *SocksProxyTestSuite) TestIncorrectSchema() {