SplitHTTP h3 panics on v2rayNG

[mmmray]

#3554 (comment)

[ll11l1lIllIl1lll]

经测试在 xray 配置设置 h3 SplitHTTP 后，在 Chromium 126.0.6478.126 通过 http/1.1 或 h2 或通过 --origin-to-force-quic-on= 强制要求的 h3 都无法复现这个问题。

[dyhkwong]

When using proxy chain (not supported under current structure), conn is cnc.connection, not *internet.PacketConnWrapper.

In AndroidLibV2rayLite, where effectiveSystemDialer is altered by internet.UseAlternativeSystemDialer, conn is *net.UDPConn.

So type assertion in

Xray-core/transport/internet/splithttp/dialer.go

Line 108 in b277bac

return quic.DialEarly(ctx, conn.(*internet.PacketConnWrapper).Conn.(*net.UDPConn), udpAddr, tlsCfg, cfg)

will just fail and you need hack like

Xray-core/transport/internet/quic/dialer.go

Lines 151 to 161 in b277bac

	var udpConn *net.UDPConn
	switch conn := rawConn.(type) {
	case *net.UDPConn:
	udpConn = conn
	case *internet.PacketConnWrapper:
	udpConn = conn.Conn.(*net.UDPConn)
	default:
	// TODO: Support sockopt for QUIC
	rawConn.Close()
	return nil, errors.New("QUIC with sockopt is unsupported").AtWarning()
	}

or maybe wrap net.Conn to fit net.PacketConn.

And, net.ListenUDP need to be replaced by internet.ListenSystemPacket in https://github.com/XTLS/Xray-core/pull/3554/files#diff-c0d111c26ebe8b3a6a004695c434204a67726f8b6af44d01b601c60573e45601R283 so that internet.RegisterListenerController can take effect.

[ll11l1lIllIl1lll]

我想我可能知道指的是什么了，虽然我 PR 的代码就是一坨混乱的 (我不应侮辱任何事物)，但是感觉不是从 Xray-core 触发的。我可能要去 quic-go 交 PR （虽然我应该没有这个能力去修，试试）

panic: connection already exists

goroutine 18747 [running]:
github.com/quic-go/quic-go.(*connMultiplexer).AddConn(0xc00046c060, {0x7fca9f8c98c0?, 0xc0007087e0?})
        github.com/quic-go/[email protected]/multiplexer.go:59 +0x165
github.com/quic-go/quic-go.(*Transport).init.func1()
        github.com/quic-go/[email protected]/transport.go:266 +0x3f0
sync.(*Once).doSlow(0xc0004ec498?, 0xe?)
        sync/once.go:74 +0xc2
sync.(*Once).Do(...)
        sync/once.go:65
github.com/quic-go/quic-go.(*Transport).init(0xc0002c4d00, 0x98?)
        github.com/quic-go/[email protected]/transport.go:225 +0x4d
github.com/quic-go/quic-go.(*Transport).dial(0xc0002c4d00, {0x1558628, 0xc000959450}, {0x154eb90, 0xc000968240}, {0x0, 0x0}, 0xc000d5e9c0, 0xc0012c2000, 0x1)
        github.com/quic-go/[email protected]/transport.go:212 +0xa5
github.com/quic-go/quic-go.(*Transport).DialEarly(...)
        github.com/quic-go/[email protected]/transport.go:204
github.com/quic-go/quic-go.DialEarly({0x1558628, 0xc000959450}, {0x155edf0?, 0xc0007087e0?}, {0x154eb90, 0xc000968240}, 0xc000d5e9c0, 0xc0012c2000)
        github.com/quic-go/[email protected]/client.go:95 +0x12a
github.com/xtls/xray-core/transport/internet/splithttp.getHTTPClient.func2({0x1558628, 0xc000959450}, {0xc001851b60?, 0xc001cdb920?}, 0xc000d5e9c0, 0xc0012c2000)
        github.com/xtls/xray-core/transport/internet/splithttp/dialer.go:129 +0x19d
github.com/quic-go/quic-go/http3.(*RoundTripper).dial(0xc000e5f260, {0x1558628, 0xc000959450}, {0xc0005b4930, 0x30})
        github.com/quic-go/[email protected]/http3/roundtrip.go:312 +0x27a
github.com/quic-go/quic-go/http3.(*RoundTripper).getClient.func1()
        github.com/quic-go/[email protected]/http3/roundtrip.go:249 +0x77
created by github.com/quic-go/quic-go/http3.(*RoundTripper).getClient in goroutine 18746
        github.com/quic-go/[email protected]/http3/roundtrip.go:246 +0x289

[dyhkwong]

这是另一个问题了跟 #3554 (comment) 不是同一个。这个早就有人交过了 quic-go 不收（说是因为 API 迁移故意 panic 的，要至少保留一年）。 但是你获取了底层的 net.UDPConn（像 https://github.com/XTLS/Xray-core/blob/b277bacdf6bc64f251626ef59b8cea7de5fe9dd2/transport/internet/quic/dialer.go#L151-L161 那样写）的话应该是不会报这个 panic 的。

[mmmray]

Thank you @dyhkwong @ll11l1lIllIl1lll for fixing this quickly!

Are there any recommendations for how to prevent this situation in the future? Can write a testcase that calls UseAlternativeSystemDialer and runs the splithttp transport tests, but I'm not sure if it's meaningful since it's separate from the v2rayNG one.

[dyhkwong]

Are there any recommendations for how to prevent this situation in the future? Can write a testcase that calls UseAlternativeSystemDialer and runs the splithttp transport tests, but I'm not sure if it's meaningful since it's separate from the v2rayNG one.

Because effectiveSystemDialer is alterable, internet.DialSystem may return anything that wrapped as a net.Conn. v2rayNG is only a special case. For dialerProxy and other cases You can wrap net.Conn to fit net.PacketConn.

Note that you need to generate some random LocalAddr() to bypass the annoying panic check in quic-go/quic-go@8189e75#diff-4c6aaadced390f3ce9bec0a9c9bb5203d5fa85df79023e3e0eec423dc9baa946R48-R62. quic-go says this panic check needs to be kept for a year and refuses to remove it (see quic-go#4079).

[RPRX]

@dyhkwong 我也发现了 #3560 (comment) ，才发现你们说的就是这个，蚌