Freedom: Cache UDP resolve result

[Fangliding]

#4800 (comment)
道理上udp就不该有fqdn当目标(((((

[Fangliding]

早看到了 已经改了
不过我回头问了一下 这和你遇到的东西没关系 这只是一个小小的refine

[patterniha]

@Fangliding

line 380 and 381 should be swapped

[patterniha]

@Fangliding

line 389 and 390 should be swapped

[Fangliding]

这两行换不换有什么区别么

[patterniha]

yes ip.domain() is panic

[patterniha]

道理上udp就不该有fqdn当目标(((((

@Fangliding

or we have multiple-IP or we have only-one-domain on single-UDP.

if we have multiple-domain it is impossible to connect response to request.

for example if we have two domain example-1.com and example-2.com, and both resolved to "2.2.2.2" so if we receive response from "2.2.2.2" this is for example-1.com or example-2.com ???

this is logically impossible, and has nothing to do with code.

///

So we never have multiple-domain on single-UDP and there is no need for resolvedUDPAddr-map, only single variable is enough(of course, I haven't deleted it from my PR yet.)

///

[Meo597]

TestCommanderListHandlers 极其容易炸

rebase 到 v25.6.8 fbae89d017aee5947598d0a82945b316fdc688ab 就好了

[patterniha]

@Fangliding

I separated my PR from your PR.‍

this PR fix WriteMultiBuffer and my PR fix ReadMultiBuffer, and these two PR will solve the UDP problem in Freedom when address is domain or fakedns.

also, you missed a few things in your code:

1. even if hasStrategy is false, we should load and store from cache.

2. if ForceIP is true, and built-in-DNS fail to resolve-IP, we should not fallback to system-DNS (net.ResolveUDPAddr)

////////////////////

@RPRX

these two PR should merge as soon as possible, and new version should be released.

I think you are aware of the sensitive situation in Iran. we are at war and the internet is very limited.

[RPRX]

话说现在伊朗基本断外网了还能 Serverless-for-Iran 吗

[patterniha]

Now that Iran is basically disconnected from the Internet, is Serverless-for-Iran still available?

only youtube and porn sites (of course, I have to change the DNS, I'm waiting for the freedom-UDP problem to be resolved)

///

but the reverse(bridge-portal) method is working great on home-internet with static-IP.

[patterniha]

The internet is back to normal.
but ipv6 and UDP are still not available.

[patterniha]

also, you missed a few things in your code:

1. even if hasStrategy is false, we should load and store from cache.
2. if ForceIP is true, and built-in-DNS fail to resolve-IP, we should not fallback to system-DNS (net.ResolveUDPAddr)

/////////////////

@Fangliding

Fix 1 and 2, thx.

[Fangliding]

It's done

[Fangliding]

算了还是改成之前那样吧 我只是不想把continue写两次
这个use真是烦死了。。

[patterniha]

Fix 0 len outbounds

How can the length of outbounds be zero?

And if it is zero, ‍‍NewPacketWriter will not be executed at all:

we have:

Xray-core/proxy/freedom/freedom.go

Lines 105 to 107 in 1976d02

	func (h *Handler) Process(ctx context.Context, link *transport.Link, dialer internet.Dialer) error {
	outbounds := session.OutboundsFromContext(ctx)
	ob := outbounds[len(outbounds)-1]

before that and it will panic.

[Fangliding]

Fix 0 len outbounds

How can the length of outbounds be zero?

And if it is zero, ‍‍NewPacketWriter will not be executed at all:

we have:

Xray-core/proxy/freedom/freedom.go

Lines 105 to 107 in 1976d02

	func (h *Handler) Process(ctx context.Context, link *transport.Link, dialer internet.Dialer) error {
	outbounds := session.OutboundsFromContext(ctx)
	ob := outbounds[len(outbounds)-1]

before that and it will panic.

我这么写可能清楚点 那个ctx可能不是上面看到的ctx 你可以自己用先前的版本试试 就是我在测你昨天那个pr的时候遇到的

[patterniha]

i tested and see len(outbounds) is always 0 and your initial-set-map does not perform at all.

NewPacketWriter/NewPacketReader ctx is newCtx define in line-174 and has no outbounds.

so we should copy ctx-outbounds to newCtx-outbounds.

///

I fix it in #4816.

[Fangliding]

就这样就行了 不要破坏timeoutonly的语义

[patterniha]

Forget what I said, I hadn't seen your last force-pushed

[Jolymmiles]

@RPRX does we really need create map for each connection? its a lot of ram(

[Jolymmiles]

[Avaritia-ZOV]

I agree. High consumption is observed when using tls with flow-udp443.

[Jolymmiles]

@Fangliding @yuhan6665

[patterniha]

@Fangliding implement this because we should not send packets to multiple IPs, and we should cache first-sent-IP, but:

1. In non-cone, address is fixed and In cone, we don't have multiple-domains, only initial-request can be domain(like stun)
2. even we want to support multi-domains, it is not enough to handle it only in write, we should design reverse-lookup and handle it in read too.

as a result, there is no need for map, we will always have at most one domain, and that is initial-request-domain, so simple cache is enough and there is no need for map cache.

also, "override.go" converts only the first fakeIP to domain and we don't have multiple-domain at all.

[patterniha]

@Fangliding is right, the ram problem is not related to this, but anyway i think there is no need for map cache.