DNS outbound: Fix some issues

[patterniha]

1. because we don't have UplinkOnly/DownlinkOnly, after outbound-connection is closed(response returned), request function(and as a result inbound-connection) remain open for a long time(especially when the network is TCP)

///

2. we have a weird code in:
   

   Xray-core/proxy/dns/dns.go

   Line 393 in 4064f8d

   return len(b), nil

   it should be: return 0, err(dns-query is not sent but we are waiting for a response!)

[RPRX]

b, err := reader.ReadMessage() 没有传入 ctx，不还是会在这一行卡着吗？

[patterniha]

b, err := reader.ReadMessage()If ctx is not passed in, won't it still be stuck on this line?

after connIdle-timeout exceed -> ctx is canceled ->task.run ends ->Process ends ->

1. if link.reader is pipe, we have interrupt(reader) after Process in proxyman/outbound/handler.go
2. else if it is inbound-reader(dispatchLink) -> dispatchLink ends -> inbound-Process ends -> inbound-conn closed -> reader closed

so in both case reader is closed and there is no problem.

[patterniha]

ready.

[patterniha]

also, there is no need for go h.handleIPQuery(id, qType, domain, writer) or go h.rejectNonIPQuery(id, qType, domain, writer), there is no need to run them in new goroutine, we can simply do h.handleIPQuery(id, qType, domain, writer) or h.rejectNonIPQuery(id, qType, domain, writer), i will fix change it .

[Fangliding]

好好的异步查询响应本来就可以乱序回又突发奇想了是吧。。。 你改点什么就是fix 然后说我fix了好几十个bug

[patterniha]

i said there is no need to run in new goroutine, write instantly returned.
so it is better not to make a new goroutine, this is not a bug, just optimization, i didn't say this a bug.

it is better to call it "change" instead of "fix"

[patterniha]

also, there is no need for go h.handleIPQuery(id, qType, domain, writer) or go h.rejectNonIPQuery(id, qType, domain, writer), there is no need to run them in new goroutine, we can simply do h.handleIPQuery(id, qType, domain, writer) or h.rejectNonIPQuery(id, qType, domain, writer), i will fix change it .

done.

[Fangliding]

i said there is no need to run in new goroutine, write instantly returned. so it is better not to make a new goroutine, this is not a bug, just optimization, i didn't say this a bug.

it is better to call it "change" instead of "fix"

你optimize了个啥 一个goroutine而已 dns本来就可以乱序处理是对的 就是些无意义改动

[patterniha]

the main bug in this PR was return len(b), nil, and after outbound-connection is closed(or dial failed), inbound-connection remained open for 300s .

also, there is no need for new goroutines, we should simplify the code.

///
@Fangliding

also, outbound-dns has userLevel option:

Xray-core/infra/conf/dns_proxy.go

Line 14 in 19f8907

UserLevel uint32 `json:"userLevel"`

but this is forgotten in CN-doc, please add it, thx.

[patterniha]

@RPRX

@Fangliding is right, because handleIPQuery use built-in-DNS and it may take a while to answer, it is better to run it in new goroutine.
but for rejectNonIPQuery there is no need for new goroutine.

so only handleIPQuery should run in goroutine, wait...

[patterniha]

so only handleIPQuery should run in goroutine, wait...

done and ready.

At first I didn't understand what @Fangliding meant