TLS ECH client: Revert echForceQuery's behavior

[patterniha]

#4949 (comment)

Yes, I just neglected this.

In case of a doh fails, the connection will be made for 6 hours and try to update the key. If it is unsuccessful within these 6 hours to get the new-key(even empty), then the connection will not be made.

[Fangliding]

这不还是dns服务器不可用的时候强制失败

[patterniha]

This is not a forced failure when the DNS server is unavailable

??? if DNS if fails, we have error other than emptyResponse, so the connection will not made. what is the problem?

[RPRX]

被你们整晕了，总之查询失败就是失败，不要管是不是 emptyResponse，然后 force-query 与否有不同的逻辑

@patterniha 你看下当前代码的逻辑是否符合 #4949 (comment) ，不符合的话我删掉最新的 commit，你重新 PR

[Fangliding]

被你们整晕了，总之查询失败就是失败，不要管是不是 emptyResponse，然后 force-query 与否有不同的逻辑

@patterniha 你看下当前代码的逻辑是否符合 #4949 (comment) ，不符合的话我删掉最新的 commit，你重新 PR

现在这样是没 force query 的话 错误是emptyresponse 连接成功 不然强制失败 以及只缓存 emptyresponse

[RPRX]

@Fangliding 那我把这个 commit 删了，你 PR 修一下吧

[patterniha]

@RPRX

let explain:

A: ech DNS query fails, get no response
B: ech DNS query is successful, and receiving response is successful, but no ech-record found.
C: ech DNS query is successful, and ech-record received.

S1: if C, connection is made with ech-TLS.
S2: if A, connection is not made.
S3: if B, if echForceQuery is true, connection is not made, otherwise connection is made with plain-SNI, also we cache emptyRecord for 5 minutes.

@RPRX

this is my logic, because ech may be used in MitM-uses, what is your logic?

please say your logic and I implemented that.

[Fangliding]

我觉得mitm继续用分片就是了 没必要硬用ech

[patterniha]

@Fangliding @RPRX

can you please say your logic exactly?

I think we can reach an agreement with three mode echForceQuery.

but I still doesn't know your exact logic.

[RPRX]

@patterniha 我们的意思是把 AB 视为同一种情况，然后走 S3 的逻辑

[patterniha]

but suppose our internet is dropped for just a seconds, so we get no response and we cache this fail for 5 minutes !!!

and we don't try to use ech-TLS, or getting new key for 5 minutes !!!

this is not help for my MitM-use and if fragment blocked, this cause connection fail for 5 minutes.

///

@RPRX

so I implement three mode echForceQuery:

full: -> equal to current true
half: -> equal to current false
none: -> equal to your logic for false.

is it OK?

[RPRX]

echForceQuery 这样吧：

• "none": 非强制，懒查询，默认值（非必须查到，任何失败都是五分钟后再尝试；对于 emptyResponse，也是等五分钟）
• "half": 非强制，积极查询（非必须查到，但每次尝试，不用等五分钟；对于 emptyResponse，等五分钟）
• "full": 强制，故积极查询（必须查到，否则不连接；对于 emptyResponse，下次重新查）

对于 "none" 和 "half"，第一次等查询，后面不等查询（但这次查到了会带 TTL 缓存，下次就能用），避免影响体验

[patterniha]

OK, please wait...

[patterniha]

ok, i only subdivide that for showing errors.

[RPRX]

还有对于 "none" 和 "half"，第一次可以等查询，后面就别等了（不知道现在是不是这个逻辑）

其实更简单的逻辑是如果 echForceQuery 填 false 就每次都尝试查询，但除了第一次，可以先建立连接