TS-4706 Truncated SNI name during escalation

[gtenev]

A fix for a problem with SSL hostname verification failing due to truncated SNI name.

[zwoop]

👍

[zwoop]

I'm generally ok with this for the immediate fixage. My only concern here is that there's now an invariant (it seems) between the client and server HttpHdr, where the caches should be invalidated together for both. That sort of feels like it then could be lifted up in the stack a bit maybe, or at least assertion that the invariant is never broken again.

Alternatively, if there's improvements that can be done here (later) such that the invalidation can be disjoint again, safely, for better performance etc., that'd be cool too. Maybe file a separate lira for this cleanup for later?

[zwoop]

[approve ci]

[gtenev]

@zwoop thanks for reviewing!

As far as can tell the escalate plugin was implemented later then the HttpHdr caching and the caching implementation does not support its use-case well. The reason we started noticing the truncated/garbage name problems is that SSL handshake changed (got stricter)

This fix is meant to solve the immediate problem of having t_state.hdr_info.server_request cache not being invalidated after the escalate plugin called TSHttpTxnRedirectUrlSet() to retry the request to a secondary origin after the primary origin failed.

This code change would invalidate (only invalidate) client request and server request HttpHdr at the same time only during HttpSM::redirect_request(), the caching state of the 2 objects would not necessarily be kept (or assumed to be) in sync (client request and server request HttpHdr were not meant to be invariant).

Filed Jira: TS-4712 to look into the HttpHdr caching use-cases and verify the HttpHdr caching functionality.

[atsci]

FreeBSD build successful! See https://ci.trafficserver.apache.org/job/Github-FreeBSD/501/ for details.

[atsci]

Linux build successful! See https://ci.trafficserver.apache.org/job/Github-Linux/398/ for details.