kern/136013: Resolver wrong diagnostics

[Libor Forst]

>Number:         136013
>Category:       kern
>Synopsis:       Resolver wrong diagnostics
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Wed Jun 24 19:30:04 UTC 2009
>Closed-Date:
>Last-Modified:
>Originator:     Libor Forst
>Release:        FreeBSD 7.1-RELEASE-p5 i386
>Organization:
Charles University, Prague
>Environment:
System: FreeBSD betynka.ms.mff.cuni.cz 7.1-RELEASE-p5 FreeBSD 7.1-RELEASE-p5 #0: Fri May 29 11:44:20 CEST 2009 root at kernun-e.ms.mff.cuni.cz:/usr/obj/usr/src/sys/YQ-MINIMAL i386


>Description:
	The res_searchN() function makes a decision about loop continuation
    after returning from the res_querydomainN() call based on two pieces
    of information: res->res_h_errno and hp->rcode. However, if the query
    was made for both A and AAAA RRs, the hp address points just to the
    first res_target structure, but the res->res_h_errno will be set by
    the result of the second query. Thus, res->res_h_errno and hp->rcode
    can be inconsistent.

>How-To-Repeat:
	At least two-domains search-list, query for a name without dots,
    the first (A) response should be NXDomain, the second (AAAA) should
    be ServFail. In this situation, the first res_querydomainN() call
    (for the NAME.DOMAIN1) returns with res->res_h_errno == TRY_AGAIN
    (according to the second query) but hp->rcode == NXDomain (because
    hp reflects the first res_target). As such, the second domain will
    not be tried (in spite of that in all three other combinations of
    results, the domain is tried).

>Fix:
	1. Both pieces of information (h_errno and rcode) must be consistent.
       I.e. the rcode should be saved e.g. into res_state structure (as
       a new res_rcode member?).
    2. The question is, whether the result (h_errno+rcode) should really
       be defined as "last query result". Why the responses NXDomain (A)
       and ServFail (AAAA) bring a different behavior than the case of
       ServFail (A) and NXDomain (AAAA)? Maybe the final result selection
       should not follow "the last" response but "the most relevant" one.

>Release-Note:
>Audit-Trail:
>Unformatted: