Difference between revisions of "Release Meeting Minutes/2011-06-21"
From K5Wiki
(New page: {{minutes|2011}} Will Fiveash, Thomas Hardjono, Greg Hudson, Carlos Garay, Nathaniel McCallum, Simo Sorce, Zhanna Tsitkova, Tom Yu [ PAD stuff. Josh Howlett's messages to krb-wg... ] ;Si...) |
|||
Line 3: | Line 3: | ||
[ PAD stuff. Josh Howlett's messages to krb-wg... ] |
[ PAD stuff. Josh Howlett's messages to krb-wg... ] |
||
+ | |||
+ | == DNS pain == |
||
;Simo: Problems in glibc related to <code>getaddrinfo()</code>. Tried to log into a host via ssh but kept requesting wrong principal. Tried turning off <code>rdns</code> (in libdefaults) etc. Finally ran gdb to discover that <code>getaddrinfo()</code> does PTR record lookup (when <code>AF_INET</code> and <code>AI_CANONNAME</code>). |
;Simo: Problems in glibc related to <code>getaddrinfo()</code>. Tried to log into a host via ssh but kept requesting wrong principal. Tried turning off <code>rdns</code> (in libdefaults) etc. Finally ran gdb to discover that <code>getaddrinfo()</code> does PTR record lookup (when <code>AF_INET</code> and <code>AI_CANONNAME</code>). |
||
Line 30: | Line 32: | ||
;Simo: OK as first step. Maybe don't set <code>AI_CANONNAME</code> at all? |
;Simo: OK as first step. Maybe don't set <code>AI_CANONNAME</code> at all? |
||
− | ;Greg: Maybe. In any case don't overload <code>rdns</code>, because the <code>getaddrinfo()</code> is intended to do a forward resolution. |
+ | ;Greg: Maybe. In any case don't overload <code>rdns</code>, because the call to <code>getaddrinfo()</code> is intended to do a forward resolution. |
;Tom: Maybe another knob for forward DNS. |
;Tom: Maybe another knob for forward DNS. |
||
Line 36: | Line 38: | ||
[ Tom will set up some test case DNS records in kerberos.org domain. ] |
[ Tom will set up some test case DNS records in kerberos.org domain. ] |
||
− | ... |
||
+ | ;Tom: Also some issues with <code>gethostname()</code>, e.g. whether it appends the domain name. |
||
+ | |||
+ | ;Greg: Most Linux distributions give just the short name. |
||
+ | |||
+ | ;Simo: Depends on configuration. |
||
+ | |||
+ | ;Greg: See [[Projects/Acceptor_Names]]. |
||
+ | |||
+ | == libverto == |
||
+ | |||
+ | ;Nathaniel: libverto. Code mostly up. https://fedorahosted.org/libverto/ |
||
+ | |||
+ | ;Tom: Too bad about idle support. [ libdispatch, etc. does some nice things with that sort of task queue thing ] |
||
+ | |||
+ | ;Nathaniel: Back end for libevent ~120 lines. Are blocks (C language extension) needed in libdispatch? |
||
+ | |||
+ | ;Tom: Only needed for interface convenience. Without blocks, you don't get the entire API (you miss the blocks-dependent interfaces), but you still get the complete functionality. |
||
+ | |||
+ | ;Nathaniel: verto.c:verto_free() ...handles some signal races during unload. |
||
+ | |||
+ | ;Nathaniel: Where to test? (Doing Linux already.) |
||
+ | |||
+ | ;Greg: Solaris. |
||
+ | |||
+ | ;Nathaniel: Nothing too GNU-specific: <code>dladdr()</code> and <code>asprintf()</code>. |
||
+ | |||
+ | == Misc. == |
||
+ | |||
+ | ;Will: Glenn was working on code... what's the rule on testing for null before calling <code>free()</code>? |
||
+ | |||
+ | ;Greg: We recently made a massive move toward not testing. [ Was needed a long time ago for platforms that had a conforming C compiler but libc was nonconforming in a way that made <code>free(NULL)</code> unsafe. ] |
||
+ | |||
+ | ;Tom: Planning patch releases for early/mid-July. |
Latest revision as of 14:48, 21 June 2011
Will Fiveash, Thomas Hardjono, Greg Hudson, Carlos Garay, Nathaniel McCallum, Simo Sorce, Zhanna Tsitkova, Tom Yu
[ PAD stuff. Josh Howlett's messages to krb-wg... ]
DNS pain
- Simo
- Problems in glibc related to
getaddrinfo()
. Tried to log into a host via ssh but kept requesting wrong principal. Tried turning offrdns
(in libdefaults) etc. Finally ran gdb to discover thatgetaddrinfo()
does PTR record lookup (whenAF_INET
andAI_CANONNAME
).
- Tom
- Does it do likewise with
AF_INET6
?
- Simo
- unknown
- Greg
- No idea why we use
AF_INET
in sn2princ?
- Tom
- Bug workaround? [ unknown ]
getaddrinfo()
seems very unportable after 10+ years...
- Greg
- No real other choice for IPv6. No obvious notes from Ken Raeburn on this situation.
- Tom
- Drop
getaddrinfo()
completely?
- Greg
- We want to have domain name appended in case the user doesn't type the FQDN. Alias resolution.
- Tom
- Another knob to turn off forward resolution?
- Greg
- Reading resolv.conf to emulate domain search list seems problematic. Long-term, KDC should set a flag "trust me for aliases".
- Simo
- Shorter term?
- Greg
- I'll ask Ken why we use
AF_INET
in sn2princ.
- Simo
- OK as first step. Maybe don't set
AI_CANONNAME
at all?
- Greg
- Maybe. In any case don't overload
rdns
, because the call togetaddrinfo()
is intended to do a forward resolution.
- Tom
- Maybe another knob for forward DNS.
[ Tom will set up some test case DNS records in kerberos.org domain. ]
- Tom
- Also some issues with
gethostname()
, e.g. whether it appends the domain name.
- Greg
- Most Linux distributions give just the short name.
- Simo
- Depends on configuration.
- Greg
- See Projects/Acceptor_Names.
libverto
- Nathaniel
- libverto. Code mostly up. https://fedorahosted.org/libverto/
- Tom
- Too bad about idle support. [ libdispatch, etc. does some nice things with that sort of task queue thing ]
- Nathaniel
- Back end for libevent ~120 lines. Are blocks (C language extension) needed in libdispatch?
- Tom
- Only needed for interface convenience. Without blocks, you don't get the entire API (you miss the blocks-dependent interfaces), but you still get the complete functionality.
- Nathaniel
- verto.c:verto_free() ...handles some signal races during unload.
- Nathaniel
- Where to test? (Doing Linux already.)
- Greg
- Solaris.
- Nathaniel
- Nothing too GNU-specific:
dladdr()
andasprintf()
.
Misc.
- Will
- Glenn was working on code... what's the rule on testing for null before calling
free()
?
- Greg
- We recently made a massive move toward not testing. [ Was needed a long time ago for platforms that had a conforming C compiler but libc was nonconforming in a way that made
free(NULL)
unsafe. ]
- Tom
- Planning patch releases for early/mid-July.