Opened 11 years ago

Closed 9 years ago

#303 closed Bug / Defect (fixed)

Reconnect fails with DNS error, when network is switched

Reported by: mihai Owned by: plaisthos
Priority: major Milestone: release 2.3.7
Component: Generic / unclassified Version: OpenVPN 2.3.1 (Community Ed)
Severity: Not set (select this one, unless your'e a OpenVPN developer) Keywords: DNS
Cc: David Sommerseth


See detailed description in bug ID 966281 here

Change History (4)

comment:1 Changed 11 years ago by Samuli Seppänen

Keywords: reconnect failure removed
Owner: set to plaisthos
Status: newassigned

This commit touches the DNS resolution stuff and is in 2.3.x but not 2.2.x. It might be the culprit, but someone needs to verify that.

comment:2 Changed 11 years ago by Samuli Seppänen

Note from cron2 in IRC:

"After the dual-stack patches [from plaisthos] are merged, this might or might not be resolved already."

Those patches still need final review.

comment:3 Changed 10 years ago by Gert Döring

Cc: David Sommerseth added

Reading through the redhat bug, I don't think the dual-stack patches will magically fix this, though I find it highly likely that it might be related to the introduction of the ipv6-transport patch in 2.3.x - changing the DNS resolving in OpenVPN, possibly even from gethostbyname() to getaddrinfo().

From the description it sounds as if getaddrinfo() would cache the content of /etc/resolv.conf, not noticing that it should use newer information and keeps querying the now stale DNS servers it has. But I'd expect it to still stat() the file regularily and re-read if needed... but this would need closer examination, possibly strace'ing a 2.2.x and 2.3.x client when changing resolv.conf behind it's back.

David, does that ring any bell with you?

comment:4 Changed 9 years ago by Gert Döring

Milestone: release 2.3.7
Resolution: fixed
Status: assignedclosed

I *think* that this could be a copy of #523 - getaddrinfo() will not magically re-read /etc/hosts, but you need to call res_init() for that, which our configure script didn't properly handle on (recent) glibc systems.

So I just claim that the fix for #523 will also fix this one, refer to it, and close this ticket.

Patch is there:

Please reopen if you think this is a different issue.

Note: See TracTickets for help on using tickets.