In message <x7r7wp7a5r.fsf@capsicum.wsrcc.com>, "Wolfgang S. Rupprecht" writes:
>
>smb@research.att.com (Steve Bellovin) writes:
>> As I've mentioned, I've had some hangs with Galeon.  Briefly, the 
>> problem occurs when there are many concurrent downloads happening, such 
>> as when I do "open whole folder in tabs"; I can bypass the problem by 
>> using a Web proxy.  Details are in pkg/24503
>
>I think I'm zero-ing in on it.  When mozilla-gtk hangs I can see that
>a resolver "A" request just went out but no reply came back in.  In
>fact, I can see 3 request go out under the same UDP port.  Normally
>when pages are clicked more slowly and the images all come from the
>same site, the DNS requests all go out of an ever-decrementing port
>number.  While, in theory, DNS has its own transaction ID (which is
>unique even when galeon wedges), the fact that things wedge when a
>non-unique port is used is somewhat suspicious.  I wonder if named()
>is getting confused when it sees the same
>src-host/dest-host/src-port/dest-port quad used for different
>requests.
>
>This jives with your observation that it doesn't happen when you use a
>proxy (because you'd only ever try to resolve a single IP for one
>hostname - that of the proxy.)
>
>(This was tested on a "client" netbsd-current machine.  Testing it on
>my server had so many DNS requests that it was hard to tell which
>program was asking for what.  Running it on a quiet client let me pick
>through the noise a bit better.)

Interesting -- I had suspected something thread-related, because the 
it's not just the downloads that hang; I can't kill tabs or windows, or 
get to any menus.

Hmm -- is the resolver it's using thread-safe?


		--Steve Bellovin, http://www.research.att.com/~smb