Gordon Ross wrote:
> James Carlson wrote:
>
>> Neither taking an interface (or route) down nor adding an ipfilter entry
>> are really equivalent to FIN. FIN starts an orderly shutdown, which
>> gives you a zero-length read. Timeouts (due to missing routes or
>> dropped packets) end up causing a reset connection -- as though you
>> received RST -- and can result in SIGPIPE, depending on how things play out.
>
> I don't really care what error the recv loop gets. The typical
> real-life cases are: (1) server shutdown, we get a FIN, and
> (2) server crash, and we get an RST when it comes back up.
>
> It would be good enough for this test if smbiod gets just about
> any error from the recv call, i.e. ECONNRESET, EHOSTUNREACH
All the more reason to have multiple testing scenarios.
>> If you just don't care, you could use the undocumented
>> TCP_IOC_ABORT_CONN ioctl to abort the connection.
>
> Would I need to send this ioctl down the particular socket that
> I want to force closed? Or how would I identify the victim?
It's not documented, but a quick google search finds:
http://typo.submonkey.net/pages/tcpdrop-solaris
It's PSARC/2001/292, which is currently unpublished, but shouldn't be.
>> Such a proxy might allow you to do other useful things, such as writing
>> a journal for debug purposes or simulating hard-to-create test
>> configurations.
>
> A good suggestion, but I'm hoping for an easier solution.
Don't forget netcat, which is easier.
But I don't think of the proxy as particularly hard. It probably
doesn't need to be more than 1000 lines of code in C; perhaps less if
you pick something higher level.
> Someone suggested ipf(1M), which seems like an interesting idea.
> Maybe that would let me cause a RST for connections to some IP?
> (I'm not too familiar with the capabilities of ipf.)
Yes. See ipf(4) -- which is inexplicably left out of OpenSolaris, but
isn't too hard to find on the web. You want the "return-rst" option.
>> Isn't smbiod already integrated? Shouldn't there already be a test
>> suite with error simulations ... ?
>
> Yes, but this was allowed in with an open test CR: 6856802
> Need tests for CIFS client auto-reconnect
Isn't there a test for abnormal client shutdown? There should be, and
I'd think that test (whatever it is) would be a good starting place for
a reconnect test.
--
James Carlson 42.703N 71.076W <[email protected]>
_______________________________________________
networking-discuss mailing list
[email protected]
