RxRPC: 3 issues found in my example code

[Tim Smith]

On Monday 28 April 2014 13:47:42 Leon Liu wrote:
> Hi Tim,
> 
> Thank you so much for your very quick response and patch!
> 
> I've tried the patch. It did fixed the issue #1. I know it maybe a temporary
> fixing, whatever, I observed two new issues:
> 
> 1. The fixing increases probability of hitting issue #2 (rxrpc stack return
> "local error 103, rx busy") dramatically. Probability changes from 1/100000
> to 1/5000. You have to turn rxrpc log off to reproduce the issue.

I can generate plenty of "BUSY" packets out of the server by turning the 
client logging off and leaving the server logging on, and making lots and lots 
of calls. Actually about 1 in 10 new calls gets me a "BUSY". These are 
happening because the accept queue is full, but I don't get the same symptom 
at the client side that you do.

Your logs go "Rx BUSY" followed by "protocol error" which suggests 
out_clientflag is true on the connection. Which is really odd.

I just get the local connection abort I expect.

> Occasionally, this can be fixed by re-insmod af-rxrpc.ko at sender.
> 2. After hitting issue #2 a lot of times, receiver kernel hangs. This was
> reproduced 3 times.

On the other hand, if I turn off all the logging and blast it with calls, I 
can reliably hang the *server* after a while. Methinks that bears 
investigation... a lockup that manifests one way can manifest another.

-- 
Tim Smith <tim at electronghost.co.uk>
Oh, marvelous. You're going to kill me. What a finely-tuned response to the 
situation.
    -- The Doctor
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.infradead.org/pipermail/linux-afs/attachments/20140504/5ebdb1ab/attachment.sig>