On Tue, 11 Sep 2007, Stephen Borrill wrote:
> Here's a repeatable (happens every hour or so) panic on NetBSD 3.1_STABLE. 
> This has been dictated on the phone by a customer, so might not be exact. 
> This machine has 2x 512MB DIMMs, we've run memtest86 w/o probs and tried 
> various combinations of RAM sticks and slots. This machine and software is 
> pretty much the same as other machines at other sites which are not having 
> this problem.
>
> uvm fault: kernel page fault trap
> stopped in pid 20.1 pagedaemon at NetBSD:pool_reclaim+0x92:mov1
> db> bt
> pool_reclaim at NetBSD:pool_reclaim+0x92
> pool_drain at NetBSD:pool_drain+0x45
> uvm_pageout at NetBSD:uvm_pageout+0x11e
>
> Any idea where to start looking?

Heh (replying to my own mail). Upon further investigation, the customer in 
question has a few PCs checking the OpenOffice.org update site 1000 times 
a second (and being denied by squid access rules otherwise it would have 
saturated their 2Mb 'net connection). Out of 79345566 requests in the 
squid logs, 79216874 were for this page with about 1.1MB per second of 
network traffic just being squid access denied pages.

So a fairly severe unintentional DoS attack and so sorry for bothering you 
all. However, I'd be interested in opinions on whether NetBSD should have 
been able to stand up to this.

-- 
Stephen