Re: kernel/2211: PF / Kernel problem, it filtered all ports

[Daniel Hartmeier <daniel@benzedrine.cx>]

I repeated the procedure a couple of times with variations, and finally
got it to panic:

panic: malloc out of space in kmem_map
_panic
_malloc
_bpf_allocbufs
_bpf_setif
_bpfioctl
_specioctl
_VOP_ioctl
_vn_ioctl
_sys_ioctl
_syscall

This is on the 64MB i386. I have to set all state timeouts to high
values (pfctl -t tcp.first=9999 etc.) and then run nmap until it fails
(around 62000 states). Then stop the scan and run just 'nmap 10.1.1.1' a
couple of times. Sometimes it just fails with various messages
(variations of 'No route to host'). After a couple of tries, I get the
above panic.

How does the trace look? Is is probably a bug in pf, like not freeing
mbufs? Or the pool handling within pf? Or is pf just the part that
exhausts the resources, and other pool usage would cause the same thing?

Daniel