pf: copy out rather than m_pullup() in pf_test_eth_rule()

[kp]

Summary

Don't change the mbuf chain layout. We've encountered alignment issues
in the tcp syncookie code on armv7, which are triggered by the
m_pullup() here.

Sponsored by: Rubicon Communications, LLC ("Netgate")

[mjg]

sys/netpfil/pf/pf.c

3920–3921

now that m does not change this assignment is redundant

more importantly though for non v4/v6 packets src and dst will be NULL and code below will be reached. that's asking for a remotely-induced crash down the road.

[kp]

sys/netpfil/pf/pf.c

3920–3921

I'll remove the redundant assignment.

src/dst are not even NULL, they're uninitialised. This hasn't actually changed with this patch, but we check af to see if we have an address we can compare to or not. It'd probably be safer to initialise src and dst to NULL, but that still wouldn't prevent a panic if we mess up the af handling.

[kp]

• remove now redundant update of *m0 and e

[mjg]

sys/netpfil/pf/pf.c

3920–3921

this definitely needs to be sorted out in a different patch, this is too error prone for any future changes

[kp]

sys/netpfil/pf/pf.c

3920–3921

See D35573