if_ipsec(4): more carefully validate invalid configuration input

[kib]

Summary

if_ipsec(4): protect against user supplying unknown address family

if_ipsec(4): handle situations where there are no policy or SADB entry for if

Both issues were found by doing approximately the following sequence of commands on IPv4-only kernel:

ifconfig ipsec0 create reqid 0
ifconfig ipsec0 inet tunnel 1.1.1.2 1.1.1.1
ifconfig ipsec0 up
ifconfig ipsec0 reqid 0

Sponsored by: NVIDIA Networking

[• hselasky]

Looks good to me.

[ae]

sys/net/if_ipsec.c

670

I'm not sure why we should use EHOSTUNREACH here?

[kib]

sys/net/if_ipsec.c

670

I selected some error, somewhat inspired by ipsecN_allocsa(). What error code should I use there?

[melifaro]

sys/net/if_ipsec.c

816

Would it be possible to use a bit more descriptive name?
As it's a list inside a hash, maybe something like ifl or iflist would be an alternative?

[• hselasky]

sys/net/if_ipsec.c

816

Would it be possible to use a bit more descriptive name?
As it's a list inside a hash, maybe something like ifl or iflist would be an alternative?

ph
phead

--HPS

[kib]

sys/net/if_ipsec.c

816

I changed the name of the local var to iflist. Will upload the new patch after the error value for SIOCGIFPSRCADDR_IN6 is decided.

[ae]

sys/net/if_ipsec.c

670

If you mean ipsec[46]_allocsa() from netipsec/ipsec_output.c, then it is appropriate to get such error when you are sending packet. But I think it can be confusing when you get "no route to host" trying to get interface configuration.

Now I'm not sure EADDRNOTAVAIL is also good here :)
Maybe ENETDOWN or even ENXIO will be better?

[kib]

sys/net/if_ipsec.c

670

ENETDOWN is IMO in the same category as EHOSTUNREACH.
Let it be ENXIO then.

[kib]

Change error code to ENXIO.
Rename local var.