if_wg: avoid sleeping under the net epoch

[kevans]

Summary

No sleeping allowed here, so avoid it. Collect the subset of data we
want inside of the epoch, as we'll need extra allocations when we add
items to the nvlist.

[markj]

sys/dev/if_wg/module/module.c

74

Why NOISE_KEY_SIZE instead of WG_KEY_SIZE?

404–407

Perhaps assert that we're in the net epoch here?

434–436

How do you guarantee that the length of the list doesn't increase between the first loop and this one?

[kevans]

sys/dev/if_wg/module/module.c

74

Stole from the r_public member of struct noise_remote that we're copying into here; the distinction doesn't seem to make much sense to me, as we do use them interchangeably elsewhere (see: noise_local_keys() definition then usage over in wg_cloneattach()).

I'll switch it over to WG_KEY_SIZE, though.

434–436

Whoops, brain-o; will fix to stop at exp->aip_count and update it after.

[kevans]

Address feedback from @markj

[markj]

sys/dev/if_wg/module/module.c

499

Is it intentional that an error part-way though will result in this function returning success with a truncated list?

533

nvl is leaked in this error path, not a new bug though.

[markj]

I think this is ok, my last questions are around preexisting behaviour.