Fix sleepq_add panic happening with too wide net epoch.

[melifaro]

Summary

Fix panic described in kern/250413 introduced by D25445.

[ae]

It seems the only solution here is taking ifnet reference. I'm not sure about PCB, probably it can not disappear here.

sys/netinet/in_mcast.c

2093

IN_MULTI_LOCK() is sx lock, so you still can sleep while in epoch.

[melifaro]

Reflect ae@ comments.

[kib]

head/sys/netinet/in_mcast.c

2085 ↗

(On Diff #78366)

I believe it would be not too hard to move this EXIT to before corresponding breaks inside switch. Current structure where ENTER is done inside switch, and you expect all cases escape with epoch entered, is too weird/fragile.