in_pcb: delay crfree() down into UMA dtor

[glebius]

Summary

inpcb lookups, which check inp_cred, work with pcbs that potentially went
through in_pcbfree(). So inp_cred should stay valid until SMR guarantees
its invisibility to lookups.

Submitted by: markj

[markj]

Seems ok, just a couple of minor optional notes.

We should really use SMR_POINTER and the other smr_types.h machinery to wrap inp_cred. This would require some churn though, since inp_cred is referenced directly in many places.

sys/netinet/in_pcb.c

1863

Perhaps set inp->inp_cred = NULL at least if INVARIANTS is defined.

1879

IMHO it is more logical to group UMA zone methods together, i.e., I would group this with inpcb_fini().

[glebius]

sys/netinet/in_pcb.c

1863

I can do it.

1879

Do you agree if I move inpcb_fini() down to here, rather than move inpcb_dtor() up? I'd like to put the whole destruction sequence sequential and readable from top to bottom?

[markj]

sys/netinet/in_pcb.c

1879

Fine with me.