> | > What are unsafe interrupts in
> | > this context though? Is splsoftnet() good enough?
> | 
> | i think so.
> | 
> | it'd be better to use macros like UIDINFO_LOCK so that it can be
> | changed later.
> 
> Well, unfortunately struct uidinfo is used for # of processes and
> # of locks accounting, where locking with splsoftnet() does not make
> a lot of sense. I will add the spl protection in chgsbsize() though.

i don't think it works, as far as the lock is shared between subsystems.
(consider a network interrupt when the lock is held by vfs code.)

YAMAMOTO Takashi