Are there any conditions under which a fault in copy_xx_user
_won't_ be trapped and returned as a error (incomplete copy)?
Under 2.1.x none should ever be missed.
I'm looking over an oops report that went
sys_newfstat->do_page_fault->die_if_kernel->do_exit etc.,
presumably by way of copy_xxx_user.
Sounds like a wild kernel pointer, not a missed copy_*_user()
exception.
If there are some faults that aren't being trapped, this would have
major ramifications for cleanup after errors ..
Is this under 2.0.x or 2.1.x, you didn't specify?