Re: [kernel-hardening] Re: [PATCH v9 1/4] syscalls: Verify address limit before returning to user-mode

[Al Viro <viro@ZenIV.linux.org.uk>]

On Wed, May 10, 2017 at 04:21:37AM +0100, Al Viro wrote:
> On Wed, May 10, 2017 at 04:12:54AM +0100, Al Viro wrote:
> 
> > Broken commit: "net: don't play with address limits in kernel_recvmsg".
> > It would be OK if it was only about data.  Unfortunately, that's not
> > true in one case: svc_udp_recvfrom() wants ->msg_control.
> > 
> > Another delicate place: you can't assume that write() always advances
> > file position by its (positive) return value.  btrfs stuff is sensitive
> > to that.
> > 
> > ashmem probably _is_ OK with demanding ->read_iter(), but I'm not sure
> > about blind asma->file->f_pos += ret.  That's begging for races.  Actually,
> > scratch that - it *is* racy.
> 
> kvec_length(): please, don't.  I would rather have the last remaining
> iov_length() gone...   What do you need it for, anyway?  You have only
> two users and both have the count passed to them (as *count and *cnt resp.)

fcntl stuff: I've decided not to put something similar into work.compat
since I couldn't decide what to do with compat stuff - word-by-word copy
from userland converting to struct flock + conversion to posix_lock +
actual work + conversion to flock + word-by-word copy to userland...  Smells
like we might be better off with compat_flock_to_posix_lock() et.al.
I'm still not sure; played a bit one way and another and dediced to drop
it for now.  Hell knows...