Re: [PATCH] Initial support for struct vfs_cred [0/1]

[Trond Myklebust <trond.myklebust@fys.uio.no>]

>>>>> " " == Luca Barbieri <ldb@ldb.ods.org> writes:

     > But you'll need to modify the declaration of the various
     > function pointers whose implementations might need credentials
     > and modify all functions that call them and deal with
     > permissions.  Instead with my proposal the credentials are

Yes... And this is a useful activity in itself, as the existence of
all these hacks that temporarily change uid/gid/whatever... show.

     > automatically immutable across the syscall without needing to
     > worry at all about locks, counts and sharing.

I still have no opinion about your proposal for implementing CLONE_CRED.

What I fail to see is why you appear to insist it would be
incompatible with the idea of copy-on-write VFS credentials (which I
explained are interesting for other purposes).

I also fail to understand why you insist that we need to drop the idea
of copy-on-write credentials in order to optimize for this fringe case
in which somebody calls sys_access() or exec with euid != fsuid.

Now repeat after me

  "changing fsuid/fsgid is *not* the common case that needs optimization."

  Trond