Re: [PATCH] Audit: EINTR instead of kernel private return codes in audit records
On Wed, 2007-11-14 at 16:17 -0500, Steve Grubb wrote:
On Wednesday 14 November 2007 16:07:42 Eric Paris wrote:
> It should be slightly faster <snip>
>
> We would also be picking up ENOIOCTLCMD but that shoudln't be seen on
> this code path, so I guess it doesn't matter.
Right, the note in the top of include/linux/errno.h says these should not be
seen by userspace. So, ENOIOCTLCMD is invalid if we ever saw it. But what
about the NFSv3 errnos (in the same file)? Do we need to worry about those?
So I started looking at these and I like my patch the way I sent it.
Although ENOIOCTLCMD, ENOTSUPP, ETOOSMALL, ESERVERFAULT, EIOCBQUEUED,
and EIOCBRETRY are not supposed to be valid return codes first glance
leads me to believe they could all slip back into userspace. None of
them are later rewritten after we collect the return code in the audit
system. It would be absolutely wrong for the audit system to translate
even ENOIOCTLCMD to anything else if this is what was actually returned
to the process.
I suggest the patch be added exactly the way I sent it, as the speed up
suggested (using >= && <=) could give incorrect audit messages vs what
actually happened.
-Eric