Hi Paul,

> Op 01-06-2026 18:27 CEST schreef Paul Moore <[email protected]>:
> 
>  
> On Sun, May 31, 2026 at 10:56 PM NeilBrown <[email protected]> wrote:
> >
> > I don't think this is sufficient.
> > I think (and we should confirm we people who know about auditing) that
> > the audit call should come before the permission check so that there is
> > an audit record on failed attempts.  In that case the
> > audit_inode_child() all needs to be much earlier.
> 
> As you likely already know, audit_inode_child() exists simply to
> record information about the file/inode access, the audit record is
> generated later (and can is dependent on other things/config).  For
> that reason, yes, it's generally a good idea to call
> audit_inode_child() as soon as we have the file/inode information.
> Auditors want to know what the user/system attempted to do, even if
> the operation failed; in some cases the log of the failed operations
> are more useful than the successful ones.
> 

But currently for mkdir(2), mknod(2), open(2) with O_CREAT, we are calling
audit_inode_child() twice. Once before actual creation (mostly in
may_create_dentry()), and once afterwards in fsnotify_*(). Is that the
desired behavior?

> 
> -- 
> paul-moore.com

Best,
Jori.

Reply via email to