[PATCH] audit: check the length of userspace generated audit records

[Steve Grubb]

On Monday, April 20, 2020 8:18:06 PM EDT Richard Guy Briggs wrote:
> On 2020-04-20 17:56, Lenny Bruzenak wrote:
> > On 4/20/20 5:29 PM, Paul Moore wrote:
> > > On Mon, Apr 20, 2020 at 5:56 PM Richard Guy Briggs<rgb at redhat.com>  
wrote:
> > > > On 2020-04-20 17:36, Paul Moore wrote:
> > > > > Commit 756125289285 ("audit: always check the netlink payload
> > > > > length
> > > > > in audit_receive_msg()") fixed a number of missing message length
> > > > > checks, but forgot to check the length of userspace generated audit
> > > > > records.  The good news is that you need CAP_AUDIT_WRITE to submit
> > > > > userspace audit records, which is generally only given to trusted
> > > > > processes, so the impact should be limited.
> > > > > 
> > > > > Cc:stable at vger.kernel.org
> > > > > Fixes: 756125289285 ("audit: always check the netlink payload
> > > > > length in audit_receive_msg()")
> > > > > Reported-by:syzbot+49e69b4d71a420ceda3e at syzkaller.appspotmail.com
> > > > > Signed-off-by: Paul Moore<paul at paul-moore.com>
> > > > > ---
> > > > > 
> > > > >   kernel/audit.c |    3 +++
> > > > >   1 file changed, 3 insertions(+)
> > > > > 
> > > > > diff --git a/kernel/audit.c b/kernel/audit.c
> > > > > index b69c8b460341..87f31bf1f0a0 100644
> > > > > --- a/kernel/audit.c
> > > > > +++ b/kernel/audit.c
> > > > > @@ -1326,6 +1326,9 @@ static int audit_receive_msg(struct sk_buff
> > > > > *skb, struct nlmsghdr *nlh)> > > > 
> > > > >        case AUDIT_FIRST_USER_MSG2 ... AUDIT_LAST_USER_MSG2:
> > > > >                if (!audit_enabled && msg_type != AUDIT_USER_AVC)
> > > > >                
> > > > >                        return 0;
> > > > > 
> > > > > +             /* exit early if there isn't at least one character
> > > > > to print */ +             if (data_len < 2)
> > > > > +                     return -EINVAL;
> > > > 
> > > > Don't we want to issue the record even if the message is empty?  If a
> > > > len of 1 is passed in, it will properly set str[0] = '\0' and str
> > > > points
> > > > to a location with a null that prints nothing between the single
> > > > quotes
> > > > of "msg=''".  So, I think that should be "if (data_len < 1)".
> > > > 
> > > > Am I missing something?
> > > 
> > > I've got no problem with allowing an empty message so long as there is
> > > a valid use for such a message.  Can anyone think of a valid reason
> > > for having an empty userspace generated record?
> > 
> > Not really. Using the libaudit interface(s), even if an empty message
> > string is sent, and handled in the lib call(s), I believe it will have
> > minimum contextual info, e.g. "exe=... hostname=... ", etc.
> > 
> > I can't think of a valid reason myself, assuming IIUC.
> 
> But even with an empty message, there is still pid, uid, auid, ses, subj
> added by the kernel with its own message type.  I could see a valid type
> of user message created that has no need for any of those fields added
> by audit_log_user_message(), calling audit_send_user_message() directly
> (but mind you, it appears to be deprecated to call it directly).

There is no valid reason. Even allowing just 1 character is also invalid and 
a waste of disk space. But I guess you need to draw the line somewhere.

-Steve