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

On 4/20/20 5:29 PM, Paul Moore wrote: > On Mon, Apr 20, 2020 at 5:56 PM Richard Guy Briggs&lt;rgb(a)redhat.com&gt; 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@vger.kernel.org > > > Fixes: 756125289285 ("audit: always check the netlink payload length in audit_receive_msg()") > > > Reported-by:syzbot+49e69b4d71a420ceda3e@syzkaller.appspotmail.com > > > Signed-off-by: Paul Moore&lt;paul(a)paul-moore.com&gt; > > > --- > > > 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.

LCB