netdev - Re: [net-next] tcp: Sanitize CMSG flags and reserved args in tcp_zerocopy

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 15 Feb 2021 08:04:11 -0700
From:   David Ahern <dsahern@...il.com>
To:     Dan Carpenter <dan.carpenter@...cle.com>, kbuild@...ts.01.org,
        Arjun Roy <arjunroy.kdev@...il.com>, davem@...emloft.net,
        netdev@...r.kernel.org
Cc:     lkp@...el.com, kbuild-all@...ts.01.org, arjunroy@...gle.com,
        edumazet@...gle.com, soheil@...gle.com,
        David Ahern <dsahern@...il.com>,
        Leon Romanovsky <leon@...nel.org>,
        Jakub Kicinski <kuba@...nel.org>
Subject: Re: [net-next] tcp: Sanitize CMSG flags and reserved args in
 tcp_zerocopy_receive.

On 2/15/21 5:03 AM, Dan Carpenter wrote:
> Hi Arjun,
> 
> url:    https://github.com/0day-ci/linux/commits/Arjun-Roy/tcp-Sanitize-CMSG-flags-and-reserved-args-in-tcp_zerocopy_receive/20210212-052537
> base:   https://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git e4b62cf7559f2ef9a022de235e5a09a8d7ded520
> config: x86_64-randconfig-m001-20210209 (attached as .config)
> compiler: gcc-9 (Debian 9.3.0-15) 9.3.0
> 
> If you fix the issue, kindly add following tag as appropriate
> Reported-by: kernel test robot <lkp@...el.com>
> Reported-by: Dan Carpenter <dan.carpenter@...cle.com>
> 
> smatch warnings:
> net/ipv4/tcp.c:4158 do_tcp_getsockopt() warn: check for integer overflow 'len'
> 
> vim +/len +4158 net/ipv4/tcp.c
> 
> 3fdadf7d27e3fb Dmitry Mishin            2006-03-20  3896  static int do_tcp_getsockopt(struct sock *sk, int level,
> 3fdadf7d27e3fb Dmitry Mishin            2006-03-20  3897  		int optname, char __user *optval, int __user *optlen)
> ^1da177e4c3f41 Linus Torvalds           2005-04-16  3898  {
> 295f7324ff8d9e Arnaldo Carvalho de Melo 2005-08-09  3899  	struct inet_connection_sock *icsk = inet_csk(sk);
> ^1da177e4c3f41 Linus Torvalds           2005-04-16  3900  	struct tcp_sock *tp = tcp_sk(sk);
> 6fa251663069e0 Nikolay Borisov          2016-02-03  3901  	struct net *net = sock_net(sk);
> ^1da177e4c3f41 Linus Torvalds           2005-04-16  3902  	int val, len;
> 
> "len" is int.
> 
> [ snip ]
> 05255b823a6173 Eric Dumazet             2018-04-27  4146  #ifdef CONFIG_MMU
> 05255b823a6173 Eric Dumazet             2018-04-27  4147  	case TCP_ZEROCOPY_RECEIVE: {
> 7eeba1706eba6d Arjun Roy                2021-01-20  4148  		struct scm_timestamping_internal tss;
> e0fecb289ad3fd Arjun Roy                2020-12-10  4149  		struct tcp_zerocopy_receive zc = {};
> 05255b823a6173 Eric Dumazet             2018-04-27  4150  		int err;
> 05255b823a6173 Eric Dumazet             2018-04-27  4151  
> 05255b823a6173 Eric Dumazet             2018-04-27  4152  		if (get_user(len, optlen))
> 05255b823a6173 Eric Dumazet             2018-04-27  4153  			return -EFAULT;
> c8856c05145490 Arjun Roy                2020-02-14  4154  		if (len < offsetofend(struct tcp_zerocopy_receive, length))
> 05255b823a6173 Eric Dumazet             2018-04-27  4155  			return -EINVAL;
> 
> 
> The problem is that negative values of "len" are type promoted to high
> positive values.  So the fix is to write this as:
> 
> 	if (len < 0 || len < offsetofend(struct tcp_zerocopy_receive, length))
> 		return -EINVAL;
> 
> 110912bdf28392 Arjun Roy                2021-02-11  4156  		if (unlikely(len > sizeof(zc))) {
> 110912bdf28392 Arjun Roy                2021-02-11  4157  			err = check_zeroed_user(optval + sizeof(zc),
> 110912bdf28392 Arjun Roy                2021-02-11 @4158  						len - sizeof(zc));
>                                                                                                         ^^^^^^^^^^^^^^^^
> Potentially "len - a negative value".
> 
> 

get_user(len, optlen) is called multiple times in that function. len < 0
was checked after the first one at the top.

Also, maybe I am missing something here, but offsetofend can not return
a negative value, so this checks catches len < 0 as well:

	if (len < offsetofend(struct tcp_zerocopy_receive, length))
		return -EINVAL;