Re: [PATCH ver.2] gtp: fix use-after-free and null-ptr-deref in gtp_genl_dump_pdp()

[Pablo Neira Ayuso]

On Wed, Feb 14, 2024 at 08:06:12PM +0300, kovalev@xxxxxxxxxxxx wrote:
> 14.02.2024 19:49, Pablo Neira Ayuso wrote:
> > On Wed, Feb 14, 2024 at 07:27:33PM +0300, kovalev@xxxxxxxxxxxx wrote:
> > > From: Vasiliy Kovalev <kovalev@xxxxxxxxxxxx>
> > > 
> > > The gtp_net_ops pernet operations structure for the subsystem must be
> > > registered before registering the generic netlink family.
> > Thanks for finding a remedy for this.
> > 
> > If your fix is correct, (I didn't test your patch yet) then maybe this
> > needs to be fixed in a few more spots in the tree?
> > 
> > net/devlink/core.c-static int __init devlink_init(void)
> > net/devlink/core.c-{
> > net/devlink/core.c-     int err;
> > net/devlink/core.c-
> > net/devlink/core.c-     err = genl_register_family(&devlink_nl_family);
> > net/devlink/core.c-     if (err)
> > net/devlink/core.c-             goto out;
> > net/devlink/core.c:     err = register_pernet_subsys(&devlink_pernet_ops);
> > net/devlink/core.c-     if (err)
> > 
> > net/handshake/netlink.c-        ret = genl_register_family(&handshake_nl_family);
> > net/handshake/netlink.c-        if (ret) {
> > net/handshake/netlink.c-                pr_warn("handshake: netlink registration failed (%d)\n", ret);
> > net/handshake/netlink.c-                handshake_req_hash_destroy();
> > net/handshake/netlink.c-                return ret;
> > net/handshake/netlink.c-        }
> > net/handshake/netlink.c-
> > net/handshake/netlink.c-        /*
> > net/handshake/netlink.c-         * ORDER: register_pernet_subsys must be done last.
> > net/handshake/netlink.c-         *
> > net/handshake/netlink.c-         *      If initialization does not make it past pernet_subsys
> > net/handshake/netlink.c-         *      registration, then handshake_net_id will remain 0. That
> > net/handshake/netlink.c-         *      shunts the handshake consumer API to return ENOTSUPP
> > net/handshake/netlink.c-         *      to prevent it from dereferencing something that hasn't
> > net/handshake/netlink.c-         *      been allocated.
> > net/handshake/netlink.c-         */
> > net/handshake/netlink.c:        ret = register_pernet_subsys(&handshake_genl_net_ops);
> > 
> > net/ipv4/tcp_metrics.c: ret = register_pernet_subsys(&tcp_net_metrics_ops);
> > net/ipv4/tcp_metrics.c- if (ret < 0)
> > net/ipv4/tcp_metrics.c-         panic("Could not register tcp_net_metrics_ops\n");
> > net/ipv4/tcp_metrics.c-
> > net/ipv4/tcp_metrics.c- ret = genl_register_family(&tcp_metrics_nl_family);
> > net/ipv4/tcp_metrics.c- if (ret < 0)
> > net/ipv4/tcp_metrics.c-         panic("Could not register tcp_metrics generic netlink\n");
> > net/ipv4/tcp_metrics.c-}
> > 
> > net/ipv6/ioam6.c-int __init ioam6_init(void)
> > net/ipv6/ioam6.c-{
> > net/ipv6/ioam6.c:       int err = register_pernet_subsys(&ioam6_net_ops);
> > net/ipv6/ioam6.c-       if (err)
> > net/ipv6/ioam6.c-               goto out;
> > net/ipv6/ioam6.c-
> > net/ipv6/ioam6.c-       err = genl_register_family(&ioam6_genl_family);
> > net/ipv6/ioam6.c-       if (err)
> > net/ipv6/ioam6.c-               goto out_unregister_pernet_subsys;
> > 
> > net/ipv6/seg6.c-        err = genl_register_family(&seg6_genl_family);
> > net/ipv6/seg6.c-        if (err)
> > net/ipv6/seg6.c-                goto out;
> > net/ipv6/seg6.c-
> > net/ipv6/seg6.c:        err = register_pernet_subsys(&ip6_segments_ops);
> > net/ipv6/seg6.c-        if (err)
> > net/ipv6/seg6.c-                goto out_unregister_genl;
> > 
> > net/netlink/genetlink.c-        err = genl_register_family(&genl_ctrl);
> > net/netlink/genetlink.c-        if (err < 0)
> > net/netlink/genetlink.c-                goto problem;
> > net/netlink/genetlink.c-
> > net/netlink/genetlink.c:        err = register_pernet_subsys(&genl_pernet_ops);
> > net/netlink/genetlink.c-        if (err)
> > net/netlink/genetlink.c-                goto problem;
> 
> Most likely, judging by the backtrace, the bug is the same [1]:
> 
> Call Trace:
>  <TASK>
>  genl_dumpit+0x119/0x220 net/netlink/genetlink.c:1025
>  netlink_dump+0x588/0xca0 net/netlink/af_netlink.c:2264
>  __netlink_dump_start+0x6d0/0x9c0 net/netlink/af_netlink.c:2370
>  genl_family_rcv_msg_dumpit+0x1e1/0x2d0 net/netlink/genetlink.c:1074
>  genl_family_rcv_msg net/netlink/genetlink.c:1190 [inline]
>  genl_rcv_msg+0x470/0x800 net/netlink/genetlink.c:1208
>  netlink_rcv_skb+0x16b/0x440 net/netlink/af_netlink.c:2543
>  genl_rcv+0x28/0x40 net/netlink/genetlink.c:1217
>  netlink_unicast_kernel net/netlink/af_netlink.c:1341 [inline]
>  netlink_unicast+0x53b/0x810 net/netlink/af_netlink.c:1367
>  netlink_sendmsg+0x8b7/0xd70 net/netlink/af_netlink.c:1908
>  sock_sendmsg_nosec net/socket.c:730 [inline]
>  __sock_sendmsg+0xd5/0x180 net/socket.c:745
>  ____sys_sendmsg+0x6ac/0x940 net/socket.c:2584
>  ___sys_sendmsg+0x135/0x1d0 net/socket.c:2638
>  __sys_sendmsg+0x117/0x1e0 net/socket.c:2667
>  do_syscall_x64 arch/x86/entry/common.c:52 [inline]
>  do_syscall_64+0xd3/0x250 arch/x86/entry/common.c:83
>  entry_SYSCALL_64_after_hwframe+0x63/0x6b
> RIP: 0033:0x7f35d567cda9

Ok, then a series of fixes probably needs to happen so each maintain
can review and apply them.

Maybe some of these subsystems above can only be compiled built-in, so
that cannot trigger.

In any case, are you up to pick on that series?

Thanks.