]> git.baikalelectronics.ru Git - kernel.git/commit
igmp: acquire pmc lock for ip_mc_clear_src()
authorWANG Cong <xiyou.wangcong@gmail.com>
Mon, 12 Jun 2017 16:52:26 +0000 (09:52 -0700)
committerDavid S. Miller <davem@davemloft.net>
Tue, 13 Jun 2017 16:51:37 +0000 (12:51 -0400)
commit72f6aa4cafe49d850f1861b9b3335b1f8f5f4ada
treeb1e63ae6f5f6cb466f5a064f700b803d90656d6e
parentad798202b5c37935fc2a2cc9cc8519f370c7d54f
igmp: acquire pmc lock for ip_mc_clear_src()

Andrey reported a use-after-free in add_grec():

        for (psf = *psf_list; psf; psf = psf_next) {
...
                psf_next = psf->sf_next;

where the struct ip_sf_list's were already freed by:

 kfree+0xe8/0x2b0 mm/slub.c:3882
 ip_mc_clear_src+0x69/0x1c0 net/ipv4/igmp.c:2078
 ip_mc_dec_group+0x19a/0x470 net/ipv4/igmp.c:1618
 ip_mc_drop_socket+0x145/0x230 net/ipv4/igmp.c:2609
 inet_release+0x4e/0x1c0 net/ipv4/af_inet.c:411
 sock_release+0x8d/0x1e0 net/socket.c:597
 sock_close+0x16/0x20 net/socket.c:1072

This happens because we don't hold pmc->lock in ip_mc_clear_src()
and a parallel mr_ifc_timer timer could jump in and access them.

The RCU lock is there but it is merely for pmc itself, this
spinlock could actually ensure we don't access them in parallel.

Thanks to Eric and Long for discussion on this bug.

Reported-by: Andrey Konovalov <andreyknvl@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Xin Long <lucien.xin@gmail.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Reviewed-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/igmp.c