[Patch net] 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 <redacted>
Cc: Eric Dumazet <redacted>
Cc: Xin Long <lucien.xin@gmail.com>
Signed-off-by: Cong Wang <redacted>
---
 net/ipv4/igmp.c | 21 +++++++++++++--------
 1 file changed, 13 insertions(+), 8 deletions(-)

diff --git a/net/ipv4/igmp.c b/net/ipv4/igmp.c
index 44fd86d..8f6b5bb 100644
--- a/net/ipv4/igmp.c
+++ b/net/ipv4/igmp.c

@@ -2071,21 +2071,26 @@ static int ip_mc_add_src(struct in_device *in_dev, __be32 *pmca, int sfmode,
 
 static void ip_mc_clear_src(struct ip_mc_list *pmc)
 {
-	struct ip_sf_list *psf, *nextpsf;
+	struct ip_sf_list *psf, *nextpsf, *tomb, *sources;
 
-	for (psf = pmc->tomb; psf; psf = nextpsf) {
+	spin_lock_bh(&pmc->lock);
+	tomb = pmc->tomb;
+	pmc->tomb = NULL;
+	sources = pmc->sources;
+	pmc->sources = NULL;
+	pmc->sfmode = MCAST_EXCLUDE;
+	pmc->sfcount[MCAST_INCLUDE] = 0;
+	pmc->sfcount[MCAST_EXCLUDE] = 1;
+	spin_unlock_bh(&pmc->lock);
+
+	for (psf = tomb; psf; psf = nextpsf) {
 		nextpsf = psf->sf_next;
 		kfree(psf);
 	}
-	pmc->tomb = NULL;
-	for (psf = pmc->sources; psf; psf = nextpsf) {
+	for (psf = sources; psf; psf = nextpsf) {
 		nextpsf = psf->sf_next;
 		kfree(psf);
 	}
-	pmc->sources = NULL;
-	pmc->sfmode = MCAST_EXCLUDE;
-	pmc->sfcount[MCAST_INCLUDE] = 0;
-	pmc->sfcount[MCAST_EXCLUDE] = 1;
 }
 
 /* Join a multicast group

-- 
2.5.5

On Tue, Jun 13, 2017 at 12:52 AM, Cong Wang [off-list ref] wrote:

Hi, Cong.

how about in ip_check_mc_rcu():
        for (psf = im->sources; psf; psf = psf->sf_next) {
               if (psf->sf_inaddr == src_addr)
                           break;
         }

I didn't see spinlock for it, is it safe to access them in parallel ?
or these two places would never be in parallel ?

I've already checked elsewhere, all other places where it accesses
or traverses im->sources are protected by this spinlock.

-       pmc->sources = NULL;
-       pmc->sfmode = MCAST_EXCLUDE;
-       pmc->sfcount[MCAST_INCLUDE] = 0;
-       pmc->sfcount[MCAST_EXCLUDE] = 1;
 }

 /* Join a multicast group
--
2.5.5

On Mon, Jun 12, 2017 at 11:30 AM, Xin Long [off-list ref] wrote:

Hi, Cong.

how about in ip_check_mc_rcu():
        for (psf = im->sources; psf; psf = psf->sf_next) {
               if (psf->sf_inaddr == src_addr)
                           break;
         }

I didn't see spinlock for it, is it safe to access them in parallel ?
or these two places would never be in parallel ?

That is a different bug which needs more work, therefore
I defer it to net-next. And I already explained to you why
it needs more work than just a call_rcu().

On Tue, Jun 13, 2017 at 2:35 AM, Cong Wang [off-list ref] wrote:

On Mon, Jun 12, 2017 at 11:30 AM, Xin Long [off-list ref] wrote:

quoted

Hi, Cong.

how about in ip_check_mc_rcu():
        for (psf = im->sources; psf; psf = psf->sf_next) {
               if (psf->sf_inaddr == src_addr)
                           break;
         }

I didn't see spinlock for it, is it safe to access them in parallel ?
or these two places would never be in parallel ?

That is a different bug which needs more work, therefore
I defer it to net-next. And I already explained to you why
it needs more work than just a call_rcu().

Okay, thanks Cong.

Reviewed-by: Xin Long <lucien.xin@gmail.com>

From: Cong Wang <redacted>
Date: Mon, 12 Jun 2017 09:52:26 -0700

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 <redacted>
Cc: Eric Dumazet <redacted>
Cc: Xin Long <lucien.xin@gmail.com>
Signed-off-by: Cong Wang <redacted>

Applied and queued up for -stable, thanks.