igmp: convert struct ip_sf_list to RCU

Commit 23d2b94043ca ("igmp: Add ip_mc_list lock in ip_check_mc_rcu")
added spin_lock_bh(&im->lock) to ip_check_mc_rcu() to prevent a
use-after-free while iterating im->sources during concurrent deletions.

However, ip_check_mc_rcu() is called from RCU read-side critical
sections in packet receive and route lookup fast paths (e.g.
__mkroute_output(), ip_route_input_rcu(), and __udp4_lib_rcv()).

When igmpv3_send_cr() or igmpv3_send_report() holds &pmc->lock and
calls add_grec() -> igmpv3_newpack() -> ip_route_output_ports(),
an XFRM policy matching a multicast destination triggers
xfrm_tmpl_resolve_one() -> xfrm4_get_saddr() -> __mkroute_output() ->
ip_check_mc_rcu(). This attempts to acquire &im->lock while &pmc->lock
is already held on the same CPU, triggering a lockdep recursive locking
warning / deadlock.

Fix this by converting IPv4 struct ip_sf_list to RCU, mirroring the
IPv6 implementation in net/ipv6/mcast.c:

1. Add struct rcu_head to struct ip_sf_list and annotate sf_next,
   sources, and tomb as __rcu pointers.
2. Use rcu_assign_pointer() and kfree_rcu() for list updates and
   deletions.
3. Remove spin_lock_bh(&im->lock) from ip_check_mc_rcu() and traverse
   im->sources locklessly with for_each_psf_rcu(), reading and writing
   counter fields with READ_ONCE() and WRITE_ONCE().

Note: RCU conversion of /proc/net/mcfilter will be done in a
separate patch.

Fixes: 23d2b94043ca ("igmp: Add ip_mc_list lock in ip_check_mc_rcu")
Reported-by: syzbot+3d99fb01bcd740f2fc1e@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=3d99fb01bcd740f2fc1e
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Ido Schimmel <idosch@nvidia.com>
Link: https://patch.msgid.link/20260827160656.903003-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

1 files changed, 4 insertions, 3 deletions

diff --git a/include/linux/igmp.h b/include/linux/igmp.h
index 3a2d35a9f307..a0cf0398519f 100644
--- a/include/linux/igmp.h
+++ b/include/linux/igmp.h
@@ -57,20 +57,21 @@ struct ip_mc_socklist {
 };
 
 struct ip_sf_list {
-	struct ip_sf_list	*sf_next;
+	struct ip_sf_list __rcu	*sf_next;
 	unsigned long		sf_count[2];	/* include/exclude counts */
 	__be32			sf_inaddr;
 	unsigned char		sf_gsresp;	/* include in g & s response? */
 	unsigned char		sf_oldin;	/* change state */
 	unsigned char		sf_crcount;	/* retrans. left to send */
+	struct rcu_head		rcu;
 };
 
 struct ip_mc_list {
 	struct in_device	*interface;
 	__be32			multiaddr;
 	unsigned int		sfmode;
-	struct ip_sf_list	*sources;
-	struct ip_sf_list	*tomb;
+	struct ip_sf_list __rcu	*sources;
+	struct ip_sf_list __rcu	*tomb;
 	unsigned long		sfcount[2];
 	union {
 		struct ip_mc_list *next;