DragonFly submit List (threaded) for 2008-02
DragonFly BSD
DragonFly submit List (threaded) for 2008-02
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

arprequest() serialization

From: "Nuno Antunes" <nuno.antunes@xxxxxxxxx>
Date: Sun, 10 Feb 2008 02:03:30 +0000

Hi all,

Please review/comment the following.

arprequest() calls ifp->if_output() without locally grabbing the
respective serializer, so ASSERT_SERIALIZED at the beginning of the
Grab the serializer at arp_rtrequest() when it calls arprequest().

diff --git a/sys/netinet/if_ether.c b/sys/netinet/if_ether.c
index b85cb03..dcc11b8 100644
--- a/sys/netinet/if_ether.c
+++ b/sys/netinet/if_ether.c
@@ -223,11 +223,14 @@ arp_rtrequest(int req, struct rtentry *rt,
struct rt_addrinfo *info)
                /* Announce a new entry if requested. */
-               if (rt->rt_flags & RTF_ANNOUNCE)
+               if (rt->rt_flags & RTF_ANNOUNCE) {
+                       lwkt_serialize_enter(rt->rt_ifp->if_serializer);
+                       lwkt_serialize_exit(rt->rt_ifp->if_serializer);
+               }
        case RTM_RESOLVE:
                if (gate->sa_family != AF_LINK ||
@@ -324,6 +327,8 @@ arprequest(struct ifnet *ifp, struct in_addr *sip,
struct in_addr *tip,
        struct sockaddr sa;
        u_short ar_hrd;

+       ASSERT_SERIALIZED(ifp->if_serializer);
        if ((m = m_gethdr(MB_DONTWAIT, MT_DATA)) == NULL)
        m->m_pkthdr.rcvif = (struct ifnet *)NULL;

I thought about grabbing the serializer locally at arprequest()
instead, but this function is also called by ether_output() which
already holds the serializer. That would cause a recursive
serialization. Would it be better to do this and drop the serializer
in ether_output() around the call to arprequest() instead of the
change proposed in the patch above? There would be other places where
the serializer would have to be droped too (almost every other call to


[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]