]> git.baikalelectronics.ru Git - kernel.git/commitdiff
packet: fix send path when running with proto == 0
authorDaniel Borkmann <dborkman@redhat.com>
Fri, 6 Dec 2013 10:36:15 +0000 (11:36 +0100)
committerDavid S. Miller <davem@davemloft.net>
Tue, 10 Dec 2013 01:09:20 +0000 (20:09 -0500)
Commit 998e7f9a9095 introduced a cached dev pointer, that gets
hooked into register_prot_hook(), __unregister_prot_hook() to
update the device used for the send path.

We need to fix this up, as otherwise this will not work with
sockets created with protocol = 0, plus with sll_protocol = 0
passed via sockaddr_ll when doing the bind.

So instead, assign the pointer directly. The compiler can inline
these helper functions automagically.

While at it, also assume the cached dev fast-path as likely(),
and document this variant of socket creation as it seems it is
not widely used (seems not even the author of TX_RING was aware
of that in his reference example [1]). Tested with reproducer
from 998e7f9a9095.

 [1] http://wiki.ipxwarzone.com/index.php5?title=Linux_packet_mmap#Example

Fixes: 998e7f9a9095 ("packet: fix use after free race in send path when dev is released")
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Tested-by: Salam Noureddine <noureddine@aristanetworks.com>
Tested-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Documentation/networking/packet_mmap.txt
net/packet/af_packet.c

index c01223628a87ae5e522b714c8a3a86edda15dea5..8e48e3b142275ff1500df754301d39c7f1639d71 100644 (file)
@@ -123,6 +123,16 @@ Transmission process is similar to capture as shown below.
 [shutdown]  close() --------> destruction of the transmission socket and
                               deallocation of all associated resources.
 
+Socket creation and destruction is also straight forward, and is done
+the same way as in capturing described in the previous paragraph:
+
+ int fd = socket(PF_PACKET, mode, 0);
+
+The protocol can optionally be 0 in case we only want to transmit
+via this socket, which avoids an expensive call to packet_rcv().
+In this case, you also need to bind(2) the TX_RING with sll_protocol = 0
+set. Otherwise, htons(ETH_P_ALL) or any other protocol, for example.
+
 Binding the socket to your network interface is mandatory (with zero copy) to
 know the header size of frames used in the circular buffer.
 
index ba2548bd85bf7d42b26e25ae9a2626b729fc8be9..88cfbc189558f75b4b508b849827eea18c3be7c0 100644 (file)
@@ -237,6 +237,30 @@ struct packet_skb_cb {
 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 
+static struct net_device *packet_cached_dev_get(struct packet_sock *po)
+{
+       struct net_device *dev;
+
+       rcu_read_lock();
+       dev = rcu_dereference(po->cached_dev);
+       if (likely(dev))
+               dev_hold(dev);
+       rcu_read_unlock();
+
+       return dev;
+}
+
+static void packet_cached_dev_assign(struct packet_sock *po,
+                                    struct net_device *dev)
+{
+       rcu_assign_pointer(po->cached_dev, dev);
+}
+
+static void packet_cached_dev_reset(struct packet_sock *po)
+{
+       RCU_INIT_POINTER(po->cached_dev, NULL);
+}
+
 /* register_prot_hook must be invoked with the po->bind_lock held,
  * or from a context in which asynchronous accesses to the packet
  * socket is not possible (packet_create()).
@@ -246,12 +270,10 @@ static void register_prot_hook(struct sock *sk)
        struct packet_sock *po = pkt_sk(sk);
 
        if (!po->running) {
-               if (po->fanout) {
+               if (po->fanout)
                        __fanout_link(sk, po);
-               } else {
+               else
                        dev_add_pack(&po->prot_hook);
-                       rcu_assign_pointer(po->cached_dev, po->prot_hook.dev);
-               }
 
                sock_hold(sk);
                po->running = 1;
@@ -270,12 +292,11 @@ static void __unregister_prot_hook(struct sock *sk, bool sync)
        struct packet_sock *po = pkt_sk(sk);
 
        po->running = 0;
-       if (po->fanout) {
+
+       if (po->fanout)
                __fanout_unlink(sk, po);
-       } else {
+       else
                __dev_remove_pack(&po->prot_hook);
-               RCU_INIT_POINTER(po->cached_dev, NULL);
-       }
 
        __sock_put(sk);
 
@@ -2059,19 +2080,6 @@ static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
        return tp_len;
 }
 
-static struct net_device *packet_cached_dev_get(struct packet_sock *po)
-{
-       struct net_device *dev;
-
-       rcu_read_lock();
-       dev = rcu_dereference(po->cached_dev);
-       if (dev)
-               dev_hold(dev);
-       rcu_read_unlock();
-
-       return dev;
-}
-
 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
 {
        struct sk_buff *skb;
@@ -2088,7 +2096,7 @@ static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
 
        mutex_lock(&po->pg_vec_lock);
 
-       if (saddr == NULL) {
+       if (likely(saddr == NULL)) {
                dev     = packet_cached_dev_get(po);
                proto   = po->num;
                addr    = NULL;
@@ -2242,7 +2250,7 @@ static int packet_snd(struct socket *sock,
         *      Get and verify the address.
         */
 
-       if (saddr == NULL) {
+       if (likely(saddr == NULL)) {
                dev     = packet_cached_dev_get(po);
                proto   = po->num;
                addr    = NULL;
@@ -2451,6 +2459,8 @@ static int packet_release(struct socket *sock)
 
        spin_lock(&po->bind_lock);
        unregister_prot_hook(sk, false);
+       packet_cached_dev_reset(po);
+
        if (po->prot_hook.dev) {
                dev_put(po->prot_hook.dev);
                po->prot_hook.dev = NULL;
@@ -2506,14 +2516,17 @@ static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protoc
 
        spin_lock(&po->bind_lock);
        unregister_prot_hook(sk, true);
+
        po->num = protocol;
        po->prot_hook.type = protocol;
        if (po->prot_hook.dev)
                dev_put(po->prot_hook.dev);
-       po->prot_hook.dev = dev;
 
+       po->prot_hook.dev = dev;
        po->ifindex = dev ? dev->ifindex : 0;
 
+       packet_cached_dev_assign(po, dev);
+
        if (protocol == 0)
                goto out_unlock;
 
@@ -2626,7 +2639,8 @@ static int packet_create(struct net *net, struct socket *sock, int protocol,
        po = pkt_sk(sk);
        sk->sk_family = PF_PACKET;
        po->num = proto;
-       RCU_INIT_POINTER(po->cached_dev, NULL);
+
+       packet_cached_dev_reset(po);
 
        sk->sk_destruct = packet_sock_destruct;
        sk_refcnt_debug_inc(sk);
@@ -3337,6 +3351,7 @@ static int packet_notifier(struct notifier_block *this,
                                                sk->sk_error_report(sk);
                                }
                                if (msg == NETDEV_UNREGISTER) {
+                                       packet_cached_dev_reset(po);
                                        po->ifindex = -1;
                                        if (po->prot_hook.dev)
                                                dev_put(po->prot_hook.dev);