Linux 3.2.102
[pandora-kernel.git] / net / packet / af_packet.c
index 82a6f34..34cc788 100644 (file)
@@ -195,6 +195,7 @@ struct tpacket_kbdq_core {
        char            *pkblk_start;
        char            *pkblk_end;
        int             kblk_size;
+       unsigned int    max_frame_len;
        unsigned int    knum_blocks;
        uint64_t        knxt_seq_num;
        char            *prev;
@@ -295,6 +296,7 @@ struct packet_sock {
        unsigned int            tp_reserve;
        unsigned int            tp_loss:1;
        unsigned int            tp_tstamp;
+       struct net_device __rcu *cached_dev;
        struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 };
 
@@ -350,11 +352,15 @@ static void __fanout_link(struct sock *sk, struct packet_sock *po);
 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;
        }
@@ -372,10 +378,13 @@ 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);
 
        if (sync) {
@@ -497,9 +506,9 @@ static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 
        pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
 
-       spin_lock(&rb_queue->lock);
+       spin_lock_bh(&rb_queue->lock);
        pkc->delete_blk_timer = 1;
-       spin_unlock(&rb_queue->lock);
+       spin_unlock_bh(&rb_queue->lock);
 
        prb_del_retire_blk_timer(pkc);
 }
@@ -608,6 +617,7 @@ static void init_prb_bdqc(struct packet_sock *po,
        p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
        p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 
+       p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
        prb_init_ft_ops(p1, req_u);
        prb_setup_retire_blk_timer(po, tx_ring);
        prb_open_block(p1, pbd);
@@ -813,37 +823,27 @@ static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 
        smp_rmb();
 
-       if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
-
-               /* We could have just memset this but we will lose the
-                * flexibility of making the priv area sticky
-                */
-               BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
-               BLOCK_NUM_PKTS(pbd1) = 0;
-               BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
-               getnstimeofday(&ts);
-               h1->ts_first_pkt.ts_sec = ts.tv_sec;
-               h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
-               pkc1->pkblk_start = (char *)pbd1;
-               pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
-               BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
-               BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
-               BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
-               pbd1->version = pkc1->version;
-               pkc1->prev = pkc1->nxt_offset;
-               pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
-               prb_thaw_queue(pkc1);
-               _prb_refresh_rx_retire_blk_timer(pkc1);
-
-               smp_wmb();
-
-               return;
-       }
+       /* We could have just memset this but we will lose the
+        * flexibility of making the priv area sticky
+        */
+       BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
+       BLOCK_NUM_PKTS(pbd1) = 0;
+       BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
+       getnstimeofday(&ts);
+       h1->ts_first_pkt.ts_sec = ts.tv_sec;
+       h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
+       pkc1->pkblk_start = (char *)pbd1;
+       pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
+                                   BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
+       BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
+       BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
+       pbd1->version = pkc1->version;
+       pkc1->prev = pkc1->nxt_offset;
+       pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
+       prb_thaw_queue(pkc1);
+       _prb_refresh_rx_retire_blk_timer(pkc1);
 
-       WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
-               pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
-       dump_stack();
-       BUG();
+       smp_wmb();
 }
 
 /*
@@ -934,10 +934,6 @@ static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
                prb_close_block(pkc, pbd, po, status);
                return;
        }
-
-       WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
-       dump_stack();
-       BUG();
 }
 
 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
@@ -1176,16 +1172,6 @@ static void packet_sock_destruct(struct sock *sk)
        sk_refcnt_debug_dec(sk);
 }
 
-static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
-{
-       int x = atomic_read(&f->rr_cur) + 1;
-
-       if (x >= num)
-               x = 0;
-
-       return x;
-}
-
 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
 {
        u32 idx, hash = skb->rxhash;
@@ -1197,13 +1183,9 @@ static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *s
 
 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
 {
-       int cur, old;
+       unsigned int val = atomic_inc_return(&f->rr_cur);
 
-       cur = atomic_read(&f->rr_cur);
-       while ((old = atomic_cmpxchg(&f->rr_cur, cur,
-                                    fanout_rr_next(f, num))) != cur)
-               cur = old;
-       return f->arr[cur];
+       return f->arr[val % num];
 }
 
 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
@@ -1217,7 +1199,7 @@ static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
                             struct packet_type *pt, struct net_device *orig_dev)
 {
        struct packet_fanout *f = pt->af_packet_priv;
-       unsigned int num = f->num_members;
+       unsigned int num = ACCESS_ONCE(f->num_members);
        struct packet_sock *po;
        struct sock *sk;
 
@@ -1262,6 +1244,8 @@ static void __fanout_link(struct sock *sk, struct packet_sock *po)
        f->arr[f->num_members] = sk;
        smp_wmb();
        f->num_members++;
+       if (f->num_members == 1)
+               dev_add_pack(&f->prot_hook);
        spin_unlock(&f->lock);
 }
 
@@ -1278,9 +1262,19 @@ static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
        BUG_ON(i >= f->num_members);
        f->arr[i] = f->arr[f->num_members - 1];
        f->num_members--;
+       if (f->num_members == 0)
+               __dev_remove_pack(&f->prot_hook);
        spin_unlock(&f->lock);
 }
 
+bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
+{
+       if (sk->sk_family != PF_PACKET)
+               return false;
+
+       return ptype->af_packet_priv == pkt_sk(sk)->fanout;
+}
+
 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
 {
        struct packet_sock *po = pkt_sk(sk);
@@ -1298,13 +1292,12 @@ static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
                return -EINVAL;
        }
 
-       if (!po->running)
-               return -EINVAL;
+       mutex_lock(&fanout_mutex);
 
+       err = -EALREADY;
        if (po->fanout)
-               return -EALREADY;
+               goto out;
 
-       mutex_lock(&fanout_mutex);
        match = NULL;
        list_for_each_entry(f, &fanout_list, list) {
                if (f->id == id &&
@@ -1333,11 +1326,14 @@ static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
                match->prot_hook.dev = po->prot_hook.dev;
                match->prot_hook.func = packet_rcv_fanout;
                match->prot_hook.af_packet_priv = match;
-               dev_add_pack(&match->prot_hook);
+               match->prot_hook.id_match = match_fanout_group;
                list_add(&match->list, &fanout_list);
        }
        err = -EINVAL;
-       if (match->type == type &&
+
+       spin_lock(&po->bind_lock);
+       if (po->running &&
+           match->type == type &&
            match->prot_hook.type == po->prot_hook.type &&
            match->prot_hook.dev == po->prot_hook.dev) {
                err = -ENOSPC;
@@ -1349,29 +1345,41 @@ static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
                        err = 0;
                }
        }
+       spin_unlock(&po->bind_lock);
+
+       if (err && !atomic_read(&match->sk_ref)) {
+               list_del(&match->list);
+               kfree(match);
+       }
+
 out:
        mutex_unlock(&fanout_mutex);
        return err;
 }
 
-static void fanout_release(struct sock *sk)
+/* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
+ * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
+ * It is the responsibility of the caller to call fanout_release_data() and
+ * free the returned packet_fanout (after synchronize_net())
+ */
+static struct packet_fanout *fanout_release(struct sock *sk)
 {
        struct packet_sock *po = pkt_sk(sk);
        struct packet_fanout *f;
 
+       mutex_lock(&fanout_mutex);
        f = po->fanout;
-       if (!f)
-               return;
-
-       po->fanout = NULL;
+       if (f) {
+               po->fanout = NULL;
 
-       mutex_lock(&fanout_mutex);
-       if (atomic_dec_and_test(&f->sk_ref)) {
-               list_del(&f->list);
-               dev_remove_pack(&f->prot_hook);
-               kfree(f);
+               if (atomic_dec_and_test(&f->sk_ref))
+                       list_del(&f->list);
+               else
+                       f = NULL;
        }
        mutex_unlock(&fanout_mutex);
+
+       return f;
 }
 
 static const struct proto_ops packet_ops;
@@ -1630,8 +1638,7 @@ static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
        if (snaplen > res)
                snaplen = res;
 
-       if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
-           (unsigned)sk->sk_rcvbuf)
+       if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
                goto drop_n_acct;
 
        if (skb_shared(skb)) {
@@ -1762,8 +1769,7 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
        if (po->tp_version <= TPACKET_V2) {
                if (macoff + snaplen > po->rx_ring.frame_size) {
                        if (po->copy_thresh &&
-                               atomic_read(&sk->sk_rmem_alloc) + skb->truesize
-                               < (unsigned)sk->sk_rcvbuf) {
+                           atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
                                if (skb_shared(skb)) {
                                        copy_skb = skb_clone(skb, GFP_ATOMIC);
                                } else {
@@ -1777,6 +1783,18 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
                        if ((int)snaplen < 0)
                                snaplen = 0;
                }
+       } else if (unlikely(macoff + snaplen >
+                           GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
+               u32 nval;
+
+               nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
+               pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
+                           snaplen, nval, macoff);
+               snaplen = nval;
+               if (unlikely((int)snaplen < 0)) {
+                       snaplen = 0;
+                       macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
+               }
        }
        spin_lock(&sk->sk_receive_queue.lock);
        h.raw = packet_current_rx_frame(po, skb,
@@ -1933,7 +1951,6 @@ static void tpacket_destruct_skb(struct sk_buff *skb)
 
        if (likely(po->tx_ring.pg_vec)) {
                ph = skb_shinfo(skb)->destructor_arg;
-               BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
                BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
                atomic_dec(&po->tx_ring.pending);
                __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
@@ -2040,12 +2057,24 @@ 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;
        struct net_device *dev;
        __be16 proto;
-       bool need_rls_dev = false;
        int err, reserve = 0;
        void *ph;
        struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
@@ -2058,7 +2087,7 @@ static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
 
        err = -EBUSY;
        if (saddr == NULL) {
-               dev = po->prot_hook.dev;
+               dev     = packet_cached_dev_get(po);
                proto   = po->num;
                addr    = NULL;
        } else {
@@ -2072,19 +2101,17 @@ static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
                proto   = saddr->sll_protocol;
                addr    = saddr->sll_addr;
                dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
-               need_rls_dev = true;
        }
 
        err = -ENXIO;
        if (unlikely(dev == NULL))
                goto out;
-
-       reserve = dev->hard_header_len;
-
        err = -ENETDOWN;
        if (unlikely(!(dev->flags & IFF_UP)))
                goto out_put;
 
+       reserve = dev->hard_header_len;
+
        size_max = po->tx_ring.frame_size
                - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
 
@@ -2160,8 +2187,7 @@ out_status:
        __packet_set_status(po, ph, status);
        kfree_skb(skb);
 out_put:
-       if (need_rls_dev)
-               dev_put(dev);
+       dev_put(dev);
 out:
        mutex_unlock(&po->pg_vec_lock);
        return err;
@@ -2199,13 +2225,13 @@ static int packet_snd(struct socket *sock,
        struct sk_buff *skb;
        struct net_device *dev;
        __be16 proto;
-       bool need_rls_dev = false;
        unsigned char *addr;
        int err, reserve = 0;
        struct virtio_net_hdr vnet_hdr = { 0 };
        int offset = 0;
        int vnet_hdr_len;
        struct packet_sock *po = pkt_sk(sk);
+       bool has_vnet_hdr = false;
        unsigned short gso_type = 0;
 
        /*
@@ -2213,7 +2239,7 @@ static int packet_snd(struct socket *sock,
         */
 
        if (saddr == NULL) {
-               dev = po->prot_hook.dev;
+               dev     = packet_cached_dev_get(po);
                proto   = po->num;
                addr    = NULL;
        } else {
@@ -2225,21 +2251,20 @@ static int packet_snd(struct socket *sock,
                proto   = saddr->sll_protocol;
                addr    = saddr->sll_addr;
                dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
-               need_rls_dev = true;
        }
 
        err = -ENXIO;
-       if (dev == NULL)
+       if (unlikely(dev == NULL))
                goto out_unlock;
-       if (sock->type == SOCK_RAW)
-               reserve = dev->hard_header_len;
-
        err = -ENETDOWN;
-       if (!(dev->flags & IFF_UP))
+       if (unlikely(!(dev->flags & IFF_UP)))
                goto out_unlock;
 
+       if (sock->type == SOCK_RAW)
+               reserve = dev->hard_header_len;
        if (po->has_vnet_hdr) {
                vnet_hdr_len = sizeof(vnet_hdr);
+               has_vnet_hdr = true;
 
                err = -EINVAL;
                if (len < vnet_hdr_len)
@@ -2331,7 +2356,7 @@ static int packet_snd(struct socket *sock,
        skb->priority = sk->sk_priority;
        skb->mark = sk->sk_mark;
 
-       if (po->has_vnet_hdr) {
+       if (has_vnet_hdr) {
                if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
                        if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
                                                  vnet_hdr.csum_offset)) {
@@ -2358,15 +2383,14 @@ static int packet_snd(struct socket *sock,
        if (err > 0 && (err = net_xmit_errno(err)) != 0)
                goto out_unlock;
 
-       if (need_rls_dev)
-               dev_put(dev);
+       dev_put(dev);
 
        return len;
 
 out_free:
        kfree_skb(skb);
 out_unlock:
-       if (dev && need_rls_dev)
+       if (dev)
                dev_put(dev);
 out:
        return err;
@@ -2392,6 +2416,7 @@ static int packet_release(struct socket *sock)
 {
        struct sock *sk = sock->sk;
        struct packet_sock *po;
+       struct packet_fanout *f;
        struct net *net;
        union tpacket_req_u req_u;
 
@@ -2416,17 +2441,22 @@ static int packet_release(struct socket *sock)
 
        packet_flush_mclist(sk);
 
-       memset(&req_u, 0, sizeof(req_u));
-
-       if (po->rx_ring.pg_vec)
+       if (po->rx_ring.pg_vec) {
+               memset(&req_u, 0, sizeof(req_u));
                packet_set_ring(sk, &req_u, 1, 0);
+       }
 
-       if (po->tx_ring.pg_vec)
+       if (po->tx_ring.pg_vec) {
+               memset(&req_u, 0, sizeof(req_u));
                packet_set_ring(sk, &req_u, 1, 1);
+       }
 
-       fanout_release(sk);
+       f = fanout_release(sk);
 
        synchronize_net();
+
+       kfree(f);
+
        /*
         *      Now the socket is dead. No more input will appear.
         */
@@ -2446,29 +2476,77 @@ static int packet_release(struct socket *sock)
  *     Attach a packet hook.
  */
 
-static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
+static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
+                         __be16 protocol)
 {
        struct packet_sock *po = pkt_sk(sk);
-
-       if (po->fanout)
-               return -EINVAL;
+       struct net_device *dev_curr;
+       struct net_device *dev = NULL;
+       int ret = 0;
+       bool unlisted = false;
 
        lock_sock(sk);
 
        spin_lock(&po->bind_lock);
-       unregister_prot_hook(sk, true);
+       rcu_read_lock();
+
+       if (po->fanout) {
+               ret = -EINVAL;
+               goto out_unlock;
+       }
+
+       if (name) {
+               dev = dev_get_by_name_rcu(sock_net(sk), name);
+               if (!dev) {
+                       ret = -ENODEV;
+                       goto out_unlock;
+               }
+       } else if (ifindex) {
+               dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
+               if (!dev) {
+                       ret = -ENODEV;
+                       goto out_unlock;
+               }
+       }
+
+       if (dev)
+               dev_hold(dev);
+
+       dev_curr = po->prot_hook.dev;
+
+       if (po->running) {
+               rcu_read_unlock();
+               /* prevents packet_notifier() from calling
+                * register_prot_hook()
+                */
+               po->num = 0;
+               __unregister_prot_hook(sk, true);
+               rcu_read_lock();
+               dev_curr = po->prot_hook.dev;
+               if (dev)
+                       unlisted = !dev_get_by_index_rcu(sock_net(sk),
+                                                        dev->ifindex);
+       }
+       BUG_ON(po->running);
        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->ifindex = dev ? dev->ifindex : 0;
+       if (unlikely(unlisted)) {
+               dev_put(dev);
+               po->prot_hook.dev = NULL;
+               po->ifindex = -1;
+       } else {
+               po->prot_hook.dev = dev;
+               po->ifindex = dev ? dev->ifindex : 0;
+       }
+
+       if (dev_curr)
+               dev_put(dev_curr);
 
        if (protocol == 0)
                goto out_unlock;
 
-       if (!dev || (dev->flags & IFF_UP)) {
+       if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
                register_prot_hook(sk);
        } else {
                sk->sk_err = ENETDOWN;
@@ -2477,9 +2555,10 @@ static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protoc
        }
 
 out_unlock:
+       rcu_read_unlock();
        spin_unlock(&po->bind_lock);
        release_sock(sk);
-       return 0;
+       return ret;
 }
 
 /*
@@ -2491,8 +2570,6 @@ static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
 {
        struct sock *sk = sock->sk;
        char name[15];
-       struct net_device *dev;
-       int err = -ENODEV;
 
        /*
         *      Check legality
@@ -2502,19 +2579,13 @@ static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
                return -EINVAL;
        strlcpy(name, uaddr->sa_data, sizeof(name));
 
-       dev = dev_get_by_name(sock_net(sk), name);
-       if (dev)
-               err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
-       return err;
+       return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
 }
 
 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 {
        struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
        struct sock *sk = sock->sk;
-       struct net_device *dev = NULL;
-       int err;
-
 
        /*
         *      Check legality
@@ -2525,16 +2596,8 @@ static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len
        if (sll->sll_family != AF_PACKET)
                return -EINVAL;
 
-       if (sll->sll_ifindex) {
-               err = -ENODEV;
-               dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
-               if (dev == NULL)
-                       goto out;
-       }
-       err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
-
-out:
-       return err;
+       return packet_do_bind(sk, NULL, sll->sll_ifindex,
+                             sll->sll_protocol ? : pkt_sk(sk)->num);
 }
 
 static struct proto packet_proto = {
@@ -2577,6 +2640,7 @@ 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);
 
        sk->sk_destruct = packet_sock_destruct;
        sk_refcnt_debug_inc(sk);
@@ -2665,7 +2729,6 @@ static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
        struct sock *sk = sock->sk;
        struct sk_buff *skb;
        int copied, err;
-       struct sockaddr_ll *sll;
        int vnet_hdr_len = 0;
 
        err = -EINVAL;
@@ -2748,22 +2811,10 @@ static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
                        goto out_free;
        }
 
-       /*
-        *      If the address length field is there to be filled in, we fill
-        *      it in now.
-        */
-
-       sll = &PACKET_SKB_CB(skb)->sa.ll;
-       if (sock->type == SOCK_PACKET)
-               msg->msg_namelen = sizeof(struct sockaddr_pkt);
-       else
-               msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
-
-       /*
-        *      You lose any data beyond the buffer you gave. If it worries a
-        *      user program they can ask the device for its MTU anyway.
+       /* You lose any data beyond the buffer you gave. If it worries
+        * a user program they can ask the device for its MTU
+        * anyway.
         */
-
        copied = skb->len;
        if (copied > len) {
                copied = len;
@@ -2776,9 +2827,20 @@ static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
 
        sock_recv_ts_and_drops(msg, sk, skb);
 
-       if (msg->msg_name)
+       if (msg->msg_name) {
+               /* If the address length field is there to be filled
+                * in, we fill it in now.
+                */
+               if (sock->type == SOCK_PACKET) {
+                       msg->msg_namelen = sizeof(struct sockaddr_pkt);
+               } else {
+                       struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
+                       msg->msg_namelen = sll->sll_halen +
+                               offsetof(struct sockaddr_ll, sll_addr);
+               }
                memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
                       msg->msg_namelen);
+       }
 
        if (pkt_sk(sk)->auxdata) {
                struct tpacket_auxdata aux;
@@ -2822,12 +2884,11 @@ static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
                return -EOPNOTSUPP;
 
        uaddr->sa_family = AF_PACKET;
+       memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
        rcu_read_lock();
        dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
        if (dev)
-               strncpy(uaddr->sa_data, dev->name, 14);
-       else
-               memset(uaddr->sa_data, 0, 14);
+               strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
        rcu_read_unlock();
        *uaddr_len = sizeof(*uaddr);
 
@@ -3082,19 +3143,25 @@ packet_setsockopt(struct socket *sock, int level, int optname, char __user *optv
 
                if (optlen != sizeof(val))
                        return -EINVAL;
-               if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
-                       return -EBUSY;
                if (copy_from_user(&val, optval, sizeof(val)))
                        return -EFAULT;
                switch (val) {
                case TPACKET_V1:
                case TPACKET_V2:
                case TPACKET_V3:
-                       po->tp_version = val;
-                       return 0;
+                       break;
                default:
                        return -EINVAL;
                }
+               lock_sock(sk);
+               if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
+                       ret = -EBUSY;
+               } else {
+                       po->tp_version = val;
+                       ret = 0;
+               }
+               release_sock(sk);
+               return ret;
        }
        case PACKET_RESERVE:
        {
@@ -3102,12 +3169,19 @@ packet_setsockopt(struct socket *sock, int level, int optname, char __user *optv
 
                if (optlen != sizeof(val))
                        return -EINVAL;
-               if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
-                       return -EBUSY;
                if (copy_from_user(&val, optval, sizeof(val)))
                        return -EFAULT;
-               po->tp_reserve = val;
-               return 0;
+               if (val > INT_MAX)
+                       return -EINVAL;
+               lock_sock(sk);
+               if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
+                       ret = -EBUSY;
+               } else {
+                       po->tp_reserve = val;
+                       ret = 0;
+               }
+               release_sock(sk);
+               return ret;
        }
        case PACKET_LOSS:
        {
@@ -3565,6 +3639,7 @@ static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
        /* Added to avoid minimal code churn */
        struct tpacket_req *req = &req_u->req;
 
+       lock_sock(sk);
        /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
        if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
                WARN(1, "Tx-ring is not supported.\n");
@@ -3605,6 +3680,10 @@ static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
                        goto out;
                if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
                        goto out;
+               if (po->tp_version >= TPACKET_V3 &&
+                   req->tp_block_size <=
+                         BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv))
+                       goto out;
                if (unlikely(req->tp_frame_size < po->tp_hdrlen +
                                        po->tp_reserve))
                        goto out;
@@ -3614,6 +3693,8 @@ static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
                rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
                if (unlikely(rb->frames_per_block <= 0))
                        goto out;
+               if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
+                       goto out;
                if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
                                        req->tp_frame_nr))
                        goto out;
@@ -3642,7 +3723,6 @@ static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
                        goto out;
        }
 
-       lock_sock(sk);
 
        /* Detach socket from network */
        spin_lock(&po->bind_lock);
@@ -3691,11 +3771,11 @@ static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
                if (!tx_ring)
                        prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
        }
-       release_sock(sk);
 
        if (pg_vec)
                free_pg_vec(pg_vec, order, req->tp_block_nr);
 out:
+       release_sock(sk);
        return err;
 }