net: spread __net_init, __net_exit
[safe/jmp/linux-2.6] / net / packet / af_packet.c
index de73bcb..a97acfe 100644 (file)
@@ -39,6 +39,7 @@
  *                                     will simply extend the hardware address
  *                                     byte arrays at the end of sockaddr_ll
  *                                     and packet_mreq.
+ *             Johann Baudy    :       Added TX RING.
  *
  *             This program is free software; you can redistribute it and/or
  *             modify it under the terms of the GNU General Public License
@@ -77,6 +78,8 @@
 #include <linux/poll.h>
 #include <linux/module.h>
 #include <linux/init.h>
+#include <linux/mutex.h>
+#include <linux/if_vlan.h>
 
 #ifdef CONFIG_INET
 #include <net/inet_common.h>
@@ -135,8 +138,7 @@ dev->hard_header == NULL (ll header is added by device, we cannot control it)
 
 /* Private packet socket structures. */
 
-struct packet_mclist
-{
+struct packet_mclist {
        struct packet_mclist    *next;
        int                     ifindex;
        int                     count;
@@ -147,8 +149,7 @@ struct packet_mclist
 /* identical to struct packet_mreq except it has
  * a longer address field.
  */
-struct packet_mreq_max
-{
+struct packet_mreq_max {
        int             mr_ifindex;
        unsigned short  mr_type;
        unsigned short  mr_alen;
@@ -156,7 +157,25 @@ struct packet_mreq_max
 };
 
 #ifdef CONFIG_PACKET_MMAP
-static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing);
+static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
+               int closing, int tx_ring);
+
+struct packet_ring_buffer {
+       char                    **pg_vec;
+       unsigned int            head;
+       unsigned int            frames_per_block;
+       unsigned int            frame_size;
+       unsigned int            frame_max;
+
+       unsigned int            pg_vec_order;
+       unsigned int            pg_vec_pages;
+       unsigned int            pg_vec_len;
+
+       atomic_t                pending;
+};
+
+struct packet_sock;
+static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 #endif
 
 static void packet_flush_mclist(struct sock *sk);
@@ -166,15 +185,12 @@ struct packet_sock {
        struct sock             sk;
        struct tpacket_stats    stats;
 #ifdef CONFIG_PACKET_MMAP
-       char *                  *pg_vec;
-       unsigned int            head;
-       unsigned int            frames_per_block;
-       unsigned int            frame_size;
-       unsigned int            frame_max;
+       struct packet_ring_buffer       rx_ring;
+       struct packet_ring_buffer       tx_ring;
        int                     copy_thresh;
 #endif
-       struct packet_type      prot_hook;
        spinlock_t              bind_lock;
+       struct mutex            pg_vec_lock;
        unsigned int            running:1,      /* prot_hook is attached*/
                                auxdata:1,
                                origdev:1;
@@ -183,13 +199,12 @@ struct packet_sock {
        struct packet_mclist    *mclist;
 #ifdef CONFIG_PACKET_MMAP
        atomic_t                mapped;
-       unsigned int            pg_vec_order;
-       unsigned int            pg_vec_pages;
-       unsigned int            pg_vec_len;
        enum tpacket_versions   tp_version;
        unsigned int            tp_hdrlen;
        unsigned int            tp_reserve;
+       unsigned int            tp_loss:1;
 #endif
+       struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 };
 
 struct packet_skb_cb {
@@ -204,36 +219,33 @@ struct packet_skb_cb {
 
 #ifdef CONFIG_PACKET_MMAP
 
-static void *packet_lookup_frame(struct packet_sock *po, unsigned int position,
-                                int status)
+static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 {
-       unsigned int pg_vec_pos, frame_offset;
        union {
                struct tpacket_hdr *h1;
                struct tpacket2_hdr *h2;
                void *raw;
        } h;
 
-       pg_vec_pos = position / po->frames_per_block;
-       frame_offset = position % po->frames_per_block;
-
-       h.raw = po->pg_vec[pg_vec_pos] + (frame_offset * po->frame_size);
+       h.raw = frame;
        switch (po->tp_version) {
        case TPACKET_V1:
-               if (status != h.h1->tp_status ? TP_STATUS_USER :
-                                               TP_STATUS_KERNEL)
-                       return NULL;
+               h.h1->tp_status = status;
+               flush_dcache_page(virt_to_page(&h.h1->tp_status));
                break;
        case TPACKET_V2:
-               if (status != h.h2->tp_status ? TP_STATUS_USER :
-                                               TP_STATUS_KERNEL)
-                       return NULL;
+               h.h2->tp_status = status;
+               flush_dcache_page(virt_to_page(&h.h2->tp_status));
                break;
+       default:
+               pr_err("TPACKET version not supported\n");
+               BUG();
        }
-       return h.raw;
+
+       smp_wmb();
 }
 
-static void __packet_set_status(struct packet_sock *po, void *frame, int status)
+static int __packet_get_status(struct packet_sock *po, void *frame)
 {
        union {
                struct tpacket_hdr *h1;
@@ -241,16 +253,66 @@ static void __packet_set_status(struct packet_sock *po, void *frame, int status)
                void *raw;
        } h;
 
+       smp_rmb();
+
        h.raw = frame;
        switch (po->tp_version) {
        case TPACKET_V1:
-               h.h1->tp_status = status;
-               break;
+               flush_dcache_page(virt_to_page(&h.h1->tp_status));
+               return h.h1->tp_status;
        case TPACKET_V2:
-               h.h2->tp_status = status;
-               break;
+               flush_dcache_page(virt_to_page(&h.h2->tp_status));
+               return h.h2->tp_status;
+       default:
+               pr_err("TPACKET version not supported\n");
+               BUG();
+               return 0;
        }
 }
+
+static void *packet_lookup_frame(struct packet_sock *po,
+               struct packet_ring_buffer *rb,
+               unsigned int position,
+               int status)
+{
+       unsigned int pg_vec_pos, frame_offset;
+       union {
+               struct tpacket_hdr *h1;
+               struct tpacket2_hdr *h2;
+               void *raw;
+       } h;
+
+       pg_vec_pos = position / rb->frames_per_block;
+       frame_offset = position % rb->frames_per_block;
+
+       h.raw = rb->pg_vec[pg_vec_pos] + (frame_offset * rb->frame_size);
+
+       if (status != __packet_get_status(po, h.raw))
+               return NULL;
+
+       return h.raw;
+}
+
+static inline void *packet_current_frame(struct packet_sock *po,
+               struct packet_ring_buffer *rb,
+               int status)
+{
+       return packet_lookup_frame(po, rb, rb->head, status);
+}
+
+static inline void *packet_previous_frame(struct packet_sock *po,
+               struct packet_ring_buffer *rb,
+               int status)
+{
+       unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
+       return packet_lookup_frame(po, rb, previous, status);
+}
+
+static inline void packet_increment_head(struct packet_ring_buffer *buff)
+{
+       buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
+}
+
 #endif
 
 static inline struct packet_sock *pkt_sk(struct sock *sk)
@@ -260,11 +322,11 @@ static inline struct packet_sock *pkt_sk(struct sock *sk)
 
 static void packet_sock_destruct(struct sock *sk)
 {
-       BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
-       BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
+       WARN_ON(atomic_read(&sk->sk_rmem_alloc));
+       WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 
        if (!sock_flag(sk, SOCK_DEAD)) {
-               printk("Attempt to release alive packet socket: %p\n", sk);
+               pr_err("Attempt to release alive packet socket: %p\n", sk);
                return;
        }
 
@@ -276,7 +338,8 @@ static const struct proto_ops packet_ops;
 
 static const struct proto_ops packet_ops_spkt;
 
-static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,  struct packet_type *pt, struct net_device *orig_dev)
+static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
+                          struct packet_type *pt, struct net_device *orig_dev)
 {
        struct sock *sk;
        struct sockaddr_pkt *spkt;
@@ -302,15 +365,15 @@ static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,  struct
        if (skb->pkt_type == PACKET_LOOPBACK)
                goto out;
 
-       if (dev_net(dev) != sock_net(sk))
+       if (!net_eq(dev_net(dev), sock_net(sk)))
                goto out;
 
-       if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
+       skb = skb_share_check(skb, GFP_ATOMIC);
+       if (skb == NULL)
                goto oom;
 
        /* drop any routing info */
-       dst_release(skb->dst);
-       skb->dst = NULL;
+       skb_dst_drop(skb);
 
        /* drop conntrack reference */
        nf_reset(skb);
@@ -332,7 +395,7 @@ static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,  struct
         *      to prevent sockets using all the memory up.
         */
 
-       if (sock_queue_rcv_skb(sk,skb) == 0)
+       if (sock_queue_rcv_skb(sk, skb) == 0)
                return 0;
 
 out:
@@ -351,32 +414,32 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
                               struct msghdr *msg, size_t len)
 {
        struct sock *sk = sock->sk;
-       struct sockaddr_pkt *saddr=(struct sockaddr_pkt *)msg->msg_name;
-       struct sk_buff *skb;
+       struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
+       struct sk_buff *skb = NULL;
        struct net_device *dev;
-       __be16 proto=0;
+       __be16 proto = 0;
        int err;
 
        /*
         *      Get and verify the address.
         */
 
-       if (saddr)
-       {
+       if (saddr) {
                if (msg->msg_namelen < sizeof(struct sockaddr))
-                       return(-EINVAL);
-               if (msg->msg_namelen==sizeof(struct sockaddr_pkt))
-                       proto=saddr->spkt_protocol;
-       }
-       else
-               return(-ENOTCONN);      /* SOCK_PACKET must be sent giving an address */
+                       return -EINVAL;
+               if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
+                       proto = saddr->spkt_protocol;
+       } else
+               return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
 
        /*
         *      Find the device first to size check it
         */
 
        saddr->spkt_device[13] = 0;
-       dev = dev_get_by_name(sock_net(sk), saddr->spkt_device);
+retry:
+       rcu_read_lock();
+       dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
        err = -ENODEV;
        if (dev == NULL)
                goto out_unlock;
@@ -386,66 +449,56 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
                goto out_unlock;
 
        /*
-        *      You may not queue a frame bigger than the mtu. This is the lowest level
-        *      raw protocol and you must do your own fragmentation at this level.
+        * You may not queue a frame bigger than the mtu. This is the lowest level
+        * raw protocol and you must do your own fragmentation at this level.
         */
 
        err = -EMSGSIZE;
        if (len > dev->mtu + dev->hard_header_len)
                goto out_unlock;
 
-       err = -ENOBUFS;
-       skb = sock_wmalloc(sk, len + LL_RESERVED_SPACE(dev), 0, GFP_KERNEL);
-
-       /*
-        *      If the write buffer is full, then tough. At this level the user gets to
-        *      deal with the problem - do your own algorithmic backoffs. That's far
-        *      more flexible.
-        */
-
-       if (skb == NULL)
-               goto out_unlock;
-
-       /*
-        *      Fill it in
-        */
-
-       /* FIXME: Save some space for broken drivers that write a
-        * hard header at transmission time by themselves. PPP is the
-        * notable one here. This should really be fixed at the driver level.
-        */
-       skb_reserve(skb, LL_RESERVED_SPACE(dev));
-       skb_reset_network_header(skb);
-
-       /* Try to align data part correctly */
-       if (dev->header_ops) {
-               skb->data -= dev->hard_header_len;
-               skb->tail -= dev->hard_header_len;
-               if (len < dev->hard_header_len)
-                       skb_reset_network_header(skb);
+       if (!skb) {
+               size_t reserved = LL_RESERVED_SPACE(dev);
+               unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
+
+               rcu_read_unlock();
+               skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
+               if (skb == NULL)
+                       return -ENOBUFS;
+               /* FIXME: Save some space for broken drivers that write a hard
+                * header at transmission time by themselves. PPP is the notable
+                * one here. This should really be fixed at the driver level.
+                */
+               skb_reserve(skb, reserved);
+               skb_reset_network_header(skb);
+
+               /* Try to align data part correctly */
+               if (hhlen) {
+                       skb->data -= hhlen;
+                       skb->tail -= hhlen;
+                       if (len < hhlen)
+                               skb_reset_network_header(skb);
+               }
+               err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
+               if (err)
+                       goto out_free;
+               goto retry;
        }
 
-       /* Returns -EFAULT on error */
-       err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
+
        skb->protocol = proto;
        skb->dev = dev;
        skb->priority = sk->sk_priority;
-       if (err)
-               goto out_free;
-
-       /*
-        *      Now send it
-        */
+       skb->mark = sk->sk_mark;
 
        dev_queue_xmit(skb);
-       dev_put(dev);
-       return(len);
+       rcu_read_unlock();
+       return len;
 
+out_unlock:
+       rcu_read_unlock();
 out_free:
        kfree_skb(skb);
-out_unlock:
-       if (dev)
-               dev_put(dev);
        return err;
 }
 
@@ -475,12 +528,13 @@ static inline unsigned int run_filter(struct sk_buff *skb, struct sock *sk,
    we will not harm anyone.
  */
 
-static int packet_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
+static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
+                     struct packet_type *pt, struct net_device *orig_dev)
 {
        struct sock *sk;
        struct sockaddr_ll *sll;
        struct packet_sock *po;
-       u8 * skb_head = skb->data;
+       u8 *skb_head = skb->data;
        int skb_len = skb->len;
        unsigned int snaplen, res;
 
@@ -490,7 +544,7 @@ static int packet_rcv(struct sk_buff *skb, struct net_device *dev, struct packet
        sk = pt->af_packet_priv;
        po = pkt_sk(sk);
 
-       if (dev_net(dev) != sock_net(sk))
+       if (!net_eq(dev_net(dev), sock_net(sk)))
                goto drop;
 
        skb->dev = dev;
@@ -558,23 +612,21 @@ static int packet_rcv(struct sk_buff *skb, struct net_device *dev, struct packet
 
        skb_set_owner_r(skb, sk);
        skb->dev = NULL;
-       dst_release(skb->dst);
-       skb->dst = NULL;
+       skb_dst_drop(skb);
 
        /* drop conntrack reference */
        nf_reset(skb);
 
        spin_lock(&sk->sk_receive_queue.lock);
        po->stats.tp_packets++;
+       skb->dropcount = atomic_read(&sk->sk_drops);
        __skb_queue_tail(&sk->sk_receive_queue, skb);
        spin_unlock(&sk->sk_receive_queue.lock);
        sk->sk_data_ready(sk, skb->len);
        return 0;
 
 drop_n_acct:
-       spin_lock(&sk->sk_receive_queue.lock);
-       po->stats.tp_drops++;
-       spin_unlock(&sk->sk_receive_queue.lock);
+       po->stats.tp_drops = atomic_inc_return(&sk->sk_drops);
 
 drop_n_restore:
        if (skb_head != skb->data && skb_shared(skb)) {
@@ -582,12 +634,13 @@ drop_n_restore:
                skb->len = skb_len;
        }
 drop:
-       kfree_skb(skb);
+       consume_skb(skb);
        return 0;
 }
 
 #ifdef CONFIG_PACKET_MMAP
-static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
+static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
+                      struct packet_type *pt, struct net_device *orig_dev)
 {
        struct sock *sk;
        struct packet_sock *po;
@@ -597,7 +650,7 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packe
                struct tpacket2_hdr *h2;
                void *raw;
        } h;
-       u8 * skb_head = skb->data;
+       u8 *skb_head = skb->data;
        int skb_len = skb->len;
        unsigned int snaplen, res;
        unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
@@ -612,7 +665,7 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packe
        sk = pt->af_packet_priv;
        po = pkt_sk(sk);
 
-       if (dev_net(dev) != sock_net(sk))
+       if (!net_eq(dev_net(dev), sock_net(sk)))
                goto drop;
 
        if (dev->header_ops) {
@@ -646,7 +699,7 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packe
                macoff = netoff - maclen;
        }
 
-       if (macoff + snaplen > po->frame_size) {
+       if (macoff + snaplen > po->rx_ring.frame_size) {
                if (po->copy_thresh &&
                    atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
                    (unsigned)sk->sk_rcvbuf) {
@@ -659,16 +712,16 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packe
                        if (copy_skb)
                                skb_set_owner_r(copy_skb, sk);
                }
-               snaplen = po->frame_size - macoff;
+               snaplen = po->rx_ring.frame_size - macoff;
                if ((int)snaplen < 0)
                        snaplen = 0;
        }
 
        spin_lock(&sk->sk_receive_queue.lock);
-       h.raw = packet_lookup_frame(po, po->head, TP_STATUS_KERNEL);
+       h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
        if (!h.raw)
                goto ring_is_full;
-       po->head = po->head != po->frame_max ? po->head+1 : 0;
+       packet_increment_head(&po->rx_ring);
        po->stats.tp_packets++;
        if (copy_skb) {
                status |= TP_STATUS_COPY;
@@ -705,7 +758,7 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packe
                        getnstimeofday(&ts);
                h.h2->tp_sec = ts.tv_sec;
                h.h2->tp_nsec = ts.tv_nsec;
-               h.h2->tp_vlan_tci = skb->vlan_tci;
+               h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
                hdrlen = sizeof(*h.h2);
                break;
        default:
@@ -725,7 +778,6 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packe
 
        __packet_set_status(po, h.raw, status);
        smp_mb();
-
        {
                struct page *p_start, *p_end;
                u8 *h_end = h.raw + macoff + snaplen - 1;
@@ -754,19 +806,252 @@ ring_is_full:
        spin_unlock(&sk->sk_receive_queue.lock);
 
        sk->sk_data_ready(sk, 0);
-       if (copy_skb)
-               kfree_skb(copy_skb);
+       kfree_skb(copy_skb);
        goto drop_n_restore;
 }
 
-#endif
+static void tpacket_destruct_skb(struct sk_buff *skb)
+{
+       struct packet_sock *po = pkt_sk(skb->sk);
+       void *ph;
 
+       BUG_ON(skb == NULL);
 
-static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
+       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);
+       }
+
+       sock_wfree(skb);
+}
+
+static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
+               void *frame, struct net_device *dev, int size_max,
+               __be16 proto, unsigned char *addr)
+{
+       union {
+               struct tpacket_hdr *h1;
+               struct tpacket2_hdr *h2;
+               void *raw;
+       } ph;
+       int to_write, offset, len, tp_len, nr_frags, len_max;
+       struct socket *sock = po->sk.sk_socket;
+       struct page *page;
+       void *data;
+       int err;
+
+       ph.raw = frame;
+
+       skb->protocol = proto;
+       skb->dev = dev;
+       skb->priority = po->sk.sk_priority;
+       skb->mark = po->sk.sk_mark;
+       skb_shinfo(skb)->destructor_arg = ph.raw;
+
+       switch (po->tp_version) {
+       case TPACKET_V2:
+               tp_len = ph.h2->tp_len;
+               break;
+       default:
+               tp_len = ph.h1->tp_len;
+               break;
+       }
+       if (unlikely(tp_len > size_max)) {
+               pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
+               return -EMSGSIZE;
+       }
+
+       skb_reserve(skb, LL_RESERVED_SPACE(dev));
+       skb_reset_network_header(skb);
+
+       data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
+       to_write = tp_len;
+
+       if (sock->type == SOCK_DGRAM) {
+               err = dev_hard_header(skb, dev, ntohs(proto), addr,
+                               NULL, tp_len);
+               if (unlikely(err < 0))
+                       return -EINVAL;
+       } else if (dev->hard_header_len) {
+               /* net device doesn't like empty head */
+               if (unlikely(tp_len <= dev->hard_header_len)) {
+                       pr_err("packet size is too short (%d < %d)\n",
+                              tp_len, dev->hard_header_len);
+                       return -EINVAL;
+               }
+
+               skb_push(skb, dev->hard_header_len);
+               err = skb_store_bits(skb, 0, data,
+                               dev->hard_header_len);
+               if (unlikely(err))
+                       return err;
+
+               data += dev->hard_header_len;
+               to_write -= dev->hard_header_len;
+       }
+
+       err = -EFAULT;
+       page = virt_to_page(data);
+       offset = offset_in_page(data);
+       len_max = PAGE_SIZE - offset;
+       len = ((to_write > len_max) ? len_max : to_write);
+
+       skb->data_len = to_write;
+       skb->len += to_write;
+       skb->truesize += to_write;
+       atomic_add(to_write, &po->sk.sk_wmem_alloc);
+
+       while (likely(to_write)) {
+               nr_frags = skb_shinfo(skb)->nr_frags;
+
+               if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
+                       pr_err("Packet exceed the number of skb frags(%lu)\n",
+                              MAX_SKB_FRAGS);
+                       return -EFAULT;
+               }
+
+               flush_dcache_page(page);
+               get_page(page);
+               skb_fill_page_desc(skb,
+                               nr_frags,
+                               page++, offset, len);
+               to_write -= len;
+               offset = 0;
+               len_max = PAGE_SIZE;
+               len = ((to_write > len_max) ? len_max : to_write);
+       }
+
+       return tp_len;
+}
+
+static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
+{
+       struct socket *sock;
+       struct sk_buff *skb;
+       struct net_device *dev;
+       __be16 proto;
+       int ifindex, err, reserve = 0;
+       void *ph;
+       struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
+       int tp_len, size_max;
+       unsigned char *addr;
+       int len_sum = 0;
+       int status = 0;
+
+       sock = po->sk.sk_socket;
+
+       mutex_lock(&po->pg_vec_lock);
+
+       err = -EBUSY;
+       if (saddr == NULL) {
+               ifindex = po->ifindex;
+               proto   = po->num;
+               addr    = NULL;
+       } else {
+               err = -EINVAL;
+               if (msg->msg_namelen < sizeof(struct sockaddr_ll))
+                       goto out;
+               if (msg->msg_namelen < (saddr->sll_halen
+                                       + offsetof(struct sockaddr_ll,
+                                               sll_addr)))
+                       goto out;
+               ifindex = saddr->sll_ifindex;
+               proto   = saddr->sll_protocol;
+               addr    = saddr->sll_addr;
+       }
+
+       dev = dev_get_by_index(sock_net(&po->sk), ifindex);
+       err = -ENXIO;
+       if (unlikely(dev == NULL))
+               goto out;
+
+       reserve = dev->hard_header_len;
+
+       err = -ENETDOWN;
+       if (unlikely(!(dev->flags & IFF_UP)))
+               goto out_put;
+
+       size_max = po->tx_ring.frame_size
+               - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
+
+       if (size_max > dev->mtu + reserve)
+               size_max = dev->mtu + reserve;
+
+       do {
+               ph = packet_current_frame(po, &po->tx_ring,
+                               TP_STATUS_SEND_REQUEST);
+
+               if (unlikely(ph == NULL)) {
+                       schedule();
+                       continue;
+               }
+
+               status = TP_STATUS_SEND_REQUEST;
+               skb = sock_alloc_send_skb(&po->sk,
+                               LL_ALLOCATED_SPACE(dev)
+                               + sizeof(struct sockaddr_ll),
+                               0, &err);
+
+               if (unlikely(skb == NULL))
+                       goto out_status;
+
+               tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
+                               addr);
+
+               if (unlikely(tp_len < 0)) {
+                       if (po->tp_loss) {
+                               __packet_set_status(po, ph,
+                                               TP_STATUS_AVAILABLE);
+                               packet_increment_head(&po->tx_ring);
+                               kfree_skb(skb);
+                               continue;
+                       } else {
+                               status = TP_STATUS_WRONG_FORMAT;
+                               err = tp_len;
+                               goto out_status;
+                       }
+               }
+
+               skb->destructor = tpacket_destruct_skb;
+               __packet_set_status(po, ph, TP_STATUS_SENDING);
+               atomic_inc(&po->tx_ring.pending);
+
+               status = TP_STATUS_SEND_REQUEST;
+               err = dev_queue_xmit(skb);
+               if (unlikely(err > 0 && (err = net_xmit_errno(err)) != 0))
+                       goto out_xmit;
+               packet_increment_head(&po->tx_ring);
+               len_sum += tp_len;
+       } while (likely((ph != NULL) ||
+                       ((!(msg->msg_flags & MSG_DONTWAIT)) &&
+                        (atomic_read(&po->tx_ring.pending))))
+               );
+
+       err = len_sum;
+       goto out_put;
+
+out_xmit:
+       skb->destructor = sock_wfree;
+       atomic_dec(&po->tx_ring.pending);
+out_status:
+       __packet_set_status(po, ph, status);
+       kfree_skb(skb);
+out_put:
+       dev_put(dev);
+out:
+       mutex_unlock(&po->pg_vec_lock);
+       return err;
+}
+#endif
+
+static int packet_snd(struct socket *sock,
                          struct msghdr *msg, size_t len)
 {
        struct sock *sk = sock->sk;
-       struct sockaddr_ll *saddr=(struct sockaddr_ll *)msg->msg_name;
+       struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
        struct sk_buff *skb;
        struct net_device *dev;
        __be16 proto;
@@ -812,7 +1097,7 @@ static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
 
        skb = sock_alloc_send_skb(sk, len + LL_ALLOCATED_SPACE(dev),
                                msg->msg_flags & MSG_DONTWAIT, &err);
-       if (skb==NULL)
+       if (skb == NULL)
                goto out_unlock;
 
        skb_reserve(skb, LL_RESERVED_SPACE(dev));
@@ -824,13 +1109,14 @@ static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
                goto out_free;
 
        /* Returns -EFAULT on error */
-       err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
+       err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
        if (err)
                goto out_free;
 
        skb->protocol = proto;
        skb->dev = dev;
        skb->priority = sk->sk_priority;
+       skb->mark = sk->sk_mark;
 
        /*
         *      Now send it
@@ -842,7 +1128,7 @@ static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
 
        dev_put(dev);
 
-       return(len);
+       return len;
 
 out_free:
        kfree_skb(skb);
@@ -853,6 +1139,19 @@ out:
        return err;
 }
 
+static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
+               struct msghdr *msg, size_t len)
+{
+#ifdef CONFIG_PACKET_MMAP
+       struct sock *sk = sock->sk;
+       struct packet_sock *po = pkt_sk(sk);
+       if (po->tx_ring.pg_vec)
+               return tpacket_snd(po, msg);
+       else
+#endif
+               return packet_snd(sock, msg, len);
+}
+
 /*
  *     Close a PACKET socket. This is fairly simple. We immediately go
  *     to 'closed' state and remove our protocol entry in the device list.
@@ -863,6 +1162,9 @@ static int packet_release(struct socket *sock)
        struct sock *sk = sock->sk;
        struct packet_sock *po;
        struct net *net;
+#ifdef CONFIG_PACKET_MMAP
+       struct tpacket_req req;
+#endif
 
        if (!sk)
                return 0;
@@ -872,6 +1174,7 @@ static int packet_release(struct socket *sock)
 
        write_lock_bh(&net->packet.sklist_lock);
        sk_del_node_init(sk);
+       sock_prot_inuse_add(net, sk->sk_prot, -1);
        write_unlock_bh(&net->packet.sklist_lock);
 
        /*
@@ -891,11 +1194,13 @@ static int packet_release(struct socket *sock)
        packet_flush_mclist(sk);
 
 #ifdef CONFIG_PACKET_MMAP
-       if (po->pg_vec) {
-               struct tpacket_req req;
-               memset(&req, 0, sizeof(req));
-               packet_set_ring(sk, &req, 1);
-       }
+       memset(&req, 0, sizeof(req));
+
+       if (po->rx_ring.pg_vec)
+               packet_set_ring(sk, &req, 1, 0);
+
+       if (po->tx_ring.pg_vec)
+               packet_set_ring(sk, &req, 1, 1);
 #endif
 
        /*
@@ -966,9 +1271,10 @@ out_unlock:
  *     Bind a packet socket to a device
  */
 
-static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, int addr_len)
+static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
+                           int addr_len)
 {
-       struct sock *sk=sock->sk;
+       struct sock *sk = sock->sk;
        char name[15];
        struct net_device *dev;
        int err = -ENODEV;
@@ -979,7 +1285,7 @@ static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, int add
 
        if (addr_len != sizeof(struct sockaddr))
                return -EINVAL;
-       strlcpy(name,uaddr->sa_data,sizeof(name));
+       strlcpy(name, uaddr->sa_data, sizeof(name));
 
        dev = dev_get_by_name(sock_net(sk), name);
        if (dev) {
@@ -991,8 +1297,8 @@ static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, int add
 
 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 sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
+       struct sock *sk = sock->sk;
        struct net_device *dev = NULL;
        int err;
 
@@ -1030,7 +1336,8 @@ static struct proto packet_proto = {
  *     Create a packet of type SOCK_PACKET.
  */
 
-static int packet_create(struct net *net, struct socket *sock, int protocol)
+static int packet_create(struct net *net, struct socket *sock, int protocol,
+                        int kern)
 {
        struct sock *sk;
        struct packet_sock *po;
@@ -1068,6 +1375,7 @@ static int packet_create(struct net *net, struct socket *sock, int protocol)
         */
 
        spin_lock_init(&po->bind_lock);
+       mutex_init(&po->pg_vec_lock);
        po->prot_hook.func = packet_rcv;
 
        if (sock->type == SOCK_PACKET)
@@ -1084,8 +1392,9 @@ static int packet_create(struct net *net, struct socket *sock, int protocol)
 
        write_lock_bh(&net->packet.sklist_lock);
        sk_add_node(sk, &net->packet.sklist);
+       sock_prot_inuse_add(net, &packet_proto, 1);
        write_unlock_bh(&net->packet.sklist_lock);
-       return(0);
+       return 0;
 out:
        return err;
 }
@@ -1122,7 +1431,7 @@ static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
         *      but then it will block.
         */
 
-       skb=skb_recv_datagram(sk,flags,flags&MSG_DONTWAIT,&err);
+       skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
 
        /*
         *      An error occurred so return it. Because skb_recv_datagram()
@@ -1150,17 +1459,16 @@ static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
         */
 
        copied = skb->len;
-       if (copied > len)
-       {
-               copied=len;
-               msg->msg_flags|=MSG_TRUNC;
+       if (copied > len) {
+               copied = len;
+               msg->msg_flags |= MSG_TRUNC;
        }
 
        err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
        if (err)
                goto out_free;
 
-       sock_recv_timestamp(msg, sk, skb);
+       sock_recv_ts_and_drops(msg, sk, skb);
 
        if (msg->msg_name)
                memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
@@ -1176,7 +1484,7 @@ static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
                aux.tp_snaplen = skb->len;
                aux.tp_mac = 0;
                aux.tp_net = skb_network_offset(skb);
-               aux.tp_vlan_tci = skb->vlan_tci;
+               aux.tp_vlan_tci = vlan_tx_tag_get(skb);
 
                put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
        }
@@ -1203,12 +1511,13 @@ static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
                return -EOPNOTSUPP;
 
        uaddr->sa_family = AF_PACKET;
-       dev = dev_get_by_index(sock_net(sk), pkt_sk(sk)->ifindex);
-       if (dev) {
+       rcu_read_lock();
+       dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
+       if (dev)
                strlcpy(uaddr->sa_data, dev->name, 15);
-               dev_put(dev);
-       } else
+       else
                memset(uaddr->sa_data, 0, 14);
+       rcu_read_unlock();
        *uaddr_len = sizeof(*uaddr);
 
        return 0;
@@ -1220,7 +1529,7 @@ static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
        struct net_device *dev;
        struct sock *sk = sock->sk;
        struct packet_sock *po = pkt_sk(sk);
-       struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
+       DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
 
        if (peer)
                return -EOPNOTSUPP;
@@ -1228,16 +1537,17 @@ static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
        sll->sll_family = AF_PACKET;
        sll->sll_ifindex = po->ifindex;
        sll->sll_protocol = po->num;
-       dev = dev_get_by_index(sock_net(sk), po->ifindex);
+       rcu_read_lock();
+       dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
        if (dev) {
                sll->sll_hatype = dev->type;
                sll->sll_halen = dev->addr_len;
                memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
-               dev_put(dev);
        } else {
                sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
                sll->sll_halen = 0;
        }
+       rcu_read_unlock();
        *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
 
        return 0;
@@ -1249,9 +1559,9 @@ static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
        switch (i->type) {
        case PACKET_MR_MULTICAST:
                if (what > 0)
-                       dev_mc_add(dev, i->addr, i->alen, 0);
+                       return dev_mc_add(dev, i->addr, i->alen, 0);
                else
-                       dev_mc_delete(dev, i->addr, i->alen, 0);
+                       return dev_mc_delete(dev, i->addr, i->alen, 0);
                break;
        case PACKET_MR_PROMISC:
                return dev_set_promiscuity(dev, what);
@@ -1259,14 +1569,21 @@ static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
        case PACKET_MR_ALLMULTI:
                return dev_set_allmulti(dev, what);
                break;
-       default:;
+       case PACKET_MR_UNICAST:
+               if (what > 0)
+                       return dev_unicast_add(dev, i->addr);
+               else
+                       return dev_unicast_delete(dev, i->addr);
+               break;
+       default:
+               break;
        }
        return 0;
 }
 
 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
 {
-       for ( ; i; i=i->next) {
+       for ( ; i; i = i->next) {
                if (i->ifindex == dev->ifindex)
                        packet_dev_mc(dev, i, what);
        }
@@ -1340,11 +1657,9 @@ static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
                        if (--ml->count == 0) {
                                struct net_device *dev;
                                *mlp = ml->next;
-                               dev = dev_get_by_index(sock_net(sk), ml->ifindex);
-                               if (dev) {
+                               dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
+                               if (dev)
                                        packet_dev_mc(dev, ml, -1);
-                                       dev_put(dev);
-                               }
                                kfree(ml);
                        }
                        rtnl_unlock();
@@ -1368,17 +1683,16 @@ static void packet_flush_mclist(struct sock *sk)
                struct net_device *dev;
 
                po->mclist = ml->next;
-               if ((dev = dev_get_by_index(sock_net(sk), ml->ifindex)) != NULL) {
+               dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
+               if (dev != NULL)
                        packet_dev_mc(dev, ml, -1);
-                       dev_put(dev);
-               }
                kfree(ml);
        }
        rtnl_unlock();
 }
 
 static int
-packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, int optlen)
+packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
 {
        struct sock *sk = sock->sk;
        struct packet_sock *po = pkt_sk(sk);
@@ -1387,7 +1701,7 @@ packet_setsockopt(struct socket *sock, int level, int optname, char __user *optv
        if (level != SOL_PACKET)
                return -ENOPROTOOPT;
 
-       switch(optname) {
+       switch (optname) {
        case PACKET_ADD_MEMBERSHIP:
        case PACKET_DROP_MEMBERSHIP:
        {
@@ -1398,7 +1712,7 @@ packet_setsockopt(struct socket *sock, int level, int optname, char __user *optv
                        return -EINVAL;
                if (len > sizeof(mreq))
                        len = sizeof(mreq);
-               if (copy_from_user(&mreq,optval,len))
+               if (copy_from_user(&mreq, optval, len))
                        return -EFAULT;
                if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
                        return -EINVAL;
@@ -1411,22 +1725,23 @@ packet_setsockopt(struct socket *sock, int level, int optname, char __user *optv
 
 #ifdef CONFIG_PACKET_MMAP
        case PACKET_RX_RING:
+       case PACKET_TX_RING:
        {
                struct tpacket_req req;
 
-               if (optlen<sizeof(req))
+               if (optlen < sizeof(req))
                        return -EINVAL;
-               if (copy_from_user(&req,optval,sizeof(req)))
+               if (copy_from_user(&req, optval, sizeof(req)))
                        return -EFAULT;
-               return packet_set_ring(sk, &req, 0);
+               return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
        }
        case PACKET_COPY_THRESH:
        {
                int val;
 
-               if (optlen!=sizeof(val))
+               if (optlen != sizeof(val))
                        return -EINVAL;
-               if (copy_from_user(&val,optval,sizeof(val)))
+               if (copy_from_user(&val, optval, sizeof(val)))
                        return -EFAULT;
 
                pkt_sk(sk)->copy_thresh = val;
@@ -1438,7 +1753,7 @@ packet_setsockopt(struct socket *sock, int level, int optname, char __user *optv
 
                if (optlen != sizeof(val))
                        return -EINVAL;
-               if (po->pg_vec)
+               if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
                        return -EBUSY;
                if (copy_from_user(&val, optval, sizeof(val)))
                        return -EFAULT;
@@ -1457,13 +1772,26 @@ packet_setsockopt(struct socket *sock, int level, int optname, char __user *optv
 
                if (optlen != sizeof(val))
                        return -EINVAL;
-               if (po->pg_vec)
+               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;
        }
+       case PACKET_LOSS:
+       {
+               unsigned int val;
+
+               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_loss = !!val;
+               return 0;
+       }
 #endif
        case PACKET_AUXDATA:
        {
@@ -1513,7 +1841,7 @@ static int packet_getsockopt(struct socket *sock, int level, int optname,
        if (len < 0)
                return -EINVAL;
 
-       switch(optname) {
+       switch (optname) {
        case PACKET_STATISTICS:
                if (len > sizeof(struct tpacket_stats))
                        len = sizeof(struct tpacket_stats);
@@ -1569,6 +1897,12 @@ static int packet_getsockopt(struct socket *sock, int level, int optname,
                val = po->tp_reserve;
                data = &val;
                break;
+       case PACKET_LOSS:
+               if (len > sizeof(unsigned int))
+                       len = sizeof(unsigned int);
+               val = po->tp_loss;
+               data = &val;
+               break;
 #endif
        default:
                return -ENOPROTOOPT;
@@ -1639,51 +1973,52 @@ static int packet_ioctl(struct socket *sock, unsigned int cmd,
 {
        struct sock *sk = sock->sk;
 
-       switch(cmd) {
-               case SIOCOUTQ:
-               {
-                       int amount = atomic_read(&sk->sk_wmem_alloc);
-                       return put_user(amount, (int __user *)arg);
-               }
-               case SIOCINQ:
-               {
-                       struct sk_buff *skb;
-                       int amount = 0;
-
-                       spin_lock_bh(&sk->sk_receive_queue.lock);
-                       skb = skb_peek(&sk->sk_receive_queue);
-                       if (skb)
-                               amount = skb->len;
-                       spin_unlock_bh(&sk->sk_receive_queue.lock);
-                       return put_user(amount, (int __user *)arg);
-               }
-               case SIOCGSTAMP:
-                       return sock_get_timestamp(sk, (struct timeval __user *)arg);
-               case SIOCGSTAMPNS:
-                       return sock_get_timestampns(sk, (struct timespec __user *)arg);
+       switch (cmd) {
+       case SIOCOUTQ:
+       {
+               int amount = sk_wmem_alloc_get(sk);
+
+               return put_user(amount, (int __user *)arg);
+       }
+       case SIOCINQ:
+       {
+               struct sk_buff *skb;
+               int amount = 0;
+
+               spin_lock_bh(&sk->sk_receive_queue.lock);
+               skb = skb_peek(&sk->sk_receive_queue);
+               if (skb)
+                       amount = skb->len;
+               spin_unlock_bh(&sk->sk_receive_queue.lock);
+               return put_user(amount, (int __user *)arg);
+       }
+       case SIOCGSTAMP:
+               return sock_get_timestamp(sk, (struct timeval __user *)arg);
+       case SIOCGSTAMPNS:
+               return sock_get_timestampns(sk, (struct timespec __user *)arg);
 
 #ifdef CONFIG_INET
-               case SIOCADDRT:
-               case SIOCDELRT:
-               case SIOCDARP:
-               case SIOCGARP:
-               case SIOCSARP:
-               case SIOCGIFADDR:
-               case SIOCSIFADDR:
-               case SIOCGIFBRDADDR:
-               case SIOCSIFBRDADDR:
-               case SIOCGIFNETMASK:
-               case SIOCSIFNETMASK:
-               case SIOCGIFDSTADDR:
-               case SIOCSIFDSTADDR:
-               case SIOCSIFFLAGS:
-                       if (sock_net(sk) != &init_net)
-                               return -ENOIOCTLCMD;
-                       return inet_dgram_ops.ioctl(sock, cmd, arg);
+       case SIOCADDRT:
+       case SIOCDELRT:
+       case SIOCDARP:
+       case SIOCGARP:
+       case SIOCSARP:
+       case SIOCGIFADDR:
+       case SIOCSIFADDR:
+       case SIOCGIFBRDADDR:
+       case SIOCSIFBRDADDR:
+       case SIOCGIFNETMASK:
+       case SIOCSIFNETMASK:
+       case SIOCGIFDSTADDR:
+       case SIOCSIFDSTADDR:
+       case SIOCSIFFLAGS:
+               if (!net_eq(sock_net(sk), &init_net))
+                       return -ENOIOCTLCMD;
+               return inet_dgram_ops.ioctl(sock, cmd, arg);
 #endif
 
-               default:
-                       return -ENOIOCTLCMD;
+       default:
+               return -ENOIOCTLCMD;
        }
        return 0;
 }
@@ -1693,7 +2028,7 @@ static int packet_ioctl(struct socket *sock, unsigned int cmd,
 #define packet_poll datagram_poll
 #else
 
-static unsigned int packet_poll(struct file * file, struct socket *sock,
+static unsigned int packet_poll(struct file *file, struct socket *sock,
                                poll_table *wait)
 {
        struct sock *sk = sock->sk;
@@ -1701,13 +2036,17 @@ static unsigned int packet_poll(struct file * file, struct socket *sock,
        unsigned int mask = datagram_poll(file, sock, wait);
 
        spin_lock_bh(&sk->sk_receive_queue.lock);
-       if (po->pg_vec) {
-               unsigned last = po->head ? po->head-1 : po->frame_max;
-
-               if (packet_lookup_frame(po, last, TP_STATUS_USER))
+       if (po->rx_ring.pg_vec) {
+               if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
                        mask |= POLLIN | POLLRDNORM;
        }
        spin_unlock_bh(&sk->sk_receive_queue.lock);
+       spin_lock_bh(&sk->sk_write_queue.lock);
+       if (po->tx_ring.pg_vec) {
+               if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
+                       mask |= POLLOUT | POLLWRNORM;
+       }
+       spin_unlock_bh(&sk->sk_write_queue.lock);
        return mask;
 }
 
@@ -1719,7 +2058,7 @@ static unsigned int packet_poll(struct file * file, struct socket *sock,
 static void packet_mm_open(struct vm_area_struct *vma)
 {
        struct file *file = vma->vm_file;
-       struct socket * sock = file->private_data;
+       struct socket *sock = file->private_data;
        struct sock *sk = sock->sk;
 
        if (sk)
@@ -1729,16 +2068,16 @@ static void packet_mm_open(struct vm_area_struct *vma)
 static void packet_mm_close(struct vm_area_struct *vma)
 {
        struct file *file = vma->vm_file;
-       struct socket * sock = file->private_data;
+       struct socket *sock = file->private_data;
        struct sock *sk = sock->sk;
 
        if (sk)
                atomic_dec(&pkt_sk(sk)->mapped);
 }
 
-static struct vm_operations_struct packet_mmap_ops = {
-       .open = packet_mm_open,
-       .close =packet_mm_close,
+static const struct vm_operations_struct packet_mmap_ops = {
+       .open   =       packet_mm_open,
+       .close  =       packet_mm_close,
 };
 
 static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
@@ -1754,8 +2093,9 @@ static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
 
 static inline char *alloc_one_pg_vec_page(unsigned long order)
 {
-       return (char *) __get_free_pages(GFP_KERNEL | __GFP_COMP | __GFP_ZERO,
-                                        order);
+       gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO | __GFP_NOWARN;
+
+       return (char *) __get_free_pages(gfp_flags, order);
 }
 
 static char **alloc_pg_vec(struct tpacket_req *req, int order)
@@ -1783,21 +2123,33 @@ out_free_pgvec:
        goto out;
 }
 
-static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing)
+static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
+               int closing, int tx_ring)
 {
        char **pg_vec = NULL;
        struct packet_sock *po = pkt_sk(sk);
        int was_running, order = 0;
+       struct packet_ring_buffer *rb;
+       struct sk_buff_head *rb_queue;
        __be16 num;
-       int err = 0;
+       int err;
 
-       if (req->tp_block_nr) {
-               int i;
+       rb = tx_ring ? &po->tx_ring : &po->rx_ring;
+       rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
 
-               /* Sanity tests and some calculations */
+       err = -EBUSY;
+       if (!closing) {
+               if (atomic_read(&po->mapped))
+                       goto out;
+               if (atomic_read(&rb->pending))
+                       goto out;
+       }
 
-               if (unlikely(po->pg_vec))
-                       return -EBUSY;
+       if (req->tp_block_nr) {
+               /* Sanity tests and some calculations */
+               err = -EBUSY;
+               if (unlikely(rb->pg_vec))
+                       goto out;
 
                switch (po->tp_version) {
                case TPACKET_V1:
@@ -1808,42 +2160,35 @@ static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing
                        break;
                }
 
+               err = -EINVAL;
                if (unlikely((int)req->tp_block_size <= 0))
-                       return -EINVAL;
+                       goto out;
                if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
-                       return -EINVAL;
+                       goto out;
                if (unlikely(req->tp_frame_size < po->tp_hdrlen +
-                                                 po->tp_reserve))
-                       return -EINVAL;
+                                       po->tp_reserve))
+                       goto out;
                if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
-                       return -EINVAL;
+                       goto out;
 
-               po->frames_per_block = req->tp_block_size/req->tp_frame_size;
-               if (unlikely(po->frames_per_block <= 0))
-                       return -EINVAL;
-               if (unlikely((po->frames_per_block * req->tp_block_nr) !=
-                            req->tp_frame_nr))
-                       return -EINVAL;
+               rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
+               if (unlikely(rb->frames_per_block <= 0))
+                       goto out;
+               if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
+                                       req->tp_frame_nr))
+                       goto out;
 
                err = -ENOMEM;
                order = get_order(req->tp_block_size);
                pg_vec = alloc_pg_vec(req, order);
                if (unlikely(!pg_vec))
                        goto out;
-
-               for (i = 0; i < req->tp_block_nr; i++) {
-                       void *ptr = pg_vec[i];
-                       int k;
-
-                       for (k = 0; k < po->frames_per_block; k++) {
-                               __packet_set_status(po, ptr, TP_STATUS_KERNEL);
-                               ptr += req->tp_frame_size;
-                       }
-               }
-               /* Done */
-       } else {
+       }
+       /* Done */
+       else {
+               err = -EINVAL;
                if (unlikely(req->tp_frame_nr))
-                       return -EINVAL;
+                       goto out;
        }
 
        lock_sock(sk);
@@ -1863,27 +2208,30 @@ static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing
        synchronize_net();
 
        err = -EBUSY;
+       mutex_lock(&po->pg_vec_lock);
        if (closing || atomic_read(&po->mapped) == 0) {
                err = 0;
 #define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
-
-               spin_lock_bh(&sk->sk_receive_queue.lock);
-               pg_vec = XC(po->pg_vec, pg_vec);
-               po->frame_max = (req->tp_frame_nr - 1);
-               po->head = 0;
-               po->frame_size = req->tp_frame_size;
-               spin_unlock_bh(&sk->sk_receive_queue.lock);
-
-               order = XC(po->pg_vec_order, order);
-               req->tp_block_nr = XC(po->pg_vec_len, req->tp_block_nr);
-
-               po->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
-               po->prot_hook.func = po->pg_vec ? tpacket_rcv : packet_rcv;
-               skb_queue_purge(&sk->sk_receive_queue);
+               spin_lock_bh(&rb_queue->lock);
+               pg_vec = XC(rb->pg_vec, pg_vec);
+               rb->frame_max = (req->tp_frame_nr - 1);
+               rb->head = 0;
+               rb->frame_size = req->tp_frame_size;
+               spin_unlock_bh(&rb_queue->lock);
+
+               order = XC(rb->pg_vec_order, order);
+               req->tp_block_nr = XC(rb->pg_vec_len, req->tp_block_nr);
+
+               rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
+               po->prot_hook.func = (po->rx_ring.pg_vec) ?
+                                               tpacket_rcv : packet_rcv;
+               skb_queue_purge(rb_queue);
 #undef XC
                if (atomic_read(&po->mapped))
-                       printk(KERN_DEBUG "packet_mmap: vma is busy: %d\n", atomic_read(&po->mapped));
+                       pr_err("packet_mmap: vma is busy: %d\n",
+                              atomic_read(&po->mapped));
        }
+       mutex_unlock(&po->pg_vec_lock);
 
        spin_lock(&po->bind_lock);
        if (was_running && !po->running) {
@@ -1902,11 +2250,13 @@ out:
        return err;
 }
 
-static int packet_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
+static int packet_mmap(struct file *file, struct socket *sock,
+               struct vm_area_struct *vma)
 {
        struct sock *sk = sock->sk;
        struct packet_sock *po = pkt_sk(sk);
-       unsigned long size;
+       unsigned long size, expected_size;
+       struct packet_ring_buffer *rb;
        unsigned long start;
        int err = -EINVAL;
        int i;
@@ -1914,32 +2264,49 @@ static int packet_mmap(struct file *file, struct socket *sock, struct vm_area_st
        if (vma->vm_pgoff)
                return -EINVAL;
 
-       size = vma->vm_end - vma->vm_start;
+       mutex_lock(&po->pg_vec_lock);
 
-       lock_sock(sk);
-       if (po->pg_vec == NULL)
+       expected_size = 0;
+       for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
+               if (rb->pg_vec) {
+                       expected_size += rb->pg_vec_len
+                                               * rb->pg_vec_pages
+                                               * PAGE_SIZE;
+               }
+       }
+
+       if (expected_size == 0)
                goto out;
-       if (size != po->pg_vec_len*po->pg_vec_pages*PAGE_SIZE)
+
+       size = vma->vm_end - vma->vm_start;
+       if (size != expected_size)
                goto out;
 
        start = vma->vm_start;
-       for (i = 0; i < po->pg_vec_len; i++) {
-               struct page *page = virt_to_page(po->pg_vec[i]);
-               int pg_num;
-
-               for (pg_num = 0; pg_num < po->pg_vec_pages; pg_num++, page++) {
-                       err = vm_insert_page(vma, start, page);
-                       if (unlikely(err))
-                               goto out;
-                       start += PAGE_SIZE;
+       for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
+               if (rb->pg_vec == NULL)
+                       continue;
+
+               for (i = 0; i < rb->pg_vec_len; i++) {
+                       struct page *page = virt_to_page(rb->pg_vec[i]);
+                       int pg_num;
+
+                       for (pg_num = 0; pg_num < rb->pg_vec_pages;
+                                       pg_num++, page++) {
+                               err = vm_insert_page(vma, start, page);
+                               if (unlikely(err))
+                                       goto out;
+                               start += PAGE_SIZE;
+                       }
                }
        }
+
        atomic_inc(&po->mapped);
        vma->vm_ops = &packet_mmap_ops;
        err = 0;
 
 out:
-       release_sock(sk);
+       mutex_unlock(&po->pg_vec_lock);
        return err;
 }
 #endif
@@ -1987,14 +2354,14 @@ static const struct proto_ops packet_ops = {
        .sendpage =     sock_no_sendpage,
 };
 
-static struct net_proto_family packet_family_ops = {
+static const struct net_proto_family packet_family_ops = {
        .family =       PF_PACKET,
        .create =       packet_create,
        .owner  =       THIS_MODULE,
 };
 
 static struct notifier_block packet_netdev_notifier = {
-       .notifier_call =packet_notifier,
+       .notifier_call =        packet_notifier,
 };
 
 #ifdef CONFIG_PROC_FS
@@ -2024,7 +2391,7 @@ static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
        ++*pos;
        return  (v == SEQ_START_TOKEN)
                ? sk_head(&net->packet.sklist)
-               : sk_next((struct sock*)v) ;
+               : sk_next((struct sock *)v) ;
 }
 
 static void packet_seq_stop(struct seq_file *seq, void *v)
@@ -2052,7 +2419,7 @@ static int packet_seq_show(struct seq_file *seq, void *v)
                           po->running,
                           atomic_read(&s->sk_rmem_alloc),
                           sock_i_uid(s),
-                          sock_i_ino(s) );
+                          sock_i_ino(s));
        }
 
        return 0;
@@ -2081,7 +2448,7 @@ static const struct file_operations packet_seq_fops = {
 
 #endif
 
-static int packet_net_init(struct net *net)
+static int __net_init packet_net_init(struct net *net)
 {
        rwlock_init(&net->packet.sklist_lock);
        INIT_HLIST_HEAD(&net->packet.sklist);
@@ -2092,7 +2459,7 @@ static int packet_net_init(struct net *net)
        return 0;
 }
 
-static void packet_net_exit(struct net *net)
+static void __net_exit packet_net_exit(struct net *net)
 {
        proc_net_remove(net, "packet");
 }