net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *
  44  *              This program is free software; you can redistribute it and/or
  45  *              modify it under the terms of the GNU General Public License
  46  *              as published by the Free Software Foundation; either version
  47  *              2 of the License, or (at your option) any later version.
  48  *
  49  */
  50
  51 #include <linux/types.h>
  52 #include <linux/mm.h>
  53 #include <linux/capability.h>
  54 #include <linux/fcntl.h>
  55 #include <linux/socket.h>
  56 #include <linux/in.h>
  57 #include <linux/inet.h>
  58 #include <linux/netdevice.h>
  59 #include <linux/if_packet.h>
  60 #include <linux/wireless.h>
  61 #include <linux/kernel.h>
  62 #include <linux/kmod.h>
  63 #include <linux/slab.h>
  64 #include <linux/vmalloc.h>
  65 #include <net/net_namespace.h>
  66 #include <net/ip.h>
  67 #include <net/protocol.h>
  68 #include <linux/skbuff.h>
  69 #include <net/sock.h>
  70 #include <linux/errno.h>
  71 #include <linux/timer.h>
  72 #include <asm/system.h>
  73 #include <asm/uaccess.h>
  74 #include <asm/ioctls.h>
  75 #include <asm/page.h>
  76 #include <asm/cacheflush.h>
  77 #include <asm/io.h>
  78 #include <linux/proc_fs.h>
  79 #include <linux/seq_file.h>
  80 #include <linux/poll.h>
  81 #include <linux/module.h>
  82 #include <linux/init.h>
  83 #include <linux/mutex.h>
  84 #include <linux/if_vlan.h>
  85 #include <linux/virtio_net.h>
  86 #include <linux/errqueue.h>
  87 #include <linux/net_tstamp.h>
  88
  89 #ifdef CONFIG_INET
  90 #include <net/inet_common.h>
  91 #endif
  92
  93 /*
  94    Assumptions:
  95    - if device has no dev->hard_header routine, it adds and removes ll header
  96      inside itself. In this case ll header is invisible outside of device,
  97      but higher levels still should reserve dev->hard_header_len.
  98      Some devices are enough clever to reallocate skb, when header
  99      will not fit to reserved space (tunnel), another ones are silly
 100      (PPP).
 101    - packet socket receives packets with pulled ll header,
 102      so that SOCK_RAW should push it back.
 103
 104 On receive:
 105 -----------
 106
 107 Incoming, dev->hard_header!=NULL
 108    mac_header -> ll header
 109    data       -> data
 110
 111 Outgoing, dev->hard_header!=NULL
 112    mac_header -> ll header
 113    data       -> ll header
 114
 115 Incoming, dev->hard_header==NULL
 116    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 117                  header.  PPP makes it, that is wrong, because introduce
 118                  assymetry between rx and tx paths.
 119    data       -> data
 120
 121 Outgoing, dev->hard_header==NULL
 122    mac_header -> data. ll header is still not built!
 123    data       -> data
 124
 125 Resume
 126   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 127
 128
 129 On transmit:
 130 ------------
 131
 132 dev->hard_header != NULL
 133    mac_header -> ll header
 134    data       -> ll header
 135
 136 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 137    mac_header -> data
 138    data       -> data
 139
 140    We should set nh.raw on output to correct posistion,
 141    packet classifier depends on it.
 142  */
 143
 144 /* Private packet socket structures. */
 145
 146 struct packet_mclist {
 147         struct packet_mclist    *next;
 148         int                     ifindex;
 149         int                     count;
 150         unsigned short          type;
 151         unsigned short          alen;
 152         unsigned char           addr[MAX_ADDR_LEN];
 153 };
 154 /* identical to struct packet_mreq except it has
 155  * a longer address field.
 156  */
 157 struct packet_mreq_max {
 158         int             mr_ifindex;
 159         unsigned short  mr_type;
 160         unsigned short  mr_alen;
 161         unsigned char   mr_address[MAX_ADDR_LEN];
 162 };
 163
 164 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
 165                 int closing, int tx_ring);
 166
 167 struct pgv {
 168         char *buffer;
 169 };
 170
 171 struct packet_ring_buffer {
 172         struct pgv              *pg_vec;
 173         unsigned int            head;
 174         unsigned int            frames_per_block;
 175         unsigned int            frame_size;
 176         unsigned int            frame_max;
 177
 178         unsigned int            pg_vec_order;
 179         unsigned int            pg_vec_pages;
 180         unsigned int            pg_vec_len;
 181
 182         atomic_t                pending;
 183 };
 184
 185 struct packet_sock;
 186 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 187
 188 static void packet_flush_mclist(struct sock *sk);
 189
 190 struct packet_sock {
 191         /* struct sock has to be the first member of packet_sock */
 192         struct sock             sk;
 193         struct tpacket_stats    stats;
 194         struct packet_ring_buffer       rx_ring;
 195         struct packet_ring_buffer       tx_ring;
 196         int                     copy_thresh;
 197         spinlock_t              bind_lock;
 198         struct mutex            pg_vec_lock;
 199         unsigned int            running:1,      /* prot_hook is attached*/
 200                                 auxdata:1,
 201                                 origdev:1,
 202                                 has_vnet_hdr:1;
 203         int                     ifindex;        /* bound device         */
 204         __be16                  num;
 205         struct packet_mclist    *mclist;
 206         atomic_t                mapped;
 207         enum tpacket_versions   tp_version;
 208         unsigned int            tp_hdrlen;
 209         unsigned int            tp_reserve;
 210         unsigned int            tp_loss:1;
 211         unsigned int            tp_tstamp;
 212         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 213 };
 214
 215 struct packet_skb_cb {
 216         unsigned int origlen;
 217         union {
 218                 struct sockaddr_pkt pkt;
 219                 struct sockaddr_ll ll;
 220         } sa;
 221 };
 222
 223 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 224
 225 static inline __pure struct page *pgv_to_page(void *addr)
 226 {
 227         if (is_vmalloc_addr(addr))
 228                 return vmalloc_to_page(addr);
 229         return virt_to_page(addr);
 230 }
 231
 232 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 233 {
 234         union {
 235                 struct tpacket_hdr *h1;
 236                 struct tpacket2_hdr *h2;
 237                 void *raw;
 238         } h;
 239
 240         h.raw = frame;
 241         switch (po->tp_version) {
 242         case TPACKET_V1:
 243                 h.h1->tp_status = status;
 244                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 245                 break;
 246         case TPACKET_V2:
 247                 h.h2->tp_status = status;
 248                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 249                 break;
 250         default:
 251                 pr_err("TPACKET version not supported\n");
 252                 BUG();
 253         }
 254
 255         smp_wmb();
 256 }
 257
 258 static int __packet_get_status(struct packet_sock *po, void *frame)
 259 {
 260         union {
 261                 struct tpacket_hdr *h1;
 262                 struct tpacket2_hdr *h2;
 263                 void *raw;
 264         } h;
 265
 266         smp_rmb();
 267
 268         h.raw = frame;
 269         switch (po->tp_version) {
 270         case TPACKET_V1:
 271                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 272                 return h.h1->tp_status;
 273         case TPACKET_V2:
 274                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 275                 return h.h2->tp_status;
 276         default:
 277                 pr_err("TPACKET version not supported\n");
 278                 BUG();
 279                 return 0;
 280         }
 281 }
 282
 283 static void *packet_lookup_frame(struct packet_sock *po,
 284                 struct packet_ring_buffer *rb,
 285                 unsigned int position,
 286                 int status)
 287 {
 288         unsigned int pg_vec_pos, frame_offset;
 289         union {
 290                 struct tpacket_hdr *h1;
 291                 struct tpacket2_hdr *h2;
 292                 void *raw;
 293         } h;
 294
 295         pg_vec_pos = position / rb->frames_per_block;
 296         frame_offset = position % rb->frames_per_block;
 297
 298         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 299                 (frame_offset * rb->frame_size);
 300
 301         if (status != __packet_get_status(po, h.raw))
 302                 return NULL;
 303
 304         return h.raw;
 305 }
 306
 307 static inline void *packet_current_frame(struct packet_sock *po,
 308                 struct packet_ring_buffer *rb,
 309                 int status)
 310 {
 311         return packet_lookup_frame(po, rb, rb->head, status);
 312 }
 313
 314 static inline void *packet_previous_frame(struct packet_sock *po,
 315                 struct packet_ring_buffer *rb,
 316                 int status)
 317 {
 318         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
 319         return packet_lookup_frame(po, rb, previous, status);
 320 }
 321
 322 static inline void packet_increment_head(struct packet_ring_buffer *buff)
 323 {
 324         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
 325 }
 326
 327 static inline struct packet_sock *pkt_sk(struct sock *sk)
 328 {
 329         return (struct packet_sock *)sk;
 330 }
 331
 332 static void packet_sock_destruct(struct sock *sk)
 333 {
 334         skb_queue_purge(&sk->sk_error_queue);
 335
 336         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 337         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 338
 339         if (!sock_flag(sk, SOCK_DEAD)) {
 340                 pr_err("Attempt to release alive packet socket: %p\n", sk);
 341                 return;
 342         }
 343
 344         sk_refcnt_debug_dec(sk);
 345 }
 346
 347
 348 static const struct proto_ops packet_ops;
 349
 350 static const struct proto_ops packet_ops_spkt;
 351
 352 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
 353                            struct packet_type *pt, struct net_device *orig_dev)
 354 {
 355         struct sock *sk;
 356         struct sockaddr_pkt *spkt;
 357
 358         /*
 359          *      When we registered the protocol we saved the socket in the data
 360          *      field for just this event.
 361          */
 362
 363         sk = pt->af_packet_priv;
 364
 365         /*
 366          *      Yank back the headers [hope the device set this
 367          *      right or kerboom...]
 368          *
 369          *      Incoming packets have ll header pulled,
 370          *      push it back.
 371          *
 372          *      For outgoing ones skb->data == skb_mac_header(skb)
 373          *      so that this procedure is noop.
 374          */
 375
 376         if (skb->pkt_type == PACKET_LOOPBACK)
 377                 goto out;
 378
 379         if (!net_eq(dev_net(dev), sock_net(sk)))
 380                 goto out;
 381
 382         skb = skb_share_check(skb, GFP_ATOMIC);
 383         if (skb == NULL)
 384                 goto oom;
 385
 386         /* drop any routing info */
 387         skb_dst_drop(skb);
 388
 389         /* drop conntrack reference */
 390         nf_reset(skb);
 391
 392         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
 393
 394         skb_push(skb, skb->data - skb_mac_header(skb));
 395
 396         /*
 397          *      The SOCK_PACKET socket receives _all_ frames.
 398          */
 399
 400         spkt->spkt_family = dev->type;
 401         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
 402         spkt->spkt_protocol = skb->protocol;
 403
 404         /*
 405          *      Charge the memory to the socket. This is done specifically
 406          *      to prevent sockets using all the memory up.
 407          */
 408
 409         if (sock_queue_rcv_skb(sk, skb) == 0)
 410                 return 0;
 411
 412 out:
 413         kfree_skb(skb);
 414 oom:
 415         return 0;
 416 }
 417
 418
 419 /*
 420  *      Output a raw packet to a device layer. This bypasses all the other
 421  *      protocol layers and you must therefore supply it with a complete frame
 422  */
 423
 424 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
 425                                struct msghdr *msg, size_t len)
 426 {
 427         struct sock *sk = sock->sk;
 428         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
 429         struct sk_buff *skb = NULL;
 430         struct net_device *dev;
 431         __be16 proto = 0;
 432         int err;
 433
 434         /*
 435          *      Get and verify the address.
 436          */
 437
 438         if (saddr) {
 439                 if (msg->msg_namelen < sizeof(struct sockaddr))
 440                         return -EINVAL;
 441                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
 442                         proto = saddr->spkt_protocol;
 443         } else
 444                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
 445
 446         /*
 447          *      Find the device first to size check it
 448          */
 449
 450         saddr->spkt_device[13] = 0;
 451 retry:
 452         rcu_read_lock();
 453         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
 454         err = -ENODEV;
 455         if (dev == NULL)
 456                 goto out_unlock;
 457
 458         err = -ENETDOWN;
 459         if (!(dev->flags & IFF_UP))
 460                 goto out_unlock;
 461
 462         /*
 463          * You may not queue a frame bigger than the mtu. This is the lowest level
 464          * raw protocol and you must do your own fragmentation at this level.
 465          */
 466
 467         err = -EMSGSIZE;
 468         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
 469                 goto out_unlock;
 470
 471         if (!skb) {
 472                 size_t reserved = LL_RESERVED_SPACE(dev);
 473                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
 474
 475                 rcu_read_unlock();
 476                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
 477                 if (skb == NULL)
 478                         return -ENOBUFS;
 479                 /* FIXME: Save some space for broken drivers that write a hard
 480                  * header at transmission time by themselves. PPP is the notable
 481                  * one here. This should really be fixed at the driver level.
 482                  */
 483                 skb_reserve(skb, reserved);
 484                 skb_reset_network_header(skb);
 485
 486                 /* Try to align data part correctly */
 487                 if (hhlen) {
 488                         skb->data -= hhlen;
 489                         skb->tail -= hhlen;
 490                         if (len < hhlen)
 491                                 skb_reset_network_header(skb);
 492                 }
 493                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
 494                 if (err)
 495                         goto out_free;
 496                 goto retry;
 497         }
 498
 499         if (len > (dev->mtu + dev->hard_header_len)) {
 500                 /* Earlier code assumed this would be a VLAN pkt,
 501                  * double-check this now that we have the actual
 502                  * packet in hand.
 503                  */
 504                 struct ethhdr *ehdr;
 505                 skb_reset_mac_header(skb);
 506                 ehdr = eth_hdr(skb);
 507                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
 508                         err = -EMSGSIZE;
 509                         goto out_unlock;
 510                 }
 511         }
 512
 513         skb->protocol = proto;
 514         skb->dev = dev;
 515         skb->priority = sk->sk_priority;
 516         skb->mark = sk->sk_mark;
 517         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
 518         if (err < 0)
 519                 goto out_unlock;
 520
 521         dev_queue_xmit(skb);
 522         rcu_read_unlock();
 523         return len;
 524
 525 out_unlock:
 526         rcu_read_unlock();
 527 out_free:
 528         kfree_skb(skb);
 529         return err;
 530 }
 531
 532 static inline unsigned int run_filter(const struct sk_buff *skb,
 533                                       const struct sock *sk,
 534                                       unsigned int res)
 535 {
 536         struct sk_filter *filter;
 537
 538         rcu_read_lock();
 539         filter = rcu_dereference(sk->sk_filter);
 540         if (filter != NULL)
 541                 res = SK_RUN_FILTER(filter, skb);
 542         rcu_read_unlock();
 543
 544         return res;
 545 }
 546
 547 /*
 548  * This function makes lazy skb cloning in hope that most of packets
 549  * are discarded by BPF.
 550  *
 551  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
 552  * and skb->cb are mangled. It works because (and until) packets
 553  * falling here are owned by current CPU. Output packets are cloned
 554  * by dev_queue_xmit_nit(), input packets are processed by net_bh
 555  * sequencially, so that if we return skb to original state on exit,
 556  * we will not harm anyone.
 557  */
 558
 559 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
 560                       struct packet_type *pt, struct net_device *orig_dev)
 561 {
 562         struct sock *sk;
 563         struct sockaddr_ll *sll;
 564         struct packet_sock *po;
 565         u8 *skb_head = skb->data;
 566         int skb_len = skb->len;
 567         unsigned int snaplen, res;
 568
 569         if (skb->pkt_type == PACKET_LOOPBACK)
 570                 goto drop;
 571
 572         sk = pt->af_packet_priv;
 573         po = pkt_sk(sk);
 574
 575         if (!net_eq(dev_net(dev), sock_net(sk)))
 576                 goto drop;
 577
 578         skb->dev = dev;
 579
 580         if (dev->header_ops) {
 581                 /* The device has an explicit notion of ll header,
 582                  * exported to higher levels.
 583                  *
 584                  * Otherwise, the device hides details of its frame
 585                  * structure, so that corresponding packet head is
 586                  * never delivered to user.
 587                  */
 588                 if (sk->sk_type != SOCK_DGRAM)
 589                         skb_push(skb, skb->data - skb_mac_header(skb));
 590                 else if (skb->pkt_type == PACKET_OUTGOING) {
 591                         /* Special case: outgoing packets have ll header at head */
 592                         skb_pull(skb, skb_network_offset(skb));
 593                 }
 594         }
 595
 596         snaplen = skb->len;
 597
 598         res = run_filter(skb, sk, snaplen);
 599         if (!res)
 600                 goto drop_n_restore;
 601         if (snaplen > res)
 602                 snaplen = res;
 603
 604         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
 605             (unsigned)sk->sk_rcvbuf)
 606                 goto drop_n_acct;
 607
 608         if (skb_shared(skb)) {
 609                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
 610                 if (nskb == NULL)
 611                         goto drop_n_acct;
 612
 613                 if (skb_head != skb->data) {
 614                         skb->data = skb_head;
 615                         skb->len = skb_len;
 616                 }
 617                 kfree_skb(skb);
 618                 skb = nskb;
 619         }
 620
 621         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
 622                      sizeof(skb->cb));
 623
 624         sll = &PACKET_SKB_CB(skb)->sa.ll;
 625         sll->sll_family = AF_PACKET;
 626         sll->sll_hatype = dev->type;
 627         sll->sll_protocol = skb->protocol;
 628         sll->sll_pkttype = skb->pkt_type;
 629         if (unlikely(po->origdev))
 630                 sll->sll_ifindex = orig_dev->ifindex;
 631         else
 632                 sll->sll_ifindex = dev->ifindex;
 633
 634         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
 635
 636         PACKET_SKB_CB(skb)->origlen = skb->len;
 637
 638         if (pskb_trim(skb, snaplen))
 639                 goto drop_n_acct;
 640
 641         skb_set_owner_r(skb, sk);
 642         skb->dev = NULL;
 643         skb_dst_drop(skb);
 644
 645         /* drop conntrack reference */
 646         nf_reset(skb);
 647
 648         spin_lock(&sk->sk_receive_queue.lock);
 649         po->stats.tp_packets++;
 650         skb->dropcount = atomic_read(&sk->sk_drops);
 651         __skb_queue_tail(&sk->sk_receive_queue, skb);
 652         spin_unlock(&sk->sk_receive_queue.lock);
 653         sk->sk_data_ready(sk, skb->len);
 654         return 0;
 655
 656 drop_n_acct:
 657         spin_lock(&sk->sk_receive_queue.lock);
 658         po->stats.tp_drops++;
 659         atomic_inc(&sk->sk_drops);
 660         spin_unlock(&sk->sk_receive_queue.lock);
 661
 662 drop_n_restore:
 663         if (skb_head != skb->data && skb_shared(skb)) {
 664                 skb->data = skb_head;
 665                 skb->len = skb_len;
 666         }
 667 drop:
 668         consume_skb(skb);
 669         return 0;
 670 }
 671
 672 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 673                        struct packet_type *pt, struct net_device *orig_dev)
 674 {
 675         struct sock *sk;
 676         struct packet_sock *po;
 677         struct sockaddr_ll *sll;
 678         union {
 679                 struct tpacket_hdr *h1;
 680                 struct tpacket2_hdr *h2;
 681                 void *raw;
 682         } h;
 683         u8 *skb_head = skb->data;
 684         int skb_len = skb->len;
 685         unsigned int snaplen, res;
 686         unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
 687         unsigned short macoff, netoff, hdrlen;
 688         struct sk_buff *copy_skb = NULL;
 689         struct timeval tv;
 690         struct timespec ts;
 691         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 692
 693         if (skb->pkt_type == PACKET_LOOPBACK)
 694                 goto drop;
 695
 696         sk = pt->af_packet_priv;
 697         po = pkt_sk(sk);
 698
 699         if (!net_eq(dev_net(dev), sock_net(sk)))
 700                 goto drop;
 701
 702         if (dev->header_ops) {
 703                 if (sk->sk_type != SOCK_DGRAM)
 704                         skb_push(skb, skb->data - skb_mac_header(skb));
 705                 else if (skb->pkt_type == PACKET_OUTGOING) {
 706                         /* Special case: outgoing packets have ll header at head */
 707                         skb_pull(skb, skb_network_offset(skb));
 708                 }
 709         }
 710
 711         if (skb->ip_summed == CHECKSUM_PARTIAL)
 712                 status |= TP_STATUS_CSUMNOTREADY;
 713
 714         snaplen = skb->len;
 715
 716         res = run_filter(skb, sk, snaplen);
 717         if (!res)
 718                 goto drop_n_restore;
 719         if (snaplen > res)
 720                 snaplen = res;
 721
 722         if (sk->sk_type == SOCK_DGRAM) {
 723                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
 724                                   po->tp_reserve;
 725         } else {
 726                 unsigned maclen = skb_network_offset(skb);
 727                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
 728                                        (maclen < 16 ? 16 : maclen)) +
 729                         po->tp_reserve;
 730                 macoff = netoff - maclen;
 731         }
 732
 733         if (macoff + snaplen > po->rx_ring.frame_size) {
 734                 if (po->copy_thresh &&
 735                     atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
 736                     (unsigned)sk->sk_rcvbuf) {
 737                         if (skb_shared(skb)) {
 738                                 copy_skb = skb_clone(skb, GFP_ATOMIC);
 739                         } else {
 740                                 copy_skb = skb_get(skb);
 741                                 skb_head = skb->data;
 742                         }
 743                         if (copy_skb)
 744                                 skb_set_owner_r(copy_skb, sk);
 745                 }
 746                 snaplen = po->rx_ring.frame_size - macoff;
 747                 if ((int)snaplen < 0)
 748                         snaplen = 0;
 749         }
 750
 751         spin_lock(&sk->sk_receive_queue.lock);
 752         h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
 753         if (!h.raw)
 754                 goto ring_is_full;
 755         packet_increment_head(&po->rx_ring);
 756         po->stats.tp_packets++;
 757         if (copy_skb) {
 758                 status |= TP_STATUS_COPY;
 759                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
 760         }
 761         if (!po->stats.tp_drops)
 762                 status &= ~TP_STATUS_LOSING;
 763         spin_unlock(&sk->sk_receive_queue.lock);
 764
 765         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
 766
 767         switch (po->tp_version) {
 768         case TPACKET_V1:
 769                 h.h1->tp_len = skb->len;
 770                 h.h1->tp_snaplen = snaplen;
 771                 h.h1->tp_mac = macoff;
 772                 h.h1->tp_net = netoff;
 773                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
 774                                 && shhwtstamps->syststamp.tv64)
 775                         tv = ktime_to_timeval(shhwtstamps->syststamp);
 776                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
 777                                 && shhwtstamps->hwtstamp.tv64)
 778                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
 779                 else if (skb->tstamp.tv64)
 780                         tv = ktime_to_timeval(skb->tstamp);
 781                 else
 782                         do_gettimeofday(&tv);
 783                 h.h1->tp_sec = tv.tv_sec;
 784                 h.h1->tp_usec = tv.tv_usec;
 785                 hdrlen = sizeof(*h.h1);
 786                 break;
 787         case TPACKET_V2:
 788                 h.h2->tp_len = skb->len;
 789                 h.h2->tp_snaplen = snaplen;
 790                 h.h2->tp_mac = macoff;
 791                 h.h2->tp_net = netoff;
 792                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
 793                                 && shhwtstamps->syststamp.tv64)
 794                         ts = ktime_to_timespec(shhwtstamps->syststamp);
 795                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
 796                                 && shhwtstamps->hwtstamp.tv64)
 797                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
 798                 else if (skb->tstamp.tv64)
 799                         ts = ktime_to_timespec(skb->tstamp);
 800                 else
 801                         getnstimeofday(&ts);
 802                 h.h2->tp_sec = ts.tv_sec;
 803                 h.h2->tp_nsec = ts.tv_nsec;
 804                 if (vlan_tx_tag_present(skb)) {
 805                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
 806                         status |= TP_STATUS_VLAN_VALID;
 807                 } else {
 808                         h.h2->tp_vlan_tci = 0;
 809                 }
 810                 h.h2->tp_padding = 0;
 811                 hdrlen = sizeof(*h.h2);
 812                 break;
 813         default:
 814                 BUG();
 815         }
 816
 817         sll = h.raw + TPACKET_ALIGN(hdrlen);
 818         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
 819         sll->sll_family = AF_PACKET;
 820         sll->sll_hatype = dev->type;
 821         sll->sll_protocol = skb->protocol;
 822         sll->sll_pkttype = skb->pkt_type;
 823         if (unlikely(po->origdev))
 824                 sll->sll_ifindex = orig_dev->ifindex;
 825         else
 826                 sll->sll_ifindex = dev->ifindex;
 827
 828         __packet_set_status(po, h.raw, status);
 829         smp_mb();
 830 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 831         {
 832                 u8 *start, *end;
 833
 834                 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw + macoff + snaplen);
 835                 for (start = h.raw; start < end; start += PAGE_SIZE)
 836                         flush_dcache_page(pgv_to_page(start));
 837         }
 838 #endif
 839
 840         sk->sk_data_ready(sk, 0);
 841
 842 drop_n_restore:
 843         if (skb_head != skb->data && skb_shared(skb)) {
 844                 skb->data = skb_head;
 845                 skb->len = skb_len;
 846         }
 847 drop:
 848         kfree_skb(skb);
 849         return 0;
 850
 851 ring_is_full:
 852         po->stats.tp_drops++;
 853         spin_unlock(&sk->sk_receive_queue.lock);
 854
 855         sk->sk_data_ready(sk, 0);
 856         kfree_skb(copy_skb);
 857         goto drop_n_restore;
 858 }
 859
 860 static void tpacket_destruct_skb(struct sk_buff *skb)
 861 {
 862         struct packet_sock *po = pkt_sk(skb->sk);
 863         void *ph;
 864
 865         BUG_ON(skb == NULL);
 866
 867         if (likely(po->tx_ring.pg_vec)) {
 868                 ph = skb_shinfo(skb)->destructor_arg;
 869                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
 870                 atomic_dec(&po->tx_ring.pending);
 871                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
 872         }
 873
 874         sock_wfree(skb);
 875 }
 876
 877 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
 878                 void *frame, struct net_device *dev, int size_max,
 879                 __be16 proto, unsigned char *addr)
 880 {
 881         union {
 882                 struct tpacket_hdr *h1;
 883                 struct tpacket2_hdr *h2;
 884                 void *raw;
 885         } ph;
 886         int to_write, offset, len, tp_len, nr_frags, len_max;
 887         struct socket *sock = po->sk.sk_socket;
 888         struct page *page;
 889         void *data;
 890         int err;
 891
 892         ph.raw = frame;
 893
 894         skb->protocol = proto;
 895         skb->dev = dev;
 896         skb->priority = po->sk.sk_priority;
 897         skb->mark = po->sk.sk_mark;
 898         skb_shinfo(skb)->destructor_arg = ph.raw;
 899
 900         switch (po->tp_version) {
 901         case TPACKET_V2:
 902                 tp_len = ph.h2->tp_len;
 903                 break;
 904         default:
 905                 tp_len = ph.h1->tp_len;
 906                 break;
 907         }
 908         if (unlikely(tp_len > size_max)) {
 909                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
 910                 return -EMSGSIZE;
 911         }
 912
 913         skb_reserve(skb, LL_RESERVED_SPACE(dev));
 914         skb_reset_network_header(skb);
 915
 916         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
 917         to_write = tp_len;
 918
 919         if (sock->type == SOCK_DGRAM) {
 920                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
 921                                 NULL, tp_len);
 922                 if (unlikely(err < 0))
 923                         return -EINVAL;
 924         } else if (dev->hard_header_len) {
 925                 /* net device doesn't like empty head */
 926                 if (unlikely(tp_len <= dev->hard_header_len)) {
 927                         pr_err("packet size is too short (%d < %d)\n",
 928                                tp_len, dev->hard_header_len);
 929                         return -EINVAL;
 930                 }
 931
 932                 skb_push(skb, dev->hard_header_len);
 933                 err = skb_store_bits(skb, 0, data,
 934                                 dev->hard_header_len);
 935                 if (unlikely(err))
 936                         return err;
 937
 938                 data += dev->hard_header_len;
 939                 to_write -= dev->hard_header_len;
 940         }
 941
 942         err = -EFAULT;
 943         offset = offset_in_page(data);
 944         len_max = PAGE_SIZE - offset;
 945         len = ((to_write > len_max) ? len_max : to_write);
 946
 947         skb->data_len = to_write;
 948         skb->len += to_write;
 949         skb->truesize += to_write;
 950         atomic_add(to_write, &po->sk.sk_wmem_alloc);
 951
 952         while (likely(to_write)) {
 953                 nr_frags = skb_shinfo(skb)->nr_frags;
 954
 955                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
 956                         pr_err("Packet exceed the number of skb frags(%lu)\n",
 957                                MAX_SKB_FRAGS);
 958                         return -EFAULT;
 959                 }
 960
 961                 page = pgv_to_page(data);
 962                 data += len;
 963                 flush_dcache_page(page);
 964                 get_page(page);
 965                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
 966                 to_write -= len;
 967                 offset = 0;
 968                 len_max = PAGE_SIZE;
 969                 len = ((to_write > len_max) ? len_max : to_write);
 970         }
 971
 972         return tp_len;
 973 }
 974
 975 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
 976 {
 977         struct sk_buff *skb;
 978         struct net_device *dev;
 979         __be16 proto;
 980         int ifindex, err, reserve = 0;
 981         void *ph;
 982         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
 983         int tp_len, size_max;
 984         unsigned char *addr;
 985         int len_sum = 0;
 986         int status = 0;
 987
 988         mutex_lock(&po->pg_vec_lock);
 989
 990         err = -EBUSY;
 991         if (saddr == NULL) {
 992                 ifindex = po->ifindex;
 993                 proto   = po->num;
 994                 addr    = NULL;
 995         } else {
 996                 err = -EINVAL;
 997                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
 998                         goto out;
 999                 if (msg->msg_namelen < (saddr->sll_halen
1000                                         + offsetof(struct sockaddr_ll,
1001                                                 sll_addr)))
1002                         goto out;
1003                 ifindex = saddr->sll_ifindex;
1004                 proto   = saddr->sll_protocol;
1005                 addr    = saddr->sll_addr;
1006         }
1007
1008         dev = dev_get_by_index(sock_net(&po->sk), ifindex);
1009         err = -ENXIO;
1010         if (unlikely(dev == NULL))
1011                 goto out;
1012
1013         reserve = dev->hard_header_len;
1014
1015         err = -ENETDOWN;
1016         if (unlikely(!(dev->flags & IFF_UP)))
1017                 goto out_put;
1018
1019         size_max = po->tx_ring.frame_size
1020                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
1021
1022         if (size_max > dev->mtu + reserve)
1023                 size_max = dev->mtu + reserve;
1024
1025         do {
1026                 ph = packet_current_frame(po, &po->tx_ring,
1027                                 TP_STATUS_SEND_REQUEST);
1028
1029                 if (unlikely(ph == NULL)) {
1030                         schedule();
1031                         continue;
1032                 }
1033
1034                 status = TP_STATUS_SEND_REQUEST;
1035                 skb = sock_alloc_send_skb(&po->sk,
1036                                 LL_ALLOCATED_SPACE(dev)
1037                                 + sizeof(struct sockaddr_ll),
1038                                 0, &err);
1039
1040                 if (unlikely(skb == NULL))
1041                         goto out_status;
1042
1043                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
1044                                 addr);
1045
1046                 if (unlikely(tp_len < 0)) {
1047                         if (po->tp_loss) {
1048                                 __packet_set_status(po, ph,
1049                                                 TP_STATUS_AVAILABLE);
1050                                 packet_increment_head(&po->tx_ring);
1051                                 kfree_skb(skb);
1052                                 continue;
1053                         } else {
1054                                 status = TP_STATUS_WRONG_FORMAT;
1055                                 err = tp_len;
1056                                 goto out_status;
1057                         }
1058                 }
1059
1060                 skb->destructor = tpacket_destruct_skb;
1061                 __packet_set_status(po, ph, TP_STATUS_SENDING);
1062                 atomic_inc(&po->tx_ring.pending);
1063
1064                 status = TP_STATUS_SEND_REQUEST;
1065                 err = dev_queue_xmit(skb);
1066                 if (unlikely(err > 0)) {
1067                         err = net_xmit_errno(err);
1068                         if (err && __packet_get_status(po, ph) ==
1069                                    TP_STATUS_AVAILABLE) {
1070                                 /* skb was destructed already */
1071                                 skb = NULL;
1072                                 goto out_status;
1073                         }
1074                         /*
1075                          * skb was dropped but not destructed yet;
1076                          * let's treat it like congestion or err < 0
1077                          */
1078                         err = 0;
1079                 }
1080                 packet_increment_head(&po->tx_ring);
1081                 len_sum += tp_len;
1082         } while (likely((ph != NULL) ||
1083                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
1084                          (atomic_read(&po->tx_ring.pending))))
1085                 );
1086
1087         err = len_sum;
1088         goto out_put;
1089
1090 out_status:
1091         __packet_set_status(po, ph, status);
1092         kfree_skb(skb);
1093 out_put:
1094         dev_put(dev);
1095 out:
1096         mutex_unlock(&po->pg_vec_lock);
1097         return err;
1098 }
1099
1100 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
1101                                                size_t reserve, size_t len,
1102                                                size_t linear, int noblock,
1103                                                int *err)
1104 {
1105         struct sk_buff *skb;
1106
1107         /* Under a page?  Don't bother with paged skb. */
1108         if (prepad + len < PAGE_SIZE || !linear)
1109                 linear = len;
1110
1111         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1112                                    err);
1113         if (!skb)
1114                 return NULL;
1115
1116         skb_reserve(skb, reserve);
1117         skb_put(skb, linear);
1118         skb->data_len = len - linear;
1119         skb->len += len - linear;
1120
1121         return skb;
1122 }
1123
1124 static int packet_snd(struct socket *sock,
1125                           struct msghdr *msg, size_t len)
1126 {
1127         struct sock *sk = sock->sk;
1128         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1129         struct sk_buff *skb;
1130         struct net_device *dev;
1131         __be16 proto;
1132         unsigned char *addr;
1133         int ifindex, err, reserve = 0;
1134         struct virtio_net_hdr vnet_hdr = { 0 };
1135         int offset = 0;
1136         int vnet_hdr_len;
1137         struct packet_sock *po = pkt_sk(sk);
1138         unsigned short gso_type = 0;
1139
1140         /*
1141          *      Get and verify the address.
1142          */
1143
1144         if (saddr == NULL) {
1145                 ifindex = po->ifindex;
1146                 proto   = po->num;
1147                 addr    = NULL;
1148         } else {
1149                 err = -EINVAL;
1150                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1151                         goto out;
1152                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1153                         goto out;
1154                 ifindex = saddr->sll_ifindex;
1155                 proto   = saddr->sll_protocol;
1156                 addr    = saddr->sll_addr;
1157         }
1158
1159
1160         dev = dev_get_by_index(sock_net(sk), ifindex);
1161         err = -ENXIO;
1162         if (dev == NULL)
1163                 goto out_unlock;
1164         if (sock->type == SOCK_RAW)
1165                 reserve = dev->hard_header_len;
1166
1167         err = -ENETDOWN;
1168         if (!(dev->flags & IFF_UP))
1169                 goto out_unlock;
1170
1171         if (po->has_vnet_hdr) {
1172                 vnet_hdr_len = sizeof(vnet_hdr);
1173
1174                 err = -EINVAL;
1175                 if (len < vnet_hdr_len)
1176                         goto out_unlock;
1177
1178                 len -= vnet_hdr_len;
1179
1180                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
1181                                        vnet_hdr_len);
1182                 if (err < 0)
1183                         goto out_unlock;
1184
1185                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1186                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
1187                       vnet_hdr.hdr_len))
1188                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
1189                                                  vnet_hdr.csum_offset + 2;
1190
1191                 err = -EINVAL;
1192                 if (vnet_hdr.hdr_len > len)
1193                         goto out_unlock;
1194
1195                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1196                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1197                         case VIRTIO_NET_HDR_GSO_TCPV4:
1198                                 gso_type = SKB_GSO_TCPV4;
1199                                 break;
1200                         case VIRTIO_NET_HDR_GSO_TCPV6:
1201                                 gso_type = SKB_GSO_TCPV6;
1202                                 break;
1203                         case VIRTIO_NET_HDR_GSO_UDP:
1204                                 gso_type = SKB_GSO_UDP;
1205                                 break;
1206                         default:
1207                                 goto out_unlock;
1208                         }
1209
1210                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1211                                 gso_type |= SKB_GSO_TCP_ECN;
1212
1213                         if (vnet_hdr.gso_size == 0)
1214                                 goto out_unlock;
1215
1216                 }
1217         }
1218
1219         err = -EMSGSIZE;
1220         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
1221                 goto out_unlock;
1222
1223         err = -ENOBUFS;
1224         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
1225                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
1226                                msg->msg_flags & MSG_DONTWAIT, &err);
1227         if (skb == NULL)
1228                 goto out_unlock;
1229
1230         skb_set_network_header(skb, reserve);
1231
1232         err = -EINVAL;
1233         if (sock->type == SOCK_DGRAM &&
1234             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
1235                 goto out_free;
1236
1237         /* Returns -EFAULT on error */
1238         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1239         if (err)
1240                 goto out_free;
1241         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1242         if (err < 0)
1243                 goto out_free;
1244
1245         if (!gso_type && (len > dev->mtu + reserve)) {
1246                 /* Earlier code assumed this would be a VLAN pkt,
1247                  * double-check this now that we have the actual
1248                  * packet in hand.
1249                  */
1250                 struct ethhdr *ehdr;
1251                 skb_reset_mac_header(skb);
1252                 ehdr = eth_hdr(skb);
1253                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1254                         err = -EMSGSIZE;
1255                         goto out_free;
1256                 }
1257         }
1258
1259         skb->protocol = proto;
1260         skb->dev = dev;
1261         skb->priority = sk->sk_priority;
1262         skb->mark = sk->sk_mark;
1263
1264         if (po->has_vnet_hdr) {
1265                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1266                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
1267                                                   vnet_hdr.csum_offset)) {
1268                                 err = -EINVAL;
1269                                 goto out_free;
1270                         }
1271                 }
1272
1273                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
1274                 skb_shinfo(skb)->gso_type = gso_type;
1275
1276                 /* Header must be checked, and gso_segs computed. */
1277                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1278                 skb_shinfo(skb)->gso_segs = 0;
1279
1280                 len += vnet_hdr_len;
1281         }
1282
1283         /*
1284          *      Now send it
1285          */
1286
1287         err = dev_queue_xmit(skb);
1288         if (err > 0 && (err = net_xmit_errno(err)) != 0)
1289                 goto out_unlock;
1290
1291         dev_put(dev);
1292
1293         return len;
1294
1295 out_free:
1296         kfree_skb(skb);
1297 out_unlock:
1298         if (dev)
1299                 dev_put(dev);
1300 out:
1301         return err;
1302 }
1303
1304 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1305                 struct msghdr *msg, size_t len)
1306 {
1307         struct sock *sk = sock->sk;
1308         struct packet_sock *po = pkt_sk(sk);
1309         if (po->tx_ring.pg_vec)
1310                 return tpacket_snd(po, msg);
1311         else
1312                 return packet_snd(sock, msg, len);
1313 }
1314
1315 /*
1316  *      Close a PACKET socket. This is fairly simple. We immediately go
1317  *      to 'closed' state and remove our protocol entry in the device list.
1318  */
1319
1320 static int packet_release(struct socket *sock)
1321 {
1322         struct sock *sk = sock->sk;
1323         struct packet_sock *po;
1324         struct net *net;
1325         struct tpacket_req req;
1326
1327         if (!sk)
1328                 return 0;
1329
1330         net = sock_net(sk);
1331         po = pkt_sk(sk);
1332
1333         spin_lock_bh(&net->packet.sklist_lock);
1334         sk_del_node_init_rcu(sk);
1335         sock_prot_inuse_add(net, sk->sk_prot, -1);
1336         spin_unlock_bh(&net->packet.sklist_lock);
1337
1338         spin_lock(&po->bind_lock);
1339         if (po->running) {
1340                 /*
1341                  * Remove from protocol table
1342                  */
1343                 po->running = 0;
1344                 po->num = 0;
1345                 __dev_remove_pack(&po->prot_hook);
1346                 __sock_put(sk);
1347         }
1348         spin_unlock(&po->bind_lock);
1349
1350         packet_flush_mclist(sk);
1351
1352         if (po->rx_ring.pg_vec) {
1353                 memset(&req, 0, sizeof(req));
1354                 packet_set_ring(sk, &req, 1, 0);
1355         }
1356
1357         if (po->tx_ring.pg_vec) {
1358                 memset(&req, 0, sizeof(req));
1359                 packet_set_ring(sk, &req, 1, 1);
1360         }
1361
1362         synchronize_net();
1363         /*
1364          *      Now the socket is dead. No more input will appear.
1365          */
1366         sock_orphan(sk);
1367         sock->sk = NULL;
1368
1369         /* Purge queues */
1370
1371         skb_queue_purge(&sk->sk_receive_queue);
1372         sk_refcnt_debug_release(sk);
1373
1374         sock_put(sk);
1375         return 0;
1376 }
1377
1378 /*
1379  *      Attach a packet hook.
1380  */
1381
1382 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1383 {
1384         struct packet_sock *po = pkt_sk(sk);
1385         /*
1386          *      Detach an existing hook if present.
1387          */
1388
1389         lock_sock(sk);
1390
1391         spin_lock(&po->bind_lock);
1392         if (po->running) {
1393                 __sock_put(sk);
1394                 po->running = 0;
1395                 po->num = 0;
1396                 spin_unlock(&po->bind_lock);
1397                 dev_remove_pack(&po->prot_hook);
1398                 spin_lock(&po->bind_lock);
1399         }
1400
1401         po->num = protocol;
1402         po->prot_hook.type = protocol;
1403         po->prot_hook.dev = dev;
1404
1405         po->ifindex = dev ? dev->ifindex : 0;
1406
1407         if (protocol == 0)
1408                 goto out_unlock;
1409
1410         if (!dev || (dev->flags & IFF_UP)) {
1411                 dev_add_pack(&po->prot_hook);
1412                 sock_hold(sk);
1413                 po->running = 1;
1414         } else {
1415                 sk->sk_err = ENETDOWN;
1416                 if (!sock_flag(sk, SOCK_DEAD))
1417                         sk->sk_error_report(sk);
1418         }
1419
1420 out_unlock:
1421         spin_unlock(&po->bind_lock);
1422         release_sock(sk);
1423         return 0;
1424 }
1425
1426 /*
1427  *      Bind a packet socket to a device
1428  */
1429
1430 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1431                             int addr_len)
1432 {
1433         struct sock *sk = sock->sk;
1434         char name[15];
1435         struct net_device *dev;
1436         int err = -ENODEV;
1437
1438         /*
1439          *      Check legality
1440          */
1441
1442         if (addr_len != sizeof(struct sockaddr))
1443                 return -EINVAL;
1444         strlcpy(name, uaddr->sa_data, sizeof(name));
1445
1446         dev = dev_get_by_name(sock_net(sk), name);
1447         if (dev) {
1448                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1449                 dev_put(dev);
1450         }
1451         return err;
1452 }
1453
1454 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1455 {
1456         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1457         struct sock *sk = sock->sk;
1458         struct net_device *dev = NULL;
1459         int err;
1460
1461
1462         /*
1463          *      Check legality
1464          */
1465
1466         if (addr_len < sizeof(struct sockaddr_ll))
1467                 return -EINVAL;
1468         if (sll->sll_family != AF_PACKET)
1469                 return -EINVAL;
1470
1471         if (sll->sll_ifindex) {
1472                 err = -ENODEV;
1473                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1474                 if (dev == NULL)
1475                         goto out;
1476         }
1477         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1478         if (dev)
1479                 dev_put(dev);
1480
1481 out:
1482         return err;
1483 }
1484
1485 static struct proto packet_proto = {
1486         .name     = "PACKET",
1487         .owner    = THIS_MODULE,
1488         .obj_size = sizeof(struct packet_sock),
1489 };
1490
1491 /*
1492  *      Create a packet of type SOCK_PACKET.
1493  */
1494
1495 static int packet_create(struct net *net, struct socket *sock, int protocol,
1496                          int kern)
1497 {
1498         struct sock *sk;
1499         struct packet_sock *po;
1500         __be16 proto = (__force __be16)protocol; /* weird, but documented */
1501         int err;
1502
1503         if (!capable(CAP_NET_RAW))
1504                 return -EPERM;
1505         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1506             sock->type != SOCK_PACKET)
1507                 return -ESOCKTNOSUPPORT;
1508
1509         sock->state = SS_UNCONNECTED;
1510
1511         err = -ENOBUFS;
1512         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1513         if (sk == NULL)
1514                 goto out;
1515
1516         sock->ops = &packet_ops;
1517         if (sock->type == SOCK_PACKET)
1518                 sock->ops = &packet_ops_spkt;
1519
1520         sock_init_data(sock, sk);
1521
1522         po = pkt_sk(sk);
1523         sk->sk_family = PF_PACKET;
1524         po->num = proto;
1525
1526         sk->sk_destruct = packet_sock_destruct;
1527         sk_refcnt_debug_inc(sk);
1528
1529         /*
1530          *      Attach a protocol block
1531          */
1532
1533         spin_lock_init(&po->bind_lock);
1534         mutex_init(&po->pg_vec_lock);
1535         po->prot_hook.func = packet_rcv;
1536
1537         if (sock->type == SOCK_PACKET)
1538                 po->prot_hook.func = packet_rcv_spkt;
1539
1540         po->prot_hook.af_packet_priv = sk;
1541
1542         if (proto) {
1543                 po->prot_hook.type = proto;
1544                 dev_add_pack(&po->prot_hook);
1545                 sock_hold(sk);
1546                 po->running = 1;
1547         }
1548
1549         spin_lock_bh(&net->packet.sklist_lock);
1550         sk_add_node_rcu(sk, &net->packet.sklist);
1551         sock_prot_inuse_add(net, &packet_proto, 1);
1552         spin_unlock_bh(&net->packet.sklist_lock);
1553
1554         return 0;
1555 out:
1556         return err;
1557 }
1558
1559 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
1560 {
1561         struct sock_exterr_skb *serr;
1562         struct sk_buff *skb, *skb2;
1563         int copied, err;
1564
1565         err = -EAGAIN;
1566         skb = skb_dequeue(&sk->sk_error_queue);
1567         if (skb == NULL)
1568                 goto out;
1569
1570         copied = skb->len;
1571         if (copied > len) {
1572                 msg->msg_flags |= MSG_TRUNC;
1573                 copied = len;
1574         }
1575         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1576         if (err)
1577                 goto out_free_skb;
1578
1579         sock_recv_timestamp(msg, sk, skb);
1580
1581         serr = SKB_EXT_ERR(skb);
1582         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
1583                  sizeof(serr->ee), &serr->ee);
1584
1585         msg->msg_flags |= MSG_ERRQUEUE;
1586         err = copied;
1587
1588         /* Reset and regenerate socket error */
1589         spin_lock_bh(&sk->sk_error_queue.lock);
1590         sk->sk_err = 0;
1591         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
1592                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
1593                 spin_unlock_bh(&sk->sk_error_queue.lock);
1594                 sk->sk_error_report(sk);
1595         } else
1596                 spin_unlock_bh(&sk->sk_error_queue.lock);
1597
1598 out_free_skb:
1599         kfree_skb(skb);
1600 out:
1601         return err;
1602 }
1603
1604 /*
1605  *      Pull a packet from our receive queue and hand it to the user.
1606  *      If necessary we block.
1607  */
1608
1609 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1610                           struct msghdr *msg, size_t len, int flags)
1611 {
1612         struct sock *sk = sock->sk;
1613         struct sk_buff *skb;
1614         int copied, err;
1615         struct sockaddr_ll *sll;
1616         int vnet_hdr_len = 0;
1617
1618         err = -EINVAL;
1619         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
1620                 goto out;
1621
1622 #if 0
1623         /* What error should we return now? EUNATTACH? */
1624         if (pkt_sk(sk)->ifindex < 0)
1625                 return -ENODEV;
1626 #endif
1627
1628         if (flags & MSG_ERRQUEUE) {
1629                 err = packet_recv_error(sk, msg, len);
1630                 goto out;
1631         }
1632
1633         /*
1634          *      Call the generic datagram receiver. This handles all sorts
1635          *      of horrible races and re-entrancy so we can forget about it
1636          *      in the protocol layers.
1637          *
1638          *      Now it will return ENETDOWN, if device have just gone down,
1639          *      but then it will block.
1640          */
1641
1642         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1643
1644         /*
1645          *      An error occurred so return it. Because skb_recv_datagram()
1646          *      handles the blocking we don't see and worry about blocking
1647          *      retries.
1648          */
1649
1650         if (skb == NULL)
1651                 goto out;
1652
1653         if (pkt_sk(sk)->has_vnet_hdr) {
1654                 struct virtio_net_hdr vnet_hdr = { 0 };
1655
1656                 err = -EINVAL;
1657                 vnet_hdr_len = sizeof(vnet_hdr);
1658                 if (len < vnet_hdr_len)
1659                         goto out_free;
1660
1661                 len -= vnet_hdr_len;
1662
1663                 if (skb_is_gso(skb)) {
1664                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1665
1666                         /* This is a hint as to how much should be linear. */
1667                         vnet_hdr.hdr_len = skb_headlen(skb);
1668                         vnet_hdr.gso_size = sinfo->gso_size;
1669                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1670                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1671                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1672                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1673                         else if (sinfo->gso_type & SKB_GSO_UDP)
1674                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1675                         else if (sinfo->gso_type & SKB_GSO_FCOE)
1676                                 goto out_free;
1677                         else
1678                                 BUG();
1679                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1680                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1681                 } else
1682                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1683
1684                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1685                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1686                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
1687                         vnet_hdr.csum_offset = skb->csum_offset;
1688                 } /* else everything is zero */
1689
1690                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
1691                                      vnet_hdr_len);
1692                 if (err < 0)
1693                         goto out_free;
1694         }
1695
1696         /*
1697          *      If the address length field is there to be filled in, we fill
1698          *      it in now.
1699          */
1700
1701         sll = &PACKET_SKB_CB(skb)->sa.ll;
1702         if (sock->type == SOCK_PACKET)
1703                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1704         else
1705                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1706
1707         /*
1708          *      You lose any data beyond the buffer you gave. If it worries a
1709          *      user program they can ask the device for its MTU anyway.
1710          */
1711
1712         copied = skb->len;
1713         if (copied > len) {
1714                 copied = len;
1715                 msg->msg_flags |= MSG_TRUNC;
1716         }
1717
1718         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1719         if (err)
1720                 goto out_free;
1721
1722         sock_recv_ts_and_drops(msg, sk, skb);
1723
1724         if (msg->msg_name)
1725                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1726                        msg->msg_namelen);
1727
1728         if (pkt_sk(sk)->auxdata) {
1729                 struct tpacket_auxdata aux;
1730
1731                 aux.tp_status = TP_STATUS_USER;
1732                 if (skb->ip_summed == CHECKSUM_PARTIAL)
1733                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1734                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1735                 aux.tp_snaplen = skb->len;
1736                 aux.tp_mac = 0;
1737                 aux.tp_net = skb_network_offset(skb);
1738                 if (vlan_tx_tag_present(skb)) {
1739                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
1740                         aux.tp_status |= TP_STATUS_VLAN_VALID;
1741                 } else {
1742                         aux.tp_vlan_tci = 0;
1743                 }
1744                 aux.tp_padding = 0;
1745                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1746         }
1747
1748         /*
1749          *      Free or return the buffer as appropriate. Again this
1750          *      hides all the races and re-entrancy issues from us.
1751          */
1752         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
1753
1754 out_free:
1755         skb_free_datagram(sk, skb);
1756 out:
1757         return err;
1758 }
1759
1760 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1761                                int *uaddr_len, int peer)
1762 {
1763         struct net_device *dev;
1764         struct sock *sk = sock->sk;
1765
1766         if (peer)
1767                 return -EOPNOTSUPP;
1768
1769         uaddr->sa_family = AF_PACKET;
1770         rcu_read_lock();
1771         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
1772         if (dev)
1773                 strncpy(uaddr->sa_data, dev->name, 14);
1774         else
1775                 memset(uaddr->sa_data, 0, 14);
1776         rcu_read_unlock();
1777         *uaddr_len = sizeof(*uaddr);
1778
1779         return 0;
1780 }
1781
1782 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1783                           int *uaddr_len, int peer)
1784 {
1785         struct net_device *dev;
1786         struct sock *sk = sock->sk;
1787         struct packet_sock *po = pkt_sk(sk);
1788         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
1789
1790         if (peer)
1791                 return -EOPNOTSUPP;
1792
1793         sll->sll_family = AF_PACKET;
1794         sll->sll_ifindex = po->ifindex;
1795         sll->sll_protocol = po->num;
1796         sll->sll_pkttype = 0;
1797         rcu_read_lock();
1798         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
1799         if (dev) {
1800                 sll->sll_hatype = dev->type;
1801                 sll->sll_halen = dev->addr_len;
1802                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1803         } else {
1804                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
1805                 sll->sll_halen = 0;
1806         }
1807         rcu_read_unlock();
1808         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1809
1810         return 0;
1811 }
1812
1813 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1814                          int what)
1815 {
1816         switch (i->type) {
1817         case PACKET_MR_MULTICAST:
1818                 if (i->alen != dev->addr_len)
1819                         return -EINVAL;
1820                 if (what > 0)
1821                         return dev_mc_add(dev, i->addr);
1822                 else
1823                         return dev_mc_del(dev, i->addr);
1824                 break;
1825         case PACKET_MR_PROMISC:
1826                 return dev_set_promiscuity(dev, what);
1827                 break;
1828         case PACKET_MR_ALLMULTI:
1829                 return dev_set_allmulti(dev, what);
1830                 break;
1831         case PACKET_MR_UNICAST:
1832                 if (i->alen != dev->addr_len)
1833                         return -EINVAL;
1834                 if (what > 0)
1835                         return dev_uc_add(dev, i->addr);
1836                 else
1837                         return dev_uc_del(dev, i->addr);
1838                 break;
1839         default:
1840                 break;
1841         }
1842         return 0;
1843 }
1844
1845 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1846 {
1847         for ( ; i; i = i->next) {
1848                 if (i->ifindex == dev->ifindex)
1849                         packet_dev_mc(dev, i, what);
1850         }
1851 }
1852
1853 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1854 {
1855         struct packet_sock *po = pkt_sk(sk);
1856         struct packet_mclist *ml, *i;
1857         struct net_device *dev;
1858         int err;
1859
1860         rtnl_lock();
1861
1862         err = -ENODEV;
1863         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1864         if (!dev)
1865                 goto done;
1866
1867         err = -EINVAL;
1868         if (mreq->mr_alen > dev->addr_len)
1869                 goto done;
1870
1871         err = -ENOBUFS;
1872         i = kmalloc(sizeof(*i), GFP_KERNEL);
1873         if (i == NULL)
1874                 goto done;
1875
1876         err = 0;
1877         for (ml = po->mclist; ml; ml = ml->next) {
1878                 if (ml->ifindex == mreq->mr_ifindex &&
1879                     ml->type == mreq->mr_type &&
1880                     ml->alen == mreq->mr_alen &&
1881                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1882                         ml->count++;
1883                         /* Free the new element ... */
1884                         kfree(i);
1885                         goto done;
1886                 }
1887         }
1888
1889         i->type = mreq->mr_type;
1890         i->ifindex = mreq->mr_ifindex;
1891         i->alen = mreq->mr_alen;
1892         memcpy(i->addr, mreq->mr_address, i->alen);
1893         i->count = 1;
1894         i->next = po->mclist;
1895         po->mclist = i;
1896         err = packet_dev_mc(dev, i, 1);
1897         if (err) {
1898                 po->mclist = i->next;
1899                 kfree(i);
1900         }
1901
1902 done:
1903         rtnl_unlock();
1904         return err;
1905 }
1906
1907 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1908 {
1909         struct packet_mclist *ml, **mlp;
1910
1911         rtnl_lock();
1912
1913         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1914                 if (ml->ifindex == mreq->mr_ifindex &&
1915                     ml->type == mreq->mr_type &&
1916                     ml->alen == mreq->mr_alen &&
1917                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1918                         if (--ml->count == 0) {
1919                                 struct net_device *dev;
1920                                 *mlp = ml->next;
1921                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1922                                 if (dev)
1923                                         packet_dev_mc(dev, ml, -1);
1924                                 kfree(ml);
1925                         }
1926                         rtnl_unlock();
1927                         return 0;
1928                 }
1929         }
1930         rtnl_unlock();
1931         return -EADDRNOTAVAIL;
1932 }
1933
1934 static void packet_flush_mclist(struct sock *sk)
1935 {
1936         struct packet_sock *po = pkt_sk(sk);
1937         struct packet_mclist *ml;
1938
1939         if (!po->mclist)
1940                 return;
1941
1942         rtnl_lock();
1943         while ((ml = po->mclist) != NULL) {
1944                 struct net_device *dev;
1945
1946                 po->mclist = ml->next;
1947                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1948                 if (dev != NULL)
1949                         packet_dev_mc(dev, ml, -1);
1950                 kfree(ml);
1951         }
1952         rtnl_unlock();
1953 }
1954
1955 static int
1956 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1957 {
1958         struct sock *sk = sock->sk;
1959         struct packet_sock *po = pkt_sk(sk);
1960         int ret;
1961
1962         if (level != SOL_PACKET)
1963                 return -ENOPROTOOPT;
1964
1965         switch (optname) {
1966         case PACKET_ADD_MEMBERSHIP:
1967         case PACKET_DROP_MEMBERSHIP:
1968         {
1969                 struct packet_mreq_max mreq;
1970                 int len = optlen;
1971                 memset(&mreq, 0, sizeof(mreq));
1972                 if (len < sizeof(struct packet_mreq))
1973                         return -EINVAL;
1974                 if (len > sizeof(mreq))
1975                         len = sizeof(mreq);
1976                 if (copy_from_user(&mreq, optval, len))
1977                         return -EFAULT;
1978                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1979                         return -EINVAL;
1980                 if (optname == PACKET_ADD_MEMBERSHIP)
1981                         ret = packet_mc_add(sk, &mreq);
1982                 else
1983                         ret = packet_mc_drop(sk, &mreq);
1984                 return ret;
1985         }
1986
1987         case PACKET_RX_RING:
1988         case PACKET_TX_RING:
1989         {
1990                 struct tpacket_req req;
1991
1992                 if (optlen < sizeof(req))
1993                         return -EINVAL;
1994                 if (pkt_sk(sk)->has_vnet_hdr)
1995                         return -EINVAL;
1996                 if (copy_from_user(&req, optval, sizeof(req)))
1997                         return -EFAULT;
1998                 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
1999         }
2000         case PACKET_COPY_THRESH:
2001         {
2002                 int val;
2003
2004                 if (optlen != sizeof(val))
2005                         return -EINVAL;
2006                 if (copy_from_user(&val, optval, sizeof(val)))
2007                         return -EFAULT;
2008
2009                 pkt_sk(sk)->copy_thresh = val;
2010                 return 0;
2011         }
2012         case PACKET_VERSION:
2013         {
2014                 int val;
2015
2016                 if (optlen != sizeof(val))
2017                         return -EINVAL;
2018                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2019                         return -EBUSY;
2020                 if (copy_from_user(&val, optval, sizeof(val)))
2021                         return -EFAULT;
2022                 switch (val) {
2023                 case TPACKET_V1:
2024                 case TPACKET_V2:
2025                         po->tp_version = val;
2026                         return 0;
2027                 default:
2028                         return -EINVAL;
2029                 }
2030         }
2031         case PACKET_RESERVE:
2032         {
2033                 unsigned int val;
2034
2035                 if (optlen != sizeof(val))
2036                         return -EINVAL;
2037                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2038                         return -EBUSY;
2039                 if (copy_from_user(&val, optval, sizeof(val)))
2040                         return -EFAULT;
2041                 po->tp_reserve = val;
2042                 return 0;
2043         }
2044         case PACKET_LOSS:
2045         {
2046                 unsigned int val;
2047
2048                 if (optlen != sizeof(val))
2049                         return -EINVAL;
2050                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2051                         return -EBUSY;
2052                 if (copy_from_user(&val, optval, sizeof(val)))
2053                         return -EFAULT;
2054                 po->tp_loss = !!val;
2055                 return 0;
2056         }
2057         case PACKET_AUXDATA:
2058         {
2059                 int val;
2060
2061                 if (optlen < sizeof(val))
2062                         return -EINVAL;
2063                 if (copy_from_user(&val, optval, sizeof(val)))
2064                         return -EFAULT;
2065
2066                 po->auxdata = !!val;
2067                 return 0;
2068         }
2069         case PACKET_ORIGDEV:
2070         {
2071                 int val;
2072
2073                 if (optlen < sizeof(val))
2074                         return -EINVAL;
2075                 if (copy_from_user(&val, optval, sizeof(val)))
2076                         return -EFAULT;
2077
2078                 po->origdev = !!val;
2079                 return 0;
2080         }
2081         case PACKET_VNET_HDR:
2082         {
2083                 int val;
2084
2085                 if (sock->type != SOCK_RAW)
2086                         return -EINVAL;
2087                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2088                         return -EBUSY;
2089                 if (optlen < sizeof(val))
2090                         return -EINVAL;
2091                 if (copy_from_user(&val, optval, sizeof(val)))
2092                         return -EFAULT;
2093
2094                 po->has_vnet_hdr = !!val;
2095                 return 0;
2096         }
2097         case PACKET_TIMESTAMP:
2098         {
2099                 int val;
2100
2101                 if (optlen != sizeof(val))
2102                         return -EINVAL;
2103                 if (copy_from_user(&val, optval, sizeof(val)))
2104                         return -EFAULT;
2105
2106                 po->tp_tstamp = val;
2107                 return 0;
2108         }
2109         default:
2110                 return -ENOPROTOOPT;
2111         }
2112 }
2113
2114 static int packet_getsockopt(struct socket *sock, int level, int optname,
2115                              char __user *optval, int __user *optlen)
2116 {
2117         int len;
2118         int val;
2119         struct sock *sk = sock->sk;
2120         struct packet_sock *po = pkt_sk(sk);
2121         void *data;
2122         struct tpacket_stats st;
2123
2124         if (level != SOL_PACKET)
2125                 return -ENOPROTOOPT;
2126
2127         if (get_user(len, optlen))
2128                 return -EFAULT;
2129
2130         if (len < 0)
2131                 return -EINVAL;
2132
2133         switch (optname) {
2134         case PACKET_STATISTICS:
2135                 if (len > sizeof(struct tpacket_stats))
2136                         len = sizeof(struct tpacket_stats);
2137                 spin_lock_bh(&sk->sk_receive_queue.lock);
2138                 st = po->stats;
2139                 memset(&po->stats, 0, sizeof(st));
2140                 spin_unlock_bh(&sk->sk_receive_queue.lock);
2141                 st.tp_packets += st.tp_drops;
2142
2143                 data = &st;
2144                 break;
2145         case PACKET_AUXDATA:
2146                 if (len > sizeof(int))
2147                         len = sizeof(int);
2148                 val = po->auxdata;
2149
2150                 data = &val;
2151                 break;
2152         case PACKET_ORIGDEV:
2153                 if (len > sizeof(int))
2154                         len = sizeof(int);
2155                 val = po->origdev;
2156
2157                 data = &val;
2158                 break;
2159         case PACKET_VNET_HDR:
2160                 if (len > sizeof(int))
2161                         len = sizeof(int);
2162                 val = po->has_vnet_hdr;
2163
2164                 data = &val;
2165                 break;
2166         case PACKET_VERSION:
2167                 if (len > sizeof(int))
2168                         len = sizeof(int);
2169                 val = po->tp_version;
2170                 data = &val;
2171                 break;
2172         case PACKET_HDRLEN:
2173                 if (len > sizeof(int))
2174                         len = sizeof(int);
2175                 if (copy_from_user(&val, optval, len))
2176                         return -EFAULT;
2177                 switch (val) {
2178                 case TPACKET_V1:
2179                         val = sizeof(struct tpacket_hdr);
2180                         break;
2181                 case TPACKET_V2:
2182                         val = sizeof(struct tpacket2_hdr);
2183                         break;
2184                 default:
2185                         return -EINVAL;
2186                 }
2187                 data = &val;
2188                 break;
2189         case PACKET_RESERVE:
2190                 if (len > sizeof(unsigned int))
2191                         len = sizeof(unsigned int);
2192                 val = po->tp_reserve;
2193                 data = &val;
2194                 break;
2195         case PACKET_LOSS:
2196                 if (len > sizeof(unsigned int))
2197                         len = sizeof(unsigned int);
2198                 val = po->tp_loss;
2199                 data = &val;
2200                 break;
2201         case PACKET_TIMESTAMP:
2202                 if (len > sizeof(int))
2203                         len = sizeof(int);
2204                 val = po->tp_tstamp;
2205                 data = &val;
2206                 break;
2207         default:
2208                 return -ENOPROTOOPT;
2209         }
2210
2211         if (put_user(len, optlen))
2212                 return -EFAULT;
2213         if (copy_to_user(optval, data, len))
2214                 return -EFAULT;
2215         return 0;
2216 }
2217
2218
2219 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
2220 {
2221         struct sock *sk;
2222         struct hlist_node *node;
2223         struct net_device *dev = data;
2224         struct net *net = dev_net(dev);
2225
2226         rcu_read_lock();
2227         sk_for_each_rcu(sk, node, &net->packet.sklist) {
2228                 struct packet_sock *po = pkt_sk(sk);
2229
2230                 switch (msg) {
2231                 case NETDEV_UNREGISTER:
2232                         if (po->mclist)
2233                                 packet_dev_mclist(dev, po->mclist, -1);
2234                         /* fallthrough */
2235
2236                 case NETDEV_DOWN:
2237                         if (dev->ifindex == po->ifindex) {
2238                                 spin_lock(&po->bind_lock);
2239                                 if (po->running) {
2240                                         __dev_remove_pack(&po->prot_hook);
2241                                         __sock_put(sk);
2242                                         po->running = 0;
2243                                         sk->sk_err = ENETDOWN;
2244                                         if (!sock_flag(sk, SOCK_DEAD))
2245                                                 sk->sk_error_report(sk);
2246                                 }
2247                                 if (msg == NETDEV_UNREGISTER) {
2248                                         po->ifindex = -1;
2249                                         po->prot_hook.dev = NULL;
2250                                 }
2251                                 spin_unlock(&po->bind_lock);
2252                         }
2253                         break;
2254                 case NETDEV_UP:
2255                         if (dev->ifindex == po->ifindex) {
2256                                 spin_lock(&po->bind_lock);
2257                                 if (po->num && !po->running) {
2258                                         dev_add_pack(&po->prot_hook);
2259                                         sock_hold(sk);
2260                                         po->running = 1;
2261                                 }
2262                                 spin_unlock(&po->bind_lock);
2263                         }
2264                         break;
2265                 }
2266         }
2267         rcu_read_unlock();
2268         return NOTIFY_DONE;
2269 }
2270
2271
2272 static int packet_ioctl(struct socket *sock, unsigned int cmd,
2273                         unsigned long arg)
2274 {
2275         struct sock *sk = sock->sk;
2276
2277         switch (cmd) {
2278         case SIOCOUTQ:
2279         {
2280                 int amount = sk_wmem_alloc_get(sk);
2281
2282                 return put_user(amount, (int __user *)arg);
2283         }
2284         case SIOCINQ:
2285         {
2286                 struct sk_buff *skb;
2287                 int amount = 0;
2288
2289                 spin_lock_bh(&sk->sk_receive_queue.lock);
2290                 skb = skb_peek(&sk->sk_receive_queue);
2291                 if (skb)
2292                         amount = skb->len;
2293                 spin_unlock_bh(&sk->sk_receive_queue.lock);
2294                 return put_user(amount, (int __user *)arg);
2295         }
2296         case SIOCGSTAMP:
2297                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2298         case SIOCGSTAMPNS:
2299                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2300
2301 #ifdef CONFIG_INET
2302         case SIOCADDRT:
2303         case SIOCDELRT:
2304         case SIOCDARP:
2305         case SIOCGARP:
2306         case SIOCSARP:
2307         case SIOCGIFADDR:
2308         case SIOCSIFADDR:
2309         case SIOCGIFBRDADDR:
2310         case SIOCSIFBRDADDR:
2311         case SIOCGIFNETMASK:
2312         case SIOCSIFNETMASK:
2313         case SIOCGIFDSTADDR:
2314         case SIOCSIFDSTADDR:
2315         case SIOCSIFFLAGS:
2316                 return inet_dgram_ops.ioctl(sock, cmd, arg);
2317 #endif
2318
2319         default:
2320                 return -ENOIOCTLCMD;
2321         }
2322         return 0;
2323 }
2324
2325 static unsigned int packet_poll(struct file *file, struct socket *sock,
2326                                 poll_table *wait)
2327 {
2328         struct sock *sk = sock->sk;
2329         struct packet_sock *po = pkt_sk(sk);
2330         unsigned int mask = datagram_poll(file, sock, wait);
2331
2332         spin_lock_bh(&sk->sk_receive_queue.lock);
2333         if (po->rx_ring.pg_vec) {
2334                 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2335                         mask |= POLLIN | POLLRDNORM;
2336         }
2337         spin_unlock_bh(&sk->sk_receive_queue.lock);
2338         spin_lock_bh(&sk->sk_write_queue.lock);
2339         if (po->tx_ring.pg_vec) {
2340                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2341                         mask |= POLLOUT | POLLWRNORM;
2342         }
2343         spin_unlock_bh(&sk->sk_write_queue.lock);
2344         return mask;
2345 }
2346
2347
2348 /* Dirty? Well, I still did not learn better way to account
2349  * for user mmaps.
2350  */
2351
2352 static void packet_mm_open(struct vm_area_struct *vma)
2353 {
2354         struct file *file = vma->vm_file;
2355         struct socket *sock = file->private_data;
2356         struct sock *sk = sock->sk;
2357
2358         if (sk)
2359                 atomic_inc(&pkt_sk(sk)->mapped);
2360 }
2361
2362 static void packet_mm_close(struct vm_area_struct *vma)
2363 {
2364         struct file *file = vma->vm_file;
2365         struct socket *sock = file->private_data;
2366         struct sock *sk = sock->sk;
2367
2368         if (sk)
2369                 atomic_dec(&pkt_sk(sk)->mapped);
2370 }
2371
2372 static const struct vm_operations_struct packet_mmap_ops = {
2373         .open   =       packet_mm_open,
2374         .close  =       packet_mm_close,
2375 };
2376
2377 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
2378                         unsigned int len)
2379 {
2380         int i;
2381
2382         for (i = 0; i < len; i++) {
2383                 if (likely(pg_vec[i].buffer)) {
2384                         if (is_vmalloc_addr(pg_vec[i].buffer))
2385                                 vfree(pg_vec[i].buffer);
2386                         else
2387                                 free_pages((unsigned long)pg_vec[i].buffer,
2388                                            order);
2389                         pg_vec[i].buffer = NULL;
2390                 }
2391         }
2392         kfree(pg_vec);
2393 }
2394
2395 static inline char *alloc_one_pg_vec_page(unsigned long order)
2396 {
2397         char *buffer = NULL;
2398         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
2399                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
2400
2401         buffer = (char *) __get_free_pages(gfp_flags, order);
2402
2403         if (buffer)
2404                 return buffer;
2405
2406         /*
2407          * __get_free_pages failed, fall back to vmalloc
2408          */
2409         buffer = vzalloc((1 << order) * PAGE_SIZE);
2410
2411         if (buffer)
2412                 return buffer;
2413
2414         /*
2415          * vmalloc failed, lets dig into swap here
2416          */
2417         gfp_flags &= ~__GFP_NORETRY;
2418         buffer = (char *)__get_free_pages(gfp_flags, order);
2419         if (buffer)
2420                 return buffer;
2421
2422         /*
2423          * complete and utter failure
2424          */
2425         return NULL;
2426 }
2427
2428 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
2429 {
2430         unsigned int block_nr = req->tp_block_nr;
2431         struct pgv *pg_vec;
2432         int i;
2433
2434         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
2435         if (unlikely(!pg_vec))
2436                 goto out;
2437
2438         for (i = 0; i < block_nr; i++) {
2439                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
2440                 if (unlikely(!pg_vec[i].buffer))
2441                         goto out_free_pgvec;
2442         }
2443
2444 out:
2445         return pg_vec;
2446
2447 out_free_pgvec:
2448         free_pg_vec(pg_vec, order, block_nr);
2449         pg_vec = NULL;
2450         goto out;
2451 }
2452
2453 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2454                 int closing, int tx_ring)
2455 {
2456         struct pgv *pg_vec = NULL;
2457         struct packet_sock *po = pkt_sk(sk);
2458         int was_running, order = 0;
2459         struct packet_ring_buffer *rb;
2460         struct sk_buff_head *rb_queue;
2461         __be16 num;
2462         int err;
2463
2464         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2465         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2466
2467         err = -EBUSY;
2468         if (!closing) {
2469                 if (atomic_read(&po->mapped))
2470                         goto out;
2471                 if (atomic_read(&rb->pending))
2472                         goto out;
2473         }
2474
2475         if (req->tp_block_nr) {
2476                 /* Sanity tests and some calculations */
2477                 err = -EBUSY;
2478                 if (unlikely(rb->pg_vec))
2479                         goto out;
2480
2481                 switch (po->tp_version) {
2482                 case TPACKET_V1:
2483                         po->tp_hdrlen = TPACKET_HDRLEN;
2484                         break;
2485                 case TPACKET_V2:
2486                         po->tp_hdrlen = TPACKET2_HDRLEN;
2487                         break;
2488                 }
2489
2490                 err = -EINVAL;
2491                 if (unlikely((int)req->tp_block_size <= 0))
2492                         goto out;
2493                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2494                         goto out;
2495                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2496                                         po->tp_reserve))
2497                         goto out;
2498                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2499                         goto out;
2500
2501                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2502                 if (unlikely(rb->frames_per_block <= 0))
2503                         goto out;
2504                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2505                                         req->tp_frame_nr))
2506                         goto out;
2507
2508                 err = -ENOMEM;
2509                 order = get_order(req->tp_block_size);
2510                 pg_vec = alloc_pg_vec(req, order);
2511                 if (unlikely(!pg_vec))
2512                         goto out;
2513         }
2514         /* Done */
2515         else {
2516                 err = -EINVAL;
2517                 if (unlikely(req->tp_frame_nr))
2518                         goto out;
2519         }
2520
2521         lock_sock(sk);
2522
2523         /* Detach socket from network */
2524         spin_lock(&po->bind_lock);
2525         was_running = po->running;
2526         num = po->num;
2527         if (was_running) {
2528                 __dev_remove_pack(&po->prot_hook);
2529                 po->num = 0;
2530                 po->running = 0;
2531                 __sock_put(sk);
2532         }
2533         spin_unlock(&po->bind_lock);
2534
2535         synchronize_net();
2536
2537         err = -EBUSY;
2538         mutex_lock(&po->pg_vec_lock);
2539         if (closing || atomic_read(&po->mapped) == 0) {
2540                 err = 0;
2541                 spin_lock_bh(&rb_queue->lock);
2542                 swap(rb->pg_vec, pg_vec);
2543                 rb->frame_max = (req->tp_frame_nr - 1);
2544                 rb->head = 0;
2545                 rb->frame_size = req->tp_frame_size;
2546                 spin_unlock_bh(&rb_queue->lock);
2547
2548                 swap(rb->pg_vec_order, order);
2549                 swap(rb->pg_vec_len, req->tp_block_nr);
2550
2551                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2552                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2553                                                 tpacket_rcv : packet_rcv;
2554                 skb_queue_purge(rb_queue);
2555                 if (atomic_read(&po->mapped))
2556                         pr_err("packet_mmap: vma is busy: %d\n",
2557                                atomic_read(&po->mapped));
2558         }
2559         mutex_unlock(&po->pg_vec_lock);
2560
2561         spin_lock(&po->bind_lock);
2562         if (was_running && !po->running) {
2563                 sock_hold(sk);
2564                 po->running = 1;
2565                 po->num = num;
2566                 dev_add_pack(&po->prot_hook);
2567         }
2568         spin_unlock(&po->bind_lock);
2569
2570         release_sock(sk);
2571
2572         if (pg_vec)
2573                 free_pg_vec(pg_vec, order, req->tp_block_nr);
2574 out:
2575         return err;
2576 }
2577
2578 static int packet_mmap(struct file *file, struct socket *sock,
2579                 struct vm_area_struct *vma)
2580 {
2581         struct sock *sk = sock->sk;
2582         struct packet_sock *po = pkt_sk(sk);
2583         unsigned long size, expected_size;
2584         struct packet_ring_buffer *rb;
2585         unsigned long start;
2586         int err = -EINVAL;
2587         int i;
2588
2589         if (vma->vm_pgoff)
2590                 return -EINVAL;
2591
2592         mutex_lock(&po->pg_vec_lock);
2593
2594         expected_size = 0;
2595         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2596                 if (rb->pg_vec) {
2597                         expected_size += rb->pg_vec_len
2598                                                 * rb->pg_vec_pages
2599                                                 * PAGE_SIZE;
2600                 }
2601         }
2602
2603         if (expected_size == 0)
2604                 goto out;
2605
2606         size = vma->vm_end - vma->vm_start;
2607         if (size != expected_size)
2608                 goto out;
2609
2610         start = vma->vm_start;
2611         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2612                 if (rb->pg_vec == NULL)
2613                         continue;
2614
2615                 for (i = 0; i < rb->pg_vec_len; i++) {
2616                         struct page *page;
2617                         void *kaddr = rb->pg_vec[i].buffer;
2618                         int pg_num;
2619
2620                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
2621                                 page = pgv_to_page(kaddr);
2622                                 err = vm_insert_page(vma, start, page);
2623                                 if (unlikely(err))
2624                                         goto out;
2625                                 start += PAGE_SIZE;
2626                                 kaddr += PAGE_SIZE;
2627                         }
2628                 }
2629         }
2630
2631         atomic_inc(&po->mapped);
2632         vma->vm_ops = &packet_mmap_ops;
2633         err = 0;
2634
2635 out:
2636         mutex_unlock(&po->pg_vec_lock);
2637         return err;
2638 }
2639
2640 static const struct proto_ops packet_ops_spkt = {
2641         .family =       PF_PACKET,
2642         .owner =        THIS_MODULE,
2643         .release =      packet_release,
2644         .bind =         packet_bind_spkt,
2645         .connect =      sock_no_connect,
2646         .socketpair =   sock_no_socketpair,
2647         .accept =       sock_no_accept,
2648         .getname =      packet_getname_spkt,
2649         .poll =         datagram_poll,
2650         .ioctl =        packet_ioctl,
2651         .listen =       sock_no_listen,
2652         .shutdown =     sock_no_shutdown,
2653         .setsockopt =   sock_no_setsockopt,
2654         .getsockopt =   sock_no_getsockopt,
2655         .sendmsg =      packet_sendmsg_spkt,
2656         .recvmsg =      packet_recvmsg,
2657         .mmap =         sock_no_mmap,
2658         .sendpage =     sock_no_sendpage,
2659 };
2660
2661 static const struct proto_ops packet_ops = {
2662         .family =       PF_PACKET,
2663         .owner =        THIS_MODULE,
2664         .release =      packet_release,
2665         .bind =         packet_bind,
2666         .connect =      sock_no_connect,
2667         .socketpair =   sock_no_socketpair,
2668         .accept =       sock_no_accept,
2669         .getname =      packet_getname,
2670         .poll =         packet_poll,
2671         .ioctl =        packet_ioctl,
2672         .listen =       sock_no_listen,
2673         .shutdown =     sock_no_shutdown,
2674         .setsockopt =   packet_setsockopt,
2675         .getsockopt =   packet_getsockopt,
2676         .sendmsg =      packet_sendmsg,
2677         .recvmsg =      packet_recvmsg,
2678         .mmap =         packet_mmap,
2679         .sendpage =     sock_no_sendpage,
2680 };
2681
2682 static const struct net_proto_family packet_family_ops = {
2683         .family =       PF_PACKET,
2684         .create =       packet_create,
2685         .owner  =       THIS_MODULE,
2686 };
2687
2688 static struct notifier_block packet_netdev_notifier = {
2689         .notifier_call =        packet_notifier,
2690 };
2691
2692 #ifdef CONFIG_PROC_FS
2693
2694 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
2695         __acquires(RCU)
2696 {
2697         struct net *net = seq_file_net(seq);
2698
2699         rcu_read_lock();
2700         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
2701 }
2702
2703 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2704 {
2705         struct net *net = seq_file_net(seq);
2706         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
2707 }
2708
2709 static void packet_seq_stop(struct seq_file *seq, void *v)
2710         __releases(RCU)
2711 {
2712         rcu_read_unlock();
2713 }
2714
2715 static int packet_seq_show(struct seq_file *seq, void *v)
2716 {
2717         if (v == SEQ_START_TOKEN)
2718                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
2719         else {
2720                 struct sock *s = sk_entry(v);
2721                 const struct packet_sock *po = pkt_sk(s);
2722
2723                 seq_printf(seq,
2724                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
2725                            s,
2726                            atomic_read(&s->sk_refcnt),
2727                            s->sk_type,
2728                            ntohs(po->num),
2729                            po->ifindex,
2730                            po->running,
2731                            atomic_read(&s->sk_rmem_alloc),
2732                            sock_i_uid(s),
2733                            sock_i_ino(s));
2734         }
2735
2736         return 0;
2737 }
2738
2739 static const struct seq_operations packet_seq_ops = {
2740         .start  = packet_seq_start,
2741         .next   = packet_seq_next,
2742         .stop   = packet_seq_stop,
2743         .show   = packet_seq_show,
2744 };
2745
2746 static int packet_seq_open(struct inode *inode, struct file *file)
2747 {
2748         return seq_open_net(inode, file, &packet_seq_ops,
2749                             sizeof(struct seq_net_private));
2750 }
2751
2752 static const struct file_operations packet_seq_fops = {
2753         .owner          = THIS_MODULE,
2754         .open           = packet_seq_open,
2755         .read           = seq_read,
2756         .llseek         = seq_lseek,
2757         .release        = seq_release_net,
2758 };
2759
2760 #endif
2761
2762 static int __net_init packet_net_init(struct net *net)
2763 {
2764         spin_lock_init(&net->packet.sklist_lock);
2765         INIT_HLIST_HEAD(&net->packet.sklist);
2766
2767         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2768                 return -ENOMEM;
2769
2770         return 0;
2771 }
2772
2773 static void __net_exit packet_net_exit(struct net *net)
2774 {
2775         proc_net_remove(net, "packet");
2776 }
2777
2778 static struct pernet_operations packet_net_ops = {
2779         .init = packet_net_init,
2780         .exit = packet_net_exit,
2781 };
2782
2783
2784 static void __exit packet_exit(void)
2785 {
2786         unregister_netdevice_notifier(&packet_netdev_notifier);
2787         unregister_pernet_subsys(&packet_net_ops);
2788         sock_unregister(PF_PACKET);
2789         proto_unregister(&packet_proto);
2790 }
2791
2792 static int __init packet_init(void)
2793 {
2794         int rc = proto_register(&packet_proto, 0);
2795
2796         if (rc != 0)
2797                 goto out;
2798
2799         sock_register(&packet_family_ops);
2800         register_pernet_subsys(&packet_net_ops);
2801         register_netdevice_notifier(&packet_netdev_notifier);
2802 out:
2803         return rc;
2804 }
2805
2806 module_init(packet_init);
2807 module_exit(packet_exit);
2808 MODULE_LICENSE("GPL");
2809 MODULE_ALIAS_NETPROTO(PF_PACKET);