SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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  *
  43  *              This program is free software; you can redistribute it and/or
  44  *              modify it under the terms of the GNU General Public License
  45  *              as published by the Free Software Foundation; either version
  46  *              2 of the License, or (at your option) any later version.
  47  *
  48  */
  49
  50 #include <linux/types.h>
  51 #include <linux/mm.h>
  52 #include <linux/capability.h>
  53 #include <linux/fcntl.h>
  54 #include <linux/socket.h>
  55 #include <linux/in.h>
  56 #include <linux/inet.h>
  57 #include <linux/netdevice.h>
  58 #include <linux/if_packet.h>
  59 #include <linux/wireless.h>
  60 #include <linux/kernel.h>
  61 #include <linux/kmod.h>
  62 #include <net/net_namespace.h>
  63 #include <net/ip.h>
  64 #include <net/protocol.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/errno.h>
  68 #include <linux/timer.h>
  69 #include <asm/system.h>
  70 #include <asm/uaccess.h>
  71 #include <asm/ioctls.h>
  72 #include <asm/page.h>
  73 #include <asm/cacheflush.h>
  74 #include <asm/io.h>
  75 #include <linux/proc_fs.h>
  76 #include <linux/seq_file.h>
  77 #include <linux/poll.h>
  78 #include <linux/module.h>
  79 #include <linux/init.h>
  80
  81 #ifdef CONFIG_INET
  82 #include <net/inet_common.h>
  83 #endif
  84
  85 /*
  86    Assumptions:
  87    - if device has no dev->hard_header routine, it adds and removes ll header
  88      inside itself. In this case ll header is invisible outside of device,
  89      but higher levels still should reserve dev->hard_header_len.
  90      Some devices are enough clever to reallocate skb, when header
  91      will not fit to reserved space (tunnel), another ones are silly
  92      (PPP).
  93    - packet socket receives packets with pulled ll header,
  94      so that SOCK_RAW should push it back.
  95
  96 On receive:
  97 -----------
  98
  99 Incoming, dev->hard_header!=NULL
 100    mac_header -> ll header
 101    data       -> data
 102
 103 Outgoing, dev->hard_header!=NULL
 104    mac_header -> ll header
 105    data       -> ll header
 106
 107 Incoming, dev->hard_header==NULL
 108    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 109                  header.  PPP makes it, that is wrong, because introduce
 110                  assymetry between rx and tx paths.
 111    data       -> data
 112
 113 Outgoing, dev->hard_header==NULL
 114    mac_header -> data. ll header is still not built!
 115    data       -> data
 116
 117 Resume
 118   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 119
 120
 121 On transmit:
 122 ------------
 123
 124 dev->hard_header != NULL
 125    mac_header -> ll header
 126    data       -> ll header
 127
 128 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 129    mac_header -> data
 130    data       -> data
 131
 132    We should set nh.raw on output to correct posistion,
 133    packet classifier depends on it.
 134  */
 135
 136 /* Private packet socket structures. */
 137
 138 struct packet_mclist
 139 {
 140         struct packet_mclist    *next;
 141         int                     ifindex;
 142         int                     count;
 143         unsigned short          type;
 144         unsigned short          alen;
 145         unsigned char           addr[MAX_ADDR_LEN];
 146 };
 147 /* identical to struct packet_mreq except it has
 148  * a longer address field.
 149  */
 150 struct packet_mreq_max
 151 {
 152         int             mr_ifindex;
 153         unsigned short  mr_type;
 154         unsigned short  mr_alen;
 155         unsigned char   mr_address[MAX_ADDR_LEN];
 156 };
 157
 158 #ifdef CONFIG_PACKET_MMAP
 159 static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing);
 160 #endif
 161
 162 static void packet_flush_mclist(struct sock *sk);
 163
 164 struct packet_sock {
 165         /* struct sock has to be the first member of packet_sock */
 166         struct sock             sk;
 167         struct tpacket_stats    stats;
 168 #ifdef CONFIG_PACKET_MMAP
 169         char *                  *pg_vec;
 170         unsigned int            head;
 171         unsigned int            frames_per_block;
 172         unsigned int            frame_size;
 173         unsigned int            frame_max;
 174         int                     copy_thresh;
 175 #endif
 176         struct packet_type      prot_hook;
 177         spinlock_t              bind_lock;
 178         unsigned int            running:1,      /* prot_hook is attached*/
 179                                 auxdata:1,
 180                                 origdev:1;
 181         int                     ifindex;        /* bound device         */
 182         __be16                  num;
 183         struct packet_mclist    *mclist;
 184 #ifdef CONFIG_PACKET_MMAP
 185         atomic_t                mapped;
 186         unsigned int            pg_vec_order;
 187         unsigned int            pg_vec_pages;
 188         unsigned int            pg_vec_len;
 189         enum tpacket_versions   tp_version;
 190         unsigned int            tp_hdrlen;
 191 #endif
 192 };
 193
 194 struct packet_skb_cb {
 195         unsigned int origlen;
 196         union {
 197                 struct sockaddr_pkt pkt;
 198                 struct sockaddr_ll ll;
 199         } sa;
 200 };
 201
 202 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 203
 204 #ifdef CONFIG_PACKET_MMAP
 205
 206 static void *packet_lookup_frame(struct packet_sock *po, unsigned int position,
 207                                  int status)
 208 {
 209         unsigned int pg_vec_pos, frame_offset;
 210         union {
 211                 struct tpacket_hdr *h1;
 212                 struct tpacket2_hdr *h2;
 213                 void *raw;
 214         } h;
 215
 216         pg_vec_pos = position / po->frames_per_block;
 217         frame_offset = position % po->frames_per_block;
 218
 219         h.raw = po->pg_vec[pg_vec_pos] + (frame_offset * po->frame_size);
 220         switch (po->tp_version) {
 221         case TPACKET_V1:
 222                 if (status != h.h1->tp_status ? TP_STATUS_USER :
 223                                                 TP_STATUS_KERNEL)
 224                         return NULL;
 225                 break;
 226         case TPACKET_V2:
 227                 if (status != h.h2->tp_status ? TP_STATUS_USER :
 228                                                 TP_STATUS_KERNEL)
 229                         return NULL;
 230                 break;
 231         }
 232         return h.raw;
 233 }
 234
 235 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 236 {
 237         union {
 238                 struct tpacket_hdr *h1;
 239                 struct tpacket2_hdr *h2;
 240                 void *raw;
 241         } h;
 242
 243         h.raw = frame;
 244         switch (po->tp_version) {
 245         case TPACKET_V1:
 246                 h.h1->tp_status = status;
 247                 break;
 248         case TPACKET_V2:
 249                 h.h2->tp_status = status;
 250                 break;
 251         }
 252 }
 253 #endif
 254
 255 static inline struct packet_sock *pkt_sk(struct sock *sk)
 256 {
 257         return (struct packet_sock *)sk;
 258 }
 259
 260 static void packet_sock_destruct(struct sock *sk)
 261 {
 262         BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
 263         BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
 264
 265         if (!sock_flag(sk, SOCK_DEAD)) {
 266                 printk("Attempt to release alive packet socket: %p\n", sk);
 267                 return;
 268         }
 269
 270         sk_refcnt_debug_dec(sk);
 271 }
 272
 273
 274 static const struct proto_ops packet_ops;
 275
 276 static const struct proto_ops packet_ops_spkt;
 277
 278 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,  struct packet_type *pt, struct net_device *orig_dev)
 279 {
 280         struct sock *sk;
 281         struct sockaddr_pkt *spkt;
 282
 283         /*
 284          *      When we registered the protocol we saved the socket in the data
 285          *      field for just this event.
 286          */
 287
 288         sk = pt->af_packet_priv;
 289
 290         /*
 291          *      Yank back the headers [hope the device set this
 292          *      right or kerboom...]
 293          *
 294          *      Incoming packets have ll header pulled,
 295          *      push it back.
 296          *
 297          *      For outgoing ones skb->data == skb_mac_header(skb)
 298          *      so that this procedure is noop.
 299          */
 300
 301         if (skb->pkt_type == PACKET_LOOPBACK)
 302                 goto out;
 303
 304         if (dev_net(dev) != sock_net(sk))
 305                 goto out;
 306
 307         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 308                 goto oom;
 309
 310         /* drop any routing info */
 311         dst_release(skb->dst);
 312         skb->dst = NULL;
 313
 314         /* drop conntrack reference */
 315         nf_reset(skb);
 316
 317         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
 318
 319         skb_push(skb, skb->data - skb_mac_header(skb));
 320
 321         /*
 322          *      The SOCK_PACKET socket receives _all_ frames.
 323          */
 324
 325         spkt->spkt_family = dev->type;
 326         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
 327         spkt->spkt_protocol = skb->protocol;
 328
 329         /*
 330          *      Charge the memory to the socket. This is done specifically
 331          *      to prevent sockets using all the memory up.
 332          */
 333
 334         if (sock_queue_rcv_skb(sk,skb) == 0)
 335                 return 0;
 336
 337 out:
 338         kfree_skb(skb);
 339 oom:
 340         return 0;
 341 }
 342
 343
 344 /*
 345  *      Output a raw packet to a device layer. This bypasses all the other
 346  *      protocol layers and you must therefore supply it with a complete frame
 347  */
 348
 349 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
 350                                struct msghdr *msg, size_t len)
 351 {
 352         struct sock *sk = sock->sk;
 353         struct sockaddr_pkt *saddr=(struct sockaddr_pkt *)msg->msg_name;
 354         struct sk_buff *skb;
 355         struct net_device *dev;
 356         __be16 proto=0;
 357         int err;
 358
 359         /*
 360          *      Get and verify the address.
 361          */
 362
 363         if (saddr)
 364         {
 365                 if (msg->msg_namelen < sizeof(struct sockaddr))
 366                         return(-EINVAL);
 367                 if (msg->msg_namelen==sizeof(struct sockaddr_pkt))
 368                         proto=saddr->spkt_protocol;
 369         }
 370         else
 371                 return(-ENOTCONN);      /* SOCK_PACKET must be sent giving an address */
 372
 373         /*
 374          *      Find the device first to size check it
 375          */
 376
 377         saddr->spkt_device[13] = 0;
 378         dev = dev_get_by_name(sock_net(sk), saddr->spkt_device);
 379         err = -ENODEV;
 380         if (dev == NULL)
 381                 goto out_unlock;
 382
 383         err = -ENETDOWN;
 384         if (!(dev->flags & IFF_UP))
 385                 goto out_unlock;
 386
 387         /*
 388          *      You may not queue a frame bigger than the mtu. This is the lowest level
 389          *      raw protocol and you must do your own fragmentation at this level.
 390          */
 391
 392         err = -EMSGSIZE;
 393         if (len > dev->mtu + dev->hard_header_len)
 394                 goto out_unlock;
 395
 396         err = -ENOBUFS;
 397         skb = sock_wmalloc(sk, len + LL_RESERVED_SPACE(dev), 0, GFP_KERNEL);
 398
 399         /*
 400          *      If the write buffer is full, then tough. At this level the user gets to
 401          *      deal with the problem - do your own algorithmic backoffs. That's far
 402          *      more flexible.
 403          */
 404
 405         if (skb == NULL)
 406                 goto out_unlock;
 407
 408         /*
 409          *      Fill it in
 410          */
 411
 412         /* FIXME: Save some space for broken drivers that write a
 413          * hard header at transmission time by themselves. PPP is the
 414          * notable one here. This should really be fixed at the driver level.
 415          */
 416         skb_reserve(skb, LL_RESERVED_SPACE(dev));
 417         skb_reset_network_header(skb);
 418
 419         /* Try to align data part correctly */
 420         if (dev->header_ops) {
 421                 skb->data -= dev->hard_header_len;
 422                 skb->tail -= dev->hard_header_len;
 423                 if (len < dev->hard_header_len)
 424                         skb_reset_network_header(skb);
 425         }
 426
 427         /* Returns -EFAULT on error */
 428         err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
 429         skb->protocol = proto;
 430         skb->dev = dev;
 431         skb->priority = sk->sk_priority;
 432         if (err)
 433                 goto out_free;
 434
 435         /*
 436          *      Now send it
 437          */
 438
 439         dev_queue_xmit(skb);
 440         dev_put(dev);
 441         return(len);
 442
 443 out_free:
 444         kfree_skb(skb);
 445 out_unlock:
 446         if (dev)
 447                 dev_put(dev);
 448         return err;
 449 }
 450
 451 static inline unsigned int run_filter(struct sk_buff *skb, struct sock *sk,
 452                                       unsigned int res)
 453 {
 454         struct sk_filter *filter;
 455
 456         rcu_read_lock_bh();
 457         filter = rcu_dereference(sk->sk_filter);
 458         if (filter != NULL)
 459                 res = sk_run_filter(skb, filter->insns, filter->len);
 460         rcu_read_unlock_bh();
 461
 462         return res;
 463 }
 464
 465 /*
 466    This function makes lazy skb cloning in hope that most of packets
 467    are discarded by BPF.
 468
 469    Note tricky part: we DO mangle shared skb! skb->data, skb->len
 470    and skb->cb are mangled. It works because (and until) packets
 471    falling here are owned by current CPU. Output packets are cloned
 472    by dev_queue_xmit_nit(), input packets are processed by net_bh
 473    sequencially, so that if we return skb to original state on exit,
 474    we will not harm anyone.
 475  */
 476
 477 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 478 {
 479         struct sock *sk;
 480         struct sockaddr_ll *sll;
 481         struct packet_sock *po;
 482         u8 * skb_head = skb->data;
 483         int skb_len = skb->len;
 484         unsigned int snaplen, res;
 485
 486         if (skb->pkt_type == PACKET_LOOPBACK)
 487                 goto drop;
 488
 489         sk = pt->af_packet_priv;
 490         po = pkt_sk(sk);
 491
 492         if (dev_net(dev) != sock_net(sk))
 493                 goto drop;
 494
 495         skb->dev = dev;
 496
 497         if (dev->header_ops) {
 498                 /* The device has an explicit notion of ll header,
 499                    exported to higher levels.
 500
 501                    Otherwise, the device hides datails of it frame
 502                    structure, so that corresponding packet head
 503                    never delivered to user.
 504                  */
 505                 if (sk->sk_type != SOCK_DGRAM)
 506                         skb_push(skb, skb->data - skb_mac_header(skb));
 507                 else if (skb->pkt_type == PACKET_OUTGOING) {
 508                         /* Special case: outgoing packets have ll header at head */
 509                         skb_pull(skb, skb_network_offset(skb));
 510                 }
 511         }
 512
 513         snaplen = skb->len;
 514
 515         res = run_filter(skb, sk, snaplen);
 516         if (!res)
 517                 goto drop_n_restore;
 518         if (snaplen > res)
 519                 snaplen = res;
 520
 521         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
 522             (unsigned)sk->sk_rcvbuf)
 523                 goto drop_n_acct;
 524
 525         if (skb_shared(skb)) {
 526                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
 527                 if (nskb == NULL)
 528                         goto drop_n_acct;
 529
 530                 if (skb_head != skb->data) {
 531                         skb->data = skb_head;
 532                         skb->len = skb_len;
 533                 }
 534                 kfree_skb(skb);
 535                 skb = nskb;
 536         }
 537
 538         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
 539                      sizeof(skb->cb));
 540
 541         sll = &PACKET_SKB_CB(skb)->sa.ll;
 542         sll->sll_family = AF_PACKET;
 543         sll->sll_hatype = dev->type;
 544         sll->sll_protocol = skb->protocol;
 545         sll->sll_pkttype = skb->pkt_type;
 546         if (unlikely(po->origdev))
 547                 sll->sll_ifindex = orig_dev->ifindex;
 548         else
 549                 sll->sll_ifindex = dev->ifindex;
 550
 551         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
 552
 553         PACKET_SKB_CB(skb)->origlen = skb->len;
 554
 555         if (pskb_trim(skb, snaplen))
 556                 goto drop_n_acct;
 557
 558         skb_set_owner_r(skb, sk);
 559         skb->dev = NULL;
 560         dst_release(skb->dst);
 561         skb->dst = NULL;
 562
 563         /* drop conntrack reference */
 564         nf_reset(skb);
 565
 566         spin_lock(&sk->sk_receive_queue.lock);
 567         po->stats.tp_packets++;
 568         __skb_queue_tail(&sk->sk_receive_queue, skb);
 569         spin_unlock(&sk->sk_receive_queue.lock);
 570         sk->sk_data_ready(sk, skb->len);
 571         return 0;
 572
 573 drop_n_acct:
 574         spin_lock(&sk->sk_receive_queue.lock);
 575         po->stats.tp_drops++;
 576         spin_unlock(&sk->sk_receive_queue.lock);
 577
 578 drop_n_restore:
 579         if (skb_head != skb->data && skb_shared(skb)) {
 580                 skb->data = skb_head;
 581                 skb->len = skb_len;
 582         }
 583 drop:
 584         kfree_skb(skb);
 585         return 0;
 586 }
 587
 588 #ifdef CONFIG_PACKET_MMAP
 589 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 590 {
 591         struct sock *sk;
 592         struct packet_sock *po;
 593         struct sockaddr_ll *sll;
 594         union {
 595                 struct tpacket_hdr *h1;
 596                 struct tpacket2_hdr *h2;
 597                 void *raw;
 598         } h;
 599         u8 * skb_head = skb->data;
 600         int skb_len = skb->len;
 601         unsigned int snaplen, res;
 602         unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
 603         unsigned short macoff, netoff, hdrlen;
 604         struct sk_buff *copy_skb = NULL;
 605         struct timeval tv;
 606         struct timespec ts;
 607
 608         if (skb->pkt_type == PACKET_LOOPBACK)
 609                 goto drop;
 610
 611         sk = pt->af_packet_priv;
 612         po = pkt_sk(sk);
 613
 614         if (dev_net(dev) != sock_net(sk))
 615                 goto drop;
 616
 617         if (dev->header_ops) {
 618                 if (sk->sk_type != SOCK_DGRAM)
 619                         skb_push(skb, skb->data - skb_mac_header(skb));
 620                 else if (skb->pkt_type == PACKET_OUTGOING) {
 621                         /* Special case: outgoing packets have ll header at head */
 622                         skb_pull(skb, skb_network_offset(skb));
 623                 }
 624         }
 625
 626         if (skb->ip_summed == CHECKSUM_PARTIAL)
 627                 status |= TP_STATUS_CSUMNOTREADY;
 628
 629         snaplen = skb->len;
 630
 631         res = run_filter(skb, sk, snaplen);
 632         if (!res)
 633                 goto drop_n_restore;
 634         if (snaplen > res)
 635                 snaplen = res;
 636
 637         if (sk->sk_type == SOCK_DGRAM) {
 638                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16;
 639         } else {
 640                 unsigned maclen = skb_network_offset(skb);
 641                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
 642                                        (maclen < 16 ? 16 : maclen));
 643                 macoff = netoff - maclen;
 644         }
 645
 646         if (macoff + snaplen > po->frame_size) {
 647                 if (po->copy_thresh &&
 648                     atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
 649                     (unsigned)sk->sk_rcvbuf) {
 650                         if (skb_shared(skb)) {
 651                                 copy_skb = skb_clone(skb, GFP_ATOMIC);
 652                         } else {
 653                                 copy_skb = skb_get(skb);
 654                                 skb_head = skb->data;
 655                         }
 656                         if (copy_skb)
 657                                 skb_set_owner_r(copy_skb, sk);
 658                 }
 659                 snaplen = po->frame_size - macoff;
 660                 if ((int)snaplen < 0)
 661                         snaplen = 0;
 662         }
 663
 664         spin_lock(&sk->sk_receive_queue.lock);
 665         h.raw = packet_lookup_frame(po, po->head, TP_STATUS_KERNEL);
 666         if (!h.raw)
 667                 goto ring_is_full;
 668         po->head = po->head != po->frame_max ? po->head+1 : 0;
 669         po->stats.tp_packets++;
 670         if (copy_skb) {
 671                 status |= TP_STATUS_COPY;
 672                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
 673         }
 674         if (!po->stats.tp_drops)
 675                 status &= ~TP_STATUS_LOSING;
 676         spin_unlock(&sk->sk_receive_queue.lock);
 677
 678         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
 679
 680         switch (po->tp_version) {
 681         case TPACKET_V1:
 682                 h.h1->tp_len = skb->len;
 683                 h.h1->tp_snaplen = snaplen;
 684                 h.h1->tp_mac = macoff;
 685                 h.h1->tp_net = netoff;
 686                 if (skb->tstamp.tv64)
 687                         tv = ktime_to_timeval(skb->tstamp);
 688                 else
 689                         do_gettimeofday(&tv);
 690                 h.h1->tp_sec = tv.tv_sec;
 691                 h.h1->tp_usec = tv.tv_usec;
 692                 hdrlen = sizeof(*h.h1);
 693                 break;
 694         case TPACKET_V2:
 695                 h.h2->tp_len = skb->len;
 696                 h.h2->tp_snaplen = snaplen;
 697                 h.h2->tp_mac = macoff;
 698                 h.h2->tp_net = netoff;
 699                 if (skb->tstamp.tv64)
 700                         ts = ktime_to_timespec(skb->tstamp);
 701                 else
 702                         getnstimeofday(&ts);
 703                 h.h2->tp_sec = ts.tv_sec;
 704                 h.h2->tp_nsec = ts.tv_nsec;
 705                 hdrlen = sizeof(*h.h2);
 706                 break;
 707         default:
 708                 BUG();
 709         }
 710
 711         sll = h.raw + TPACKET_ALIGN(hdrlen);
 712         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
 713         sll->sll_family = AF_PACKET;
 714         sll->sll_hatype = dev->type;
 715         sll->sll_protocol = skb->protocol;
 716         sll->sll_pkttype = skb->pkt_type;
 717         if (unlikely(po->origdev))
 718                 sll->sll_ifindex = orig_dev->ifindex;
 719         else
 720                 sll->sll_ifindex = dev->ifindex;
 721
 722         __packet_set_status(po, h.raw, status);
 723         smp_mb();
 724
 725         {
 726                 struct page *p_start, *p_end;
 727                 u8 *h_end = h.raw + macoff + snaplen - 1;
 728
 729                 p_start = virt_to_page(h.raw);
 730                 p_end = virt_to_page(h_end);
 731                 while (p_start <= p_end) {
 732                         flush_dcache_page(p_start);
 733                         p_start++;
 734                 }
 735         }
 736
 737         sk->sk_data_ready(sk, 0);
 738
 739 drop_n_restore:
 740         if (skb_head != skb->data && skb_shared(skb)) {
 741                 skb->data = skb_head;
 742                 skb->len = skb_len;
 743         }
 744 drop:
 745         kfree_skb(skb);
 746         return 0;
 747
 748 ring_is_full:
 749         po->stats.tp_drops++;
 750         spin_unlock(&sk->sk_receive_queue.lock);
 751
 752         sk->sk_data_ready(sk, 0);
 753         if (copy_skb)
 754                 kfree_skb(copy_skb);
 755         goto drop_n_restore;
 756 }
 757
 758 #endif
 759
 760
 761 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
 762                           struct msghdr *msg, size_t len)
 763 {
 764         struct sock *sk = sock->sk;
 765         struct sockaddr_ll *saddr=(struct sockaddr_ll *)msg->msg_name;
 766         struct sk_buff *skb;
 767         struct net_device *dev;
 768         __be16 proto;
 769         unsigned char *addr;
 770         int ifindex, err, reserve = 0;
 771
 772         /*
 773          *      Get and verify the address.
 774          */
 775
 776         if (saddr == NULL) {
 777                 struct packet_sock *po = pkt_sk(sk);
 778
 779                 ifindex = po->ifindex;
 780                 proto   = po->num;
 781                 addr    = NULL;
 782         } else {
 783                 err = -EINVAL;
 784                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
 785                         goto out;
 786                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
 787                         goto out;
 788                 ifindex = saddr->sll_ifindex;
 789                 proto   = saddr->sll_protocol;
 790                 addr    = saddr->sll_addr;
 791         }
 792
 793
 794         dev = dev_get_by_index(sock_net(sk), ifindex);
 795         err = -ENXIO;
 796         if (dev == NULL)
 797                 goto out_unlock;
 798         if (sock->type == SOCK_RAW)
 799                 reserve = dev->hard_header_len;
 800
 801         err = -ENETDOWN;
 802         if (!(dev->flags & IFF_UP))
 803                 goto out_unlock;
 804
 805         err = -EMSGSIZE;
 806         if (len > dev->mtu+reserve)
 807                 goto out_unlock;
 808
 809         skb = sock_alloc_send_skb(sk, len + LL_ALLOCATED_SPACE(dev),
 810                                 msg->msg_flags & MSG_DONTWAIT, &err);
 811         if (skb==NULL)
 812                 goto out_unlock;
 813
 814         skb_reserve(skb, LL_RESERVED_SPACE(dev));
 815         skb_reset_network_header(skb);
 816
 817         err = -EINVAL;
 818         if (sock->type == SOCK_DGRAM &&
 819             dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len) < 0)
 820                 goto out_free;
 821
 822         /* Returns -EFAULT on error */
 823         err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
 824         if (err)
 825                 goto out_free;
 826
 827         skb->protocol = proto;
 828         skb->dev = dev;
 829         skb->priority = sk->sk_priority;
 830
 831         /*
 832          *      Now send it
 833          */
 834
 835         err = dev_queue_xmit(skb);
 836         if (err > 0 && (err = net_xmit_errno(err)) != 0)
 837                 goto out_unlock;
 838
 839         dev_put(dev);
 840
 841         return(len);
 842
 843 out_free:
 844         kfree_skb(skb);
 845 out_unlock:
 846         if (dev)
 847                 dev_put(dev);
 848 out:
 849         return err;
 850 }
 851
 852 /*
 853  *      Close a PACKET socket. This is fairly simple. We immediately go
 854  *      to 'closed' state and remove our protocol entry in the device list.
 855  */
 856
 857 static int packet_release(struct socket *sock)
 858 {
 859         struct sock *sk = sock->sk;
 860         struct packet_sock *po;
 861         struct net *net;
 862
 863         if (!sk)
 864                 return 0;
 865
 866         net = sock_net(sk);
 867         po = pkt_sk(sk);
 868
 869         write_lock_bh(&net->packet.sklist_lock);
 870         sk_del_node_init(sk);
 871         write_unlock_bh(&net->packet.sklist_lock);
 872
 873         /*
 874          *      Unhook packet receive handler.
 875          */
 876
 877         if (po->running) {
 878                 /*
 879                  *      Remove the protocol hook
 880                  */
 881                 dev_remove_pack(&po->prot_hook);
 882                 po->running = 0;
 883                 po->num = 0;
 884                 __sock_put(sk);
 885         }
 886
 887         packet_flush_mclist(sk);
 888
 889 #ifdef CONFIG_PACKET_MMAP
 890         if (po->pg_vec) {
 891                 struct tpacket_req req;
 892                 memset(&req, 0, sizeof(req));
 893                 packet_set_ring(sk, &req, 1);
 894         }
 895 #endif
 896
 897         /*
 898          *      Now the socket is dead. No more input will appear.
 899          */
 900
 901         sock_orphan(sk);
 902         sock->sk = NULL;
 903
 904         /* Purge queues */
 905
 906         skb_queue_purge(&sk->sk_receive_queue);
 907         sk_refcnt_debug_release(sk);
 908
 909         sock_put(sk);
 910         return 0;
 911 }
 912
 913 /*
 914  *      Attach a packet hook.
 915  */
 916
 917 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
 918 {
 919         struct packet_sock *po = pkt_sk(sk);
 920         /*
 921          *      Detach an existing hook if present.
 922          */
 923
 924         lock_sock(sk);
 925
 926         spin_lock(&po->bind_lock);
 927         if (po->running) {
 928                 __sock_put(sk);
 929                 po->running = 0;
 930                 po->num = 0;
 931                 spin_unlock(&po->bind_lock);
 932                 dev_remove_pack(&po->prot_hook);
 933                 spin_lock(&po->bind_lock);
 934         }
 935
 936         po->num = protocol;
 937         po->prot_hook.type = protocol;
 938         po->prot_hook.dev = dev;
 939
 940         po->ifindex = dev ? dev->ifindex : 0;
 941
 942         if (protocol == 0)
 943                 goto out_unlock;
 944
 945         if (!dev || (dev->flags & IFF_UP)) {
 946                 dev_add_pack(&po->prot_hook);
 947                 sock_hold(sk);
 948                 po->running = 1;
 949         } else {
 950                 sk->sk_err = ENETDOWN;
 951                 if (!sock_flag(sk, SOCK_DEAD))
 952                         sk->sk_error_report(sk);
 953         }
 954
 955 out_unlock:
 956         spin_unlock(&po->bind_lock);
 957         release_sock(sk);
 958         return 0;
 959 }
 960
 961 /*
 962  *      Bind a packet socket to a device
 963  */
 964
 965 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 966 {
 967         struct sock *sk=sock->sk;
 968         char name[15];
 969         struct net_device *dev;
 970         int err = -ENODEV;
 971
 972         /*
 973          *      Check legality
 974          */
 975
 976         if (addr_len != sizeof(struct sockaddr))
 977                 return -EINVAL;
 978         strlcpy(name,uaddr->sa_data,sizeof(name));
 979
 980         dev = dev_get_by_name(sock_net(sk), name);
 981         if (dev) {
 982                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
 983                 dev_put(dev);
 984         }
 985         return err;
 986 }
 987
 988 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 989 {
 990         struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
 991         struct sock *sk=sock->sk;
 992         struct net_device *dev = NULL;
 993         int err;
 994
 995
 996         /*
 997          *      Check legality
 998          */
 999
1000         if (addr_len < sizeof(struct sockaddr_ll))
1001                 return -EINVAL;
1002         if (sll->sll_family != AF_PACKET)
1003                 return -EINVAL;
1004
1005         if (sll->sll_ifindex) {
1006                 err = -ENODEV;
1007                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1008                 if (dev == NULL)
1009                         goto out;
1010         }
1011         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1012         if (dev)
1013                 dev_put(dev);
1014
1015 out:
1016         return err;
1017 }
1018
1019 static struct proto packet_proto = {
1020         .name     = "PACKET",
1021         .owner    = THIS_MODULE,
1022         .obj_size = sizeof(struct packet_sock),
1023 };
1024
1025 /*
1026  *      Create a packet of type SOCK_PACKET.
1027  */
1028
1029 static int packet_create(struct net *net, struct socket *sock, int protocol)
1030 {
1031         struct sock *sk;
1032         struct packet_sock *po;
1033         __be16 proto = (__force __be16)protocol; /* weird, but documented */
1034         int err;
1035
1036         if (!capable(CAP_NET_RAW))
1037                 return -EPERM;
1038         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1039             sock->type != SOCK_PACKET)
1040                 return -ESOCKTNOSUPPORT;
1041
1042         sock->state = SS_UNCONNECTED;
1043
1044         err = -ENOBUFS;
1045         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1046         if (sk == NULL)
1047                 goto out;
1048
1049         sock->ops = &packet_ops;
1050         if (sock->type == SOCK_PACKET)
1051                 sock->ops = &packet_ops_spkt;
1052
1053         sock_init_data(sock, sk);
1054
1055         po = pkt_sk(sk);
1056         sk->sk_family = PF_PACKET;
1057         po->num = proto;
1058
1059         sk->sk_destruct = packet_sock_destruct;
1060         sk_refcnt_debug_inc(sk);
1061
1062         /*
1063          *      Attach a protocol block
1064          */
1065
1066         spin_lock_init(&po->bind_lock);
1067         po->prot_hook.func = packet_rcv;
1068
1069         if (sock->type == SOCK_PACKET)
1070                 po->prot_hook.func = packet_rcv_spkt;
1071
1072         po->prot_hook.af_packet_priv = sk;
1073
1074         if (proto) {
1075                 po->prot_hook.type = proto;
1076                 dev_add_pack(&po->prot_hook);
1077                 sock_hold(sk);
1078                 po->running = 1;
1079         }
1080
1081         write_lock_bh(&net->packet.sklist_lock);
1082         sk_add_node(sk, &net->packet.sklist);
1083         write_unlock_bh(&net->packet.sklist_lock);
1084         return(0);
1085 out:
1086         return err;
1087 }
1088
1089 /*
1090  *      Pull a packet from our receive queue and hand it to the user.
1091  *      If necessary we block.
1092  */
1093
1094 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1095                           struct msghdr *msg, size_t len, int flags)
1096 {
1097         struct sock *sk = sock->sk;
1098         struct sk_buff *skb;
1099         int copied, err;
1100         struct sockaddr_ll *sll;
1101
1102         err = -EINVAL;
1103         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT))
1104                 goto out;
1105
1106 #if 0
1107         /* What error should we return now? EUNATTACH? */
1108         if (pkt_sk(sk)->ifindex < 0)
1109                 return -ENODEV;
1110 #endif
1111
1112         /*
1113          *      Call the generic datagram receiver. This handles all sorts
1114          *      of horrible races and re-entrancy so we can forget about it
1115          *      in the protocol layers.
1116          *
1117          *      Now it will return ENETDOWN, if device have just gone down,
1118          *      but then it will block.
1119          */
1120
1121         skb=skb_recv_datagram(sk,flags,flags&MSG_DONTWAIT,&err);
1122
1123         /*
1124          *      An error occurred so return it. Because skb_recv_datagram()
1125          *      handles the blocking we don't see and worry about blocking
1126          *      retries.
1127          */
1128
1129         if (skb == NULL)
1130                 goto out;
1131
1132         /*
1133          *      If the address length field is there to be filled in, we fill
1134          *      it in now.
1135          */
1136
1137         sll = &PACKET_SKB_CB(skb)->sa.ll;
1138         if (sock->type == SOCK_PACKET)
1139                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1140         else
1141                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1142
1143         /*
1144          *      You lose any data beyond the buffer you gave. If it worries a
1145          *      user program they can ask the device for its MTU anyway.
1146          */
1147
1148         copied = skb->len;
1149         if (copied > len)
1150         {
1151                 copied=len;
1152                 msg->msg_flags|=MSG_TRUNC;
1153         }
1154
1155         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1156         if (err)
1157                 goto out_free;
1158
1159         sock_recv_timestamp(msg, sk, skb);
1160
1161         if (msg->msg_name)
1162                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1163                        msg->msg_namelen);
1164
1165         if (pkt_sk(sk)->auxdata) {
1166                 struct tpacket_auxdata aux;
1167
1168                 aux.tp_status = TP_STATUS_USER;
1169                 if (skb->ip_summed == CHECKSUM_PARTIAL)
1170                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1171                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1172                 aux.tp_snaplen = skb->len;
1173                 aux.tp_mac = 0;
1174                 aux.tp_net = skb_network_offset(skb);
1175
1176                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1177         }
1178
1179         /*
1180          *      Free or return the buffer as appropriate. Again this
1181          *      hides all the races and re-entrancy issues from us.
1182          */
1183         err = (flags&MSG_TRUNC) ? skb->len : copied;
1184
1185 out_free:
1186         skb_free_datagram(sk, skb);
1187 out:
1188         return err;
1189 }
1190
1191 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1192                                int *uaddr_len, int peer)
1193 {
1194         struct net_device *dev;
1195         struct sock *sk = sock->sk;
1196
1197         if (peer)
1198                 return -EOPNOTSUPP;
1199
1200         uaddr->sa_family = AF_PACKET;
1201         dev = dev_get_by_index(sock_net(sk), pkt_sk(sk)->ifindex);
1202         if (dev) {
1203                 strlcpy(uaddr->sa_data, dev->name, 15);
1204                 dev_put(dev);
1205         } else
1206                 memset(uaddr->sa_data, 0, 14);
1207         *uaddr_len = sizeof(*uaddr);
1208
1209         return 0;
1210 }
1211
1212 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1213                           int *uaddr_len, int peer)
1214 {
1215         struct net_device *dev;
1216         struct sock *sk = sock->sk;
1217         struct packet_sock *po = pkt_sk(sk);
1218         struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
1219
1220         if (peer)
1221                 return -EOPNOTSUPP;
1222
1223         sll->sll_family = AF_PACKET;
1224         sll->sll_ifindex = po->ifindex;
1225         sll->sll_protocol = po->num;
1226         dev = dev_get_by_index(sock_net(sk), po->ifindex);
1227         if (dev) {
1228                 sll->sll_hatype = dev->type;
1229                 sll->sll_halen = dev->addr_len;
1230                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1231                 dev_put(dev);
1232         } else {
1233                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
1234                 sll->sll_halen = 0;
1235         }
1236         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1237
1238         return 0;
1239 }
1240
1241 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1242                          int what)
1243 {
1244         switch (i->type) {
1245         case PACKET_MR_MULTICAST:
1246                 if (what > 0)
1247                         dev_mc_add(dev, i->addr, i->alen, 0);
1248                 else
1249                         dev_mc_delete(dev, i->addr, i->alen, 0);
1250                 break;
1251         case PACKET_MR_PROMISC:
1252                 return dev_set_promiscuity(dev, what);
1253                 break;
1254         case PACKET_MR_ALLMULTI:
1255                 return dev_set_allmulti(dev, what);
1256                 break;
1257         default:;
1258         }
1259         return 0;
1260 }
1261
1262 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1263 {
1264         for ( ; i; i=i->next) {
1265                 if (i->ifindex == dev->ifindex)
1266                         packet_dev_mc(dev, i, what);
1267         }
1268 }
1269
1270 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1271 {
1272         struct packet_sock *po = pkt_sk(sk);
1273         struct packet_mclist *ml, *i;
1274         struct net_device *dev;
1275         int err;
1276
1277         rtnl_lock();
1278
1279         err = -ENODEV;
1280         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1281         if (!dev)
1282                 goto done;
1283
1284         err = -EINVAL;
1285         if (mreq->mr_alen > dev->addr_len)
1286                 goto done;
1287
1288         err = -ENOBUFS;
1289         i = kmalloc(sizeof(*i), GFP_KERNEL);
1290         if (i == NULL)
1291                 goto done;
1292
1293         err = 0;
1294         for (ml = po->mclist; ml; ml = ml->next) {
1295                 if (ml->ifindex == mreq->mr_ifindex &&
1296                     ml->type == mreq->mr_type &&
1297                     ml->alen == mreq->mr_alen &&
1298                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1299                         ml->count++;
1300                         /* Free the new element ... */
1301                         kfree(i);
1302                         goto done;
1303                 }
1304         }
1305
1306         i->type = mreq->mr_type;
1307         i->ifindex = mreq->mr_ifindex;
1308         i->alen = mreq->mr_alen;
1309         memcpy(i->addr, mreq->mr_address, i->alen);
1310         i->count = 1;
1311         i->next = po->mclist;
1312         po->mclist = i;
1313         err = packet_dev_mc(dev, i, 1);
1314         if (err) {
1315                 po->mclist = i->next;
1316                 kfree(i);
1317         }
1318
1319 done:
1320         rtnl_unlock();
1321         return err;
1322 }
1323
1324 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1325 {
1326         struct packet_mclist *ml, **mlp;
1327
1328         rtnl_lock();
1329
1330         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1331                 if (ml->ifindex == mreq->mr_ifindex &&
1332                     ml->type == mreq->mr_type &&
1333                     ml->alen == mreq->mr_alen &&
1334                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1335                         if (--ml->count == 0) {
1336                                 struct net_device *dev;
1337                                 *mlp = ml->next;
1338                                 dev = dev_get_by_index(sock_net(sk), ml->ifindex);
1339                                 if (dev) {
1340                                         packet_dev_mc(dev, ml, -1);
1341                                         dev_put(dev);
1342                                 }
1343                                 kfree(ml);
1344                         }
1345                         rtnl_unlock();
1346                         return 0;
1347                 }
1348         }
1349         rtnl_unlock();
1350         return -EADDRNOTAVAIL;
1351 }
1352
1353 static void packet_flush_mclist(struct sock *sk)
1354 {
1355         struct packet_sock *po = pkt_sk(sk);
1356         struct packet_mclist *ml;
1357
1358         if (!po->mclist)
1359                 return;
1360
1361         rtnl_lock();
1362         while ((ml = po->mclist) != NULL) {
1363                 struct net_device *dev;
1364
1365                 po->mclist = ml->next;
1366                 if ((dev = dev_get_by_index(sock_net(sk), ml->ifindex)) != NULL) {
1367                         packet_dev_mc(dev, ml, -1);
1368                         dev_put(dev);
1369                 }
1370                 kfree(ml);
1371         }
1372         rtnl_unlock();
1373 }
1374
1375 static int
1376 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, int optlen)
1377 {
1378         struct sock *sk = sock->sk;
1379         struct packet_sock *po = pkt_sk(sk);
1380         int ret;
1381
1382         if (level != SOL_PACKET)
1383                 return -ENOPROTOOPT;
1384
1385         switch(optname) {
1386         case PACKET_ADD_MEMBERSHIP:
1387         case PACKET_DROP_MEMBERSHIP:
1388         {
1389                 struct packet_mreq_max mreq;
1390                 int len = optlen;
1391                 memset(&mreq, 0, sizeof(mreq));
1392                 if (len < sizeof(struct packet_mreq))
1393                         return -EINVAL;
1394                 if (len > sizeof(mreq))
1395                         len = sizeof(mreq);
1396                 if (copy_from_user(&mreq,optval,len))
1397                         return -EFAULT;
1398                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1399                         return -EINVAL;
1400                 if (optname == PACKET_ADD_MEMBERSHIP)
1401                         ret = packet_mc_add(sk, &mreq);
1402                 else
1403                         ret = packet_mc_drop(sk, &mreq);
1404                 return ret;
1405         }
1406
1407 #ifdef CONFIG_PACKET_MMAP
1408         case PACKET_RX_RING:
1409         {
1410                 struct tpacket_req req;
1411
1412                 if (optlen<sizeof(req))
1413                         return -EINVAL;
1414                 if (copy_from_user(&req,optval,sizeof(req)))
1415                         return -EFAULT;
1416                 return packet_set_ring(sk, &req, 0);
1417         }
1418         case PACKET_COPY_THRESH:
1419         {
1420                 int val;
1421
1422                 if (optlen!=sizeof(val))
1423                         return -EINVAL;
1424                 if (copy_from_user(&val,optval,sizeof(val)))
1425                         return -EFAULT;
1426
1427                 pkt_sk(sk)->copy_thresh = val;
1428                 return 0;
1429         }
1430         case PACKET_VERSION:
1431         {
1432                 int val;
1433
1434                 if (optlen != sizeof(val))
1435                         return -EINVAL;
1436                 if (po->pg_vec)
1437                         return -EBUSY;
1438                 if (copy_from_user(&val, optval, sizeof(val)))
1439                         return -EFAULT;
1440                 switch (val) {
1441                 case TPACKET_V1:
1442                 case TPACKET_V2:
1443                         po->tp_version = val;
1444                         return 0;
1445                 default:
1446                         return -EINVAL;
1447                 }
1448         }
1449 #endif
1450         case PACKET_AUXDATA:
1451         {
1452                 int val;
1453
1454                 if (optlen < sizeof(val))
1455                         return -EINVAL;
1456                 if (copy_from_user(&val, optval, sizeof(val)))
1457                         return -EFAULT;
1458
1459                 po->auxdata = !!val;
1460                 return 0;
1461         }
1462         case PACKET_ORIGDEV:
1463         {
1464                 int val;
1465
1466                 if (optlen < sizeof(val))
1467                         return -EINVAL;
1468                 if (copy_from_user(&val, optval, sizeof(val)))
1469                         return -EFAULT;
1470
1471                 po->origdev = !!val;
1472                 return 0;
1473         }
1474         default:
1475                 return -ENOPROTOOPT;
1476         }
1477 }
1478
1479 static int packet_getsockopt(struct socket *sock, int level, int optname,
1480                              char __user *optval, int __user *optlen)
1481 {
1482         int len;
1483         int val;
1484         struct sock *sk = sock->sk;
1485         struct packet_sock *po = pkt_sk(sk);
1486         void *data;
1487         struct tpacket_stats st;
1488
1489         if (level != SOL_PACKET)
1490                 return -ENOPROTOOPT;
1491
1492         if (get_user(len, optlen))
1493                 return -EFAULT;
1494
1495         if (len < 0)
1496                 return -EINVAL;
1497
1498         switch(optname) {
1499         case PACKET_STATISTICS:
1500                 if (len > sizeof(struct tpacket_stats))
1501                         len = sizeof(struct tpacket_stats);
1502                 spin_lock_bh(&sk->sk_receive_queue.lock);
1503                 st = po->stats;
1504                 memset(&po->stats, 0, sizeof(st));
1505                 spin_unlock_bh(&sk->sk_receive_queue.lock);
1506                 st.tp_packets += st.tp_drops;
1507
1508                 data = &st;
1509                 break;
1510         case PACKET_AUXDATA:
1511                 if (len > sizeof(int))
1512                         len = sizeof(int);
1513                 val = po->auxdata;
1514
1515                 data = &val;
1516                 break;
1517         case PACKET_ORIGDEV:
1518                 if (len > sizeof(int))
1519                         len = sizeof(int);
1520                 val = po->origdev;
1521
1522                 data = &val;
1523                 break;
1524 #ifdef CONFIG_PACKET_MMAP
1525         case PACKET_VERSION:
1526                 if (len > sizeof(int))
1527                         len = sizeof(int);
1528                 val = po->tp_version;
1529                 data = &val;
1530                 break;
1531         case PACKET_HDRLEN:
1532                 if (len > sizeof(int))
1533                         len = sizeof(int);
1534                 if (copy_from_user(&val, optval, len))
1535                         return -EFAULT;
1536                 switch (val) {
1537                 case TPACKET_V1:
1538                         val = sizeof(struct tpacket_hdr);
1539                         break;
1540                 case TPACKET_V2:
1541                         val = sizeof(struct tpacket2_hdr);
1542                         break;
1543                 default:
1544                         return -EINVAL;
1545                 }
1546                 data = &val;
1547                 break;
1548 #endif
1549         default:
1550                 return -ENOPROTOOPT;
1551         }
1552
1553         if (put_user(len, optlen))
1554                 return -EFAULT;
1555         if (copy_to_user(optval, data, len))
1556                 return -EFAULT;
1557         return 0;
1558 }
1559
1560
1561 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
1562 {
1563         struct sock *sk;
1564         struct hlist_node *node;
1565         struct net_device *dev = data;
1566         struct net *net = dev_net(dev);
1567
1568         read_lock(&net->packet.sklist_lock);
1569         sk_for_each(sk, node, &net->packet.sklist) {
1570                 struct packet_sock *po = pkt_sk(sk);
1571
1572                 switch (msg) {
1573                 case NETDEV_UNREGISTER:
1574                         if (po->mclist)
1575                                 packet_dev_mclist(dev, po->mclist, -1);
1576                         /* fallthrough */
1577
1578                 case NETDEV_DOWN:
1579                         if (dev->ifindex == po->ifindex) {
1580                                 spin_lock(&po->bind_lock);
1581                                 if (po->running) {
1582                                         __dev_remove_pack(&po->prot_hook);
1583                                         __sock_put(sk);
1584                                         po->running = 0;
1585                                         sk->sk_err = ENETDOWN;
1586                                         if (!sock_flag(sk, SOCK_DEAD))
1587                                                 sk->sk_error_report(sk);
1588                                 }
1589                                 if (msg == NETDEV_UNREGISTER) {
1590                                         po->ifindex = -1;
1591                                         po->prot_hook.dev = NULL;
1592                                 }
1593                                 spin_unlock(&po->bind_lock);
1594                         }
1595                         break;
1596                 case NETDEV_UP:
1597                         spin_lock(&po->bind_lock);
1598                         if (dev->ifindex == po->ifindex && po->num &&
1599                             !po->running) {
1600                                 dev_add_pack(&po->prot_hook);
1601                                 sock_hold(sk);
1602                                 po->running = 1;
1603                         }
1604                         spin_unlock(&po->bind_lock);
1605                         break;
1606                 }
1607         }
1608         read_unlock(&net->packet.sklist_lock);
1609         return NOTIFY_DONE;
1610 }
1611
1612
1613 static int packet_ioctl(struct socket *sock, unsigned int cmd,
1614                         unsigned long arg)
1615 {
1616         struct sock *sk = sock->sk;
1617
1618         switch(cmd) {
1619                 case SIOCOUTQ:
1620                 {
1621                         int amount = atomic_read(&sk->sk_wmem_alloc);
1622                         return put_user(amount, (int __user *)arg);
1623                 }
1624                 case SIOCINQ:
1625                 {
1626                         struct sk_buff *skb;
1627                         int amount = 0;
1628
1629                         spin_lock_bh(&sk->sk_receive_queue.lock);
1630                         skb = skb_peek(&sk->sk_receive_queue);
1631                         if (skb)
1632                                 amount = skb->len;
1633                         spin_unlock_bh(&sk->sk_receive_queue.lock);
1634                         return put_user(amount, (int __user *)arg);
1635                 }
1636                 case SIOCGSTAMP:
1637                         return sock_get_timestamp(sk, (struct timeval __user *)arg);
1638                 case SIOCGSTAMPNS:
1639                         return sock_get_timestampns(sk, (struct timespec __user *)arg);
1640
1641 #ifdef CONFIG_INET
1642                 case SIOCADDRT:
1643                 case SIOCDELRT:
1644                 case SIOCDARP:
1645                 case SIOCGARP:
1646                 case SIOCSARP:
1647                 case SIOCGIFADDR:
1648                 case SIOCSIFADDR:
1649                 case SIOCGIFBRDADDR:
1650                 case SIOCSIFBRDADDR:
1651                 case SIOCGIFNETMASK:
1652                 case SIOCSIFNETMASK:
1653                 case SIOCGIFDSTADDR:
1654                 case SIOCSIFDSTADDR:
1655                 case SIOCSIFFLAGS:
1656                         if (sock_net(sk) != &init_net)
1657                                 return -ENOIOCTLCMD;
1658                         return inet_dgram_ops.ioctl(sock, cmd, arg);
1659 #endif
1660
1661                 default:
1662                         return -ENOIOCTLCMD;
1663         }
1664         return 0;
1665 }
1666
1667 #ifndef CONFIG_PACKET_MMAP
1668 #define packet_mmap sock_no_mmap
1669 #define packet_poll datagram_poll
1670 #else
1671
1672 static unsigned int packet_poll(struct file * file, struct socket *sock,
1673                                 poll_table *wait)
1674 {
1675         struct sock *sk = sock->sk;
1676         struct packet_sock *po = pkt_sk(sk);
1677         unsigned int mask = datagram_poll(file, sock, wait);
1678
1679         spin_lock_bh(&sk->sk_receive_queue.lock);
1680         if (po->pg_vec) {
1681                 unsigned last = po->head ? po->head-1 : po->frame_max;
1682
1683                 if (packet_lookup_frame(po, last, TP_STATUS_USER))
1684                         mask |= POLLIN | POLLRDNORM;
1685         }
1686         spin_unlock_bh(&sk->sk_receive_queue.lock);
1687         return mask;
1688 }
1689
1690
1691 /* Dirty? Well, I still did not learn better way to account
1692  * for user mmaps.
1693  */
1694
1695 static void packet_mm_open(struct vm_area_struct *vma)
1696 {
1697         struct file *file = vma->vm_file;
1698         struct socket * sock = file->private_data;
1699         struct sock *sk = sock->sk;
1700
1701         if (sk)
1702                 atomic_inc(&pkt_sk(sk)->mapped);
1703 }
1704
1705 static void packet_mm_close(struct vm_area_struct *vma)
1706 {
1707         struct file *file = vma->vm_file;
1708         struct socket * sock = file->private_data;
1709         struct sock *sk = sock->sk;
1710
1711         if (sk)
1712                 atomic_dec(&pkt_sk(sk)->mapped);
1713 }
1714
1715 static struct vm_operations_struct packet_mmap_ops = {
1716         .open = packet_mm_open,
1717         .close =packet_mm_close,
1718 };
1719
1720 static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
1721 {
1722         int i;
1723
1724         for (i = 0; i < len; i++) {
1725                 if (likely(pg_vec[i]))
1726                         free_pages((unsigned long) pg_vec[i], order);
1727         }
1728         kfree(pg_vec);
1729 }
1730
1731 static inline char *alloc_one_pg_vec_page(unsigned long order)
1732 {
1733         return (char *) __get_free_pages(GFP_KERNEL | __GFP_COMP | __GFP_ZERO,
1734                                          order);
1735 }
1736
1737 static char **alloc_pg_vec(struct tpacket_req *req, int order)
1738 {
1739         unsigned int block_nr = req->tp_block_nr;
1740         char **pg_vec;
1741         int i;
1742
1743         pg_vec = kzalloc(block_nr * sizeof(char *), GFP_KERNEL);
1744         if (unlikely(!pg_vec))
1745                 goto out;
1746
1747         for (i = 0; i < block_nr; i++) {
1748                 pg_vec[i] = alloc_one_pg_vec_page(order);
1749                 if (unlikely(!pg_vec[i]))
1750                         goto out_free_pgvec;
1751         }
1752
1753 out:
1754         return pg_vec;
1755
1756 out_free_pgvec:
1757         free_pg_vec(pg_vec, order, block_nr);
1758         pg_vec = NULL;
1759         goto out;
1760 }
1761
1762 static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing)
1763 {
1764         char **pg_vec = NULL;
1765         struct packet_sock *po = pkt_sk(sk);
1766         int was_running, order = 0;
1767         __be16 num;
1768         int err = 0;
1769
1770         if (req->tp_block_nr) {
1771                 int i;
1772
1773                 /* Sanity tests and some calculations */
1774
1775                 if (unlikely(po->pg_vec))
1776                         return -EBUSY;
1777
1778                 switch (po->tp_version) {
1779                 case TPACKET_V1:
1780                         po->tp_hdrlen = TPACKET_HDRLEN;
1781                         break;
1782                 case TPACKET_V2:
1783                         po->tp_hdrlen = TPACKET2_HDRLEN;
1784                         break;
1785                 }
1786
1787                 if (unlikely((int)req->tp_block_size <= 0))
1788                         return -EINVAL;
1789                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
1790                         return -EINVAL;
1791                 if (unlikely(req->tp_frame_size < po->tp_hdrlen))
1792                         return -EINVAL;
1793                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
1794                         return -EINVAL;
1795
1796                 po->frames_per_block = req->tp_block_size/req->tp_frame_size;
1797                 if (unlikely(po->frames_per_block <= 0))
1798                         return -EINVAL;
1799                 if (unlikely((po->frames_per_block * req->tp_block_nr) !=
1800                              req->tp_frame_nr))
1801                         return -EINVAL;
1802
1803                 err = -ENOMEM;
1804                 order = get_order(req->tp_block_size);
1805                 pg_vec = alloc_pg_vec(req, order);
1806                 if (unlikely(!pg_vec))
1807                         goto out;
1808
1809                 for (i = 0; i < req->tp_block_nr; i++) {
1810                         void *ptr = pg_vec[i];
1811                         int k;
1812
1813                         for (k = 0; k < po->frames_per_block; k++) {
1814                                 __packet_set_status(po, ptr, TP_STATUS_KERNEL);
1815                                 ptr += req->tp_frame_size;
1816                         }
1817                 }
1818                 /* Done */
1819         } else {
1820                 if (unlikely(req->tp_frame_nr))
1821                         return -EINVAL;
1822         }
1823
1824         lock_sock(sk);
1825
1826         /* Detach socket from network */
1827         spin_lock(&po->bind_lock);
1828         was_running = po->running;
1829         num = po->num;
1830         if (was_running) {
1831                 __dev_remove_pack(&po->prot_hook);
1832                 po->num = 0;
1833                 po->running = 0;
1834                 __sock_put(sk);
1835         }
1836         spin_unlock(&po->bind_lock);
1837
1838         synchronize_net();
1839
1840         err = -EBUSY;
1841         if (closing || atomic_read(&po->mapped) == 0) {
1842                 err = 0;
1843 #define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
1844
1845                 spin_lock_bh(&sk->sk_receive_queue.lock);
1846                 pg_vec = XC(po->pg_vec, pg_vec);
1847                 po->frame_max = (req->tp_frame_nr - 1);
1848                 po->head = 0;
1849                 po->frame_size = req->tp_frame_size;
1850                 spin_unlock_bh(&sk->sk_receive_queue.lock);
1851
1852                 order = XC(po->pg_vec_order, order);
1853                 req->tp_block_nr = XC(po->pg_vec_len, req->tp_block_nr);
1854
1855                 po->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
1856                 po->prot_hook.func = po->pg_vec ? tpacket_rcv : packet_rcv;
1857                 skb_queue_purge(&sk->sk_receive_queue);
1858 #undef XC
1859                 if (atomic_read(&po->mapped))
1860                         printk(KERN_DEBUG "packet_mmap: vma is busy: %d\n", atomic_read(&po->mapped));
1861         }
1862
1863         spin_lock(&po->bind_lock);
1864         if (was_running && !po->running) {
1865                 sock_hold(sk);
1866                 po->running = 1;
1867                 po->num = num;
1868                 dev_add_pack(&po->prot_hook);
1869         }
1870         spin_unlock(&po->bind_lock);
1871
1872         release_sock(sk);
1873
1874         if (pg_vec)
1875                 free_pg_vec(pg_vec, order, req->tp_block_nr);
1876 out:
1877         return err;
1878 }
1879
1880 static int packet_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1881 {
1882         struct sock *sk = sock->sk;
1883         struct packet_sock *po = pkt_sk(sk);
1884         unsigned long size;
1885         unsigned long start;
1886         int err = -EINVAL;
1887         int i;
1888
1889         if (vma->vm_pgoff)
1890                 return -EINVAL;
1891
1892         size = vma->vm_end - vma->vm_start;
1893
1894         lock_sock(sk);
1895         if (po->pg_vec == NULL)
1896                 goto out;
1897         if (size != po->pg_vec_len*po->pg_vec_pages*PAGE_SIZE)
1898                 goto out;
1899
1900         start = vma->vm_start;
1901         for (i = 0; i < po->pg_vec_len; i++) {
1902                 struct page *page = virt_to_page(po->pg_vec[i]);
1903                 int pg_num;
1904
1905                 for (pg_num = 0; pg_num < po->pg_vec_pages; pg_num++, page++) {
1906                         err = vm_insert_page(vma, start, page);
1907                         if (unlikely(err))
1908                                 goto out;
1909                         start += PAGE_SIZE;
1910                 }
1911         }
1912         atomic_inc(&po->mapped);
1913         vma->vm_ops = &packet_mmap_ops;
1914         err = 0;
1915
1916 out:
1917         release_sock(sk);
1918         return err;
1919 }
1920 #endif
1921
1922
1923 static const struct proto_ops packet_ops_spkt = {
1924         .family =       PF_PACKET,
1925         .owner =        THIS_MODULE,
1926         .release =      packet_release,
1927         .bind =         packet_bind_spkt,
1928         .connect =      sock_no_connect,
1929         .socketpair =   sock_no_socketpair,
1930         .accept =       sock_no_accept,
1931         .getname =      packet_getname_spkt,
1932         .poll =         datagram_poll,
1933         .ioctl =        packet_ioctl,
1934         .listen =       sock_no_listen,
1935         .shutdown =     sock_no_shutdown,
1936         .setsockopt =   sock_no_setsockopt,
1937         .getsockopt =   sock_no_getsockopt,
1938         .sendmsg =      packet_sendmsg_spkt,
1939         .recvmsg =      packet_recvmsg,
1940         .mmap =         sock_no_mmap,
1941         .sendpage =     sock_no_sendpage,
1942 };
1943
1944 static const struct proto_ops packet_ops = {
1945         .family =       PF_PACKET,
1946         .owner =        THIS_MODULE,
1947         .release =      packet_release,
1948         .bind =         packet_bind,
1949         .connect =      sock_no_connect,
1950         .socketpair =   sock_no_socketpair,
1951         .accept =       sock_no_accept,
1952         .getname =      packet_getname,
1953         .poll =         packet_poll,
1954         .ioctl =        packet_ioctl,
1955         .listen =       sock_no_listen,
1956         .shutdown =     sock_no_shutdown,
1957         .setsockopt =   packet_setsockopt,
1958         .getsockopt =   packet_getsockopt,
1959         .sendmsg =      packet_sendmsg,
1960         .recvmsg =      packet_recvmsg,
1961         .mmap =         packet_mmap,
1962         .sendpage =     sock_no_sendpage,
1963 };
1964
1965 static struct net_proto_family packet_family_ops = {
1966         .family =       PF_PACKET,
1967         .create =       packet_create,
1968         .owner  =       THIS_MODULE,
1969 };
1970
1971 static struct notifier_block packet_netdev_notifier = {
1972         .notifier_call =packet_notifier,
1973 };
1974
1975 #ifdef CONFIG_PROC_FS
1976 static inline struct sock *packet_seq_idx(struct net *net, loff_t off)
1977 {
1978         struct sock *s;
1979         struct hlist_node *node;
1980
1981         sk_for_each(s, node, &net->packet.sklist) {
1982                 if (!off--)
1983                         return s;
1984         }
1985         return NULL;
1986 }
1987
1988 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
1989         __acquires(seq_file_net(seq)->packet.sklist_lock)
1990 {
1991         struct net *net = seq_file_net(seq);
1992         read_lock(&net->packet.sklist_lock);
1993         return *pos ? packet_seq_idx(net, *pos - 1) : SEQ_START_TOKEN;
1994 }
1995
1996 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1997 {
1998         struct net *net = seq_file_net(seq);
1999         ++*pos;
2000         return  (v == SEQ_START_TOKEN)
2001                 ? sk_head(&net->packet.sklist)
2002                 : sk_next((struct sock*)v) ;
2003 }
2004
2005 static void packet_seq_stop(struct seq_file *seq, void *v)
2006         __releases(seq_file_net(seq)->packet.sklist_lock)
2007 {
2008         struct net *net = seq_file_net(seq);
2009         read_unlock(&net->packet.sklist_lock);
2010 }
2011
2012 static int packet_seq_show(struct seq_file *seq, void *v)
2013 {
2014         if (v == SEQ_START_TOKEN)
2015                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
2016         else {
2017                 struct sock *s = v;
2018                 const struct packet_sock *po = pkt_sk(s);
2019
2020                 seq_printf(seq,
2021                            "%p %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
2022                            s,
2023                            atomic_read(&s->sk_refcnt),
2024                            s->sk_type,
2025                            ntohs(po->num),
2026                            po->ifindex,
2027                            po->running,
2028                            atomic_read(&s->sk_rmem_alloc),
2029                            sock_i_uid(s),
2030                            sock_i_ino(s) );
2031         }
2032
2033         return 0;
2034 }
2035
2036 static const struct seq_operations packet_seq_ops = {
2037         .start  = packet_seq_start,
2038         .next   = packet_seq_next,
2039         .stop   = packet_seq_stop,
2040         .show   = packet_seq_show,
2041 };
2042
2043 static int packet_seq_open(struct inode *inode, struct file *file)
2044 {
2045         return seq_open_net(inode, file, &packet_seq_ops,
2046                             sizeof(struct seq_net_private));
2047 }
2048
2049 static const struct file_operations packet_seq_fops = {
2050         .owner          = THIS_MODULE,
2051         .open           = packet_seq_open,
2052         .read           = seq_read,
2053         .llseek         = seq_lseek,
2054         .release        = seq_release_net,
2055 };
2056
2057 #endif
2058
2059 static int packet_net_init(struct net *net)
2060 {
2061         rwlock_init(&net->packet.sklist_lock);
2062         INIT_HLIST_HEAD(&net->packet.sklist);
2063
2064         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2065                 return -ENOMEM;
2066
2067         return 0;
2068 }
2069
2070 static void packet_net_exit(struct net *net)
2071 {
2072         proc_net_remove(net, "packet");
2073 }
2074
2075 static struct pernet_operations packet_net_ops = {
2076         .init = packet_net_init,
2077         .exit = packet_net_exit,
2078 };
2079
2080
2081 static void __exit packet_exit(void)
2082 {
2083         unregister_netdevice_notifier(&packet_netdev_notifier);
2084         unregister_pernet_subsys(&packet_net_ops);
2085         sock_unregister(PF_PACKET);
2086         proto_unregister(&packet_proto);
2087 }
2088
2089 static int __init packet_init(void)
2090 {
2091         int rc = proto_register(&packet_proto, 0);
2092
2093         if (rc != 0)
2094                 goto out;
2095
2096         sock_register(&packet_family_ops);
2097         register_pernet_subsys(&packet_net_ops);
2098         register_netdevice_notifier(&packet_netdev_notifier);
2099 out:
2100         return rc;
2101 }
2102
2103 module_init(packet_init);
2104 module_exit(packet_exit);
2105 MODULE_LICENSE("GPL");
2106 MODULE_ALIAS_NETPROTO(PF_PACKET);