git.oblomov.eu Git - 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  * Version:     $Id: af_packet.c,v 1.61 2002/02/08 03:57:19 davem Exp $
   9  *
  10  * Authors:     Ross Biro
  11  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  13  *
  14  * Fixes:
  15  *              Alan Cox        :       verify_area() now used correctly
  16  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  17  *              Alan Cox        :       tidied skbuff lists.
  18  *              Alan Cox        :       Now uses generic datagram routines I
  19  *                                      added. Also fixed the peek/read crash
  20  *                                      from all old Linux datagram code.
  21  *              Alan Cox        :       Uses the improved datagram code.
  22  *              Alan Cox        :       Added NULL's for socket options.
  23  *              Alan Cox        :       Re-commented the code.
  24  *              Alan Cox        :       Use new kernel side addressing
  25  *              Rob Janssen     :       Correct MTU usage.
  26  *              Dave Platt      :       Counter leaks caused by incorrect
  27  *                                      interrupt locking and some slightly
  28  *                                      dubious gcc output. Can you read
  29  *                                      compiler: it said _VOLATILE_
  30  *      Richard Kooijman        :       Timestamp fixes.
  31  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  32  *              Alan Cox        :       sendmsg/recvmsg support.
  33  *              Alan Cox        :       Protocol setting support
  34  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  35  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  36  *      Michal Ostrowski        :       Module initialization cleanup.
  37  *         Ulises Alonso        :       Frame number limit removal and
  38  *                                      packet_set_ring memory leak.
  39  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  40  *                                      The convention is that longer addresses
  41  *                                      will simply extend the hardware address
  42  *                                      byte arrays at the end of sockaddr_ll
  43  *                                      and packet_mreq.
  44  *
  45  *              This program is free software; you can redistribute it and/or
  46  *              modify it under the terms of the GNU General Public License
  47  *              as published by the Free Software Foundation; either version
  48  *              2 of the License, or (at your option) any later version.
  49  *
  50  */
  51
  52 #include <linux/types.h>
  53 #include <linux/sched.h>
  54 #include <linux/mm.h>
  55 #include <linux/capability.h>
  56 #include <linux/fcntl.h>
  57 #include <linux/socket.h>
  58 #include <linux/in.h>
  59 #include <linux/inet.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/if_packet.h>
  62 #include <linux/wireless.h>
  63 #include <linux/kmod.h>
  64 #include <net/ip.h>
  65 #include <net/protocol.h>
  66 #include <linux/skbuff.h>
  67 #include <net/sock.h>
  68 #include <linux/errno.h>
  69 #include <linux/timer.h>
  70 #include <asm/system.h>
  71 #include <asm/uaccess.h>
  72 #include <asm/ioctls.h>
  73 #include <asm/page.h>
  74 #include <asm/cacheflush.h>
  75 #include <asm/io.h>
  76 #include <linux/proc_fs.h>
  77 #include <linux/seq_file.h>
  78 #include <linux/poll.h>
  79 #include <linux/module.h>
  80 #include <linux/init.h>
  81
  82 #ifdef CONFIG_INET
  83 #include <net/inet_common.h>
  84 #endif
  85
  86 #define CONFIG_SOCK_PACKET      1
  87
  88 /*
  89    Proposed replacement for SIOC{ADD,DEL}MULTI and
  90    IFF_PROMISC, IFF_ALLMULTI flags.
  91
  92    It is more expensive, but I believe,
  93    it is really correct solution: reentereble, safe and fault tolerant.
  94
  95    IFF_PROMISC/IFF_ALLMULTI/SIOC{ADD/DEL}MULTI are faked by keeping
  96    reference count and global flag, so that real status is
  97    (gflag|(count != 0)), so that we can use obsolete faulty interface
  98    not harming clever users.
  99  */
 100 #define CONFIG_PACKET_MULTICAST 1
 101
 102 /*
 103    Assumptions:
 104    - if device has no dev->hard_header routine, it adds and removes ll header
 105      inside itself. In this case ll header is invisible outside of device,
 106      but higher levels still should reserve dev->hard_header_len.
 107      Some devices are enough clever to reallocate skb, when header
 108      will not fit to reserved space (tunnel), another ones are silly
 109      (PPP).
 110    - packet socket receives packets with pulled ll header,
 111      so that SOCK_RAW should push it back.
 112
 113 On receive:
 114 -----------
 115
 116 Incoming, dev->hard_header!=NULL
 117    mac.raw -> ll header
 118    data    -> data
 119
 120 Outgoing, dev->hard_header!=NULL
 121    mac.raw -> ll header
 122    data    -> ll header
 123
 124 Incoming, dev->hard_header==NULL
 125    mac.raw -> UNKNOWN position. It is very likely, that it points to ll header.
 126               PPP makes it, that is wrong, because introduce assymetry
 127               between rx and tx paths.
 128    data    -> data
 129
 130 Outgoing, dev->hard_header==NULL
 131    mac.raw -> data. ll header is still not built!
 132    data    -> data
 133
 134 Resume
 135   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 136
 137
 138 On transmit:
 139 ------------
 140
 141 dev->hard_header != NULL
 142    mac.raw -> ll header
 143    data    -> ll header
 144
 145 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 146    mac.raw -> data
 147    data -> data
 148
 149    We should set nh.raw on output to correct posistion,
 150    packet classifier depends on it.
 151  */
 152
 153 /* List of all packet sockets. */
 154 static HLIST_HEAD(packet_sklist);
 155 static DEFINE_RWLOCK(packet_sklist_lock);
 156
 157 static atomic_t packet_socks_nr;
 158
 159
 160 /* Private packet socket structures. */
 161
 162 #ifdef CONFIG_PACKET_MULTICAST
 163 struct packet_mclist
 164 {
 165         struct packet_mclist    *next;
 166         int                     ifindex;
 167         int                     count;
 168         unsigned short          type;
 169         unsigned short          alen;
 170         unsigned char           addr[MAX_ADDR_LEN];
 171 };
 172 /* identical to struct packet_mreq except it has
 173  * a longer address field.
 174  */
 175 struct packet_mreq_max
 176 {
 177         int             mr_ifindex;
 178         unsigned short  mr_type;
 179         unsigned short  mr_alen;
 180         unsigned char   mr_address[MAX_ADDR_LEN];
 181 };
 182 #endif
 183 #ifdef CONFIG_PACKET_MMAP
 184 static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing);
 185 #endif
 186
 187 static void packet_flush_mclist(struct sock *sk);
 188
 189 struct packet_sock {
 190         /* struct sock has to be the first member of packet_sock */
 191         struct sock             sk;
 192         struct tpacket_stats    stats;
 193 #ifdef CONFIG_PACKET_MMAP
 194         char *                  *pg_vec;
 195         unsigned int            head;
 196         unsigned int            frames_per_block;
 197         unsigned int            frame_size;
 198         unsigned int            frame_max;
 199         int                     copy_thresh;
 200 #endif
 201         struct packet_type      prot_hook;
 202         spinlock_t              bind_lock;
 203         char                    running;        /* prot_hook is attached*/
 204         int                     ifindex;        /* bound device         */
 205         __be16                  num;
 206 #ifdef CONFIG_PACKET_MULTICAST
 207         struct packet_mclist    *mclist;
 208 #endif
 209 #ifdef CONFIG_PACKET_MMAP
 210         atomic_t                mapped;
 211         unsigned int            pg_vec_order;
 212         unsigned int            pg_vec_pages;
 213         unsigned int            pg_vec_len;
 214 #endif
 215 };
 216
 217 #ifdef CONFIG_PACKET_MMAP
 218
 219 static inline char *packet_lookup_frame(struct packet_sock *po, unsigned int position)
 220 {
 221         unsigned int pg_vec_pos, frame_offset;
 222         char *frame;
 223
 224         pg_vec_pos = position / po->frames_per_block;
 225         frame_offset = position % po->frames_per_block;
 226
 227         frame = po->pg_vec[pg_vec_pos] + (frame_offset * po->frame_size);
 228
 229         return frame;
 230 }
 231 #endif
 232
 233 static inline struct packet_sock *pkt_sk(struct sock *sk)
 234 {
 235         return (struct packet_sock *)sk;
 236 }
 237
 238 static void packet_sock_destruct(struct sock *sk)
 239 {
 240         BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
 241         BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
 242
 243         if (!sock_flag(sk, SOCK_DEAD)) {
 244                 printk("Attempt to release alive packet socket: %p\n", sk);
 245                 return;
 246         }
 247
 248         atomic_dec(&packet_socks_nr);
 249 #ifdef PACKET_REFCNT_DEBUG
 250         printk(KERN_DEBUG "PACKET socket %p is free, %d are alive\n", sk, atomic_read(&packet_socks_nr));
 251 #endif
 252 }
 253
 254
 255 static const struct proto_ops packet_ops;
 256
 257 #ifdef CONFIG_SOCK_PACKET
 258 static const struct proto_ops packet_ops_spkt;
 259
 260 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,  struct packet_type *pt, struct net_device *orig_dev)
 261 {
 262         struct sock *sk;
 263         struct sockaddr_pkt *spkt;
 264
 265         /*
 266          *      When we registered the protocol we saved the socket in the data
 267          *      field for just this event.
 268          */
 269
 270         sk = pt->af_packet_priv;
 271
 272         /*
 273          *      Yank back the headers [hope the device set this
 274          *      right or kerboom...]
 275          *
 276          *      Incoming packets have ll header pulled,
 277          *      push it back.
 278          *
 279          *      For outgoing ones skb->data == skb->mac.raw
 280          *      so that this procedure is noop.
 281          */
 282
 283         if (skb->pkt_type == PACKET_LOOPBACK)
 284                 goto out;
 285
 286         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 287                 goto oom;
 288
 289         /* drop any routing info */
 290         dst_release(skb->dst);
 291         skb->dst = NULL;
 292
 293         /* drop conntrack reference */
 294         nf_reset(skb);
 295
 296         spkt = (struct sockaddr_pkt*)skb->cb;
 297
 298         skb_push(skb, skb->data-skb->mac.raw);
 299
 300         /*
 301          *      The SOCK_PACKET socket receives _all_ frames.
 302          */
 303
 304         spkt->spkt_family = dev->type;
 305         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
 306         spkt->spkt_protocol = skb->protocol;
 307
 308         /*
 309          *      Charge the memory to the socket. This is done specifically
 310          *      to prevent sockets using all the memory up.
 311          */
 312
 313         if (sock_queue_rcv_skb(sk,skb) == 0)
 314                 return 0;
 315
 316 out:
 317         kfree_skb(skb);
 318 oom:
 319         return 0;
 320 }
 321
 322
 323 /*
 324  *      Output a raw packet to a device layer. This bypasses all the other
 325  *      protocol layers and you must therefore supply it with a complete frame
 326  */
 327
 328 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
 329                                struct msghdr *msg, size_t len)
 330 {
 331         struct sock *sk = sock->sk;
 332         struct sockaddr_pkt *saddr=(struct sockaddr_pkt *)msg->msg_name;
 333         struct sk_buff *skb;
 334         struct net_device *dev;
 335         __be16 proto=0;
 336         int err;
 337
 338         /*
 339          *      Get and verify the address.
 340          */
 341
 342         if (saddr)
 343         {
 344                 if (msg->msg_namelen < sizeof(struct sockaddr))
 345                         return(-EINVAL);
 346                 if (msg->msg_namelen==sizeof(struct sockaddr_pkt))
 347                         proto=saddr->spkt_protocol;
 348         }
 349         else
 350                 return(-ENOTCONN);      /* SOCK_PACKET must be sent giving an address */
 351
 352         /*
 353          *      Find the device first to size check it
 354          */
 355
 356         saddr->spkt_device[13] = 0;
 357         dev = dev_get_by_name(saddr->spkt_device);
 358         err = -ENODEV;
 359         if (dev == NULL)
 360                 goto out_unlock;
 361
 362         err = -ENETDOWN;
 363         if (!(dev->flags & IFF_UP))
 364                 goto out_unlock;
 365
 366         /*
 367          *      You may not queue a frame bigger than the mtu. This is the lowest level
 368          *      raw protocol and you must do your own fragmentation at this level.
 369          */
 370
 371         err = -EMSGSIZE;
 372         if (len > dev->mtu + dev->hard_header_len)
 373                 goto out_unlock;
 374
 375         err = -ENOBUFS;
 376         skb = sock_wmalloc(sk, len + LL_RESERVED_SPACE(dev), 0, GFP_KERNEL);
 377
 378         /*
 379          *      If the write buffer is full, then tough. At this level the user gets to
 380          *      deal with the problem - do your own algorithmic backoffs. That's far
 381          *      more flexible.
 382          */
 383
 384         if (skb == NULL)
 385                 goto out_unlock;
 386
 387         /*
 388          *      Fill it in
 389          */
 390
 391         /* FIXME: Save some space for broken drivers that write a
 392          * hard header at transmission time by themselves. PPP is the
 393          * notable one here. This should really be fixed at the driver level.
 394          */
 395         skb_reserve(skb, LL_RESERVED_SPACE(dev));
 396         skb->nh.raw = skb->data;
 397
 398         /* Try to align data part correctly */
 399         if (dev->hard_header) {
 400                 skb->data -= dev->hard_header_len;
 401                 skb->tail -= dev->hard_header_len;
 402                 if (len < dev->hard_header_len)
 403                         skb->nh.raw = skb->data;
 404         }
 405
 406         /* Returns -EFAULT on error */
 407         err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
 408         skb->protocol = proto;
 409         skb->dev = dev;
 410         skb->priority = sk->sk_priority;
 411         if (err)
 412                 goto out_free;
 413
 414         /*
 415          *      Now send it
 416          */
 417
 418         dev_queue_xmit(skb);
 419         dev_put(dev);
 420         return(len);
 421
 422 out_free:
 423         kfree_skb(skb);
 424 out_unlock:
 425         if (dev)
 426                 dev_put(dev);
 427         return err;
 428 }
 429 #endif
 430
 431 static inline unsigned int run_filter(struct sk_buff *skb, struct sock *sk,
 432                                       unsigned int res)
 433 {
 434         struct sk_filter *filter;
 435
 436         rcu_read_lock_bh();
 437         filter = rcu_dereference(sk->sk_filter);
 438         if (filter != NULL)
 439                 res = sk_run_filter(skb, filter->insns, filter->len);
 440         rcu_read_unlock_bh();
 441
 442         return res;
 443 }
 444
 445 /*
 446    This function makes lazy skb cloning in hope that most of packets
 447    are discarded by BPF.
 448
 449    Note tricky part: we DO mangle shared skb! skb->data, skb->len
 450    and skb->cb are mangled. It works because (and until) packets
 451    falling here are owned by current CPU. Output packets are cloned
 452    by dev_queue_xmit_nit(), input packets are processed by net_bh
 453    sequencially, so that if we return skb to original state on exit,
 454    we will not harm anyone.
 455  */
 456
 457 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 458 {
 459         struct sock *sk;
 460         struct sockaddr_ll *sll;
 461         struct packet_sock *po;
 462         u8 * skb_head = skb->data;
 463         int skb_len = skb->len;
 464         unsigned int snaplen, res;
 465
 466         if (skb->pkt_type == PACKET_LOOPBACK)
 467                 goto drop;
 468
 469         sk = pt->af_packet_priv;
 470         po = pkt_sk(sk);
 471
 472         skb->dev = dev;
 473
 474         if (dev->hard_header) {
 475                 /* The device has an explicit notion of ll header,
 476                    exported to higher levels.
 477
 478                    Otherwise, the device hides datails of it frame
 479                    structure, so that corresponding packet head
 480                    never delivered to user.
 481                  */
 482                 if (sk->sk_type != SOCK_DGRAM)
 483                         skb_push(skb, skb->data - skb->mac.raw);
 484                 else if (skb->pkt_type == PACKET_OUTGOING) {
 485                         /* Special case: outgoing packets have ll header at head */
 486                         skb_pull(skb, skb->nh.raw - skb->data);
 487                 }
 488         }
 489
 490         snaplen = skb->len;
 491
 492         res = run_filter(skb, sk, snaplen);
 493         if (!res)
 494                 goto drop_n_restore;
 495         if (snaplen > res)
 496                 snaplen = res;
 497
 498         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
 499             (unsigned)sk->sk_rcvbuf)
 500                 goto drop_n_acct;
 501
 502         if (skb_shared(skb)) {
 503                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
 504                 if (nskb == NULL)
 505                         goto drop_n_acct;
 506
 507                 if (skb_head != skb->data) {
 508                         skb->data = skb_head;
 509                         skb->len = skb_len;
 510                 }
 511                 kfree_skb(skb);
 512                 skb = nskb;
 513         }
 514
 515         sll = (struct sockaddr_ll*)skb->cb;
 516         sll->sll_family = AF_PACKET;
 517         sll->sll_hatype = dev->type;
 518         sll->sll_protocol = skb->protocol;
 519         sll->sll_pkttype = skb->pkt_type;
 520         sll->sll_ifindex = dev->ifindex;
 521         sll->sll_halen = 0;
 522
 523         if (dev->hard_header_parse)
 524                 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
 525
 526         if (pskb_trim(skb, snaplen))
 527                 goto drop_n_acct;
 528
 529         skb_set_owner_r(skb, sk);
 530         skb->dev = NULL;
 531         dst_release(skb->dst);
 532         skb->dst = NULL;
 533
 534         /* drop conntrack reference */
 535         nf_reset(skb);
 536
 537         spin_lock(&sk->sk_receive_queue.lock);
 538         po->stats.tp_packets++;
 539         __skb_queue_tail(&sk->sk_receive_queue, skb);
 540         spin_unlock(&sk->sk_receive_queue.lock);
 541         sk->sk_data_ready(sk, skb->len);
 542         return 0;
 543
 544 drop_n_acct:
 545         spin_lock(&sk->sk_receive_queue.lock);
 546         po->stats.tp_drops++;
 547         spin_unlock(&sk->sk_receive_queue.lock);
 548
 549 drop_n_restore:
 550         if (skb_head != skb->data && skb_shared(skb)) {
 551                 skb->data = skb_head;
 552                 skb->len = skb_len;
 553         }
 554 drop:
 555         kfree_skb(skb);
 556         return 0;
 557 }
 558
 559 #ifdef CONFIG_PACKET_MMAP
 560 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 561 {
 562         struct sock *sk;
 563         struct packet_sock *po;
 564         struct sockaddr_ll *sll;
 565         struct tpacket_hdr *h;
 566         u8 * skb_head = skb->data;
 567         int skb_len = skb->len;
 568         unsigned int snaplen, res;
 569         unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
 570         unsigned short macoff, netoff;
 571         struct sk_buff *copy_skb = NULL;
 572
 573         if (skb->pkt_type == PACKET_LOOPBACK)
 574                 goto drop;
 575
 576         sk = pt->af_packet_priv;
 577         po = pkt_sk(sk);
 578
 579         if (dev->hard_header) {
 580                 if (sk->sk_type != SOCK_DGRAM)
 581                         skb_push(skb, skb->data - skb->mac.raw);
 582                 else if (skb->pkt_type == PACKET_OUTGOING) {
 583                         /* Special case: outgoing packets have ll header at head */
 584                         skb_pull(skb, skb->nh.raw - skb->data);
 585                         if (skb->ip_summed == CHECKSUM_PARTIAL)
 586                                 status |= TP_STATUS_CSUMNOTREADY;
 587                 }
 588         }
 589
 590         snaplen = skb->len;
 591
 592         res = run_filter(skb, sk, snaplen);
 593         if (!res)
 594                 goto drop_n_restore;
 595         if (snaplen > res)
 596                 snaplen = res;
 597
 598         if (sk->sk_type == SOCK_DGRAM) {
 599                 macoff = netoff = TPACKET_ALIGN(TPACKET_HDRLEN) + 16;
 600         } else {
 601                 unsigned maclen = skb->nh.raw - skb->data;
 602                 netoff = TPACKET_ALIGN(TPACKET_HDRLEN + (maclen < 16 ? 16 : maclen));
 603                 macoff = netoff - maclen;
 604         }
 605
 606         if (macoff + snaplen > po->frame_size) {
 607                 if (po->copy_thresh &&
 608                     atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
 609                     (unsigned)sk->sk_rcvbuf) {
 610                         if (skb_shared(skb)) {
 611                                 copy_skb = skb_clone(skb, GFP_ATOMIC);
 612                         } else {
 613                                 copy_skb = skb_get(skb);
 614                                 skb_head = skb->data;
 615                         }
 616                         if (copy_skb)
 617                                 skb_set_owner_r(copy_skb, sk);
 618                 }
 619                 snaplen = po->frame_size - macoff;
 620                 if ((int)snaplen < 0)
 621                         snaplen = 0;
 622         }
 623
 624         spin_lock(&sk->sk_receive_queue.lock);
 625         h = (struct tpacket_hdr *)packet_lookup_frame(po, po->head);
 626
 627         if (h->tp_status)
 628                 goto ring_is_full;
 629         po->head = po->head != po->frame_max ? po->head+1 : 0;
 630         po->stats.tp_packets++;
 631         if (copy_skb) {
 632                 status |= TP_STATUS_COPY;
 633                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
 634         }
 635         if (!po->stats.tp_drops)
 636                 status &= ~TP_STATUS_LOSING;
 637         spin_unlock(&sk->sk_receive_queue.lock);
 638
 639         skb_copy_bits(skb, 0, (u8*)h + macoff, snaplen);
 640
 641         h->tp_len = skb->len;
 642         h->tp_snaplen = snaplen;
 643         h->tp_mac = macoff;
 644         h->tp_net = netoff;
 645         if (skb->tstamp.off_sec == 0) {
 646                 __net_timestamp(skb);
 647                 sock_enable_timestamp(sk);
 648         }
 649         h->tp_sec = skb->tstamp.off_sec;
 650         h->tp_usec = skb->tstamp.off_usec;
 651
 652         sll = (struct sockaddr_ll*)((u8*)h + TPACKET_ALIGN(sizeof(*h)));
 653         sll->sll_halen = 0;
 654         if (dev->hard_header_parse)
 655                 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
 656         sll->sll_family = AF_PACKET;
 657         sll->sll_hatype = dev->type;
 658         sll->sll_protocol = skb->protocol;
 659         sll->sll_pkttype = skb->pkt_type;
 660         sll->sll_ifindex = dev->ifindex;
 661
 662         h->tp_status = status;
 663         smp_mb();
 664
 665         {
 666                 struct page *p_start, *p_end;
 667                 u8 *h_end = (u8 *)h + macoff + snaplen - 1;
 668
 669                 p_start = virt_to_page(h);
 670                 p_end = virt_to_page(h_end);
 671                 while (p_start <= p_end) {
 672                         flush_dcache_page(p_start);
 673                         p_start++;
 674                 }
 675         }
 676
 677         sk->sk_data_ready(sk, 0);
 678
 679 drop_n_restore:
 680         if (skb_head != skb->data && skb_shared(skb)) {
 681                 skb->data = skb_head;
 682                 skb->len = skb_len;
 683         }
 684 drop:
 685         kfree_skb(skb);
 686         return 0;
 687
 688 ring_is_full:
 689         po->stats.tp_drops++;
 690         spin_unlock(&sk->sk_receive_queue.lock);
 691
 692         sk->sk_data_ready(sk, 0);
 693         if (copy_skb)
 694                 kfree_skb(copy_skb);
 695         goto drop_n_restore;
 696 }
 697
 698 #endif
 699
 700
 701 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
 702                           struct msghdr *msg, size_t len)
 703 {
 704         struct sock *sk = sock->sk;
 705         struct sockaddr_ll *saddr=(struct sockaddr_ll *)msg->msg_name;
 706         struct sk_buff *skb;
 707         struct net_device *dev;
 708         __be16 proto;
 709         unsigned char *addr;
 710         int ifindex, err, reserve = 0;
 711
 712         /*
 713          *      Get and verify the address.
 714          */
 715
 716         if (saddr == NULL) {
 717                 struct packet_sock *po = pkt_sk(sk);
 718
 719                 ifindex = po->ifindex;
 720                 proto   = po->num;
 721                 addr    = NULL;
 722         } else {
 723                 err = -EINVAL;
 724                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
 725                         goto out;
 726                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
 727                         goto out;
 728                 ifindex = saddr->sll_ifindex;
 729                 proto   = saddr->sll_protocol;
 730                 addr    = saddr->sll_addr;
 731         }
 732
 733
 734         dev = dev_get_by_index(ifindex);
 735         err = -ENXIO;
 736         if (dev == NULL)
 737                 goto out_unlock;
 738         if (sock->type == SOCK_RAW)
 739                 reserve = dev->hard_header_len;
 740
 741         err = -ENETDOWN;
 742         if (!(dev->flags & IFF_UP))
 743                 goto out_unlock;
 744
 745         err = -EMSGSIZE;
 746         if (len > dev->mtu+reserve)
 747                 goto out_unlock;
 748
 749         skb = sock_alloc_send_skb(sk, len + LL_RESERVED_SPACE(dev),
 750                                 msg->msg_flags & MSG_DONTWAIT, &err);
 751         if (skb==NULL)
 752                 goto out_unlock;
 753
 754         skb_reserve(skb, LL_RESERVED_SPACE(dev));
 755         skb->nh.raw = skb->data;
 756
 757         if (dev->hard_header) {
 758                 int res;
 759                 err = -EINVAL;
 760                 res = dev->hard_header(skb, dev, ntohs(proto), addr, NULL, len);
 761                 if (sock->type != SOCK_DGRAM) {
 762                         skb->tail = skb->data;
 763                         skb->len = 0;
 764                 } else if (res < 0)
 765                         goto out_free;
 766         }
 767
 768         /* Returns -EFAULT on error */
 769         err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
 770         if (err)
 771                 goto out_free;
 772
 773         skb->protocol = proto;
 774         skb->dev = dev;
 775         skb->priority = sk->sk_priority;
 776
 777         /*
 778          *      Now send it
 779          */
 780
 781         err = dev_queue_xmit(skb);
 782         if (err > 0 && (err = net_xmit_errno(err)) != 0)
 783                 goto out_unlock;
 784
 785         dev_put(dev);
 786
 787         return(len);
 788
 789 out_free:
 790         kfree_skb(skb);
 791 out_unlock:
 792         if (dev)
 793                 dev_put(dev);
 794 out:
 795         return err;
 796 }
 797
 798 /*
 799  *      Close a PACKET socket. This is fairly simple. We immediately go
 800  *      to 'closed' state and remove our protocol entry in the device list.
 801  */
 802
 803 static int packet_release(struct socket *sock)
 804 {
 805         struct sock *sk = sock->sk;
 806         struct packet_sock *po;
 807
 808         if (!sk)
 809                 return 0;
 810
 811         po = pkt_sk(sk);
 812
 813         write_lock_bh(&packet_sklist_lock);
 814         sk_del_node_init(sk);
 815         write_unlock_bh(&packet_sklist_lock);
 816
 817         /*
 818          *      Unhook packet receive handler.
 819          */
 820
 821         if (po->running) {
 822                 /*
 823                  *      Remove the protocol hook
 824                  */
 825                 dev_remove_pack(&po->prot_hook);
 826                 po->running = 0;
 827                 po->num = 0;
 828                 __sock_put(sk);
 829         }
 830
 831 #ifdef CONFIG_PACKET_MULTICAST
 832         packet_flush_mclist(sk);
 833 #endif
 834
 835 #ifdef CONFIG_PACKET_MMAP
 836         if (po->pg_vec) {
 837                 struct tpacket_req req;
 838                 memset(&req, 0, sizeof(req));
 839                 packet_set_ring(sk, &req, 1);
 840         }
 841 #endif
 842
 843         /*
 844          *      Now the socket is dead. No more input will appear.
 845          */
 846
 847         sock_orphan(sk);
 848         sock->sk = NULL;
 849
 850         /* Purge queues */
 851
 852         skb_queue_purge(&sk->sk_receive_queue);
 853
 854         sock_put(sk);
 855         return 0;
 856 }
 857
 858 /*
 859  *      Attach a packet hook.
 860  */
 861
 862 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
 863 {
 864         struct packet_sock *po = pkt_sk(sk);
 865         /*
 866          *      Detach an existing hook if present.
 867          */
 868
 869         lock_sock(sk);
 870
 871         spin_lock(&po->bind_lock);
 872         if (po->running) {
 873                 __sock_put(sk);
 874                 po->running = 0;
 875                 po->num = 0;
 876                 spin_unlock(&po->bind_lock);
 877                 dev_remove_pack(&po->prot_hook);
 878                 spin_lock(&po->bind_lock);
 879         }
 880
 881         po->num = protocol;
 882         po->prot_hook.type = protocol;
 883         po->prot_hook.dev = dev;
 884
 885         po->ifindex = dev ? dev->ifindex : 0;
 886
 887         if (protocol == 0)
 888                 goto out_unlock;
 889
 890         if (dev) {
 891                 if (dev->flags&IFF_UP) {
 892                         dev_add_pack(&po->prot_hook);
 893                         sock_hold(sk);
 894                         po->running = 1;
 895                 } else {
 896                         sk->sk_err = ENETDOWN;
 897                         if (!sock_flag(sk, SOCK_DEAD))
 898                                 sk->sk_error_report(sk);
 899                 }
 900         } else {
 901                 dev_add_pack(&po->prot_hook);
 902                 sock_hold(sk);
 903                 po->running = 1;
 904         }
 905
 906 out_unlock:
 907         spin_unlock(&po->bind_lock);
 908         release_sock(sk);
 909         return 0;
 910 }
 911
 912 /*
 913  *      Bind a packet socket to a device
 914  */
 915
 916 #ifdef CONFIG_SOCK_PACKET
 917
 918 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 919 {
 920         struct sock *sk=sock->sk;
 921         char name[15];
 922         struct net_device *dev;
 923         int err = -ENODEV;
 924
 925         /*
 926          *      Check legality
 927          */
 928
 929         if (addr_len != sizeof(struct sockaddr))
 930                 return -EINVAL;
 931         strlcpy(name,uaddr->sa_data,sizeof(name));
 932
 933         dev = dev_get_by_name(name);
 934         if (dev) {
 935                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
 936                 dev_put(dev);
 937         }
 938         return err;
 939 }
 940 #endif
 941
 942 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 943 {
 944         struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
 945         struct sock *sk=sock->sk;
 946         struct net_device *dev = NULL;
 947         int err;
 948
 949
 950         /*
 951          *      Check legality
 952          */
 953
 954         if (addr_len < sizeof(struct sockaddr_ll))
 955                 return -EINVAL;
 956         if (sll->sll_family != AF_PACKET)
 957                 return -EINVAL;
 958
 959         if (sll->sll_ifindex) {
 960                 err = -ENODEV;
 961                 dev = dev_get_by_index(sll->sll_ifindex);
 962                 if (dev == NULL)
 963                         goto out;
 964         }
 965         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
 966         if (dev)
 967                 dev_put(dev);
 968
 969 out:
 970         return err;
 971 }
 972
 973 static struct proto packet_proto = {
 974         .name     = "PACKET",
 975         .owner    = THIS_MODULE,
 976         .obj_size = sizeof(struct packet_sock),
 977 };
 978
 979 /*
 980  *      Create a packet of type SOCK_PACKET.
 981  */
 982
 983 static int packet_create(struct socket *sock, int protocol)
 984 {
 985         struct sock *sk;
 986         struct packet_sock *po;
 987         __be16 proto = (__force __be16)protocol; /* weird, but documented */
 988         int err;
 989
 990         if (!capable(CAP_NET_RAW))
 991                 return -EPERM;
 992         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW
 993 #ifdef CONFIG_SOCK_PACKET
 994             && sock->type != SOCK_PACKET
 995 #endif
 996             )
 997                 return -ESOCKTNOSUPPORT;
 998
 999         sock->state = SS_UNCONNECTED;
1000
1001         err = -ENOBUFS;
1002         sk = sk_alloc(PF_PACKET, GFP_KERNEL, &packet_proto, 1);
1003         if (sk == NULL)
1004                 goto out;
1005
1006         sock->ops = &packet_ops;
1007 #ifdef CONFIG_SOCK_PACKET
1008         if (sock->type == SOCK_PACKET)
1009                 sock->ops = &packet_ops_spkt;
1010 #endif
1011         sock_init_data(sock, sk);
1012
1013         po = pkt_sk(sk);
1014         sk->sk_family = PF_PACKET;
1015         po->num = proto;
1016
1017         sk->sk_destruct = packet_sock_destruct;
1018         atomic_inc(&packet_socks_nr);
1019
1020         /*
1021          *      Attach a protocol block
1022          */
1023
1024         spin_lock_init(&po->bind_lock);
1025         po->prot_hook.func = packet_rcv;
1026 #ifdef CONFIG_SOCK_PACKET
1027         if (sock->type == SOCK_PACKET)
1028                 po->prot_hook.func = packet_rcv_spkt;
1029 #endif
1030         po->prot_hook.af_packet_priv = sk;
1031
1032         if (proto) {
1033                 po->prot_hook.type = proto;
1034                 dev_add_pack(&po->prot_hook);
1035                 sock_hold(sk);
1036                 po->running = 1;
1037         }
1038
1039         write_lock_bh(&packet_sklist_lock);
1040         sk_add_node(sk, &packet_sklist);
1041         write_unlock_bh(&packet_sklist_lock);
1042         return(0);
1043 out:
1044         return err;
1045 }
1046
1047 /*
1048  *      Pull a packet from our receive queue and hand it to the user.
1049  *      If necessary we block.
1050  */
1051
1052 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1053                           struct msghdr *msg, size_t len, int flags)
1054 {
1055         struct sock *sk = sock->sk;
1056         struct sk_buff *skb;
1057         int copied, err;
1058         struct sockaddr_ll *sll;
1059
1060         err = -EINVAL;
1061         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT))
1062                 goto out;
1063
1064 #if 0
1065         /* What error should we return now? EUNATTACH? */
1066         if (pkt_sk(sk)->ifindex < 0)
1067                 return -ENODEV;
1068 #endif
1069
1070         /*
1071          *      Call the generic datagram receiver. This handles all sorts
1072          *      of horrible races and re-entrancy so we can forget about it
1073          *      in the protocol layers.
1074          *
1075          *      Now it will return ENETDOWN, if device have just gone down,
1076          *      but then it will block.
1077          */
1078
1079         skb=skb_recv_datagram(sk,flags,flags&MSG_DONTWAIT,&err);
1080
1081         /*
1082          *      An error occurred so return it. Because skb_recv_datagram()
1083          *      handles the blocking we don't see and worry about blocking
1084          *      retries.
1085          */
1086
1087         if (skb == NULL)
1088                 goto out;
1089
1090         /*
1091          *      If the address length field is there to be filled in, we fill
1092          *      it in now.
1093          */
1094
1095         sll = (struct sockaddr_ll*)skb->cb;
1096         if (sock->type == SOCK_PACKET)
1097                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1098         else
1099                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1100
1101         /*
1102          *      You lose any data beyond the buffer you gave. If it worries a
1103          *      user program they can ask the device for its MTU anyway.
1104          */
1105
1106         copied = skb->len;
1107         if (copied > len)
1108         {
1109                 copied=len;
1110                 msg->msg_flags|=MSG_TRUNC;
1111         }
1112
1113         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1114         if (err)
1115                 goto out_free;
1116
1117         sock_recv_timestamp(msg, sk, skb);
1118
1119         if (msg->msg_name)
1120                 memcpy(msg->msg_name, skb->cb, msg->msg_namelen);
1121
1122         /*
1123          *      Free or return the buffer as appropriate. Again this
1124          *      hides all the races and re-entrancy issues from us.
1125          */
1126         err = (flags&MSG_TRUNC) ? skb->len : copied;
1127
1128 out_free:
1129         skb_free_datagram(sk, skb);
1130 out:
1131         return err;
1132 }
1133
1134 #ifdef CONFIG_SOCK_PACKET
1135 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1136                                int *uaddr_len, int peer)
1137 {
1138         struct net_device *dev;
1139         struct sock *sk = sock->sk;
1140
1141         if (peer)
1142                 return -EOPNOTSUPP;
1143
1144         uaddr->sa_family = AF_PACKET;
1145         dev = dev_get_by_index(pkt_sk(sk)->ifindex);
1146         if (dev) {
1147                 strlcpy(uaddr->sa_data, dev->name, 15);
1148                 dev_put(dev);
1149         } else
1150                 memset(uaddr->sa_data, 0, 14);
1151         *uaddr_len = sizeof(*uaddr);
1152
1153         return 0;
1154 }
1155 #endif
1156
1157 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1158                           int *uaddr_len, int peer)
1159 {
1160         struct net_device *dev;
1161         struct sock *sk = sock->sk;
1162         struct packet_sock *po = pkt_sk(sk);
1163         struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
1164
1165         if (peer)
1166                 return -EOPNOTSUPP;
1167
1168         sll->sll_family = AF_PACKET;
1169         sll->sll_ifindex = po->ifindex;
1170         sll->sll_protocol = po->num;
1171         dev = dev_get_by_index(po->ifindex);
1172         if (dev) {
1173                 sll->sll_hatype = dev->type;
1174                 sll->sll_halen = dev->addr_len;
1175                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1176                 dev_put(dev);
1177         } else {
1178                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
1179                 sll->sll_halen = 0;
1180         }
1181         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1182
1183         return 0;
1184 }
1185
1186 #ifdef CONFIG_PACKET_MULTICAST
1187 static void packet_dev_mc(struct net_device *dev, struct packet_mclist *i, int what)
1188 {
1189         switch (i->type) {
1190         case PACKET_MR_MULTICAST:
1191                 if (what > 0)
1192                         dev_mc_add(dev, i->addr, i->alen, 0);
1193                 else
1194                         dev_mc_delete(dev, i->addr, i->alen, 0);
1195                 break;
1196         case PACKET_MR_PROMISC:
1197                 dev_set_promiscuity(dev, what);
1198                 break;
1199         case PACKET_MR_ALLMULTI:
1200                 dev_set_allmulti(dev, what);
1201                 break;
1202         default:;
1203         }
1204 }
1205
1206 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1207 {
1208         for ( ; i; i=i->next) {
1209                 if (i->ifindex == dev->ifindex)
1210                         packet_dev_mc(dev, i, what);
1211         }
1212 }
1213
1214 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1215 {
1216         struct packet_sock *po = pkt_sk(sk);
1217         struct packet_mclist *ml, *i;
1218         struct net_device *dev;
1219         int err;
1220
1221         rtnl_lock();
1222
1223         err = -ENODEV;
1224         dev = __dev_get_by_index(mreq->mr_ifindex);
1225         if (!dev)
1226                 goto done;
1227
1228         err = -EINVAL;
1229         if (mreq->mr_alen > dev->addr_len)
1230                 goto done;
1231
1232         err = -ENOBUFS;
1233         i = kmalloc(sizeof(*i), GFP_KERNEL);
1234         if (i == NULL)
1235                 goto done;
1236
1237         err = 0;
1238         for (ml = po->mclist; ml; ml = ml->next) {
1239                 if (ml->ifindex == mreq->mr_ifindex &&
1240                     ml->type == mreq->mr_type &&
1241                     ml->alen == mreq->mr_alen &&
1242                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1243                         ml->count++;
1244                         /* Free the new element ... */
1245                         kfree(i);
1246                         goto done;
1247                 }
1248         }
1249
1250         i->type = mreq->mr_type;
1251         i->ifindex = mreq->mr_ifindex;
1252         i->alen = mreq->mr_alen;
1253         memcpy(i->addr, mreq->mr_address, i->alen);
1254         i->count = 1;
1255         i->next = po->mclist;
1256         po->mclist = i;
1257         packet_dev_mc(dev, i, +1);
1258
1259 done:
1260         rtnl_unlock();
1261         return err;
1262 }
1263
1264 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1265 {
1266         struct packet_mclist *ml, **mlp;
1267
1268         rtnl_lock();
1269
1270         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1271                 if (ml->ifindex == mreq->mr_ifindex &&
1272                     ml->type == mreq->mr_type &&
1273                     ml->alen == mreq->mr_alen &&
1274                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1275                         if (--ml->count == 0) {
1276                                 struct net_device *dev;
1277                                 *mlp = ml->next;
1278                                 dev = dev_get_by_index(ml->ifindex);
1279                                 if (dev) {
1280                                         packet_dev_mc(dev, ml, -1);
1281                                         dev_put(dev);
1282                                 }
1283                                 kfree(ml);
1284                         }
1285                         rtnl_unlock();
1286                         return 0;
1287                 }
1288         }
1289         rtnl_unlock();
1290         return -EADDRNOTAVAIL;
1291 }
1292
1293 static void packet_flush_mclist(struct sock *sk)
1294 {
1295         struct packet_sock *po = pkt_sk(sk);
1296         struct packet_mclist *ml;
1297
1298         if (!po->mclist)
1299                 return;
1300
1301         rtnl_lock();
1302         while ((ml = po->mclist) != NULL) {
1303                 struct net_device *dev;
1304
1305                 po->mclist = ml->next;
1306                 if ((dev = dev_get_by_index(ml->ifindex)) != NULL) {
1307                         packet_dev_mc(dev, ml, -1);
1308                         dev_put(dev);
1309                 }
1310                 kfree(ml);
1311         }
1312         rtnl_unlock();
1313 }
1314 #endif
1315
1316 static int
1317 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, int optlen)
1318 {
1319         struct sock *sk = sock->sk;
1320         int ret;
1321
1322         if (level != SOL_PACKET)
1323                 return -ENOPROTOOPT;
1324
1325         switch(optname) {
1326 #ifdef CONFIG_PACKET_MULTICAST
1327         case PACKET_ADD_MEMBERSHIP:
1328         case PACKET_DROP_MEMBERSHIP:
1329         {
1330                 struct packet_mreq_max mreq;
1331                 int len = optlen;
1332                 memset(&mreq, 0, sizeof(mreq));
1333                 if (len < sizeof(struct packet_mreq))
1334                         return -EINVAL;
1335                 if (len > sizeof(mreq))
1336                         len = sizeof(mreq);
1337                 if (copy_from_user(&mreq,optval,len))
1338                         return -EFAULT;
1339                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1340                         return -EINVAL;
1341                 if (optname == PACKET_ADD_MEMBERSHIP)
1342                         ret = packet_mc_add(sk, &mreq);
1343                 else
1344                         ret = packet_mc_drop(sk, &mreq);
1345                 return ret;
1346         }
1347 #endif
1348 #ifdef CONFIG_PACKET_MMAP
1349         case PACKET_RX_RING:
1350         {
1351                 struct tpacket_req req;
1352
1353                 if (optlen<sizeof(req))
1354                         return -EINVAL;
1355                 if (copy_from_user(&req,optval,sizeof(req)))
1356                         return -EFAULT;
1357                 return packet_set_ring(sk, &req, 0);
1358         }
1359         case PACKET_COPY_THRESH:
1360         {
1361                 int val;
1362
1363                 if (optlen!=sizeof(val))
1364                         return -EINVAL;
1365                 if (copy_from_user(&val,optval,sizeof(val)))
1366                         return -EFAULT;
1367
1368                 pkt_sk(sk)->copy_thresh = val;
1369                 return 0;
1370         }
1371 #endif
1372         default:
1373                 return -ENOPROTOOPT;
1374         }
1375 }
1376
1377 static int packet_getsockopt(struct socket *sock, int level, int optname,
1378                              char __user *optval, int __user *optlen)
1379 {
1380         int len;
1381         struct sock *sk = sock->sk;
1382         struct packet_sock *po = pkt_sk(sk);
1383
1384         if (level != SOL_PACKET)
1385                 return -ENOPROTOOPT;
1386
1387         if (get_user(len, optlen))
1388                 return -EFAULT;
1389
1390         if (len < 0)
1391                 return -EINVAL;
1392
1393         switch(optname) {
1394         case PACKET_STATISTICS:
1395         {
1396                 struct tpacket_stats st;
1397
1398                 if (len > sizeof(struct tpacket_stats))
1399                         len = sizeof(struct tpacket_stats);
1400                 spin_lock_bh(&sk->sk_receive_queue.lock);
1401                 st = po->stats;
1402                 memset(&po->stats, 0, sizeof(st));
1403                 spin_unlock_bh(&sk->sk_receive_queue.lock);
1404                 st.tp_packets += st.tp_drops;
1405
1406                 if (copy_to_user(optval, &st, len))
1407                         return -EFAULT;
1408                 break;
1409         }
1410         default:
1411                 return -ENOPROTOOPT;
1412         }
1413
1414         if (put_user(len, optlen))
1415                 return -EFAULT;
1416         return 0;
1417 }
1418
1419
1420 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
1421 {
1422         struct sock *sk;
1423         struct hlist_node *node;
1424         struct net_device *dev = (struct net_device*)data;
1425
1426         read_lock(&packet_sklist_lock);
1427         sk_for_each(sk, node, &packet_sklist) {
1428                 struct packet_sock *po = pkt_sk(sk);
1429
1430                 switch (msg) {
1431                 case NETDEV_UNREGISTER:
1432 #ifdef CONFIG_PACKET_MULTICAST
1433                         if (po->mclist)
1434                                 packet_dev_mclist(dev, po->mclist, -1);
1435                         // fallthrough
1436 #endif
1437                 case NETDEV_DOWN:
1438                         if (dev->ifindex == po->ifindex) {
1439                                 spin_lock(&po->bind_lock);
1440                                 if (po->running) {
1441                                         __dev_remove_pack(&po->prot_hook);
1442                                         __sock_put(sk);
1443                                         po->running = 0;
1444                                         sk->sk_err = ENETDOWN;
1445                                         if (!sock_flag(sk, SOCK_DEAD))
1446                                                 sk->sk_error_report(sk);
1447                                 }
1448                                 if (msg == NETDEV_UNREGISTER) {
1449                                         po->ifindex = -1;
1450                                         po->prot_hook.dev = NULL;
1451                                 }
1452                                 spin_unlock(&po->bind_lock);
1453                         }
1454                         break;
1455                 case NETDEV_UP:
1456                         spin_lock(&po->bind_lock);
1457                         if (dev->ifindex == po->ifindex && po->num &&
1458                             !po->running) {
1459                                 dev_add_pack(&po->prot_hook);
1460                                 sock_hold(sk);
1461                                 po->running = 1;
1462                         }
1463                         spin_unlock(&po->bind_lock);
1464                         break;
1465                 }
1466         }
1467         read_unlock(&packet_sklist_lock);
1468         return NOTIFY_DONE;
1469 }
1470
1471
1472 static int packet_ioctl(struct socket *sock, unsigned int cmd,
1473                         unsigned long arg)
1474 {
1475         struct sock *sk = sock->sk;
1476
1477         switch(cmd) {
1478                 case SIOCOUTQ:
1479                 {
1480                         int amount = atomic_read(&sk->sk_wmem_alloc);
1481                         return put_user(amount, (int __user *)arg);
1482                 }
1483                 case SIOCINQ:
1484                 {
1485                         struct sk_buff *skb;
1486                         int amount = 0;
1487
1488                         spin_lock_bh(&sk->sk_receive_queue.lock);
1489                         skb = skb_peek(&sk->sk_receive_queue);
1490                         if (skb)
1491                                 amount = skb->len;
1492                         spin_unlock_bh(&sk->sk_receive_queue.lock);
1493                         return put_user(amount, (int __user *)arg);
1494                 }
1495                 case SIOCGSTAMP:
1496                         return sock_get_timestamp(sk, (struct timeval __user *)arg);
1497
1498 #ifdef CONFIG_INET
1499                 case SIOCADDRT:
1500                 case SIOCDELRT:
1501                 case SIOCDARP:
1502                 case SIOCGARP:
1503                 case SIOCSARP:
1504                 case SIOCGIFADDR:
1505                 case SIOCSIFADDR:
1506                 case SIOCGIFBRDADDR:
1507                 case SIOCSIFBRDADDR:
1508                 case SIOCGIFNETMASK:
1509                 case SIOCSIFNETMASK:
1510                 case SIOCGIFDSTADDR:
1511                 case SIOCSIFDSTADDR:
1512                 case SIOCSIFFLAGS:
1513                         return inet_dgram_ops.ioctl(sock, cmd, arg);
1514 #endif
1515
1516                 default:
1517                         return -ENOIOCTLCMD;
1518         }
1519         return 0;
1520 }
1521
1522 #ifndef CONFIG_PACKET_MMAP
1523 #define packet_mmap sock_no_mmap
1524 #define packet_poll datagram_poll
1525 #else
1526
1527 static unsigned int packet_poll(struct file * file, struct socket *sock,
1528                                 poll_table *wait)
1529 {
1530         struct sock *sk = sock->sk;
1531         struct packet_sock *po = pkt_sk(sk);
1532         unsigned int mask = datagram_poll(file, sock, wait);
1533
1534         spin_lock_bh(&sk->sk_receive_queue.lock);
1535         if (po->pg_vec) {
1536                 unsigned last = po->head ? po->head-1 : po->frame_max;
1537                 struct tpacket_hdr *h;
1538
1539                 h = (struct tpacket_hdr *)packet_lookup_frame(po, last);
1540
1541                 if (h->tp_status)
1542                         mask |= POLLIN | POLLRDNORM;
1543         }
1544         spin_unlock_bh(&sk->sk_receive_queue.lock);
1545         return mask;
1546 }
1547
1548
1549 /* Dirty? Well, I still did not learn better way to account
1550  * for user mmaps.
1551  */
1552
1553 static void packet_mm_open(struct vm_area_struct *vma)
1554 {
1555         struct file *file = vma->vm_file;
1556         struct socket * sock = file->private_data;
1557         struct sock *sk = sock->sk;
1558
1559         if (sk)
1560                 atomic_inc(&pkt_sk(sk)->mapped);
1561 }
1562
1563 static void packet_mm_close(struct vm_area_struct *vma)
1564 {
1565         struct file *file = vma->vm_file;
1566         struct socket * sock = file->private_data;
1567         struct sock *sk = sock->sk;
1568
1569         if (sk)
1570                 atomic_dec(&pkt_sk(sk)->mapped);
1571 }
1572
1573 static struct vm_operations_struct packet_mmap_ops = {
1574         .open = packet_mm_open,
1575         .close =packet_mm_close,
1576 };
1577
1578 static inline struct page *pg_vec_endpage(char *one_pg_vec, unsigned int order)
1579 {
1580         return virt_to_page(one_pg_vec + (PAGE_SIZE << order) - 1);
1581 }
1582
1583 static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
1584 {
1585         int i;
1586
1587         for (i = 0; i < len; i++) {
1588                 if (likely(pg_vec[i]))
1589                         free_pages((unsigned long) pg_vec[i], order);
1590         }
1591         kfree(pg_vec);
1592 }
1593
1594 static inline char *alloc_one_pg_vec_page(unsigned long order)
1595 {
1596         return (char *) __get_free_pages(GFP_KERNEL | __GFP_COMP | __GFP_ZERO,
1597                                          order);
1598 }
1599
1600 static char **alloc_pg_vec(struct tpacket_req *req, int order)
1601 {
1602         unsigned int block_nr = req->tp_block_nr;
1603         char **pg_vec;
1604         int i;
1605
1606         pg_vec = kzalloc(block_nr * sizeof(char *), GFP_KERNEL);
1607         if (unlikely(!pg_vec))
1608                 goto out;
1609
1610         for (i = 0; i < block_nr; i++) {
1611                 pg_vec[i] = alloc_one_pg_vec_page(order);
1612                 if (unlikely(!pg_vec[i]))
1613                         goto out_free_pgvec;
1614         }
1615
1616 out:
1617         return pg_vec;
1618
1619 out_free_pgvec:
1620         free_pg_vec(pg_vec, order, block_nr);
1621         pg_vec = NULL;
1622         goto out;
1623 }
1624
1625 static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing)
1626 {
1627         char **pg_vec = NULL;
1628         struct packet_sock *po = pkt_sk(sk);
1629         int was_running, order = 0;
1630         __be16 num;
1631         int err = 0;
1632
1633         if (req->tp_block_nr) {
1634                 int i, l;
1635
1636                 /* Sanity tests and some calculations */
1637
1638                 if (unlikely(po->pg_vec))
1639                         return -EBUSY;
1640
1641                 if (unlikely((int)req->tp_block_size <= 0))
1642                         return -EINVAL;
1643                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
1644                         return -EINVAL;
1645                 if (unlikely(req->tp_frame_size < TPACKET_HDRLEN))
1646                         return -EINVAL;
1647                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
1648                         return -EINVAL;
1649
1650                 po->frames_per_block = req->tp_block_size/req->tp_frame_size;
1651                 if (unlikely(po->frames_per_block <= 0))
1652                         return -EINVAL;
1653                 if (unlikely((po->frames_per_block * req->tp_block_nr) !=
1654                              req->tp_frame_nr))
1655                         return -EINVAL;
1656
1657                 err = -ENOMEM;
1658                 order = get_order(req->tp_block_size);
1659                 pg_vec = alloc_pg_vec(req, order);
1660                 if (unlikely(!pg_vec))
1661                         goto out;
1662
1663                 l = 0;
1664                 for (i = 0; i < req->tp_block_nr; i++) {
1665                         char *ptr = pg_vec[i];
1666                         struct tpacket_hdr *header;
1667                         int k;
1668
1669                         for (k = 0; k < po->frames_per_block; k++) {
1670                                 header = (struct tpacket_hdr *) ptr;
1671                                 header->tp_status = TP_STATUS_KERNEL;
1672                                 ptr += req->tp_frame_size;
1673                         }
1674                 }
1675                 /* Done */
1676         } else {
1677                 if (unlikely(req->tp_frame_nr))
1678                         return -EINVAL;
1679         }
1680
1681         lock_sock(sk);
1682
1683         /* Detach socket from network */
1684         spin_lock(&po->bind_lock);
1685         was_running = po->running;
1686         num = po->num;
1687         if (was_running) {
1688                 __dev_remove_pack(&po->prot_hook);
1689                 po->num = 0;
1690                 po->running = 0;
1691                 __sock_put(sk);
1692         }
1693         spin_unlock(&po->bind_lock);
1694
1695         synchronize_net();
1696
1697         err = -EBUSY;
1698         if (closing || atomic_read(&po->mapped) == 0) {
1699                 err = 0;
1700 #define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
1701
1702                 spin_lock_bh(&sk->sk_receive_queue.lock);
1703                 pg_vec = XC(po->pg_vec, pg_vec);
1704                 po->frame_max = (req->tp_frame_nr - 1);
1705                 po->head = 0;
1706                 po->frame_size = req->tp_frame_size;
1707                 spin_unlock_bh(&sk->sk_receive_queue.lock);
1708
1709                 order = XC(po->pg_vec_order, order);
1710                 req->tp_block_nr = XC(po->pg_vec_len, req->tp_block_nr);
1711
1712                 po->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
1713                 po->prot_hook.func = po->pg_vec ? tpacket_rcv : packet_rcv;
1714                 skb_queue_purge(&sk->sk_receive_queue);
1715 #undef XC
1716                 if (atomic_read(&po->mapped))
1717                         printk(KERN_DEBUG "packet_mmap: vma is busy: %d\n", atomic_read(&po->mapped));
1718         }
1719
1720         spin_lock(&po->bind_lock);
1721         if (was_running && !po->running) {
1722                 sock_hold(sk);
1723                 po->running = 1;
1724                 po->num = num;
1725                 dev_add_pack(&po->prot_hook);
1726         }
1727         spin_unlock(&po->bind_lock);
1728
1729         release_sock(sk);
1730
1731         if (pg_vec)
1732                 free_pg_vec(pg_vec, order, req->tp_block_nr);
1733 out:
1734         return err;
1735 }
1736
1737 static int packet_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1738 {
1739         struct sock *sk = sock->sk;
1740         struct packet_sock *po = pkt_sk(sk);
1741         unsigned long size;
1742         unsigned long start;
1743         int err = -EINVAL;
1744         int i;
1745
1746         if (vma->vm_pgoff)
1747                 return -EINVAL;
1748
1749         size = vma->vm_end - vma->vm_start;
1750
1751         lock_sock(sk);
1752         if (po->pg_vec == NULL)
1753                 goto out;
1754         if (size != po->pg_vec_len*po->pg_vec_pages*PAGE_SIZE)
1755                 goto out;
1756
1757         start = vma->vm_start;
1758         for (i = 0; i < po->pg_vec_len; i++) {
1759                 struct page *page = virt_to_page(po->pg_vec[i]);
1760                 int pg_num;
1761
1762                 for (pg_num = 0; pg_num < po->pg_vec_pages; pg_num++, page++) {
1763                         err = vm_insert_page(vma, start, page);
1764                         if (unlikely(err))
1765                                 goto out;
1766                         start += PAGE_SIZE;
1767                 }
1768         }
1769         atomic_inc(&po->mapped);
1770         vma->vm_ops = &packet_mmap_ops;
1771         err = 0;
1772
1773 out:
1774         release_sock(sk);
1775         return err;
1776 }
1777 #endif
1778
1779
1780 #ifdef CONFIG_SOCK_PACKET
1781 static const struct proto_ops packet_ops_spkt = {
1782         .family =       PF_PACKET,
1783         .owner =        THIS_MODULE,
1784         .release =      packet_release,
1785         .bind =         packet_bind_spkt,
1786         .connect =      sock_no_connect,
1787         .socketpair =   sock_no_socketpair,
1788         .accept =       sock_no_accept,
1789         .getname =      packet_getname_spkt,
1790         .poll =         datagram_poll,
1791         .ioctl =        packet_ioctl,
1792         .listen =       sock_no_listen,
1793         .shutdown =     sock_no_shutdown,
1794         .setsockopt =   sock_no_setsockopt,
1795         .getsockopt =   sock_no_getsockopt,
1796         .sendmsg =      packet_sendmsg_spkt,
1797         .recvmsg =      packet_recvmsg,
1798         .mmap =         sock_no_mmap,
1799         .sendpage =     sock_no_sendpage,
1800 };
1801 #endif
1802
1803 static const struct proto_ops packet_ops = {
1804         .family =       PF_PACKET,
1805         .owner =        THIS_MODULE,
1806         .release =      packet_release,
1807         .bind =         packet_bind,
1808         .connect =      sock_no_connect,
1809         .socketpair =   sock_no_socketpair,
1810         .accept =       sock_no_accept,
1811         .getname =      packet_getname,
1812         .poll =         packet_poll,
1813         .ioctl =        packet_ioctl,
1814         .listen =       sock_no_listen,
1815         .shutdown =     sock_no_shutdown,
1816         .setsockopt =   packet_setsockopt,
1817         .getsockopt =   packet_getsockopt,
1818         .sendmsg =      packet_sendmsg,
1819         .recvmsg =      packet_recvmsg,
1820         .mmap =         packet_mmap,
1821         .sendpage =     sock_no_sendpage,
1822 };
1823
1824 static struct net_proto_family packet_family_ops = {
1825         .family =       PF_PACKET,
1826         .create =       packet_create,
1827         .owner  =       THIS_MODULE,
1828 };
1829
1830 static struct notifier_block packet_netdev_notifier = {
1831         .notifier_call =packet_notifier,
1832 };
1833
1834 #ifdef CONFIG_PROC_FS
1835 static inline struct sock *packet_seq_idx(loff_t off)
1836 {
1837         struct sock *s;
1838         struct hlist_node *node;
1839
1840         sk_for_each(s, node, &packet_sklist) {
1841                 if (!off--)
1842                         return s;
1843         }
1844         return NULL;
1845 }
1846
1847 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
1848 {
1849         read_lock(&packet_sklist_lock);
1850         return *pos ? packet_seq_idx(*pos - 1) : SEQ_START_TOKEN;
1851 }
1852
1853 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1854 {
1855         ++*pos;
1856         return  (v == SEQ_START_TOKEN)
1857                 ? sk_head(&packet_sklist)
1858                 : sk_next((struct sock*)v) ;
1859 }
1860
1861 static void packet_seq_stop(struct seq_file *seq, void *v)
1862 {
1863         read_unlock(&packet_sklist_lock);
1864 }
1865
1866 static int packet_seq_show(struct seq_file *seq, void *v)
1867 {
1868         if (v == SEQ_START_TOKEN)
1869                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
1870         else {
1871                 struct sock *s = v;
1872                 const struct packet_sock *po = pkt_sk(s);
1873
1874                 seq_printf(seq,
1875                            "%p %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
1876                            s,
1877                            atomic_read(&s->sk_refcnt),
1878                            s->sk_type,
1879                            ntohs(po->num),
1880                            po->ifindex,
1881                            po->running,
1882                            atomic_read(&s->sk_rmem_alloc),
1883                            sock_i_uid(s),
1884                            sock_i_ino(s) );
1885         }
1886
1887         return 0;
1888 }
1889
1890 static struct seq_operations packet_seq_ops = {
1891         .start  = packet_seq_start,
1892         .next   = packet_seq_next,
1893         .stop   = packet_seq_stop,
1894         .show   = packet_seq_show,
1895 };
1896
1897 static int packet_seq_open(struct inode *inode, struct file *file)
1898 {
1899         return seq_open(file, &packet_seq_ops);
1900 }
1901
1902 static struct file_operations packet_seq_fops = {
1903         .owner          = THIS_MODULE,
1904         .open           = packet_seq_open,
1905         .read           = seq_read,
1906         .llseek         = seq_lseek,
1907         .release        = seq_release,
1908 };
1909
1910 #endif
1911
1912 static void __exit packet_exit(void)
1913 {
1914         proc_net_remove("packet");
1915         unregister_netdevice_notifier(&packet_netdev_notifier);
1916         sock_unregister(PF_PACKET);
1917         proto_unregister(&packet_proto);
1918 }
1919
1920 static int __init packet_init(void)
1921 {
1922         int rc = proto_register(&packet_proto, 0);
1923
1924         if (rc != 0)
1925                 goto out;
1926
1927         sock_register(&packet_family_ops);
1928         register_netdevice_notifier(&packet_netdev_notifier);
1929         proc_net_fops_create("packet", 0, &packet_seq_fops);
1930 out:
1931         return rc;
1932 }
1933
1934 module_init(packet_init);
1935 module_exit(packet_exit);
1936 MODULE_LICENSE("GPL");
1937 MODULE_ALIAS_NETPROTO(PF_PACKET);