net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              Johann Baudy    :       Added TX RING.
  43  *
  44  *              This program is free software; you can redistribute it and/or
  45  *              modify it under the terms of the GNU General Public License
  46  *              as published by the Free Software Foundation; either version
  47  *              2 of the License, or (at your option) any later version.
  48  *
  49  */
  50
  51 #include <linux/types.h>
  52 #include <linux/mm.h>
  53 #include <linux/capability.h>
  54 #include <linux/fcntl.h>
  55 #include <linux/socket.h>
  56 #include <linux/in.h>
  57 #include <linux/inet.h>
  58 #include <linux/netdevice.h>
  59 #include <linux/if_packet.h>
  60 #include <linux/wireless.h>
  61 #include <linux/kernel.h>
  62 #include <linux/kmod.h>
  63 #include <linux/slab.h>
  64 #include <linux/vmalloc.h>
  65 #include <net/net_namespace.h>
  66 #include <net/ip.h>
  67 #include <net/protocol.h>
  68 #include <linux/skbuff.h>
  69 #include <net/sock.h>
  70 #include <linux/errno.h>
  71 #include <linux/timer.h>
  72 #include <asm/system.h>
  73 #include <asm/uaccess.h>
  74 #include <asm/ioctls.h>
  75 #include <asm/page.h>
  76 #include <asm/cacheflush.h>
  77 #include <asm/io.h>
  78 #include <linux/proc_fs.h>
  79 #include <linux/seq_file.h>
  80 #include <linux/poll.h>
  81 #include <linux/module.h>
  82 #include <linux/init.h>
  83 #include <linux/mutex.h>
  84 #include <linux/if_vlan.h>
  85 #include <linux/virtio_net.h>
  86 #include <linux/errqueue.h>
  87 #include <linux/net_tstamp.h>
  88
  89 #ifdef CONFIG_INET
  90 #include <net/inet_common.h>
  91 #endif
  92
  93 /*
  94    Assumptions:
  95    - if device has no dev->hard_header routine, it adds and removes ll header
  96      inside itself. In this case ll header is invisible outside of device,
  97      but higher levels still should reserve dev->hard_header_len.
  98      Some devices are enough clever to reallocate skb, when header
  99      will not fit to reserved space (tunnel), another ones are silly
 100      (PPP).
 101    - packet socket receives packets with pulled ll header,
 102      so that SOCK_RAW should push it back.
 103
 104 On receive:
 105 -----------
 106
 107 Incoming, dev->hard_header!=NULL
 108    mac_header -> ll header
 109    data       -> data
 110
 111 Outgoing, dev->hard_header!=NULL
 112    mac_header -> ll header
 113    data       -> ll header
 114
 115 Incoming, dev->hard_header==NULL
 116    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 117                  header.  PPP makes it, that is wrong, because introduce
 118                  assymetry between rx and tx paths.
 119    data       -> data
 120
 121 Outgoing, dev->hard_header==NULL
 122    mac_header -> data. ll header is still not built!
 123    data       -> data
 124
 125 Resume
 126   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 127
 128
 129 On transmit:
 130 ------------
 131
 132 dev->hard_header != NULL
 133    mac_header -> ll header
 134    data       -> ll header
 135
 136 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 137    mac_header -> data
 138    data       -> data
 139
 140    We should set nh.raw on output to correct posistion,
 141    packet classifier depends on it.
 142  */
 143
 144 /* Private packet socket structures. */
 145
 146 struct packet_mclist {
 147         struct packet_mclist    *next;
 148         int                     ifindex;
 149         int                     count;
 150         unsigned short          type;
 151         unsigned short          alen;
 152         unsigned char           addr[MAX_ADDR_LEN];
 153 };
 154 /* identical to struct packet_mreq except it has
 155  * a longer address field.
 156  */
 157 struct packet_mreq_max {
 158         int             mr_ifindex;
 159         unsigned short  mr_type;
 160         unsigned short  mr_alen;
 161         unsigned char   mr_address[MAX_ADDR_LEN];
 162 };
 163
 164 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
 165                 int closing, int tx_ring);
 166
 167 struct pgv {
 168         char *buffer;
 169 };
 170
 171 struct packet_ring_buffer {
 172         struct pgv              *pg_vec;
 173         unsigned int            head;
 174         unsigned int            frames_per_block;
 175         unsigned int            frame_size;
 176         unsigned int            frame_max;
 177
 178         unsigned int            pg_vec_order;
 179         unsigned int            pg_vec_pages;
 180         unsigned int            pg_vec_len;
 181
 182         atomic_t                pending;
 183 };
 184
 185 struct packet_sock;
 186 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 187
 188 static void packet_flush_mclist(struct sock *sk);
 189
 190 struct packet_fanout;
 191 struct packet_sock {
 192         /* struct sock has to be the first member of packet_sock */
 193         struct sock             sk;
 194         struct packet_fanout    *fanout;
 195         struct tpacket_stats    stats;
 196         struct packet_ring_buffer       rx_ring;
 197         struct packet_ring_buffer       tx_ring;
 198         int                     copy_thresh;
 199         spinlock_t              bind_lock;
 200         struct mutex            pg_vec_lock;
 201         unsigned int            running:1,      /* prot_hook is attached*/
 202                                 auxdata:1,
 203                                 origdev:1,
 204                                 has_vnet_hdr:1;
 205         int                     ifindex;        /* bound device         */
 206         __be16                  num;
 207         struct packet_mclist    *mclist;
 208         atomic_t                mapped;
 209         enum tpacket_versions   tp_version;
 210         unsigned int            tp_hdrlen;
 211         unsigned int            tp_reserve;
 212         unsigned int            tp_loss:1;
 213         unsigned int            tp_tstamp;
 214         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 215 };
 216
 217 #define PACKET_FANOUT_MAX       256
 218
 219 struct packet_fanout {
 220 #ifdef CONFIG_NET_NS
 221         struct net              *net;
 222 #endif
 223         unsigned int            num_members;
 224         u16                     id;
 225         u8                      type;
 226         u8                      defrag;
 227         atomic_t                rr_cur;
 228         struct list_head        list;
 229         struct sock             *arr[PACKET_FANOUT_MAX];
 230         spinlock_t              lock;
 231         atomic_t                sk_ref;
 232         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
 233 };
 234
 235 struct packet_skb_cb {
 236         unsigned int origlen;
 237         union {
 238                 struct sockaddr_pkt pkt;
 239                 struct sockaddr_ll ll;
 240         } sa;
 241 };
 242
 243 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 244
 245 static inline struct packet_sock *pkt_sk(struct sock *sk)
 246 {
 247         return (struct packet_sock *)sk;
 248 }
 249
 250 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 251 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 252
 253 /* register_prot_hook must be invoked with the po->bind_lock held,
 254  * or from a context in which asynchronous accesses to the packet
 255  * socket is not possible (packet_create()).
 256  */
 257 static void register_prot_hook(struct sock *sk)
 258 {
 259         struct packet_sock *po = pkt_sk(sk);
 260         if (!po->running) {
 261                 if (po->fanout)
 262                         __fanout_link(sk, po);
 263                 else
 264                         dev_add_pack(&po->prot_hook);
 265                 sock_hold(sk);
 266                 po->running = 1;
 267         }
 268 }
 269
 270 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 271  * held.   If the sync parameter is true, we will temporarily drop
 272  * the po->bind_lock and do a synchronize_net to make sure no
 273  * asynchronous packet processing paths still refer to the elements
 274  * of po->prot_hook.  If the sync parameter is false, it is the
 275  * callers responsibility to take care of this.
 276  */
 277 static void __unregister_prot_hook(struct sock *sk, bool sync)
 278 {
 279         struct packet_sock *po = pkt_sk(sk);
 280
 281         po->running = 0;
 282         if (po->fanout)
 283                 __fanout_unlink(sk, po);
 284         else
 285                 __dev_remove_pack(&po->prot_hook);
 286         __sock_put(sk);
 287
 288         if (sync) {
 289                 spin_unlock(&po->bind_lock);
 290                 synchronize_net();
 291                 spin_lock(&po->bind_lock);
 292         }
 293 }
 294
 295 static void unregister_prot_hook(struct sock *sk, bool sync)
 296 {
 297         struct packet_sock *po = pkt_sk(sk);
 298
 299         if (po->running)
 300                 __unregister_prot_hook(sk, sync);
 301 }
 302
 303 static inline __pure struct page *pgv_to_page(void *addr)
 304 {
 305         if (is_vmalloc_addr(addr))
 306                 return vmalloc_to_page(addr);
 307         return virt_to_page(addr);
 308 }
 309
 310 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 311 {
 312         union {
 313                 struct tpacket_hdr *h1;
 314                 struct tpacket2_hdr *h2;
 315                 void *raw;
 316         } h;
 317
 318         h.raw = frame;
 319         switch (po->tp_version) {
 320         case TPACKET_V1:
 321                 h.h1->tp_status = status;
 322                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 323                 break;
 324         case TPACKET_V2:
 325                 h.h2->tp_status = status;
 326                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 327                 break;
 328         default:
 329                 pr_err("TPACKET version not supported\n");
 330                 BUG();
 331         }
 332
 333         smp_wmb();
 334 }
 335
 336 static int __packet_get_status(struct packet_sock *po, void *frame)
 337 {
 338         union {
 339                 struct tpacket_hdr *h1;
 340                 struct tpacket2_hdr *h2;
 341                 void *raw;
 342         } h;
 343
 344         smp_rmb();
 345
 346         h.raw = frame;
 347         switch (po->tp_version) {
 348         case TPACKET_V1:
 349                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 350                 return h.h1->tp_status;
 351         case TPACKET_V2:
 352                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 353                 return h.h2->tp_status;
 354         default:
 355                 pr_err("TPACKET version not supported\n");
 356                 BUG();
 357                 return 0;
 358         }
 359 }
 360
 361 static void *packet_lookup_frame(struct packet_sock *po,
 362                 struct packet_ring_buffer *rb,
 363                 unsigned int position,
 364                 int status)
 365 {
 366         unsigned int pg_vec_pos, frame_offset;
 367         union {
 368                 struct tpacket_hdr *h1;
 369                 struct tpacket2_hdr *h2;
 370                 void *raw;
 371         } h;
 372
 373         pg_vec_pos = position / rb->frames_per_block;
 374         frame_offset = position % rb->frames_per_block;
 375
 376         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 377                 (frame_offset * rb->frame_size);
 378
 379         if (status != __packet_get_status(po, h.raw))
 380                 return NULL;
 381
 382         return h.raw;
 383 }
 384
 385 static inline void *packet_current_frame(struct packet_sock *po,
 386                 struct packet_ring_buffer *rb,
 387                 int status)
 388 {
 389         return packet_lookup_frame(po, rb, rb->head, status);
 390 }
 391
 392 static inline void *packet_previous_frame(struct packet_sock *po,
 393                 struct packet_ring_buffer *rb,
 394                 int status)
 395 {
 396         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
 397         return packet_lookup_frame(po, rb, previous, status);
 398 }
 399
 400 static inline void packet_increment_head(struct packet_ring_buffer *buff)
 401 {
 402         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
 403 }
 404
 405 static void packet_sock_destruct(struct sock *sk)
 406 {
 407         skb_queue_purge(&sk->sk_error_queue);
 408
 409         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
 410         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
 411
 412         if (!sock_flag(sk, SOCK_DEAD)) {
 413                 pr_err("Attempt to release alive packet socket: %p\n", sk);
 414                 return;
 415         }
 416
 417         sk_refcnt_debug_dec(sk);
 418 }
 419
 420 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
 421 {
 422         int x = atomic_read(&f->rr_cur) + 1;
 423
 424         if (x >= num)
 425                 x = 0;
 426
 427         return x;
 428 }
 429
 430 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
 431 {
 432         u32 idx, hash = skb->rxhash;
 433
 434         idx = ((u64)hash * num) >> 32;
 435
 436         return f->arr[idx];
 437 }
 438
 439 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
 440 {
 441         int cur, old;
 442
 443         cur = atomic_read(&f->rr_cur);
 444         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
 445                                      fanout_rr_next(f, num))) != cur)
 446                 cur = old;
 447         return f->arr[cur];
 448 }
 449
 450 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
 451 {
 452         unsigned int cpu = smp_processor_id();
 453
 454         return f->arr[cpu % num];
 455 }
 456
 457 static struct sk_buff *fanout_check_defrag(struct sk_buff *skb)
 458 {
 459 #ifdef CONFIG_INET
 460         const struct iphdr *iph;
 461         u32 len;
 462
 463         if (skb->protocol != htons(ETH_P_IP))
 464                 return skb;
 465
 466         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
 467                 return skb;
 468
 469         iph = ip_hdr(skb);
 470         if (iph->ihl < 5 || iph->version != 4)
 471                 return skb;
 472         if (!pskb_may_pull(skb, iph->ihl*4))
 473                 return skb;
 474         iph = ip_hdr(skb);
 475         len = ntohs(iph->tot_len);
 476         if (skb->len < len || len < (iph->ihl * 4))
 477                 return skb;
 478
 479         if (ip_is_fragment(ip_hdr(skb))) {
 480                 skb = skb_share_check(skb, GFP_ATOMIC);
 481                 if (skb) {
 482                         if (pskb_trim_rcsum(skb, len))
 483                                 return skb;
 484                         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
 485                         if (ip_defrag(skb, IP_DEFRAG_AF_PACKET))
 486                                 return NULL;
 487                         skb->rxhash = 0;
 488                 }
 489         }
 490 #endif
 491         return skb;
 492 }
 493
 494 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
 495                              struct packet_type *pt, struct net_device *orig_dev)
 496 {
 497         struct packet_fanout *f = pt->af_packet_priv;
 498         unsigned int num = f->num_members;
 499         struct packet_sock *po;
 500         struct sock *sk;
 501
 502         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
 503             !num) {
 504                 kfree_skb(skb);
 505                 return 0;
 506         }
 507
 508         switch (f->type) {
 509         case PACKET_FANOUT_HASH:
 510         default:
 511                 if (f->defrag) {
 512                         skb = fanout_check_defrag(skb);
 513                         if (!skb)
 514                                 return 0;
 515                 }
 516                 skb_get_rxhash(skb);
 517                 sk = fanout_demux_hash(f, skb, num);
 518                 break;
 519         case PACKET_FANOUT_LB:
 520                 sk = fanout_demux_lb(f, skb, num);
 521                 break;
 522         case PACKET_FANOUT_CPU:
 523                 sk = fanout_demux_cpu(f, skb, num);
 524                 break;
 525         }
 526
 527         po = pkt_sk(sk);
 528
 529         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
 530 }
 531
 532 static DEFINE_MUTEX(fanout_mutex);
 533 static LIST_HEAD(fanout_list);
 534
 535 static void __fanout_link(struct sock *sk, struct packet_sock *po)
 536 {
 537         struct packet_fanout *f = po->fanout;
 538
 539         spin_lock(&f->lock);
 540         f->arr[f->num_members] = sk;
 541         smp_wmb();
 542         f->num_members++;
 543         spin_unlock(&f->lock);
 544 }
 545
 546 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
 547 {
 548         struct packet_fanout *f = po->fanout;
 549         int i;
 550
 551         spin_lock(&f->lock);
 552         for (i = 0; i < f->num_members; i++) {
 553                 if (f->arr[i] == sk)
 554                         break;
 555         }
 556         BUG_ON(i >= f->num_members);
 557         f->arr[i] = f->arr[f->num_members - 1];
 558         f->num_members--;
 559         spin_unlock(&f->lock);
 560 }
 561
 562 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
 563 {
 564         struct packet_sock *po = pkt_sk(sk);
 565         struct packet_fanout *f, *match;
 566         u8 type = type_flags & 0xff;
 567         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
 568         int err;
 569
 570         switch (type) {
 571         case PACKET_FANOUT_HASH:
 572         case PACKET_FANOUT_LB:
 573         case PACKET_FANOUT_CPU:
 574                 break;
 575         default:
 576                 return -EINVAL;
 577         }
 578
 579         if (!po->running)
 580                 return -EINVAL;
 581
 582         if (po->fanout)
 583                 return -EALREADY;
 584
 585         mutex_lock(&fanout_mutex);
 586         match = NULL;
 587         list_for_each_entry(f, &fanout_list, list) {
 588                 if (f->id == id &&
 589                     read_pnet(&f->net) == sock_net(sk)) {
 590                         match = f;
 591                         break;
 592                 }
 593         }
 594         err = -EINVAL;
 595         if (match && match->defrag != defrag)
 596                 goto out;
 597         if (!match) {
 598                 err = -ENOMEM;
 599                 match = kzalloc(sizeof(*match), GFP_KERNEL);
 600                 if (!match)
 601                         goto out;
 602                 write_pnet(&match->net, sock_net(sk));
 603                 match->id = id;
 604                 match->type = type;
 605                 match->defrag = defrag;
 606                 atomic_set(&match->rr_cur, 0);
 607                 INIT_LIST_HEAD(&match->list);
 608                 spin_lock_init(&match->lock);
 609                 atomic_set(&match->sk_ref, 0);
 610                 match->prot_hook.type = po->prot_hook.type;
 611                 match->prot_hook.dev = po->prot_hook.dev;
 612                 match->prot_hook.func = packet_rcv_fanout;
 613                 match->prot_hook.af_packet_priv = match;
 614                 dev_add_pack(&match->prot_hook);
 615                 list_add(&match->list, &fanout_list);
 616         }
 617         err = -EINVAL;
 618         if (match->type == type &&
 619             match->prot_hook.type == po->prot_hook.type &&
 620             match->prot_hook.dev == po->prot_hook.dev) {
 621                 err = -ENOSPC;
 622                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
 623                         __dev_remove_pack(&po->prot_hook);
 624                         po->fanout = match;
 625                         atomic_inc(&match->sk_ref);
 626                         __fanout_link(sk, po);
 627                         err = 0;
 628                 }
 629         }
 630 out:
 631         mutex_unlock(&fanout_mutex);
 632         return err;
 633 }
 634
 635 static void fanout_release(struct sock *sk)
 636 {
 637         struct packet_sock *po = pkt_sk(sk);
 638         struct packet_fanout *f;
 639
 640         f = po->fanout;
 641         if (!f)
 642                 return;
 643
 644         po->fanout = NULL;
 645
 646         mutex_lock(&fanout_mutex);
 647         if (atomic_dec_and_test(&f->sk_ref)) {
 648                 list_del(&f->list);
 649                 dev_remove_pack(&f->prot_hook);
 650                 kfree(f);
 651         }
 652         mutex_unlock(&fanout_mutex);
 653 }
 654
 655 static const struct proto_ops packet_ops;
 656
 657 static const struct proto_ops packet_ops_spkt;
 658
 659 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
 660                            struct packet_type *pt, struct net_device *orig_dev)
 661 {
 662         struct sock *sk;
 663         struct sockaddr_pkt *spkt;
 664
 665         /*
 666          *      When we registered the protocol we saved the socket in the data
 667          *      field for just this event.
 668          */
 669
 670         sk = pt->af_packet_priv;
 671
 672         /*
 673          *      Yank back the headers [hope the device set this
 674          *      right or kerboom...]
 675          *
 676          *      Incoming packets have ll header pulled,
 677          *      push it back.
 678          *
 679          *      For outgoing ones skb->data == skb_mac_header(skb)
 680          *      so that this procedure is noop.
 681          */
 682
 683         if (skb->pkt_type == PACKET_LOOPBACK)
 684                 goto out;
 685
 686         if (!net_eq(dev_net(dev), sock_net(sk)))
 687                 goto out;
 688
 689         skb = skb_share_check(skb, GFP_ATOMIC);
 690         if (skb == NULL)
 691                 goto oom;
 692
 693         /* drop any routing info */
 694         skb_dst_drop(skb);
 695
 696         /* drop conntrack reference */
 697         nf_reset(skb);
 698
 699         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
 700
 701         skb_push(skb, skb->data - skb_mac_header(skb));
 702
 703         /*
 704          *      The SOCK_PACKET socket receives _all_ frames.
 705          */
 706
 707         spkt->spkt_family = dev->type;
 708         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
 709         spkt->spkt_protocol = skb->protocol;
 710
 711         /*
 712          *      Charge the memory to the socket. This is done specifically
 713          *      to prevent sockets using all the memory up.
 714          */
 715
 716         if (sock_queue_rcv_skb(sk, skb) == 0)
 717                 return 0;
 718
 719 out:
 720         kfree_skb(skb);
 721 oom:
 722         return 0;
 723 }
 724
 725
 726 /*
 727  *      Output a raw packet to a device layer. This bypasses all the other
 728  *      protocol layers and you must therefore supply it with a complete frame
 729  */
 730
 731 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
 732                                struct msghdr *msg, size_t len)
 733 {
 734         struct sock *sk = sock->sk;
 735         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
 736         struct sk_buff *skb = NULL;
 737         struct net_device *dev;
 738         __be16 proto = 0;
 739         int err;
 740
 741         /*
 742          *      Get and verify the address.
 743          */
 744
 745         if (saddr) {
 746                 if (msg->msg_namelen < sizeof(struct sockaddr))
 747                         return -EINVAL;
 748                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
 749                         proto = saddr->spkt_protocol;
 750         } else
 751                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
 752
 753         /*
 754          *      Find the device first to size check it
 755          */
 756
 757         saddr->spkt_device[13] = 0;
 758 retry:
 759         rcu_read_lock();
 760         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
 761         err = -ENODEV;
 762         if (dev == NULL)
 763                 goto out_unlock;
 764
 765         err = -ENETDOWN;
 766         if (!(dev->flags & IFF_UP))
 767                 goto out_unlock;
 768
 769         /*
 770          * You may not queue a frame bigger than the mtu. This is the lowest level
 771          * raw protocol and you must do your own fragmentation at this level.
 772          */
 773
 774         err = -EMSGSIZE;
 775         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
 776                 goto out_unlock;
 777
 778         if (!skb) {
 779                 size_t reserved = LL_RESERVED_SPACE(dev);
 780                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
 781
 782                 rcu_read_unlock();
 783                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
 784                 if (skb == NULL)
 785                         return -ENOBUFS;
 786                 /* FIXME: Save some space for broken drivers that write a hard
 787                  * header at transmission time by themselves. PPP is the notable
 788                  * one here. This should really be fixed at the driver level.
 789                  */
 790                 skb_reserve(skb, reserved);
 791                 skb_reset_network_header(skb);
 792
 793                 /* Try to align data part correctly */
 794                 if (hhlen) {
 795                         skb->data -= hhlen;
 796                         skb->tail -= hhlen;
 797                         if (len < hhlen)
 798                                 skb_reset_network_header(skb);
 799                 }
 800                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
 801                 if (err)
 802                         goto out_free;
 803                 goto retry;
 804         }
 805
 806         if (len > (dev->mtu + dev->hard_header_len)) {
 807                 /* Earlier code assumed this would be a VLAN pkt,
 808                  * double-check this now that we have the actual
 809                  * packet in hand.
 810                  */
 811                 struct ethhdr *ehdr;
 812                 skb_reset_mac_header(skb);
 813                 ehdr = eth_hdr(skb);
 814                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
 815                         err = -EMSGSIZE;
 816                         goto out_unlock;
 817                 }
 818         }
 819
 820         skb->protocol = proto;
 821         skb->dev = dev;
 822         skb->priority = sk->sk_priority;
 823         skb->mark = sk->sk_mark;
 824         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
 825         if (err < 0)
 826                 goto out_unlock;
 827
 828         dev_queue_xmit(skb);
 829         rcu_read_unlock();
 830         return len;
 831
 832 out_unlock:
 833         rcu_read_unlock();
 834 out_free:
 835         kfree_skb(skb);
 836         return err;
 837 }
 838
 839 static inline unsigned int run_filter(const struct sk_buff *skb,
 840                                       const struct sock *sk,
 841                                       unsigned int res)
 842 {
 843         struct sk_filter *filter;
 844
 845         rcu_read_lock();
 846         filter = rcu_dereference(sk->sk_filter);
 847         if (filter != NULL)
 848                 res = SK_RUN_FILTER(filter, skb);
 849         rcu_read_unlock();
 850
 851         return res;
 852 }
 853
 854 /*
 855  * This function makes lazy skb cloning in hope that most of packets
 856  * are discarded by BPF.
 857  *
 858  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
 859  * and skb->cb are mangled. It works because (and until) packets
 860  * falling here are owned by current CPU. Output packets are cloned
 861  * by dev_queue_xmit_nit(), input packets are processed by net_bh
 862  * sequencially, so that if we return skb to original state on exit,
 863  * we will not harm anyone.
 864  */
 865
 866 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
 867                       struct packet_type *pt, struct net_device *orig_dev)
 868 {
 869         struct sock *sk;
 870         struct sockaddr_ll *sll;
 871         struct packet_sock *po;
 872         u8 *skb_head = skb->data;
 873         int skb_len = skb->len;
 874         unsigned int snaplen, res;
 875
 876         if (skb->pkt_type == PACKET_LOOPBACK)
 877                 goto drop;
 878
 879         sk = pt->af_packet_priv;
 880         po = pkt_sk(sk);
 881
 882         if (!net_eq(dev_net(dev), sock_net(sk)))
 883                 goto drop;
 884
 885         skb->dev = dev;
 886
 887         if (dev->header_ops) {
 888                 /* The device has an explicit notion of ll header,
 889                  * exported to higher levels.
 890                  *
 891                  * Otherwise, the device hides details of its frame
 892                  * structure, so that corresponding packet head is
 893                  * never delivered to user.
 894                  */
 895                 if (sk->sk_type != SOCK_DGRAM)
 896                         skb_push(skb, skb->data - skb_mac_header(skb));
 897                 else if (skb->pkt_type == PACKET_OUTGOING) {
 898                         /* Special case: outgoing packets have ll header at head */
 899                         skb_pull(skb, skb_network_offset(skb));
 900                 }
 901         }
 902
 903         snaplen = skb->len;
 904
 905         res = run_filter(skb, sk, snaplen);
 906         if (!res)
 907                 goto drop_n_restore;
 908         if (snaplen > res)
 909                 snaplen = res;
 910
 911         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
 912             (unsigned)sk->sk_rcvbuf)
 913                 goto drop_n_acct;
 914
 915         if (skb_shared(skb)) {
 916                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
 917                 if (nskb == NULL)
 918                         goto drop_n_acct;
 919
 920                 if (skb_head != skb->data) {
 921                         skb->data = skb_head;
 922                         skb->len = skb_len;
 923                 }
 924                 kfree_skb(skb);
 925                 skb = nskb;
 926         }
 927
 928         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
 929                      sizeof(skb->cb));
 930
 931         sll = &PACKET_SKB_CB(skb)->sa.ll;
 932         sll->sll_family = AF_PACKET;
 933         sll->sll_hatype = dev->type;
 934         sll->sll_protocol = skb->protocol;
 935         sll->sll_pkttype = skb->pkt_type;
 936         if (unlikely(po->origdev))
 937                 sll->sll_ifindex = orig_dev->ifindex;
 938         else
 939                 sll->sll_ifindex = dev->ifindex;
 940
 941         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
 942
 943         PACKET_SKB_CB(skb)->origlen = skb->len;
 944
 945         if (pskb_trim(skb, snaplen))
 946                 goto drop_n_acct;
 947
 948         skb_set_owner_r(skb, sk);
 949         skb->dev = NULL;
 950         skb_dst_drop(skb);
 951
 952         /* drop conntrack reference */
 953         nf_reset(skb);
 954
 955         spin_lock(&sk->sk_receive_queue.lock);
 956         po->stats.tp_packets++;
 957         skb->dropcount = atomic_read(&sk->sk_drops);
 958         __skb_queue_tail(&sk->sk_receive_queue, skb);
 959         spin_unlock(&sk->sk_receive_queue.lock);
 960         sk->sk_data_ready(sk, skb->len);
 961         return 0;
 962
 963 drop_n_acct:
 964         po->stats.tp_drops = atomic_inc_return(&sk->sk_drops);
 965
 966 drop_n_restore:
 967         if (skb_head != skb->data && skb_shared(skb)) {
 968                 skb->data = skb_head;
 969                 skb->len = skb_len;
 970         }
 971 drop:
 972         consume_skb(skb);
 973         return 0;
 974 }
 975
 976 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 977                        struct packet_type *pt, struct net_device *orig_dev)
 978 {
 979         struct sock *sk;
 980         struct packet_sock *po;
 981         struct sockaddr_ll *sll;
 982         union {
 983                 struct tpacket_hdr *h1;
 984                 struct tpacket2_hdr *h2;
 985                 void *raw;
 986         } h;
 987         u8 *skb_head = skb->data;
 988         int skb_len = skb->len;
 989         unsigned int snaplen, res;
 990         unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
 991         unsigned short macoff, netoff, hdrlen;
 992         struct sk_buff *copy_skb = NULL;
 993         struct timeval tv;
 994         struct timespec ts;
 995         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 996
 997         if (skb->pkt_type == PACKET_LOOPBACK)
 998                 goto drop;
 999
1000         sk = pt->af_packet_priv;
1001         po = pkt_sk(sk);
1002
1003         if (!net_eq(dev_net(dev), sock_net(sk)))
1004                 goto drop;
1005
1006         if (dev->header_ops) {
1007                 if (sk->sk_type != SOCK_DGRAM)
1008                         skb_push(skb, skb->data - skb_mac_header(skb));
1009                 else if (skb->pkt_type == PACKET_OUTGOING) {
1010                         /* Special case: outgoing packets have ll header at head */
1011                         skb_pull(skb, skb_network_offset(skb));
1012                 }
1013         }
1014
1015         if (skb->ip_summed == CHECKSUM_PARTIAL)
1016                 status |= TP_STATUS_CSUMNOTREADY;
1017
1018         snaplen = skb->len;
1019
1020         res = run_filter(skb, sk, snaplen);
1021         if (!res)
1022                 goto drop_n_restore;
1023         if (snaplen > res)
1024                 snaplen = res;
1025
1026         if (sk->sk_type == SOCK_DGRAM) {
1027                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1028                                   po->tp_reserve;
1029         } else {
1030                 unsigned maclen = skb_network_offset(skb);
1031                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1032                                        (maclen < 16 ? 16 : maclen)) +
1033                         po->tp_reserve;
1034                 macoff = netoff - maclen;
1035         }
1036
1037         if (macoff + snaplen > po->rx_ring.frame_size) {
1038                 if (po->copy_thresh &&
1039                     atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
1040                     (unsigned)sk->sk_rcvbuf) {
1041                         if (skb_shared(skb)) {
1042                                 copy_skb = skb_clone(skb, GFP_ATOMIC);
1043                         } else {
1044                                 copy_skb = skb_get(skb);
1045                                 skb_head = skb->data;
1046                         }
1047                         if (copy_skb)
1048                                 skb_set_owner_r(copy_skb, sk);
1049                 }
1050                 snaplen = po->rx_ring.frame_size - macoff;
1051                 if ((int)snaplen < 0)
1052                         snaplen = 0;
1053         }
1054
1055         spin_lock(&sk->sk_receive_queue.lock);
1056         h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
1057         if (!h.raw)
1058                 goto ring_is_full;
1059         packet_increment_head(&po->rx_ring);
1060         po->stats.tp_packets++;
1061         if (copy_skb) {
1062                 status |= TP_STATUS_COPY;
1063                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1064         }
1065         if (!po->stats.tp_drops)
1066                 status &= ~TP_STATUS_LOSING;
1067         spin_unlock(&sk->sk_receive_queue.lock);
1068
1069         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1070
1071         switch (po->tp_version) {
1072         case TPACKET_V1:
1073                 h.h1->tp_len = skb->len;
1074                 h.h1->tp_snaplen = snaplen;
1075                 h.h1->tp_mac = macoff;
1076                 h.h1->tp_net = netoff;
1077                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1078                                 && shhwtstamps->syststamp.tv64)
1079                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1080                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1081                                 && shhwtstamps->hwtstamp.tv64)
1082                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1083                 else if (skb->tstamp.tv64)
1084                         tv = ktime_to_timeval(skb->tstamp);
1085                 else
1086                         do_gettimeofday(&tv);
1087                 h.h1->tp_sec = tv.tv_sec;
1088                 h.h1->tp_usec = tv.tv_usec;
1089                 hdrlen = sizeof(*h.h1);
1090                 break;
1091         case TPACKET_V2:
1092                 h.h2->tp_len = skb->len;
1093                 h.h2->tp_snaplen = snaplen;
1094                 h.h2->tp_mac = macoff;
1095                 h.h2->tp_net = netoff;
1096                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1097                                 && shhwtstamps->syststamp.tv64)
1098                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1099                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1100                                 && shhwtstamps->hwtstamp.tv64)
1101                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1102                 else if (skb->tstamp.tv64)
1103                         ts = ktime_to_timespec(skb->tstamp);
1104                 else
1105                         getnstimeofday(&ts);
1106                 h.h2->tp_sec = ts.tv_sec;
1107                 h.h2->tp_nsec = ts.tv_nsec;
1108                 if (vlan_tx_tag_present(skb)) {
1109                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1110                         status |= TP_STATUS_VLAN_VALID;
1111                 } else {
1112                         h.h2->tp_vlan_tci = 0;
1113                 }
1114                 h.h2->tp_padding = 0;
1115                 hdrlen = sizeof(*h.h2);
1116                 break;
1117         default:
1118                 BUG();
1119         }
1120
1121         sll = h.raw + TPACKET_ALIGN(hdrlen);
1122         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1123         sll->sll_family = AF_PACKET;
1124         sll->sll_hatype = dev->type;
1125         sll->sll_protocol = skb->protocol;
1126         sll->sll_pkttype = skb->pkt_type;
1127         if (unlikely(po->origdev))
1128                 sll->sll_ifindex = orig_dev->ifindex;
1129         else
1130                 sll->sll_ifindex = dev->ifindex;
1131
1132         smp_mb();
1133 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1134         {
1135                 u8 *start, *end;
1136
1137                 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw + macoff + snaplen);
1138                 for (start = h.raw; start < end; start += PAGE_SIZE)
1139                         flush_dcache_page(pgv_to_page(start));
1140                 smp_wmb();
1141         }
1142 #endif
1143         __packet_set_status(po, h.raw, status);
1144
1145         sk->sk_data_ready(sk, 0);
1146
1147 drop_n_restore:
1148         if (skb_head != skb->data && skb_shared(skb)) {
1149                 skb->data = skb_head;
1150                 skb->len = skb_len;
1151         }
1152 drop:
1153         kfree_skb(skb);
1154         return 0;
1155
1156 ring_is_full:
1157         po->stats.tp_drops++;
1158         spin_unlock(&sk->sk_receive_queue.lock);
1159
1160         sk->sk_data_ready(sk, 0);
1161         kfree_skb(copy_skb);
1162         goto drop_n_restore;
1163 }
1164
1165 static void tpacket_destruct_skb(struct sk_buff *skb)
1166 {
1167         struct packet_sock *po = pkt_sk(skb->sk);
1168         void *ph;
1169
1170         BUG_ON(skb == NULL);
1171
1172         if (likely(po->tx_ring.pg_vec)) {
1173                 ph = skb_shinfo(skb)->destructor_arg;
1174                 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
1175                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1176                 atomic_dec(&po->tx_ring.pending);
1177                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1178         }
1179
1180         sock_wfree(skb);
1181 }
1182
1183 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1184                 void *frame, struct net_device *dev, int size_max,
1185                 __be16 proto, unsigned char *addr)
1186 {
1187         union {
1188                 struct tpacket_hdr *h1;
1189                 struct tpacket2_hdr *h2;
1190                 void *raw;
1191         } ph;
1192         int to_write, offset, len, tp_len, nr_frags, len_max;
1193         struct socket *sock = po->sk.sk_socket;
1194         struct page *page;
1195         void *data;
1196         int err;
1197
1198         ph.raw = frame;
1199
1200         skb->protocol = proto;
1201         skb->dev = dev;
1202         skb->priority = po->sk.sk_priority;
1203         skb->mark = po->sk.sk_mark;
1204         skb_shinfo(skb)->destructor_arg = ph.raw;
1205
1206         switch (po->tp_version) {
1207         case TPACKET_V2:
1208                 tp_len = ph.h2->tp_len;
1209                 break;
1210         default:
1211                 tp_len = ph.h1->tp_len;
1212                 break;
1213         }
1214         if (unlikely(tp_len > size_max)) {
1215                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1216                 return -EMSGSIZE;
1217         }
1218
1219         skb_reserve(skb, LL_RESERVED_SPACE(dev));
1220         skb_reset_network_header(skb);
1221
1222         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1223         to_write = tp_len;
1224
1225         if (sock->type == SOCK_DGRAM) {
1226                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1227                                 NULL, tp_len);
1228                 if (unlikely(err < 0))
1229                         return -EINVAL;
1230         } else if (dev->hard_header_len) {
1231                 /* net device doesn't like empty head */
1232                 if (unlikely(tp_len <= dev->hard_header_len)) {
1233                         pr_err("packet size is too short (%d < %d)\n",
1234                                tp_len, dev->hard_header_len);
1235                         return -EINVAL;
1236                 }
1237
1238                 skb_push(skb, dev->hard_header_len);
1239                 err = skb_store_bits(skb, 0, data,
1240                                 dev->hard_header_len);
1241                 if (unlikely(err))
1242                         return err;
1243
1244                 data += dev->hard_header_len;
1245                 to_write -= dev->hard_header_len;
1246         }
1247
1248         err = -EFAULT;
1249         offset = offset_in_page(data);
1250         len_max = PAGE_SIZE - offset;
1251         len = ((to_write > len_max) ? len_max : to_write);
1252
1253         skb->data_len = to_write;
1254         skb->len += to_write;
1255         skb->truesize += to_write;
1256         atomic_add(to_write, &po->sk.sk_wmem_alloc);
1257
1258         while (likely(to_write)) {
1259                 nr_frags = skb_shinfo(skb)->nr_frags;
1260
1261                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
1262                         pr_err("Packet exceed the number of skb frags(%lu)\n",
1263                                MAX_SKB_FRAGS);
1264                         return -EFAULT;
1265                 }
1266
1267                 page = pgv_to_page(data);
1268                 data += len;
1269                 flush_dcache_page(page);
1270                 get_page(page);
1271                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
1272                 to_write -= len;
1273                 offset = 0;
1274                 len_max = PAGE_SIZE;
1275                 len = ((to_write > len_max) ? len_max : to_write);
1276         }
1277
1278         return tp_len;
1279 }
1280
1281 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
1282 {
1283         struct sk_buff *skb;
1284         struct net_device *dev;
1285         __be16 proto;
1286         bool need_rls_dev = false;
1287         int err, reserve = 0;
1288         void *ph;
1289         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1290         int tp_len, size_max;
1291         unsigned char *addr;
1292         int len_sum = 0;
1293         int status = 0;
1294
1295         mutex_lock(&po->pg_vec_lock);
1296
1297         err = -EBUSY;
1298         if (saddr == NULL) {
1299                 dev = po->prot_hook.dev;
1300                 proto   = po->num;
1301                 addr    = NULL;
1302         } else {
1303                 err = -EINVAL;
1304                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1305                         goto out;
1306                 if (msg->msg_namelen < (saddr->sll_halen
1307                                         + offsetof(struct sockaddr_ll,
1308                                                 sll_addr)))
1309                         goto out;
1310                 proto   = saddr->sll_protocol;
1311                 addr    = saddr->sll_addr;
1312                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
1313                 need_rls_dev = true;
1314         }
1315
1316         err = -ENXIO;
1317         if (unlikely(dev == NULL))
1318                 goto out;
1319
1320         reserve = dev->hard_header_len;
1321
1322         err = -ENETDOWN;
1323         if (unlikely(!(dev->flags & IFF_UP)))
1324                 goto out_put;
1325
1326         size_max = po->tx_ring.frame_size
1327                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
1328
1329         if (size_max > dev->mtu + reserve)
1330                 size_max = dev->mtu + reserve;
1331
1332         do {
1333                 ph = packet_current_frame(po, &po->tx_ring,
1334                                 TP_STATUS_SEND_REQUEST);
1335
1336                 if (unlikely(ph == NULL)) {
1337                         schedule();
1338                         continue;
1339                 }
1340
1341                 status = TP_STATUS_SEND_REQUEST;
1342                 skb = sock_alloc_send_skb(&po->sk,
1343                                 LL_ALLOCATED_SPACE(dev)
1344                                 + sizeof(struct sockaddr_ll),
1345                                 0, &err);
1346
1347                 if (unlikely(skb == NULL))
1348                         goto out_status;
1349
1350                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
1351                                 addr);
1352
1353                 if (unlikely(tp_len < 0)) {
1354                         if (po->tp_loss) {
1355                                 __packet_set_status(po, ph,
1356                                                 TP_STATUS_AVAILABLE);
1357                                 packet_increment_head(&po->tx_ring);
1358                                 kfree_skb(skb);
1359                                 continue;
1360                         } else {
1361                                 status = TP_STATUS_WRONG_FORMAT;
1362                                 err = tp_len;
1363                                 goto out_status;
1364                         }
1365                 }
1366
1367                 skb->destructor = tpacket_destruct_skb;
1368                 __packet_set_status(po, ph, TP_STATUS_SENDING);
1369                 atomic_inc(&po->tx_ring.pending);
1370
1371                 status = TP_STATUS_SEND_REQUEST;
1372                 err = dev_queue_xmit(skb);
1373                 if (unlikely(err > 0)) {
1374                         err = net_xmit_errno(err);
1375                         if (err && __packet_get_status(po, ph) ==
1376                                    TP_STATUS_AVAILABLE) {
1377                                 /* skb was destructed already */
1378                                 skb = NULL;
1379                                 goto out_status;
1380                         }
1381                         /*
1382                          * skb was dropped but not destructed yet;
1383                          * let's treat it like congestion or err < 0
1384                          */
1385                         err = 0;
1386                 }
1387                 packet_increment_head(&po->tx_ring);
1388                 len_sum += tp_len;
1389         } while (likely((ph != NULL) ||
1390                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
1391                          (atomic_read(&po->tx_ring.pending))))
1392                 );
1393
1394         err = len_sum;
1395         goto out_put;
1396
1397 out_status:
1398         __packet_set_status(po, ph, status);
1399         kfree_skb(skb);
1400 out_put:
1401         if (need_rls_dev)
1402                 dev_put(dev);
1403 out:
1404         mutex_unlock(&po->pg_vec_lock);
1405         return err;
1406 }
1407
1408 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
1409                                                size_t reserve, size_t len,
1410                                                size_t linear, int noblock,
1411                                                int *err)
1412 {
1413         struct sk_buff *skb;
1414
1415         /* Under a page?  Don't bother with paged skb. */
1416         if (prepad + len < PAGE_SIZE || !linear)
1417                 linear = len;
1418
1419         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1420                                    err);
1421         if (!skb)
1422                 return NULL;
1423
1424         skb_reserve(skb, reserve);
1425         skb_put(skb, linear);
1426         skb->data_len = len - linear;
1427         skb->len += len - linear;
1428
1429         return skb;
1430 }
1431
1432 static int packet_snd(struct socket *sock,
1433                           struct msghdr *msg, size_t len)
1434 {
1435         struct sock *sk = sock->sk;
1436         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1437         struct sk_buff *skb;
1438         struct net_device *dev;
1439         __be16 proto;
1440         bool need_rls_dev = false;
1441         unsigned char *addr;
1442         int err, reserve = 0;
1443         struct virtio_net_hdr vnet_hdr = { 0 };
1444         int offset = 0;
1445         int vnet_hdr_len;
1446         struct packet_sock *po = pkt_sk(sk);
1447         unsigned short gso_type = 0;
1448
1449         /*
1450          *      Get and verify the address.
1451          */
1452
1453         if (saddr == NULL) {
1454                 dev = po->prot_hook.dev;
1455                 proto   = po->num;
1456                 addr    = NULL;
1457         } else {
1458                 err = -EINVAL;
1459                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1460                         goto out;
1461                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1462                         goto out;
1463                 proto   = saddr->sll_protocol;
1464                 addr    = saddr->sll_addr;
1465                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
1466                 need_rls_dev = true;
1467         }
1468
1469         err = -ENXIO;
1470         if (dev == NULL)
1471                 goto out_unlock;
1472         if (sock->type == SOCK_RAW)
1473                 reserve = dev->hard_header_len;
1474
1475         err = -ENETDOWN;
1476         if (!(dev->flags & IFF_UP))
1477                 goto out_unlock;
1478
1479         if (po->has_vnet_hdr) {
1480                 vnet_hdr_len = sizeof(vnet_hdr);
1481
1482                 err = -EINVAL;
1483                 if (len < vnet_hdr_len)
1484                         goto out_unlock;
1485
1486                 len -= vnet_hdr_len;
1487
1488                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
1489                                        vnet_hdr_len);
1490                 if (err < 0)
1491                         goto out_unlock;
1492
1493                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1494                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
1495                       vnet_hdr.hdr_len))
1496                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
1497                                                  vnet_hdr.csum_offset + 2;
1498
1499                 err = -EINVAL;
1500                 if (vnet_hdr.hdr_len > len)
1501                         goto out_unlock;
1502
1503                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1504                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1505                         case VIRTIO_NET_HDR_GSO_TCPV4:
1506                                 gso_type = SKB_GSO_TCPV4;
1507                                 break;
1508                         case VIRTIO_NET_HDR_GSO_TCPV6:
1509                                 gso_type = SKB_GSO_TCPV6;
1510                                 break;
1511                         case VIRTIO_NET_HDR_GSO_UDP:
1512                                 gso_type = SKB_GSO_UDP;
1513                                 break;
1514                         default:
1515                                 goto out_unlock;
1516                         }
1517
1518                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1519                                 gso_type |= SKB_GSO_TCP_ECN;
1520
1521                         if (vnet_hdr.gso_size == 0)
1522                                 goto out_unlock;
1523
1524                 }
1525         }
1526
1527         err = -EMSGSIZE;
1528         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
1529                 goto out_unlock;
1530
1531         err = -ENOBUFS;
1532         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
1533                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
1534                                msg->msg_flags & MSG_DONTWAIT, &err);
1535         if (skb == NULL)
1536                 goto out_unlock;
1537
1538         skb_set_network_header(skb, reserve);
1539
1540         err = -EINVAL;
1541         if (sock->type == SOCK_DGRAM &&
1542             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
1543                 goto out_free;
1544
1545         /* Returns -EFAULT on error */
1546         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1547         if (err)
1548                 goto out_free;
1549         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1550         if (err < 0)
1551                 goto out_free;
1552
1553         if (!gso_type && (len > dev->mtu + reserve)) {
1554                 /* Earlier code assumed this would be a VLAN pkt,
1555                  * double-check this now that we have the actual
1556                  * packet in hand.
1557                  */
1558                 struct ethhdr *ehdr;
1559                 skb_reset_mac_header(skb);
1560                 ehdr = eth_hdr(skb);
1561                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1562                         err = -EMSGSIZE;
1563                         goto out_free;
1564                 }
1565         }
1566
1567         skb->protocol = proto;
1568         skb->dev = dev;
1569         skb->priority = sk->sk_priority;
1570         skb->mark = sk->sk_mark;
1571
1572         if (po->has_vnet_hdr) {
1573                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1574                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
1575                                                   vnet_hdr.csum_offset)) {
1576                                 err = -EINVAL;
1577                                 goto out_free;
1578                         }
1579                 }
1580
1581                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
1582                 skb_shinfo(skb)->gso_type = gso_type;
1583
1584                 /* Header must be checked, and gso_segs computed. */
1585                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1586                 skb_shinfo(skb)->gso_segs = 0;
1587
1588                 len += vnet_hdr_len;
1589         }
1590
1591         /*
1592          *      Now send it
1593          */
1594
1595         err = dev_queue_xmit(skb);
1596         if (err > 0 && (err = net_xmit_errno(err)) != 0)
1597                 goto out_unlock;
1598
1599         if (need_rls_dev)
1600                 dev_put(dev);
1601
1602         return len;
1603
1604 out_free:
1605         kfree_skb(skb);
1606 out_unlock:
1607         if (dev && need_rls_dev)
1608                 dev_put(dev);
1609 out:
1610         return err;
1611 }
1612
1613 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1614                 struct msghdr *msg, size_t len)
1615 {
1616         struct sock *sk = sock->sk;
1617         struct packet_sock *po = pkt_sk(sk);
1618         if (po->tx_ring.pg_vec)
1619                 return tpacket_snd(po, msg);
1620         else
1621                 return packet_snd(sock, msg, len);
1622 }
1623
1624 /*
1625  *      Close a PACKET socket. This is fairly simple. We immediately go
1626  *      to 'closed' state and remove our protocol entry in the device list.
1627  */
1628
1629 static int packet_release(struct socket *sock)
1630 {
1631         struct sock *sk = sock->sk;
1632         struct packet_sock *po;
1633         struct net *net;
1634         struct tpacket_req req;
1635
1636         if (!sk)
1637                 return 0;
1638
1639         net = sock_net(sk);
1640         po = pkt_sk(sk);
1641
1642         spin_lock_bh(&net->packet.sklist_lock);
1643         sk_del_node_init_rcu(sk);
1644         sock_prot_inuse_add(net, sk->sk_prot, -1);
1645         spin_unlock_bh(&net->packet.sklist_lock);
1646
1647         spin_lock(&po->bind_lock);
1648         unregister_prot_hook(sk, false);
1649         if (po->prot_hook.dev) {
1650                 dev_put(po->prot_hook.dev);
1651                 po->prot_hook.dev = NULL;
1652         }
1653         spin_unlock(&po->bind_lock);
1654
1655         packet_flush_mclist(sk);
1656
1657         memset(&req, 0, sizeof(req));
1658
1659         if (po->rx_ring.pg_vec)
1660                 packet_set_ring(sk, &req, 1, 0);
1661
1662         if (po->tx_ring.pg_vec)
1663                 packet_set_ring(sk, &req, 1, 1);
1664
1665         fanout_release(sk);
1666
1667         synchronize_net();
1668         /*
1669          *      Now the socket is dead. No more input will appear.
1670          */
1671         sock_orphan(sk);
1672         sock->sk = NULL;
1673
1674         /* Purge queues */
1675
1676         skb_queue_purge(&sk->sk_receive_queue);
1677         sk_refcnt_debug_release(sk);
1678
1679         sock_put(sk);
1680         return 0;
1681 }
1682
1683 /*
1684  *      Attach a packet hook.
1685  */
1686
1687 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1688 {
1689         struct packet_sock *po = pkt_sk(sk);
1690
1691         if (po->fanout)
1692                 return -EINVAL;
1693
1694         lock_sock(sk);
1695
1696         spin_lock(&po->bind_lock);
1697         unregister_prot_hook(sk, true);
1698         po->num = protocol;
1699         po->prot_hook.type = protocol;
1700         if (po->prot_hook.dev)
1701                 dev_put(po->prot_hook.dev);
1702         po->prot_hook.dev = dev;
1703
1704         po->ifindex = dev ? dev->ifindex : 0;
1705
1706         if (protocol == 0)
1707                 goto out_unlock;
1708
1709         if (!dev || (dev->flags & IFF_UP)) {
1710                 register_prot_hook(sk);
1711         } else {
1712                 sk->sk_err = ENETDOWN;
1713                 if (!sock_flag(sk, SOCK_DEAD))
1714                         sk->sk_error_report(sk);
1715         }
1716
1717 out_unlock:
1718         spin_unlock(&po->bind_lock);
1719         release_sock(sk);
1720         return 0;
1721 }
1722
1723 /*
1724  *      Bind a packet socket to a device
1725  */
1726
1727 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1728                             int addr_len)
1729 {
1730         struct sock *sk = sock->sk;
1731         char name[15];
1732         struct net_device *dev;
1733         int err = -ENODEV;
1734
1735         /*
1736          *      Check legality
1737          */
1738
1739         if (addr_len != sizeof(struct sockaddr))
1740                 return -EINVAL;
1741         strlcpy(name, uaddr->sa_data, sizeof(name));
1742
1743         dev = dev_get_by_name(sock_net(sk), name);
1744         if (dev)
1745                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1746         return err;
1747 }
1748
1749 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1750 {
1751         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1752         struct sock *sk = sock->sk;
1753         struct net_device *dev = NULL;
1754         int err;
1755
1756
1757         /*
1758          *      Check legality
1759          */
1760
1761         if (addr_len < sizeof(struct sockaddr_ll))
1762                 return -EINVAL;
1763         if (sll->sll_family != AF_PACKET)
1764                 return -EINVAL;
1765
1766         if (sll->sll_ifindex) {
1767                 err = -ENODEV;
1768                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1769                 if (dev == NULL)
1770                         goto out;
1771         }
1772         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1773
1774 out:
1775         return err;
1776 }
1777
1778 static struct proto packet_proto = {
1779         .name     = "PACKET",
1780         .owner    = THIS_MODULE,
1781         .obj_size = sizeof(struct packet_sock),
1782 };
1783
1784 /*
1785  *      Create a packet of type SOCK_PACKET.
1786  */
1787
1788 static int packet_create(struct net *net, struct socket *sock, int protocol,
1789                          int kern)
1790 {
1791         struct sock *sk;
1792         struct packet_sock *po;
1793         __be16 proto = (__force __be16)protocol; /* weird, but documented */
1794         int err;
1795
1796         if (!capable(CAP_NET_RAW))
1797                 return -EPERM;
1798         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1799             sock->type != SOCK_PACKET)
1800                 return -ESOCKTNOSUPPORT;
1801
1802         sock->state = SS_UNCONNECTED;
1803
1804         err = -ENOBUFS;
1805         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1806         if (sk == NULL)
1807                 goto out;
1808
1809         sock->ops = &packet_ops;
1810         if (sock->type == SOCK_PACKET)
1811                 sock->ops = &packet_ops_spkt;
1812
1813         sock_init_data(sock, sk);
1814
1815         po = pkt_sk(sk);
1816         sk->sk_family = PF_PACKET;
1817         po->num = proto;
1818
1819         sk->sk_destruct = packet_sock_destruct;
1820         sk_refcnt_debug_inc(sk);
1821
1822         /*
1823          *      Attach a protocol block
1824          */
1825
1826         spin_lock_init(&po->bind_lock);
1827         mutex_init(&po->pg_vec_lock);
1828         po->prot_hook.func = packet_rcv;
1829
1830         if (sock->type == SOCK_PACKET)
1831                 po->prot_hook.func = packet_rcv_spkt;
1832
1833         po->prot_hook.af_packet_priv = sk;
1834
1835         if (proto) {
1836                 po->prot_hook.type = proto;
1837                 register_prot_hook(sk);
1838         }
1839
1840         spin_lock_bh(&net->packet.sklist_lock);
1841         sk_add_node_rcu(sk, &net->packet.sklist);
1842         sock_prot_inuse_add(net, &packet_proto, 1);
1843         spin_unlock_bh(&net->packet.sklist_lock);
1844
1845         return 0;
1846 out:
1847         return err;
1848 }
1849
1850 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
1851 {
1852         struct sock_exterr_skb *serr;
1853         struct sk_buff *skb, *skb2;
1854         int copied, err;
1855
1856         err = -EAGAIN;
1857         skb = skb_dequeue(&sk->sk_error_queue);
1858         if (skb == NULL)
1859                 goto out;
1860
1861         copied = skb->len;
1862         if (copied > len) {
1863                 msg->msg_flags |= MSG_TRUNC;
1864                 copied = len;
1865         }
1866         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1867         if (err)
1868                 goto out_free_skb;
1869
1870         sock_recv_timestamp(msg, sk, skb);
1871
1872         serr = SKB_EXT_ERR(skb);
1873         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
1874                  sizeof(serr->ee), &serr->ee);
1875
1876         msg->msg_flags |= MSG_ERRQUEUE;
1877         err = copied;
1878
1879         /* Reset and regenerate socket error */
1880         spin_lock_bh(&sk->sk_error_queue.lock);
1881         sk->sk_err = 0;
1882         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
1883                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
1884                 spin_unlock_bh(&sk->sk_error_queue.lock);
1885                 sk->sk_error_report(sk);
1886         } else
1887                 spin_unlock_bh(&sk->sk_error_queue.lock);
1888
1889 out_free_skb:
1890         kfree_skb(skb);
1891 out:
1892         return err;
1893 }
1894
1895 /*
1896  *      Pull a packet from our receive queue and hand it to the user.
1897  *      If necessary we block.
1898  */
1899
1900 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1901                           struct msghdr *msg, size_t len, int flags)
1902 {
1903         struct sock *sk = sock->sk;
1904         struct sk_buff *skb;
1905         int copied, err;
1906         struct sockaddr_ll *sll;
1907         int vnet_hdr_len = 0;
1908
1909         err = -EINVAL;
1910         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
1911                 goto out;
1912
1913 #if 0
1914         /* What error should we return now? EUNATTACH? */
1915         if (pkt_sk(sk)->ifindex < 0)
1916                 return -ENODEV;
1917 #endif
1918
1919         if (flags & MSG_ERRQUEUE) {
1920                 err = packet_recv_error(sk, msg, len);
1921                 goto out;
1922         }
1923
1924         /*
1925          *      Call the generic datagram receiver. This handles all sorts
1926          *      of horrible races and re-entrancy so we can forget about it
1927          *      in the protocol layers.
1928          *
1929          *      Now it will return ENETDOWN, if device have just gone down,
1930          *      but then it will block.
1931          */
1932
1933         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1934
1935         /*
1936          *      An error occurred so return it. Because skb_recv_datagram()
1937          *      handles the blocking we don't see and worry about blocking
1938          *      retries.
1939          */
1940
1941         if (skb == NULL)
1942                 goto out;
1943
1944         if (pkt_sk(sk)->has_vnet_hdr) {
1945                 struct virtio_net_hdr vnet_hdr = { 0 };
1946
1947                 err = -EINVAL;
1948                 vnet_hdr_len = sizeof(vnet_hdr);
1949                 if (len < vnet_hdr_len)
1950                         goto out_free;
1951
1952                 len -= vnet_hdr_len;
1953
1954                 if (skb_is_gso(skb)) {
1955                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1956
1957                         /* This is a hint as to how much should be linear. */
1958                         vnet_hdr.hdr_len = skb_headlen(skb);
1959                         vnet_hdr.gso_size = sinfo->gso_size;
1960                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1961                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1962                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1963                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1964                         else if (sinfo->gso_type & SKB_GSO_UDP)
1965                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1966                         else if (sinfo->gso_type & SKB_GSO_FCOE)
1967                                 goto out_free;
1968                         else
1969                                 BUG();
1970                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1971                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1972                 } else
1973                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1974
1975                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1976                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1977                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
1978                         vnet_hdr.csum_offset = skb->csum_offset;
1979                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1980                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1981                 } /* else everything is zero */
1982
1983                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
1984                                      vnet_hdr_len);
1985                 if (err < 0)
1986                         goto out_free;
1987         }
1988
1989         /*
1990          *      If the address length field is there to be filled in, we fill
1991          *      it in now.
1992          */
1993
1994         sll = &PACKET_SKB_CB(skb)->sa.ll;
1995         if (sock->type == SOCK_PACKET)
1996                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1997         else
1998                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1999
2000         /*
2001          *      You lose any data beyond the buffer you gave. If it worries a
2002          *      user program they can ask the device for its MTU anyway.
2003          */
2004
2005         copied = skb->len;
2006         if (copied > len) {
2007                 copied = len;
2008                 msg->msg_flags |= MSG_TRUNC;
2009         }
2010
2011         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2012         if (err)
2013                 goto out_free;
2014
2015         sock_recv_ts_and_drops(msg, sk, skb);
2016
2017         if (msg->msg_name)
2018                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2019                        msg->msg_namelen);
2020
2021         if (pkt_sk(sk)->auxdata) {
2022                 struct tpacket_auxdata aux;
2023
2024                 aux.tp_status = TP_STATUS_USER;
2025                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2026                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2027                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2028                 aux.tp_snaplen = skb->len;
2029                 aux.tp_mac = 0;
2030                 aux.tp_net = skb_network_offset(skb);
2031                 if (vlan_tx_tag_present(skb)) {
2032                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2033                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2034                 } else {
2035                         aux.tp_vlan_tci = 0;
2036                 }
2037                 aux.tp_padding = 0;
2038                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2039         }
2040
2041         /*
2042          *      Free or return the buffer as appropriate. Again this
2043          *      hides all the races and re-entrancy issues from us.
2044          */
2045         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2046
2047 out_free:
2048         skb_free_datagram(sk, skb);
2049 out:
2050         return err;
2051 }
2052
2053 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2054                                int *uaddr_len, int peer)
2055 {
2056         struct net_device *dev;
2057         struct sock *sk = sock->sk;
2058
2059         if (peer)
2060                 return -EOPNOTSUPP;
2061
2062         uaddr->sa_family = AF_PACKET;
2063         rcu_read_lock();
2064         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2065         if (dev)
2066                 strncpy(uaddr->sa_data, dev->name, 14);
2067         else
2068                 memset(uaddr->sa_data, 0, 14);
2069         rcu_read_unlock();
2070         *uaddr_len = sizeof(*uaddr);
2071
2072         return 0;
2073 }
2074
2075 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2076                           int *uaddr_len, int peer)
2077 {
2078         struct net_device *dev;
2079         struct sock *sk = sock->sk;
2080         struct packet_sock *po = pkt_sk(sk);
2081         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2082
2083         if (peer)
2084                 return -EOPNOTSUPP;
2085
2086         sll->sll_family = AF_PACKET;
2087         sll->sll_ifindex = po->ifindex;
2088         sll->sll_protocol = po->num;
2089         sll->sll_pkttype = 0;
2090         rcu_read_lock();
2091         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2092         if (dev) {
2093                 sll->sll_hatype = dev->type;
2094                 sll->sll_halen = dev->addr_len;
2095                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2096         } else {
2097                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2098                 sll->sll_halen = 0;
2099         }
2100         rcu_read_unlock();
2101         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2102
2103         return 0;
2104 }
2105
2106 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2107                          int what)
2108 {
2109         switch (i->type) {
2110         case PACKET_MR_MULTICAST:
2111                 if (i->alen != dev->addr_len)
2112                         return -EINVAL;
2113                 if (what > 0)
2114                         return dev_mc_add(dev, i->addr);
2115                 else
2116                         return dev_mc_del(dev, i->addr);
2117                 break;
2118         case PACKET_MR_PROMISC:
2119                 return dev_set_promiscuity(dev, what);
2120                 break;
2121         case PACKET_MR_ALLMULTI:
2122                 return dev_set_allmulti(dev, what);
2123                 break;
2124         case PACKET_MR_UNICAST:
2125                 if (i->alen != dev->addr_len)
2126                         return -EINVAL;
2127                 if (what > 0)
2128                         return dev_uc_add(dev, i->addr);
2129                 else
2130                         return dev_uc_del(dev, i->addr);
2131                 break;
2132         default:
2133                 break;
2134         }
2135         return 0;
2136 }
2137
2138 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2139 {
2140         for ( ; i; i = i->next) {
2141                 if (i->ifindex == dev->ifindex)
2142                         packet_dev_mc(dev, i, what);
2143         }
2144 }
2145
2146 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2147 {
2148         struct packet_sock *po = pkt_sk(sk);
2149         struct packet_mclist *ml, *i;
2150         struct net_device *dev;
2151         int err;
2152
2153         rtnl_lock();
2154
2155         err = -ENODEV;
2156         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2157         if (!dev)
2158                 goto done;
2159
2160         err = -EINVAL;
2161         if (mreq->mr_alen > dev->addr_len)
2162                 goto done;
2163
2164         err = -ENOBUFS;
2165         i = kmalloc(sizeof(*i), GFP_KERNEL);
2166         if (i == NULL)
2167                 goto done;
2168
2169         err = 0;
2170         for (ml = po->mclist; ml; ml = ml->next) {
2171                 if (ml->ifindex == mreq->mr_ifindex &&
2172                     ml->type == mreq->mr_type &&
2173                     ml->alen == mreq->mr_alen &&
2174                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2175                         ml->count++;
2176                         /* Free the new element ... */
2177                         kfree(i);
2178                         goto done;
2179                 }
2180         }
2181
2182         i->type = mreq->mr_type;
2183         i->ifindex = mreq->mr_ifindex;
2184         i->alen = mreq->mr_alen;
2185         memcpy(i->addr, mreq->mr_address, i->alen);
2186         i->count = 1;
2187         i->next = po->mclist;
2188         po->mclist = i;
2189         err = packet_dev_mc(dev, i, 1);
2190         if (err) {
2191                 po->mclist = i->next;
2192                 kfree(i);
2193         }
2194
2195 done:
2196         rtnl_unlock();
2197         return err;
2198 }
2199
2200 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2201 {
2202         struct packet_mclist *ml, **mlp;
2203
2204         rtnl_lock();
2205
2206         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2207                 if (ml->ifindex == mreq->mr_ifindex &&
2208                     ml->type == mreq->mr_type &&
2209                     ml->alen == mreq->mr_alen &&
2210                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2211                         if (--ml->count == 0) {
2212                                 struct net_device *dev;
2213                                 *mlp = ml->next;
2214                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2215                                 if (dev)
2216                                         packet_dev_mc(dev, ml, -1);
2217                                 kfree(ml);
2218                         }
2219                         rtnl_unlock();
2220                         return 0;
2221                 }
2222         }
2223         rtnl_unlock();
2224         return -EADDRNOTAVAIL;
2225 }
2226
2227 static void packet_flush_mclist(struct sock *sk)
2228 {
2229         struct packet_sock *po = pkt_sk(sk);
2230         struct packet_mclist *ml;
2231
2232         if (!po->mclist)
2233                 return;
2234
2235         rtnl_lock();
2236         while ((ml = po->mclist) != NULL) {
2237                 struct net_device *dev;
2238
2239                 po->mclist = ml->next;
2240                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2241                 if (dev != NULL)
2242                         packet_dev_mc(dev, ml, -1);
2243                 kfree(ml);
2244         }
2245         rtnl_unlock();
2246 }
2247
2248 static int
2249 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
2250 {
2251         struct sock *sk = sock->sk;
2252         struct packet_sock *po = pkt_sk(sk);
2253         int ret;
2254
2255         if (level != SOL_PACKET)
2256                 return -ENOPROTOOPT;
2257
2258         switch (optname) {
2259         case PACKET_ADD_MEMBERSHIP:
2260         case PACKET_DROP_MEMBERSHIP:
2261         {
2262                 struct packet_mreq_max mreq;
2263                 int len = optlen;
2264                 memset(&mreq, 0, sizeof(mreq));
2265                 if (len < sizeof(struct packet_mreq))
2266                         return -EINVAL;
2267                 if (len > sizeof(mreq))
2268                         len = sizeof(mreq);
2269                 if (copy_from_user(&mreq, optval, len))
2270                         return -EFAULT;
2271                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
2272                         return -EINVAL;
2273                 if (optname == PACKET_ADD_MEMBERSHIP)
2274                         ret = packet_mc_add(sk, &mreq);
2275                 else
2276                         ret = packet_mc_drop(sk, &mreq);
2277                 return ret;
2278         }
2279
2280         case PACKET_RX_RING:
2281         case PACKET_TX_RING:
2282         {
2283                 struct tpacket_req req;
2284
2285                 if (optlen < sizeof(req))
2286                         return -EINVAL;
2287                 if (pkt_sk(sk)->has_vnet_hdr)
2288                         return -EINVAL;
2289                 if (copy_from_user(&req, optval, sizeof(req)))
2290                         return -EFAULT;
2291                 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
2292         }
2293         case PACKET_COPY_THRESH:
2294         {
2295                 int val;
2296
2297                 if (optlen != sizeof(val))
2298                         return -EINVAL;
2299                 if (copy_from_user(&val, optval, sizeof(val)))
2300                         return -EFAULT;
2301
2302                 pkt_sk(sk)->copy_thresh = val;
2303                 return 0;
2304         }
2305         case PACKET_VERSION:
2306         {
2307                 int val;
2308
2309                 if (optlen != sizeof(val))
2310                         return -EINVAL;
2311                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2312                         return -EBUSY;
2313                 if (copy_from_user(&val, optval, sizeof(val)))
2314                         return -EFAULT;
2315                 switch (val) {
2316                 case TPACKET_V1:
2317                 case TPACKET_V2:
2318                         po->tp_version = val;
2319                         return 0;
2320                 default:
2321                         return -EINVAL;
2322                 }
2323         }
2324         case PACKET_RESERVE:
2325         {
2326                 unsigned int val;
2327
2328                 if (optlen != sizeof(val))
2329                         return -EINVAL;
2330                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2331                         return -EBUSY;
2332                 if (copy_from_user(&val, optval, sizeof(val)))
2333                         return -EFAULT;
2334                 po->tp_reserve = val;
2335                 return 0;
2336         }
2337         case PACKET_LOSS:
2338         {
2339                 unsigned int val;
2340
2341                 if (optlen != sizeof(val))
2342                         return -EINVAL;
2343                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2344                         return -EBUSY;
2345                 if (copy_from_user(&val, optval, sizeof(val)))
2346                         return -EFAULT;
2347                 po->tp_loss = !!val;
2348                 return 0;
2349         }
2350         case PACKET_AUXDATA:
2351         {
2352                 int val;
2353
2354                 if (optlen < sizeof(val))
2355                         return -EINVAL;
2356                 if (copy_from_user(&val, optval, sizeof(val)))
2357                         return -EFAULT;
2358
2359                 po->auxdata = !!val;
2360                 return 0;
2361         }
2362         case PACKET_ORIGDEV:
2363         {
2364                 int val;
2365
2366                 if (optlen < sizeof(val))
2367                         return -EINVAL;
2368                 if (copy_from_user(&val, optval, sizeof(val)))
2369                         return -EFAULT;
2370
2371                 po->origdev = !!val;
2372                 return 0;
2373         }
2374         case PACKET_VNET_HDR:
2375         {
2376                 int val;
2377
2378                 if (sock->type != SOCK_RAW)
2379                         return -EINVAL;
2380                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2381                         return -EBUSY;
2382                 if (optlen < sizeof(val))
2383                         return -EINVAL;
2384                 if (copy_from_user(&val, optval, sizeof(val)))
2385                         return -EFAULT;
2386
2387                 po->has_vnet_hdr = !!val;
2388                 return 0;
2389         }
2390         case PACKET_TIMESTAMP:
2391         {
2392                 int val;
2393
2394                 if (optlen != sizeof(val))
2395                         return -EINVAL;
2396                 if (copy_from_user(&val, optval, sizeof(val)))
2397                         return -EFAULT;
2398
2399                 po->tp_tstamp = val;
2400                 return 0;
2401         }
2402         case PACKET_FANOUT:
2403         {
2404                 int val;
2405
2406                 if (optlen != sizeof(val))
2407                         return -EINVAL;
2408                 if (copy_from_user(&val, optval, sizeof(val)))
2409                         return -EFAULT;
2410
2411                 return fanout_add(sk, val & 0xffff, val >> 16);
2412         }
2413         default:
2414                 return -ENOPROTOOPT;
2415         }
2416 }
2417
2418 static int packet_getsockopt(struct socket *sock, int level, int optname,
2419                              char __user *optval, int __user *optlen)
2420 {
2421         int len;
2422         int val;
2423         struct sock *sk = sock->sk;
2424         struct packet_sock *po = pkt_sk(sk);
2425         void *data;
2426         struct tpacket_stats st;
2427
2428         if (level != SOL_PACKET)
2429                 return -ENOPROTOOPT;
2430
2431         if (get_user(len, optlen))
2432                 return -EFAULT;
2433
2434         if (len < 0)
2435                 return -EINVAL;
2436
2437         switch (optname) {
2438         case PACKET_STATISTICS:
2439                 if (len > sizeof(struct tpacket_stats))
2440                         len = sizeof(struct tpacket_stats);
2441                 spin_lock_bh(&sk->sk_receive_queue.lock);
2442                 st = po->stats;
2443                 memset(&po->stats, 0, sizeof(st));
2444                 spin_unlock_bh(&sk->sk_receive_queue.lock);
2445                 st.tp_packets += st.tp_drops;
2446
2447                 data = &st;
2448                 break;
2449         case PACKET_AUXDATA:
2450                 if (len > sizeof(int))
2451                         len = sizeof(int);
2452                 val = po->auxdata;
2453
2454                 data = &val;
2455                 break;
2456         case PACKET_ORIGDEV:
2457                 if (len > sizeof(int))
2458                         len = sizeof(int);
2459                 val = po->origdev;
2460
2461                 data = &val;
2462                 break;
2463         case PACKET_VNET_HDR:
2464                 if (len > sizeof(int))
2465                         len = sizeof(int);
2466                 val = po->has_vnet_hdr;
2467
2468                 data = &val;
2469                 break;
2470         case PACKET_VERSION:
2471                 if (len > sizeof(int))
2472                         len = sizeof(int);
2473                 val = po->tp_version;
2474                 data = &val;
2475                 break;
2476         case PACKET_HDRLEN:
2477                 if (len > sizeof(int))
2478                         len = sizeof(int);
2479                 if (copy_from_user(&val, optval, len))
2480                         return -EFAULT;
2481                 switch (val) {
2482                 case TPACKET_V1:
2483                         val = sizeof(struct tpacket_hdr);
2484                         break;
2485                 case TPACKET_V2:
2486                         val = sizeof(struct tpacket2_hdr);
2487                         break;
2488                 default:
2489                         return -EINVAL;
2490                 }
2491                 data = &val;
2492                 break;
2493         case PACKET_RESERVE:
2494                 if (len > sizeof(unsigned int))
2495                         len = sizeof(unsigned int);
2496                 val = po->tp_reserve;
2497                 data = &val;
2498                 break;
2499         case PACKET_LOSS:
2500                 if (len > sizeof(unsigned int))
2501                         len = sizeof(unsigned int);
2502                 val = po->tp_loss;
2503                 data = &val;
2504                 break;
2505         case PACKET_TIMESTAMP:
2506                 if (len > sizeof(int))
2507                         len = sizeof(int);
2508                 val = po->tp_tstamp;
2509                 data = &val;
2510                 break;
2511         case PACKET_FANOUT:
2512                 if (len > sizeof(int))
2513                         len = sizeof(int);
2514                 val = (po->fanout ?
2515                        ((u32)po->fanout->id |
2516                         ((u32)po->fanout->type << 16)) :
2517                        0);
2518                 data = &val;
2519                 break;
2520         default:
2521                 return -ENOPROTOOPT;
2522         }
2523
2524         if (put_user(len, optlen))
2525                 return -EFAULT;
2526         if (copy_to_user(optval, data, len))
2527                 return -EFAULT;
2528         return 0;
2529 }
2530
2531
2532 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
2533 {
2534         struct sock *sk;
2535         struct hlist_node *node;
2536         struct net_device *dev = data;
2537         struct net *net = dev_net(dev);
2538
2539         rcu_read_lock();
2540         sk_for_each_rcu(sk, node, &net->packet.sklist) {
2541                 struct packet_sock *po = pkt_sk(sk);
2542
2543                 switch (msg) {
2544                 case NETDEV_UNREGISTER:
2545                         if (po->mclist)
2546                                 packet_dev_mclist(dev, po->mclist, -1);
2547                         /* fallthrough */
2548
2549                 case NETDEV_DOWN:
2550                         if (dev->ifindex == po->ifindex) {
2551                                 spin_lock(&po->bind_lock);
2552                                 if (po->running) {
2553                                         __unregister_prot_hook(sk, false);
2554                                         sk->sk_err = ENETDOWN;
2555                                         if (!sock_flag(sk, SOCK_DEAD))
2556                                                 sk->sk_error_report(sk);
2557                                 }
2558                                 if (msg == NETDEV_UNREGISTER) {
2559                                         po->ifindex = -1;
2560                                         if (po->prot_hook.dev)
2561                                                 dev_put(po->prot_hook.dev);
2562                                         po->prot_hook.dev = NULL;
2563                                 }
2564                                 spin_unlock(&po->bind_lock);
2565                         }
2566                         break;
2567                 case NETDEV_UP:
2568                         if (dev->ifindex == po->ifindex) {
2569                                 spin_lock(&po->bind_lock);
2570                                 if (po->num)
2571                                         register_prot_hook(sk);
2572                                 spin_unlock(&po->bind_lock);
2573                         }
2574                         break;
2575                 }
2576         }
2577         rcu_read_unlock();
2578         return NOTIFY_DONE;
2579 }
2580
2581
2582 static int packet_ioctl(struct socket *sock, unsigned int cmd,
2583                         unsigned long arg)
2584 {
2585         struct sock *sk = sock->sk;
2586
2587         switch (cmd) {
2588         case SIOCOUTQ:
2589         {
2590                 int amount = sk_wmem_alloc_get(sk);
2591
2592                 return put_user(amount, (int __user *)arg);
2593         }
2594         case SIOCINQ:
2595         {
2596                 struct sk_buff *skb;
2597                 int amount = 0;
2598
2599                 spin_lock_bh(&sk->sk_receive_queue.lock);
2600                 skb = skb_peek(&sk->sk_receive_queue);
2601                 if (skb)
2602                         amount = skb->len;
2603                 spin_unlock_bh(&sk->sk_receive_queue.lock);
2604                 return put_user(amount, (int __user *)arg);
2605         }
2606         case SIOCGSTAMP:
2607                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2608         case SIOCGSTAMPNS:
2609                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2610
2611 #ifdef CONFIG_INET
2612         case SIOCADDRT:
2613         case SIOCDELRT:
2614         case SIOCDARP:
2615         case SIOCGARP:
2616         case SIOCSARP:
2617         case SIOCGIFADDR:
2618         case SIOCSIFADDR:
2619         case SIOCGIFBRDADDR:
2620         case SIOCSIFBRDADDR:
2621         case SIOCGIFNETMASK:
2622         case SIOCSIFNETMASK:
2623         case SIOCGIFDSTADDR:
2624         case SIOCSIFDSTADDR:
2625         case SIOCSIFFLAGS:
2626                 return inet_dgram_ops.ioctl(sock, cmd, arg);
2627 #endif
2628
2629         default:
2630                 return -ENOIOCTLCMD;
2631         }
2632         return 0;
2633 }
2634
2635 static unsigned int packet_poll(struct file *file, struct socket *sock,
2636                                 poll_table *wait)
2637 {
2638         struct sock *sk = sock->sk;
2639         struct packet_sock *po = pkt_sk(sk);
2640         unsigned int mask = datagram_poll(file, sock, wait);
2641
2642         spin_lock_bh(&sk->sk_receive_queue.lock);
2643         if (po->rx_ring.pg_vec) {
2644                 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2645                         mask |= POLLIN | POLLRDNORM;
2646         }
2647         spin_unlock_bh(&sk->sk_receive_queue.lock);
2648         spin_lock_bh(&sk->sk_write_queue.lock);
2649         if (po->tx_ring.pg_vec) {
2650                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2651                         mask |= POLLOUT | POLLWRNORM;
2652         }
2653         spin_unlock_bh(&sk->sk_write_queue.lock);
2654         return mask;
2655 }
2656
2657
2658 /* Dirty? Well, I still did not learn better way to account
2659  * for user mmaps.
2660  */
2661
2662 static void packet_mm_open(struct vm_area_struct *vma)
2663 {
2664         struct file *file = vma->vm_file;
2665         struct socket *sock = file->private_data;
2666         struct sock *sk = sock->sk;
2667
2668         if (sk)
2669                 atomic_inc(&pkt_sk(sk)->mapped);
2670 }
2671
2672 static void packet_mm_close(struct vm_area_struct *vma)
2673 {
2674         struct file *file = vma->vm_file;
2675         struct socket *sock = file->private_data;
2676         struct sock *sk = sock->sk;
2677
2678         if (sk)
2679                 atomic_dec(&pkt_sk(sk)->mapped);
2680 }
2681
2682 static const struct vm_operations_struct packet_mmap_ops = {
2683         .open   =       packet_mm_open,
2684         .close  =       packet_mm_close,
2685 };
2686
2687 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
2688                         unsigned int len)
2689 {
2690         int i;
2691
2692         for (i = 0; i < len; i++) {
2693                 if (likely(pg_vec[i].buffer)) {
2694                         if (is_vmalloc_addr(pg_vec[i].buffer))
2695                                 vfree(pg_vec[i].buffer);
2696                         else
2697                                 free_pages((unsigned long)pg_vec[i].buffer,
2698                                            order);
2699                         pg_vec[i].buffer = NULL;
2700                 }
2701         }
2702         kfree(pg_vec);
2703 }
2704
2705 static inline char *alloc_one_pg_vec_page(unsigned long order)
2706 {
2707         char *buffer = NULL;
2708         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
2709                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
2710
2711         buffer = (char *) __get_free_pages(gfp_flags, order);
2712
2713         if (buffer)
2714                 return buffer;
2715
2716         /*
2717          * __get_free_pages failed, fall back to vmalloc
2718          */
2719         buffer = vzalloc((1 << order) * PAGE_SIZE);
2720
2721         if (buffer)
2722                 return buffer;
2723
2724         /*
2725          * vmalloc failed, lets dig into swap here
2726          */
2727         gfp_flags &= ~__GFP_NORETRY;
2728         buffer = (char *)__get_free_pages(gfp_flags, order);
2729         if (buffer)
2730                 return buffer;
2731
2732         /*
2733          * complete and utter failure
2734          */
2735         return NULL;
2736 }
2737
2738 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
2739 {
2740         unsigned int block_nr = req->tp_block_nr;
2741         struct pgv *pg_vec;
2742         int i;
2743
2744         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
2745         if (unlikely(!pg_vec))
2746                 goto out;
2747
2748         for (i = 0; i < block_nr; i++) {
2749                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
2750                 if (unlikely(!pg_vec[i].buffer))
2751                         goto out_free_pgvec;
2752         }
2753
2754 out:
2755         return pg_vec;
2756
2757 out_free_pgvec:
2758         free_pg_vec(pg_vec, order, block_nr);
2759         pg_vec = NULL;
2760         goto out;
2761 }
2762
2763 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2764                 int closing, int tx_ring)
2765 {
2766         struct pgv *pg_vec = NULL;
2767         struct packet_sock *po = pkt_sk(sk);
2768         int was_running, order = 0;
2769         struct packet_ring_buffer *rb;
2770         struct sk_buff_head *rb_queue;
2771         __be16 num;
2772         int err;
2773
2774         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2775         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2776
2777         err = -EBUSY;
2778         if (!closing) {
2779                 if (atomic_read(&po->mapped))
2780                         goto out;
2781                 if (atomic_read(&rb->pending))
2782                         goto out;
2783         }
2784
2785         if (req->tp_block_nr) {
2786                 /* Sanity tests and some calculations */
2787                 err = -EBUSY;
2788                 if (unlikely(rb->pg_vec))
2789                         goto out;
2790
2791                 switch (po->tp_version) {
2792                 case TPACKET_V1:
2793                         po->tp_hdrlen = TPACKET_HDRLEN;
2794                         break;
2795                 case TPACKET_V2:
2796                         po->tp_hdrlen = TPACKET2_HDRLEN;
2797                         break;
2798                 }
2799
2800                 err = -EINVAL;
2801                 if (unlikely((int)req->tp_block_size <= 0))
2802                         goto out;
2803                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2804                         goto out;
2805                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2806                                         po->tp_reserve))
2807                         goto out;
2808                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2809                         goto out;
2810
2811                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2812                 if (unlikely(rb->frames_per_block <= 0))
2813                         goto out;
2814                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2815                                         req->tp_frame_nr))
2816                         goto out;
2817
2818                 err = -ENOMEM;
2819                 order = get_order(req->tp_block_size);
2820                 pg_vec = alloc_pg_vec(req, order);
2821                 if (unlikely(!pg_vec))
2822                         goto out;
2823         }
2824         /* Done */
2825         else {
2826                 err = -EINVAL;
2827                 if (unlikely(req->tp_frame_nr))
2828                         goto out;
2829         }
2830
2831         lock_sock(sk);
2832
2833         /* Detach socket from network */
2834         spin_lock(&po->bind_lock);
2835         was_running = po->running;
2836         num = po->num;
2837         if (was_running) {
2838                 po->num = 0;
2839                 __unregister_prot_hook(sk, false);
2840         }
2841         spin_unlock(&po->bind_lock);
2842
2843         synchronize_net();
2844
2845         err = -EBUSY;
2846         mutex_lock(&po->pg_vec_lock);
2847         if (closing || atomic_read(&po->mapped) == 0) {
2848                 err = 0;
2849                 spin_lock_bh(&rb_queue->lock);
2850                 swap(rb->pg_vec, pg_vec);
2851                 rb->frame_max = (req->tp_frame_nr - 1);
2852                 rb->head = 0;
2853                 rb->frame_size = req->tp_frame_size;
2854                 spin_unlock_bh(&rb_queue->lock);
2855
2856                 swap(rb->pg_vec_order, order);
2857                 swap(rb->pg_vec_len, req->tp_block_nr);
2858
2859                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2860                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2861                                                 tpacket_rcv : packet_rcv;
2862                 skb_queue_purge(rb_queue);
2863                 if (atomic_read(&po->mapped))
2864                         pr_err("packet_mmap: vma is busy: %d\n",
2865                                atomic_read(&po->mapped));
2866         }
2867         mutex_unlock(&po->pg_vec_lock);
2868
2869         spin_lock(&po->bind_lock);
2870         if (was_running) {
2871                 po->num = num;
2872                 register_prot_hook(sk);
2873         }
2874         spin_unlock(&po->bind_lock);
2875
2876         release_sock(sk);
2877
2878         if (pg_vec)
2879                 free_pg_vec(pg_vec, order, req->tp_block_nr);
2880 out:
2881         return err;
2882 }
2883
2884 static int packet_mmap(struct file *file, struct socket *sock,
2885                 struct vm_area_struct *vma)
2886 {
2887         struct sock *sk = sock->sk;
2888         struct packet_sock *po = pkt_sk(sk);
2889         unsigned long size, expected_size;
2890         struct packet_ring_buffer *rb;
2891         unsigned long start;
2892         int err = -EINVAL;
2893         int i;
2894
2895         if (vma->vm_pgoff)
2896                 return -EINVAL;
2897
2898         mutex_lock(&po->pg_vec_lock);
2899
2900         expected_size = 0;
2901         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2902                 if (rb->pg_vec) {
2903                         expected_size += rb->pg_vec_len
2904                                                 * rb->pg_vec_pages
2905                                                 * PAGE_SIZE;
2906                 }
2907         }
2908
2909         if (expected_size == 0)
2910                 goto out;
2911
2912         size = vma->vm_end - vma->vm_start;
2913         if (size != expected_size)
2914                 goto out;
2915
2916         start = vma->vm_start;
2917         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2918                 if (rb->pg_vec == NULL)
2919                         continue;
2920
2921                 for (i = 0; i < rb->pg_vec_len; i++) {
2922                         struct page *page;
2923                         void *kaddr = rb->pg_vec[i].buffer;
2924                         int pg_num;
2925
2926                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
2927                                 page = pgv_to_page(kaddr);
2928                                 err = vm_insert_page(vma, start, page);
2929                                 if (unlikely(err))
2930                                         goto out;
2931                                 start += PAGE_SIZE;
2932                                 kaddr += PAGE_SIZE;
2933                         }
2934                 }
2935         }
2936
2937         atomic_inc(&po->mapped);
2938         vma->vm_ops = &packet_mmap_ops;
2939         err = 0;
2940
2941 out:
2942         mutex_unlock(&po->pg_vec_lock);
2943         return err;
2944 }
2945
2946 static const struct proto_ops packet_ops_spkt = {
2947         .family =       PF_PACKET,
2948         .owner =        THIS_MODULE,
2949         .release =      packet_release,
2950         .bind =         packet_bind_spkt,
2951         .connect =      sock_no_connect,
2952         .socketpair =   sock_no_socketpair,
2953         .accept =       sock_no_accept,
2954         .getname =      packet_getname_spkt,
2955         .poll =         datagram_poll,
2956         .ioctl =        packet_ioctl,
2957         .listen =       sock_no_listen,
2958         .shutdown =     sock_no_shutdown,
2959         .setsockopt =   sock_no_setsockopt,
2960         .getsockopt =   sock_no_getsockopt,
2961         .sendmsg =      packet_sendmsg_spkt,
2962         .recvmsg =      packet_recvmsg,
2963         .mmap =         sock_no_mmap,
2964         .sendpage =     sock_no_sendpage,
2965 };
2966
2967 static const struct proto_ops packet_ops = {
2968         .family =       PF_PACKET,
2969         .owner =        THIS_MODULE,
2970         .release =      packet_release,
2971         .bind =         packet_bind,
2972         .connect =      sock_no_connect,
2973         .socketpair =   sock_no_socketpair,
2974         .accept =       sock_no_accept,
2975         .getname =      packet_getname,
2976         .poll =         packet_poll,
2977         .ioctl =        packet_ioctl,
2978         .listen =       sock_no_listen,
2979         .shutdown =     sock_no_shutdown,
2980         .setsockopt =   packet_setsockopt,
2981         .getsockopt =   packet_getsockopt,
2982         .sendmsg =      packet_sendmsg,
2983         .recvmsg =      packet_recvmsg,
2984         .mmap =         packet_mmap,
2985         .sendpage =     sock_no_sendpage,
2986 };
2987
2988 static const struct net_proto_family packet_family_ops = {
2989         .family =       PF_PACKET,
2990         .create =       packet_create,
2991         .owner  =       THIS_MODULE,
2992 };
2993
2994 static struct notifier_block packet_netdev_notifier = {
2995         .notifier_call =        packet_notifier,
2996 };
2997
2998 #ifdef CONFIG_PROC_FS
2999
3000 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3001         __acquires(RCU)
3002 {
3003         struct net *net = seq_file_net(seq);
3004
3005         rcu_read_lock();
3006         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3007 }
3008
3009 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3010 {
3011         struct net *net = seq_file_net(seq);
3012         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3013 }
3014
3015 static void packet_seq_stop(struct seq_file *seq, void *v)
3016         __releases(RCU)
3017 {
3018         rcu_read_unlock();
3019 }
3020
3021 static int packet_seq_show(struct seq_file *seq, void *v)
3022 {
3023         if (v == SEQ_START_TOKEN)
3024                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3025         else {
3026                 struct sock *s = sk_entry(v);
3027                 const struct packet_sock *po = pkt_sk(s);
3028
3029                 seq_printf(seq,
3030                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3031                            s,
3032                            atomic_read(&s->sk_refcnt),
3033                            s->sk_type,
3034                            ntohs(po->num),
3035                            po->ifindex,
3036                            po->running,
3037                            atomic_read(&s->sk_rmem_alloc),
3038                            sock_i_uid(s),
3039                            sock_i_ino(s));
3040         }
3041
3042         return 0;
3043 }
3044
3045 static const struct seq_operations packet_seq_ops = {
3046         .start  = packet_seq_start,
3047         .next   = packet_seq_next,
3048         .stop   = packet_seq_stop,
3049         .show   = packet_seq_show,
3050 };
3051
3052 static int packet_seq_open(struct inode *inode, struct file *file)
3053 {
3054         return seq_open_net(inode, file, &packet_seq_ops,
3055                             sizeof(struct seq_net_private));
3056 }
3057
3058 static const struct file_operations packet_seq_fops = {
3059         .owner          = THIS_MODULE,
3060         .open           = packet_seq_open,
3061         .read           = seq_read,
3062         .llseek         = seq_lseek,
3063         .release        = seq_release_net,
3064 };
3065
3066 #endif
3067
3068 static int __net_init packet_net_init(struct net *net)
3069 {
3070         spin_lock_init(&net->packet.sklist_lock);
3071         INIT_HLIST_HEAD(&net->packet.sklist);
3072
3073         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3074                 return -ENOMEM;
3075
3076         return 0;
3077 }
3078
3079 static void __net_exit packet_net_exit(struct net *net)
3080 {
3081         proc_net_remove(net, "packet");
3082 }
3083
3084 static struct pernet_operations packet_net_ops = {
3085         .init = packet_net_init,
3086         .exit = packet_net_exit,
3087 };
3088
3089
3090 static void __exit packet_exit(void)
3091 {
3092         unregister_netdevice_notifier(&packet_netdev_notifier);
3093         unregister_pernet_subsys(&packet_net_ops);
3094         sock_unregister(PF_PACKET);
3095         proto_unregister(&packet_proto);
3096 }
3097
3098 static int __init packet_init(void)
3099 {
3100         int rc = proto_register(&packet_proto, 0);
3101
3102         if (rc != 0)
3103                 goto out;
3104
3105         sock_register(&packet_family_ops);
3106         register_pernet_subsys(&packet_net_ops);
3107         register_netdevice_notifier(&packet_netdev_notifier);
3108 out:
3109         return rc;
3110 }
3111
3112 module_init(packet_init);
3113 module_exit(packet_exit);
3114 MODULE_LICENSE("GPL");
3115 MODULE_ALIAS_NETPROTO(PF_PACKET);