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         spin_lock(&sk->sk_receive_queue.lock);
 965         po->stats.tp_drops++;
 966         atomic_inc(&sk->sk_drops);
 967         spin_unlock(&sk->sk_receive_queue.lock);
 968
 969 drop_n_restore:
 970         if (skb_head != skb->data && skb_shared(skb)) {
 971                 skb->data = skb_head;
 972                 skb->len = skb_len;
 973         }
 974 drop:
 975         consume_skb(skb);
 976         return 0;
 977 }
 978
 979 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 980                        struct packet_type *pt, struct net_device *orig_dev)
 981 {
 982         struct sock *sk;
 983         struct packet_sock *po;
 984         struct sockaddr_ll *sll;
 985         union {
 986                 struct tpacket_hdr *h1;
 987                 struct tpacket2_hdr *h2;
 988                 void *raw;
 989         } h;
 990         u8 *skb_head = skb->data;
 991         int skb_len = skb->len;
 992         unsigned int snaplen, res;
 993         unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
 994         unsigned short macoff, netoff, hdrlen;
 995         struct sk_buff *copy_skb = NULL;
 996         struct timeval tv;
 997         struct timespec ts;
 998         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 999
1000         if (skb->pkt_type == PACKET_LOOPBACK)
1001                 goto drop;
1002
1003         sk = pt->af_packet_priv;
1004         po = pkt_sk(sk);
1005
1006         if (!net_eq(dev_net(dev), sock_net(sk)))
1007                 goto drop;
1008
1009         if (dev->header_ops) {
1010                 if (sk->sk_type != SOCK_DGRAM)
1011                         skb_push(skb, skb->data - skb_mac_header(skb));
1012                 else if (skb->pkt_type == PACKET_OUTGOING) {
1013                         /* Special case: outgoing packets have ll header at head */
1014                         skb_pull(skb, skb_network_offset(skb));
1015                 }
1016         }
1017
1018         if (skb->ip_summed == CHECKSUM_PARTIAL)
1019                 status |= TP_STATUS_CSUMNOTREADY;
1020
1021         snaplen = skb->len;
1022
1023         res = run_filter(skb, sk, snaplen);
1024         if (!res)
1025                 goto drop_n_restore;
1026         if (snaplen > res)
1027                 snaplen = res;
1028
1029         if (sk->sk_type == SOCK_DGRAM) {
1030                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1031                                   po->tp_reserve;
1032         } else {
1033                 unsigned maclen = skb_network_offset(skb);
1034                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1035                                        (maclen < 16 ? 16 : maclen)) +
1036                         po->tp_reserve;
1037                 macoff = netoff - maclen;
1038         }
1039
1040         if (macoff + snaplen > po->rx_ring.frame_size) {
1041                 if (po->copy_thresh &&
1042                     atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
1043                     (unsigned)sk->sk_rcvbuf) {
1044                         if (skb_shared(skb)) {
1045                                 copy_skb = skb_clone(skb, GFP_ATOMIC);
1046                         } else {
1047                                 copy_skb = skb_get(skb);
1048                                 skb_head = skb->data;
1049                         }
1050                         if (copy_skb)
1051                                 skb_set_owner_r(copy_skb, sk);
1052                 }
1053                 snaplen = po->rx_ring.frame_size - macoff;
1054                 if ((int)snaplen < 0)
1055                         snaplen = 0;
1056         }
1057
1058         spin_lock(&sk->sk_receive_queue.lock);
1059         h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
1060         if (!h.raw)
1061                 goto ring_is_full;
1062         packet_increment_head(&po->rx_ring);
1063         po->stats.tp_packets++;
1064         if (copy_skb) {
1065                 status |= TP_STATUS_COPY;
1066                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1067         }
1068         if (!po->stats.tp_drops)
1069                 status &= ~TP_STATUS_LOSING;
1070         spin_unlock(&sk->sk_receive_queue.lock);
1071
1072         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1073
1074         switch (po->tp_version) {
1075         case TPACKET_V1:
1076                 h.h1->tp_len = skb->len;
1077                 h.h1->tp_snaplen = snaplen;
1078                 h.h1->tp_mac = macoff;
1079                 h.h1->tp_net = netoff;
1080                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1081                                 && shhwtstamps->syststamp.tv64)
1082                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1083                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1084                                 && shhwtstamps->hwtstamp.tv64)
1085                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1086                 else if (skb->tstamp.tv64)
1087                         tv = ktime_to_timeval(skb->tstamp);
1088                 else
1089                         do_gettimeofday(&tv);
1090                 h.h1->tp_sec = tv.tv_sec;
1091                 h.h1->tp_usec = tv.tv_usec;
1092                 hdrlen = sizeof(*h.h1);
1093                 break;
1094         case TPACKET_V2:
1095                 h.h2->tp_len = skb->len;
1096                 h.h2->tp_snaplen = snaplen;
1097                 h.h2->tp_mac = macoff;
1098                 h.h2->tp_net = netoff;
1099                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1100                                 && shhwtstamps->syststamp.tv64)
1101                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1102                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1103                                 && shhwtstamps->hwtstamp.tv64)
1104                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1105                 else if (skb->tstamp.tv64)
1106                         ts = ktime_to_timespec(skb->tstamp);
1107                 else
1108                         getnstimeofday(&ts);
1109                 h.h2->tp_sec = ts.tv_sec;
1110                 h.h2->tp_nsec = ts.tv_nsec;
1111                 if (vlan_tx_tag_present(skb)) {
1112                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1113                         status |= TP_STATUS_VLAN_VALID;
1114                 } else {
1115                         h.h2->tp_vlan_tci = 0;
1116                 }
1117                 h.h2->tp_padding = 0;
1118                 hdrlen = sizeof(*h.h2);
1119                 break;
1120         default:
1121                 BUG();
1122         }
1123
1124         sll = h.raw + TPACKET_ALIGN(hdrlen);
1125         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1126         sll->sll_family = AF_PACKET;
1127         sll->sll_hatype = dev->type;
1128         sll->sll_protocol = skb->protocol;
1129         sll->sll_pkttype = skb->pkt_type;
1130         if (unlikely(po->origdev))
1131                 sll->sll_ifindex = orig_dev->ifindex;
1132         else
1133                 sll->sll_ifindex = dev->ifindex;
1134
1135         smp_mb();
1136 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1137         {
1138                 u8 *start, *end;
1139
1140                 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw + macoff + snaplen);
1141                 for (start = h.raw; start < end; start += PAGE_SIZE)
1142                         flush_dcache_page(pgv_to_page(start));
1143                 smp_wmb();
1144         }
1145 #endif
1146         __packet_set_status(po, h.raw, status);
1147
1148         sk->sk_data_ready(sk, 0);
1149
1150 drop_n_restore:
1151         if (skb_head != skb->data && skb_shared(skb)) {
1152                 skb->data = skb_head;
1153                 skb->len = skb_len;
1154         }
1155 drop:
1156         kfree_skb(skb);
1157         return 0;
1158
1159 ring_is_full:
1160         po->stats.tp_drops++;
1161         spin_unlock(&sk->sk_receive_queue.lock);
1162
1163         sk->sk_data_ready(sk, 0);
1164         kfree_skb(copy_skb);
1165         goto drop_n_restore;
1166 }
1167
1168 static void tpacket_destruct_skb(struct sk_buff *skb)
1169 {
1170         struct packet_sock *po = pkt_sk(skb->sk);
1171         void *ph;
1172
1173         BUG_ON(skb == NULL);
1174
1175         if (likely(po->tx_ring.pg_vec)) {
1176                 ph = skb_shinfo(skb)->destructor_arg;
1177                 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
1178                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1179                 atomic_dec(&po->tx_ring.pending);
1180                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1181         }
1182
1183         sock_wfree(skb);
1184 }
1185
1186 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1187                 void *frame, struct net_device *dev, int size_max,
1188                 __be16 proto, unsigned char *addr)
1189 {
1190         union {
1191                 struct tpacket_hdr *h1;
1192                 struct tpacket2_hdr *h2;
1193                 void *raw;
1194         } ph;
1195         int to_write, offset, len, tp_len, nr_frags, len_max;
1196         struct socket *sock = po->sk.sk_socket;
1197         struct page *page;
1198         void *data;
1199         int err;
1200
1201         ph.raw = frame;
1202
1203         skb->protocol = proto;
1204         skb->dev = dev;
1205         skb->priority = po->sk.sk_priority;
1206         skb->mark = po->sk.sk_mark;
1207         skb_shinfo(skb)->destructor_arg = ph.raw;
1208
1209         switch (po->tp_version) {
1210         case TPACKET_V2:
1211                 tp_len = ph.h2->tp_len;
1212                 break;
1213         default:
1214                 tp_len = ph.h1->tp_len;
1215                 break;
1216         }
1217         if (unlikely(tp_len > size_max)) {
1218                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1219                 return -EMSGSIZE;
1220         }
1221
1222         skb_reserve(skb, LL_RESERVED_SPACE(dev));
1223         skb_reset_network_header(skb);
1224
1225         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1226         to_write = tp_len;
1227
1228         if (sock->type == SOCK_DGRAM) {
1229                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1230                                 NULL, tp_len);
1231                 if (unlikely(err < 0))
1232                         return -EINVAL;
1233         } else if (dev->hard_header_len) {
1234                 /* net device doesn't like empty head */
1235                 if (unlikely(tp_len <= dev->hard_header_len)) {
1236                         pr_err("packet size is too short (%d < %d)\n",
1237                                tp_len, dev->hard_header_len);
1238                         return -EINVAL;
1239                 }
1240
1241                 skb_push(skb, dev->hard_header_len);
1242                 err = skb_store_bits(skb, 0, data,
1243                                 dev->hard_header_len);
1244                 if (unlikely(err))
1245                         return err;
1246
1247                 data += dev->hard_header_len;
1248                 to_write -= dev->hard_header_len;
1249         }
1250
1251         err = -EFAULT;
1252         offset = offset_in_page(data);
1253         len_max = PAGE_SIZE - offset;
1254         len = ((to_write > len_max) ? len_max : to_write);
1255
1256         skb->data_len = to_write;
1257         skb->len += to_write;
1258         skb->truesize += to_write;
1259         atomic_add(to_write, &po->sk.sk_wmem_alloc);
1260
1261         while (likely(to_write)) {
1262                 nr_frags = skb_shinfo(skb)->nr_frags;
1263
1264                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
1265                         pr_err("Packet exceed the number of skb frags(%lu)\n",
1266                                MAX_SKB_FRAGS);
1267                         return -EFAULT;
1268                 }
1269
1270                 page = pgv_to_page(data);
1271                 data += len;
1272                 flush_dcache_page(page);
1273                 get_page(page);
1274                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
1275                 to_write -= len;
1276                 offset = 0;
1277                 len_max = PAGE_SIZE;
1278                 len = ((to_write > len_max) ? len_max : to_write);
1279         }
1280
1281         return tp_len;
1282 }
1283
1284 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
1285 {
1286         struct sk_buff *skb;
1287         struct net_device *dev;
1288         __be16 proto;
1289         bool need_rls_dev = false;
1290         int err, reserve = 0;
1291         void *ph;
1292         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1293         int tp_len, size_max;
1294         unsigned char *addr;
1295         int len_sum = 0;
1296         int status = 0;
1297
1298         mutex_lock(&po->pg_vec_lock);
1299
1300         err = -EBUSY;
1301         if (saddr == NULL) {
1302                 dev = po->prot_hook.dev;
1303                 proto   = po->num;
1304                 addr    = NULL;
1305         } else {
1306                 err = -EINVAL;
1307                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1308                         goto out;
1309                 if (msg->msg_namelen < (saddr->sll_halen
1310                                         + offsetof(struct sockaddr_ll,
1311                                                 sll_addr)))
1312                         goto out;
1313                 proto   = saddr->sll_protocol;
1314                 addr    = saddr->sll_addr;
1315                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
1316                 need_rls_dev = true;
1317         }
1318
1319         err = -ENXIO;
1320         if (unlikely(dev == NULL))
1321                 goto out;
1322
1323         reserve = dev->hard_header_len;
1324
1325         err = -ENETDOWN;
1326         if (unlikely(!(dev->flags & IFF_UP)))
1327                 goto out_put;
1328
1329         size_max = po->tx_ring.frame_size
1330                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
1331
1332         if (size_max > dev->mtu + reserve)
1333                 size_max = dev->mtu + reserve;
1334
1335         do {
1336                 ph = packet_current_frame(po, &po->tx_ring,
1337                                 TP_STATUS_SEND_REQUEST);
1338
1339                 if (unlikely(ph == NULL)) {
1340                         schedule();
1341                         continue;
1342                 }
1343
1344                 status = TP_STATUS_SEND_REQUEST;
1345                 skb = sock_alloc_send_skb(&po->sk,
1346                                 LL_ALLOCATED_SPACE(dev)
1347                                 + sizeof(struct sockaddr_ll),
1348                                 0, &err);
1349
1350                 if (unlikely(skb == NULL))
1351                         goto out_status;
1352
1353                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
1354                                 addr);
1355
1356                 if (unlikely(tp_len < 0)) {
1357                         if (po->tp_loss) {
1358                                 __packet_set_status(po, ph,
1359                                                 TP_STATUS_AVAILABLE);
1360                                 packet_increment_head(&po->tx_ring);
1361                                 kfree_skb(skb);
1362                                 continue;
1363                         } else {
1364                                 status = TP_STATUS_WRONG_FORMAT;
1365                                 err = tp_len;
1366                                 goto out_status;
1367                         }
1368                 }
1369
1370                 skb->destructor = tpacket_destruct_skb;
1371                 __packet_set_status(po, ph, TP_STATUS_SENDING);
1372                 atomic_inc(&po->tx_ring.pending);
1373
1374                 status = TP_STATUS_SEND_REQUEST;
1375                 err = dev_queue_xmit(skb);
1376                 if (unlikely(err > 0)) {
1377                         err = net_xmit_errno(err);
1378                         if (err && __packet_get_status(po, ph) ==
1379                                    TP_STATUS_AVAILABLE) {
1380                                 /* skb was destructed already */
1381                                 skb = NULL;
1382                                 goto out_status;
1383                         }
1384                         /*
1385                          * skb was dropped but not destructed yet;
1386                          * let's treat it like congestion or err < 0
1387                          */
1388                         err = 0;
1389                 }
1390                 packet_increment_head(&po->tx_ring);
1391                 len_sum += tp_len;
1392         } while (likely((ph != NULL) ||
1393                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
1394                          (atomic_read(&po->tx_ring.pending))))
1395                 );
1396
1397         err = len_sum;
1398         goto out_put;
1399
1400 out_status:
1401         __packet_set_status(po, ph, status);
1402         kfree_skb(skb);
1403 out_put:
1404         if (need_rls_dev)
1405                 dev_put(dev);
1406 out:
1407         mutex_unlock(&po->pg_vec_lock);
1408         return err;
1409 }
1410
1411 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
1412                                                size_t reserve, size_t len,
1413                                                size_t linear, int noblock,
1414                                                int *err)
1415 {
1416         struct sk_buff *skb;
1417
1418         /* Under a page?  Don't bother with paged skb. */
1419         if (prepad + len < PAGE_SIZE || !linear)
1420                 linear = len;
1421
1422         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1423                                    err);
1424         if (!skb)
1425                 return NULL;
1426
1427         skb_reserve(skb, reserve);
1428         skb_put(skb, linear);
1429         skb->data_len = len - linear;
1430         skb->len += len - linear;
1431
1432         return skb;
1433 }
1434
1435 static int packet_snd(struct socket *sock,
1436                           struct msghdr *msg, size_t len)
1437 {
1438         struct sock *sk = sock->sk;
1439         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1440         struct sk_buff *skb;
1441         struct net_device *dev;
1442         __be16 proto;
1443         bool need_rls_dev = false;
1444         unsigned char *addr;
1445         int err, reserve = 0;
1446         struct virtio_net_hdr vnet_hdr = { 0 };
1447         int offset = 0;
1448         int vnet_hdr_len;
1449         struct packet_sock *po = pkt_sk(sk);
1450         unsigned short gso_type = 0;
1451
1452         /*
1453          *      Get and verify the address.
1454          */
1455
1456         if (saddr == NULL) {
1457                 dev = po->prot_hook.dev;
1458                 proto   = po->num;
1459                 addr    = NULL;
1460         } else {
1461                 err = -EINVAL;
1462                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1463                         goto out;
1464                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1465                         goto out;
1466                 proto   = saddr->sll_protocol;
1467                 addr    = saddr->sll_addr;
1468                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
1469                 need_rls_dev = true;
1470         }
1471
1472         err = -ENXIO;
1473         if (dev == NULL)
1474                 goto out_unlock;
1475         if (sock->type == SOCK_RAW)
1476                 reserve = dev->hard_header_len;
1477
1478         err = -ENETDOWN;
1479         if (!(dev->flags & IFF_UP))
1480                 goto out_unlock;
1481
1482         if (po->has_vnet_hdr) {
1483                 vnet_hdr_len = sizeof(vnet_hdr);
1484
1485                 err = -EINVAL;
1486                 if (len < vnet_hdr_len)
1487                         goto out_unlock;
1488
1489                 len -= vnet_hdr_len;
1490
1491                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
1492                                        vnet_hdr_len);
1493                 if (err < 0)
1494                         goto out_unlock;
1495
1496                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1497                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
1498                       vnet_hdr.hdr_len))
1499                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
1500                                                  vnet_hdr.csum_offset + 2;
1501
1502                 err = -EINVAL;
1503                 if (vnet_hdr.hdr_len > len)
1504                         goto out_unlock;
1505
1506                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1507                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1508                         case VIRTIO_NET_HDR_GSO_TCPV4:
1509                                 gso_type = SKB_GSO_TCPV4;
1510                                 break;
1511                         case VIRTIO_NET_HDR_GSO_TCPV6:
1512                                 gso_type = SKB_GSO_TCPV6;
1513                                 break;
1514                         case VIRTIO_NET_HDR_GSO_UDP:
1515                                 gso_type = SKB_GSO_UDP;
1516                                 break;
1517                         default:
1518                                 goto out_unlock;
1519                         }
1520
1521                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1522                                 gso_type |= SKB_GSO_TCP_ECN;
1523
1524                         if (vnet_hdr.gso_size == 0)
1525                                 goto out_unlock;
1526
1527                 }
1528         }
1529
1530         err = -EMSGSIZE;
1531         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
1532                 goto out_unlock;
1533
1534         err = -ENOBUFS;
1535         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
1536                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
1537                                msg->msg_flags & MSG_DONTWAIT, &err);
1538         if (skb == NULL)
1539                 goto out_unlock;
1540
1541         skb_set_network_header(skb, reserve);
1542
1543         err = -EINVAL;
1544         if (sock->type == SOCK_DGRAM &&
1545             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
1546                 goto out_free;
1547
1548         /* Returns -EFAULT on error */
1549         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1550         if (err)
1551                 goto out_free;
1552         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1553         if (err < 0)
1554                 goto out_free;
1555
1556         if (!gso_type && (len > dev->mtu + reserve)) {
1557                 /* Earlier code assumed this would be a VLAN pkt,
1558                  * double-check this now that we have the actual
1559                  * packet in hand.
1560                  */
1561                 struct ethhdr *ehdr;
1562                 skb_reset_mac_header(skb);
1563                 ehdr = eth_hdr(skb);
1564                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1565                         err = -EMSGSIZE;
1566                         goto out_free;
1567                 }
1568         }
1569
1570         skb->protocol = proto;
1571         skb->dev = dev;
1572         skb->priority = sk->sk_priority;
1573         skb->mark = sk->sk_mark;
1574
1575         if (po->has_vnet_hdr) {
1576                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1577                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
1578                                                   vnet_hdr.csum_offset)) {
1579                                 err = -EINVAL;
1580                                 goto out_free;
1581                         }
1582                 }
1583
1584                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
1585                 skb_shinfo(skb)->gso_type = gso_type;
1586
1587                 /* Header must be checked, and gso_segs computed. */
1588                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1589                 skb_shinfo(skb)->gso_segs = 0;
1590
1591                 len += vnet_hdr_len;
1592         }
1593
1594         /*
1595          *      Now send it
1596          */
1597
1598         err = dev_queue_xmit(skb);
1599         if (err > 0 && (err = net_xmit_errno(err)) != 0)
1600                 goto out_unlock;
1601
1602         if (need_rls_dev)
1603                 dev_put(dev);
1604
1605         return len;
1606
1607 out_free:
1608         kfree_skb(skb);
1609 out_unlock:
1610         if (dev && need_rls_dev)
1611                 dev_put(dev);
1612 out:
1613         return err;
1614 }
1615
1616 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1617                 struct msghdr *msg, size_t len)
1618 {
1619         struct sock *sk = sock->sk;
1620         struct packet_sock *po = pkt_sk(sk);
1621         if (po->tx_ring.pg_vec)
1622                 return tpacket_snd(po, msg);
1623         else
1624                 return packet_snd(sock, msg, len);
1625 }
1626
1627 /*
1628  *      Close a PACKET socket. This is fairly simple. We immediately go
1629  *      to 'closed' state and remove our protocol entry in the device list.
1630  */
1631
1632 static int packet_release(struct socket *sock)
1633 {
1634         struct sock *sk = sock->sk;
1635         struct packet_sock *po;
1636         struct net *net;
1637         struct tpacket_req req;
1638
1639         if (!sk)
1640                 return 0;
1641
1642         net = sock_net(sk);
1643         po = pkt_sk(sk);
1644
1645         spin_lock_bh(&net->packet.sklist_lock);
1646         sk_del_node_init_rcu(sk);
1647         sock_prot_inuse_add(net, sk->sk_prot, -1);
1648         spin_unlock_bh(&net->packet.sklist_lock);
1649
1650         spin_lock(&po->bind_lock);
1651         unregister_prot_hook(sk, false);
1652         if (po->prot_hook.dev) {
1653                 dev_put(po->prot_hook.dev);
1654                 po->prot_hook.dev = NULL;
1655         }
1656         spin_unlock(&po->bind_lock);
1657
1658         packet_flush_mclist(sk);
1659
1660         memset(&req, 0, sizeof(req));
1661
1662         if (po->rx_ring.pg_vec)
1663                 packet_set_ring(sk, &req, 1, 0);
1664
1665         if (po->tx_ring.pg_vec)
1666                 packet_set_ring(sk, &req, 1, 1);
1667
1668         fanout_release(sk);
1669
1670         synchronize_net();
1671         /*
1672          *      Now the socket is dead. No more input will appear.
1673          */
1674         sock_orphan(sk);
1675         sock->sk = NULL;
1676
1677         /* Purge queues */
1678
1679         skb_queue_purge(&sk->sk_receive_queue);
1680         sk_refcnt_debug_release(sk);
1681
1682         sock_put(sk);
1683         return 0;
1684 }
1685
1686 /*
1687  *      Attach a packet hook.
1688  */
1689
1690 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1691 {
1692         struct packet_sock *po = pkt_sk(sk);
1693
1694         if (po->fanout)
1695                 return -EINVAL;
1696
1697         lock_sock(sk);
1698
1699         spin_lock(&po->bind_lock);
1700         unregister_prot_hook(sk, true);
1701         po->num = protocol;
1702         po->prot_hook.type = protocol;
1703         if (po->prot_hook.dev)
1704                 dev_put(po->prot_hook.dev);
1705         po->prot_hook.dev = dev;
1706
1707         po->ifindex = dev ? dev->ifindex : 0;
1708
1709         if (protocol == 0)
1710                 goto out_unlock;
1711
1712         if (!dev || (dev->flags & IFF_UP)) {
1713                 register_prot_hook(sk);
1714         } else {
1715                 sk->sk_err = ENETDOWN;
1716                 if (!sock_flag(sk, SOCK_DEAD))
1717                         sk->sk_error_report(sk);
1718         }
1719
1720 out_unlock:
1721         spin_unlock(&po->bind_lock);
1722         release_sock(sk);
1723         return 0;
1724 }
1725
1726 /*
1727  *      Bind a packet socket to a device
1728  */
1729
1730 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1731                             int addr_len)
1732 {
1733         struct sock *sk = sock->sk;
1734         char name[15];
1735         struct net_device *dev;
1736         int err = -ENODEV;
1737
1738         /*
1739          *      Check legality
1740          */
1741
1742         if (addr_len != sizeof(struct sockaddr))
1743                 return -EINVAL;
1744         strlcpy(name, uaddr->sa_data, sizeof(name));
1745
1746         dev = dev_get_by_name(sock_net(sk), name);
1747         if (dev)
1748                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1749         return err;
1750 }
1751
1752 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1753 {
1754         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1755         struct sock *sk = sock->sk;
1756         struct net_device *dev = NULL;
1757         int err;
1758
1759
1760         /*
1761          *      Check legality
1762          */
1763
1764         if (addr_len < sizeof(struct sockaddr_ll))
1765                 return -EINVAL;
1766         if (sll->sll_family != AF_PACKET)
1767                 return -EINVAL;
1768
1769         if (sll->sll_ifindex) {
1770                 err = -ENODEV;
1771                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1772                 if (dev == NULL)
1773                         goto out;
1774         }
1775         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1776
1777 out:
1778         return err;
1779 }
1780
1781 static struct proto packet_proto = {
1782         .name     = "PACKET",
1783         .owner    = THIS_MODULE,
1784         .obj_size = sizeof(struct packet_sock),
1785 };
1786
1787 /*
1788  *      Create a packet of type SOCK_PACKET.
1789  */
1790
1791 static int packet_create(struct net *net, struct socket *sock, int protocol,
1792                          int kern)
1793 {
1794         struct sock *sk;
1795         struct packet_sock *po;
1796         __be16 proto = (__force __be16)protocol; /* weird, but documented */
1797         int err;
1798
1799         if (!capable(CAP_NET_RAW))
1800                 return -EPERM;
1801         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1802             sock->type != SOCK_PACKET)
1803                 return -ESOCKTNOSUPPORT;
1804
1805         sock->state = SS_UNCONNECTED;
1806
1807         err = -ENOBUFS;
1808         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1809         if (sk == NULL)
1810                 goto out;
1811
1812         sock->ops = &packet_ops;
1813         if (sock->type == SOCK_PACKET)
1814                 sock->ops = &packet_ops_spkt;
1815
1816         sock_init_data(sock, sk);
1817
1818         po = pkt_sk(sk);
1819         sk->sk_family = PF_PACKET;
1820         po->num = proto;
1821
1822         sk->sk_destruct = packet_sock_destruct;
1823         sk_refcnt_debug_inc(sk);
1824
1825         /*
1826          *      Attach a protocol block
1827          */
1828
1829         spin_lock_init(&po->bind_lock);
1830         mutex_init(&po->pg_vec_lock);
1831         po->prot_hook.func = packet_rcv;
1832
1833         if (sock->type == SOCK_PACKET)
1834                 po->prot_hook.func = packet_rcv_spkt;
1835
1836         po->prot_hook.af_packet_priv = sk;
1837
1838         if (proto) {
1839                 po->prot_hook.type = proto;
1840                 register_prot_hook(sk);
1841         }
1842
1843         spin_lock_bh(&net->packet.sklist_lock);
1844         sk_add_node_rcu(sk, &net->packet.sklist);
1845         sock_prot_inuse_add(net, &packet_proto, 1);
1846         spin_unlock_bh(&net->packet.sklist_lock);
1847
1848         return 0;
1849 out:
1850         return err;
1851 }
1852
1853 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
1854 {
1855         struct sock_exterr_skb *serr;
1856         struct sk_buff *skb, *skb2;
1857         int copied, err;
1858
1859         err = -EAGAIN;
1860         skb = skb_dequeue(&sk->sk_error_queue);
1861         if (skb == NULL)
1862                 goto out;
1863
1864         copied = skb->len;
1865         if (copied > len) {
1866                 msg->msg_flags |= MSG_TRUNC;
1867                 copied = len;
1868         }
1869         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1870         if (err)
1871                 goto out_free_skb;
1872
1873         sock_recv_timestamp(msg, sk, skb);
1874
1875         serr = SKB_EXT_ERR(skb);
1876         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
1877                  sizeof(serr->ee), &serr->ee);
1878
1879         msg->msg_flags |= MSG_ERRQUEUE;
1880         err = copied;
1881
1882         /* Reset and regenerate socket error */
1883         spin_lock_bh(&sk->sk_error_queue.lock);
1884         sk->sk_err = 0;
1885         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
1886                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
1887                 spin_unlock_bh(&sk->sk_error_queue.lock);
1888                 sk->sk_error_report(sk);
1889         } else
1890                 spin_unlock_bh(&sk->sk_error_queue.lock);
1891
1892 out_free_skb:
1893         kfree_skb(skb);
1894 out:
1895         return err;
1896 }
1897
1898 /*
1899  *      Pull a packet from our receive queue and hand it to the user.
1900  *      If necessary we block.
1901  */
1902
1903 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1904                           struct msghdr *msg, size_t len, int flags)
1905 {
1906         struct sock *sk = sock->sk;
1907         struct sk_buff *skb;
1908         int copied, err;
1909         struct sockaddr_ll *sll;
1910         int vnet_hdr_len = 0;
1911
1912         err = -EINVAL;
1913         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
1914                 goto out;
1915
1916 #if 0
1917         /* What error should we return now? EUNATTACH? */
1918         if (pkt_sk(sk)->ifindex < 0)
1919                 return -ENODEV;
1920 #endif
1921
1922         if (flags & MSG_ERRQUEUE) {
1923                 err = packet_recv_error(sk, msg, len);
1924                 goto out;
1925         }
1926
1927         /*
1928          *      Call the generic datagram receiver. This handles all sorts
1929          *      of horrible races and re-entrancy so we can forget about it
1930          *      in the protocol layers.
1931          *
1932          *      Now it will return ENETDOWN, if device have just gone down,
1933          *      but then it will block.
1934          */
1935
1936         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1937
1938         /*
1939          *      An error occurred so return it. Because skb_recv_datagram()
1940          *      handles the blocking we don't see and worry about blocking
1941          *      retries.
1942          */
1943
1944         if (skb == NULL)
1945                 goto out;
1946
1947         if (pkt_sk(sk)->has_vnet_hdr) {
1948                 struct virtio_net_hdr vnet_hdr = { 0 };
1949
1950                 err = -EINVAL;
1951                 vnet_hdr_len = sizeof(vnet_hdr);
1952                 if (len < vnet_hdr_len)
1953                         goto out_free;
1954
1955                 len -= vnet_hdr_len;
1956
1957                 if (skb_is_gso(skb)) {
1958                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1959
1960                         /* This is a hint as to how much should be linear. */
1961                         vnet_hdr.hdr_len = skb_headlen(skb);
1962                         vnet_hdr.gso_size = sinfo->gso_size;
1963                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1964                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1965                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1966                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1967                         else if (sinfo->gso_type & SKB_GSO_UDP)
1968                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1969                         else if (sinfo->gso_type & SKB_GSO_FCOE)
1970                                 goto out_free;
1971                         else
1972                                 BUG();
1973                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1974                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1975                 } else
1976                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1977
1978                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1979                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1980                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
1981                         vnet_hdr.csum_offset = skb->csum_offset;
1982                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
1983                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
1984                 } /* else everything is zero */
1985
1986                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
1987                                      vnet_hdr_len);
1988                 if (err < 0)
1989                         goto out_free;
1990         }
1991
1992         /*
1993          *      If the address length field is there to be filled in, we fill
1994          *      it in now.
1995          */
1996
1997         sll = &PACKET_SKB_CB(skb)->sa.ll;
1998         if (sock->type == SOCK_PACKET)
1999                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2000         else
2001                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2002
2003         /*
2004          *      You lose any data beyond the buffer you gave. If it worries a
2005          *      user program they can ask the device for its MTU anyway.
2006          */
2007
2008         copied = skb->len;
2009         if (copied > len) {
2010                 copied = len;
2011                 msg->msg_flags |= MSG_TRUNC;
2012         }
2013
2014         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2015         if (err)
2016                 goto out_free;
2017
2018         sock_recv_ts_and_drops(msg, sk, skb);
2019
2020         if (msg->msg_name)
2021                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2022                        msg->msg_namelen);
2023
2024         if (pkt_sk(sk)->auxdata) {
2025                 struct tpacket_auxdata aux;
2026
2027                 aux.tp_status = TP_STATUS_USER;
2028                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2029                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2030                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2031                 aux.tp_snaplen = skb->len;
2032                 aux.tp_mac = 0;
2033                 aux.tp_net = skb_network_offset(skb);
2034                 if (vlan_tx_tag_present(skb)) {
2035                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2036                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2037                 } else {
2038                         aux.tp_vlan_tci = 0;
2039                 }
2040                 aux.tp_padding = 0;
2041                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2042         }
2043
2044         /*
2045          *      Free or return the buffer as appropriate. Again this
2046          *      hides all the races and re-entrancy issues from us.
2047          */
2048         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2049
2050 out_free:
2051         skb_free_datagram(sk, skb);
2052 out:
2053         return err;
2054 }
2055
2056 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2057                                int *uaddr_len, int peer)
2058 {
2059         struct net_device *dev;
2060         struct sock *sk = sock->sk;
2061
2062         if (peer)
2063                 return -EOPNOTSUPP;
2064
2065         uaddr->sa_family = AF_PACKET;
2066         rcu_read_lock();
2067         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2068         if (dev)
2069                 strncpy(uaddr->sa_data, dev->name, 14);
2070         else
2071                 memset(uaddr->sa_data, 0, 14);
2072         rcu_read_unlock();
2073         *uaddr_len = sizeof(*uaddr);
2074
2075         return 0;
2076 }
2077
2078 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2079                           int *uaddr_len, int peer)
2080 {
2081         struct net_device *dev;
2082         struct sock *sk = sock->sk;
2083         struct packet_sock *po = pkt_sk(sk);
2084         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2085
2086         if (peer)
2087                 return -EOPNOTSUPP;
2088
2089         sll->sll_family = AF_PACKET;
2090         sll->sll_ifindex = po->ifindex;
2091         sll->sll_protocol = po->num;
2092         sll->sll_pkttype = 0;
2093         rcu_read_lock();
2094         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2095         if (dev) {
2096                 sll->sll_hatype = dev->type;
2097                 sll->sll_halen = dev->addr_len;
2098                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2099         } else {
2100                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2101                 sll->sll_halen = 0;
2102         }
2103         rcu_read_unlock();
2104         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2105
2106         return 0;
2107 }
2108
2109 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2110                          int what)
2111 {
2112         switch (i->type) {
2113         case PACKET_MR_MULTICAST:
2114                 if (i->alen != dev->addr_len)
2115                         return -EINVAL;
2116                 if (what > 0)
2117                         return dev_mc_add(dev, i->addr);
2118                 else
2119                         return dev_mc_del(dev, i->addr);
2120                 break;
2121         case PACKET_MR_PROMISC:
2122                 return dev_set_promiscuity(dev, what);
2123                 break;
2124         case PACKET_MR_ALLMULTI:
2125                 return dev_set_allmulti(dev, what);
2126                 break;
2127         case PACKET_MR_UNICAST:
2128                 if (i->alen != dev->addr_len)
2129                         return -EINVAL;
2130                 if (what > 0)
2131                         return dev_uc_add(dev, i->addr);
2132                 else
2133                         return dev_uc_del(dev, i->addr);
2134                 break;
2135         default:
2136                 break;
2137         }
2138         return 0;
2139 }
2140
2141 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2142 {
2143         for ( ; i; i = i->next) {
2144                 if (i->ifindex == dev->ifindex)
2145                         packet_dev_mc(dev, i, what);
2146         }
2147 }
2148
2149 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2150 {
2151         struct packet_sock *po = pkt_sk(sk);
2152         struct packet_mclist *ml, *i;
2153         struct net_device *dev;
2154         int err;
2155
2156         rtnl_lock();
2157
2158         err = -ENODEV;
2159         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2160         if (!dev)
2161                 goto done;
2162
2163         err = -EINVAL;
2164         if (mreq->mr_alen > dev->addr_len)
2165                 goto done;
2166
2167         err = -ENOBUFS;
2168         i = kmalloc(sizeof(*i), GFP_KERNEL);
2169         if (i == NULL)
2170                 goto done;
2171
2172         err = 0;
2173         for (ml = po->mclist; ml; ml = ml->next) {
2174                 if (ml->ifindex == mreq->mr_ifindex &&
2175                     ml->type == mreq->mr_type &&
2176                     ml->alen == mreq->mr_alen &&
2177                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2178                         ml->count++;
2179                         /* Free the new element ... */
2180                         kfree(i);
2181                         goto done;
2182                 }
2183         }
2184
2185         i->type = mreq->mr_type;
2186         i->ifindex = mreq->mr_ifindex;
2187         i->alen = mreq->mr_alen;
2188         memcpy(i->addr, mreq->mr_address, i->alen);
2189         i->count = 1;
2190         i->next = po->mclist;
2191         po->mclist = i;
2192         err = packet_dev_mc(dev, i, 1);
2193         if (err) {
2194                 po->mclist = i->next;
2195                 kfree(i);
2196         }
2197
2198 done:
2199         rtnl_unlock();
2200         return err;
2201 }
2202
2203 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2204 {
2205         struct packet_mclist *ml, **mlp;
2206
2207         rtnl_lock();
2208
2209         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2210                 if (ml->ifindex == mreq->mr_ifindex &&
2211                     ml->type == mreq->mr_type &&
2212                     ml->alen == mreq->mr_alen &&
2213                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2214                         if (--ml->count == 0) {
2215                                 struct net_device *dev;
2216                                 *mlp = ml->next;
2217                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2218                                 if (dev)
2219                                         packet_dev_mc(dev, ml, -1);
2220                                 kfree(ml);
2221                         }
2222                         rtnl_unlock();
2223                         return 0;
2224                 }
2225         }
2226         rtnl_unlock();
2227         return -EADDRNOTAVAIL;
2228 }
2229
2230 static void packet_flush_mclist(struct sock *sk)
2231 {
2232         struct packet_sock *po = pkt_sk(sk);
2233         struct packet_mclist *ml;
2234
2235         if (!po->mclist)
2236                 return;
2237
2238         rtnl_lock();
2239         while ((ml = po->mclist) != NULL) {
2240                 struct net_device *dev;
2241
2242                 po->mclist = ml->next;
2243                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2244                 if (dev != NULL)
2245                         packet_dev_mc(dev, ml, -1);
2246                 kfree(ml);
2247         }
2248         rtnl_unlock();
2249 }
2250
2251 static int
2252 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
2253 {
2254         struct sock *sk = sock->sk;
2255         struct packet_sock *po = pkt_sk(sk);
2256         int ret;
2257
2258         if (level != SOL_PACKET)
2259                 return -ENOPROTOOPT;
2260
2261         switch (optname) {
2262         case PACKET_ADD_MEMBERSHIP:
2263         case PACKET_DROP_MEMBERSHIP:
2264         {
2265                 struct packet_mreq_max mreq;
2266                 int len = optlen;
2267                 memset(&mreq, 0, sizeof(mreq));
2268                 if (len < sizeof(struct packet_mreq))
2269                         return -EINVAL;
2270                 if (len > sizeof(mreq))
2271                         len = sizeof(mreq);
2272                 if (copy_from_user(&mreq, optval, len))
2273                         return -EFAULT;
2274                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
2275                         return -EINVAL;
2276                 if (optname == PACKET_ADD_MEMBERSHIP)
2277                         ret = packet_mc_add(sk, &mreq);
2278                 else
2279                         ret = packet_mc_drop(sk, &mreq);
2280                 return ret;
2281         }
2282
2283         case PACKET_RX_RING:
2284         case PACKET_TX_RING:
2285         {
2286                 struct tpacket_req req;
2287
2288                 if (optlen < sizeof(req))
2289                         return -EINVAL;
2290                 if (pkt_sk(sk)->has_vnet_hdr)
2291                         return -EINVAL;
2292                 if (copy_from_user(&req, optval, sizeof(req)))
2293                         return -EFAULT;
2294                 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
2295         }
2296         case PACKET_COPY_THRESH:
2297         {
2298                 int val;
2299
2300                 if (optlen != sizeof(val))
2301                         return -EINVAL;
2302                 if (copy_from_user(&val, optval, sizeof(val)))
2303                         return -EFAULT;
2304
2305                 pkt_sk(sk)->copy_thresh = val;
2306                 return 0;
2307         }
2308         case PACKET_VERSION:
2309         {
2310                 int val;
2311
2312                 if (optlen != sizeof(val))
2313                         return -EINVAL;
2314                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2315                         return -EBUSY;
2316                 if (copy_from_user(&val, optval, sizeof(val)))
2317                         return -EFAULT;
2318                 switch (val) {
2319                 case TPACKET_V1:
2320                 case TPACKET_V2:
2321                         po->tp_version = val;
2322                         return 0;
2323                 default:
2324                         return -EINVAL;
2325                 }
2326         }
2327         case PACKET_RESERVE:
2328         {
2329                 unsigned int val;
2330
2331                 if (optlen != sizeof(val))
2332                         return -EINVAL;
2333                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2334                         return -EBUSY;
2335                 if (copy_from_user(&val, optval, sizeof(val)))
2336                         return -EFAULT;
2337                 po->tp_reserve = val;
2338                 return 0;
2339         }
2340         case PACKET_LOSS:
2341         {
2342                 unsigned int val;
2343
2344                 if (optlen != sizeof(val))
2345                         return -EINVAL;
2346                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2347                         return -EBUSY;
2348                 if (copy_from_user(&val, optval, sizeof(val)))
2349                         return -EFAULT;
2350                 po->tp_loss = !!val;
2351                 return 0;
2352         }
2353         case PACKET_AUXDATA:
2354         {
2355                 int val;
2356
2357                 if (optlen < sizeof(val))
2358                         return -EINVAL;
2359                 if (copy_from_user(&val, optval, sizeof(val)))
2360                         return -EFAULT;
2361
2362                 po->auxdata = !!val;
2363                 return 0;
2364         }
2365         case PACKET_ORIGDEV:
2366         {
2367                 int val;
2368
2369                 if (optlen < sizeof(val))
2370                         return -EINVAL;
2371                 if (copy_from_user(&val, optval, sizeof(val)))
2372                         return -EFAULT;
2373
2374                 po->origdev = !!val;
2375                 return 0;
2376         }
2377         case PACKET_VNET_HDR:
2378         {
2379                 int val;
2380
2381                 if (sock->type != SOCK_RAW)
2382                         return -EINVAL;
2383                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2384                         return -EBUSY;
2385                 if (optlen < sizeof(val))
2386                         return -EINVAL;
2387                 if (copy_from_user(&val, optval, sizeof(val)))
2388                         return -EFAULT;
2389
2390                 po->has_vnet_hdr = !!val;
2391                 return 0;
2392         }
2393         case PACKET_TIMESTAMP:
2394         {
2395                 int val;
2396
2397                 if (optlen != sizeof(val))
2398                         return -EINVAL;
2399                 if (copy_from_user(&val, optval, sizeof(val)))
2400                         return -EFAULT;
2401
2402                 po->tp_tstamp = val;
2403                 return 0;
2404         }
2405         case PACKET_FANOUT:
2406         {
2407                 int val;
2408
2409                 if (optlen != sizeof(val))
2410                         return -EINVAL;
2411                 if (copy_from_user(&val, optval, sizeof(val)))
2412                         return -EFAULT;
2413
2414                 return fanout_add(sk, val & 0xffff, val >> 16);
2415         }
2416         default:
2417                 return -ENOPROTOOPT;
2418         }
2419 }
2420
2421 static int packet_getsockopt(struct socket *sock, int level, int optname,
2422                              char __user *optval, int __user *optlen)
2423 {
2424         int len;
2425         int val;
2426         struct sock *sk = sock->sk;
2427         struct packet_sock *po = pkt_sk(sk);
2428         void *data;
2429         struct tpacket_stats st;
2430
2431         if (level != SOL_PACKET)
2432                 return -ENOPROTOOPT;
2433
2434         if (get_user(len, optlen))
2435                 return -EFAULT;
2436
2437         if (len < 0)
2438                 return -EINVAL;
2439
2440         switch (optname) {
2441         case PACKET_STATISTICS:
2442                 if (len > sizeof(struct tpacket_stats))
2443                         len = sizeof(struct tpacket_stats);
2444                 spin_lock_bh(&sk->sk_receive_queue.lock);
2445                 st = po->stats;
2446                 memset(&po->stats, 0, sizeof(st));
2447                 spin_unlock_bh(&sk->sk_receive_queue.lock);
2448                 st.tp_packets += st.tp_drops;
2449
2450                 data = &st;
2451                 break;
2452         case PACKET_AUXDATA:
2453                 if (len > sizeof(int))
2454                         len = sizeof(int);
2455                 val = po->auxdata;
2456
2457                 data = &val;
2458                 break;
2459         case PACKET_ORIGDEV:
2460                 if (len > sizeof(int))
2461                         len = sizeof(int);
2462                 val = po->origdev;
2463
2464                 data = &val;
2465                 break;
2466         case PACKET_VNET_HDR:
2467                 if (len > sizeof(int))
2468                         len = sizeof(int);
2469                 val = po->has_vnet_hdr;
2470
2471                 data = &val;
2472                 break;
2473         case PACKET_VERSION:
2474                 if (len > sizeof(int))
2475                         len = sizeof(int);
2476                 val = po->tp_version;
2477                 data = &val;
2478                 break;
2479         case PACKET_HDRLEN:
2480                 if (len > sizeof(int))
2481                         len = sizeof(int);
2482                 if (copy_from_user(&val, optval, len))
2483                         return -EFAULT;
2484                 switch (val) {
2485                 case TPACKET_V1:
2486                         val = sizeof(struct tpacket_hdr);
2487                         break;
2488                 case TPACKET_V2:
2489                         val = sizeof(struct tpacket2_hdr);
2490                         break;
2491                 default:
2492                         return -EINVAL;
2493                 }
2494                 data = &val;
2495                 break;
2496         case PACKET_RESERVE:
2497                 if (len > sizeof(unsigned int))
2498                         len = sizeof(unsigned int);
2499                 val = po->tp_reserve;
2500                 data = &val;
2501                 break;
2502         case PACKET_LOSS:
2503                 if (len > sizeof(unsigned int))
2504                         len = sizeof(unsigned int);
2505                 val = po->tp_loss;
2506                 data = &val;
2507                 break;
2508         case PACKET_TIMESTAMP:
2509                 if (len > sizeof(int))
2510                         len = sizeof(int);
2511                 val = po->tp_tstamp;
2512                 data = &val;
2513                 break;
2514         case PACKET_FANOUT:
2515                 if (len > sizeof(int))
2516                         len = sizeof(int);
2517                 val = (po->fanout ?
2518                        ((u32)po->fanout->id |
2519                         ((u32)po->fanout->type << 16)) :
2520                        0);
2521                 data = &val;
2522                 break;
2523         default:
2524                 return -ENOPROTOOPT;
2525         }
2526
2527         if (put_user(len, optlen))
2528                 return -EFAULT;
2529         if (copy_to_user(optval, data, len))
2530                 return -EFAULT;
2531         return 0;
2532 }
2533
2534
2535 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
2536 {
2537         struct sock *sk;
2538         struct hlist_node *node;
2539         struct net_device *dev = data;
2540         struct net *net = dev_net(dev);
2541
2542         rcu_read_lock();
2543         sk_for_each_rcu(sk, node, &net->packet.sklist) {
2544                 struct packet_sock *po = pkt_sk(sk);
2545
2546                 switch (msg) {
2547                 case NETDEV_UNREGISTER:
2548                         if (po->mclist)
2549                                 packet_dev_mclist(dev, po->mclist, -1);
2550                         /* fallthrough */
2551
2552                 case NETDEV_DOWN:
2553                         if (dev->ifindex == po->ifindex) {
2554                                 spin_lock(&po->bind_lock);
2555                                 if (po->running) {
2556                                         __unregister_prot_hook(sk, false);
2557                                         sk->sk_err = ENETDOWN;
2558                                         if (!sock_flag(sk, SOCK_DEAD))
2559                                                 sk->sk_error_report(sk);
2560                                 }
2561                                 if (msg == NETDEV_UNREGISTER) {
2562                                         po->ifindex = -1;
2563                                         if (po->prot_hook.dev)
2564                                                 dev_put(po->prot_hook.dev);
2565                                         po->prot_hook.dev = NULL;
2566                                 }
2567                                 spin_unlock(&po->bind_lock);
2568                         }
2569                         break;
2570                 case NETDEV_UP:
2571                         if (dev->ifindex == po->ifindex) {
2572                                 spin_lock(&po->bind_lock);
2573                                 if (po->num)
2574                                         register_prot_hook(sk);
2575                                 spin_unlock(&po->bind_lock);
2576                         }
2577                         break;
2578                 }
2579         }
2580         rcu_read_unlock();
2581         return NOTIFY_DONE;
2582 }
2583
2584
2585 static int packet_ioctl(struct socket *sock, unsigned int cmd,
2586                         unsigned long arg)
2587 {
2588         struct sock *sk = sock->sk;
2589
2590         switch (cmd) {
2591         case SIOCOUTQ:
2592         {
2593                 int amount = sk_wmem_alloc_get(sk);
2594
2595                 return put_user(amount, (int __user *)arg);
2596         }
2597         case SIOCINQ:
2598         {
2599                 struct sk_buff *skb;
2600                 int amount = 0;
2601
2602                 spin_lock_bh(&sk->sk_receive_queue.lock);
2603                 skb = skb_peek(&sk->sk_receive_queue);
2604                 if (skb)
2605                         amount = skb->len;
2606                 spin_unlock_bh(&sk->sk_receive_queue.lock);
2607                 return put_user(amount, (int __user *)arg);
2608         }
2609         case SIOCGSTAMP:
2610                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2611         case SIOCGSTAMPNS:
2612                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2613
2614 #ifdef CONFIG_INET
2615         case SIOCADDRT:
2616         case SIOCDELRT:
2617         case SIOCDARP:
2618         case SIOCGARP:
2619         case SIOCSARP:
2620         case SIOCGIFADDR:
2621         case SIOCSIFADDR:
2622         case SIOCGIFBRDADDR:
2623         case SIOCSIFBRDADDR:
2624         case SIOCGIFNETMASK:
2625         case SIOCSIFNETMASK:
2626         case SIOCGIFDSTADDR:
2627         case SIOCSIFDSTADDR:
2628         case SIOCSIFFLAGS:
2629                 return inet_dgram_ops.ioctl(sock, cmd, arg);
2630 #endif
2631
2632         default:
2633                 return -ENOIOCTLCMD;
2634         }
2635         return 0;
2636 }
2637
2638 static unsigned int packet_poll(struct file *file, struct socket *sock,
2639                                 poll_table *wait)
2640 {
2641         struct sock *sk = sock->sk;
2642         struct packet_sock *po = pkt_sk(sk);
2643         unsigned int mask = datagram_poll(file, sock, wait);
2644
2645         spin_lock_bh(&sk->sk_receive_queue.lock);
2646         if (po->rx_ring.pg_vec) {
2647                 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2648                         mask |= POLLIN | POLLRDNORM;
2649         }
2650         spin_unlock_bh(&sk->sk_receive_queue.lock);
2651         spin_lock_bh(&sk->sk_write_queue.lock);
2652         if (po->tx_ring.pg_vec) {
2653                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2654                         mask |= POLLOUT | POLLWRNORM;
2655         }
2656         spin_unlock_bh(&sk->sk_write_queue.lock);
2657         return mask;
2658 }
2659
2660
2661 /* Dirty? Well, I still did not learn better way to account
2662  * for user mmaps.
2663  */
2664
2665 static void packet_mm_open(struct vm_area_struct *vma)
2666 {
2667         struct file *file = vma->vm_file;
2668         struct socket *sock = file->private_data;
2669         struct sock *sk = sock->sk;
2670
2671         if (sk)
2672                 atomic_inc(&pkt_sk(sk)->mapped);
2673 }
2674
2675 static void packet_mm_close(struct vm_area_struct *vma)
2676 {
2677         struct file *file = vma->vm_file;
2678         struct socket *sock = file->private_data;
2679         struct sock *sk = sock->sk;
2680
2681         if (sk)
2682                 atomic_dec(&pkt_sk(sk)->mapped);
2683 }
2684
2685 static const struct vm_operations_struct packet_mmap_ops = {
2686         .open   =       packet_mm_open,
2687         .close  =       packet_mm_close,
2688 };
2689
2690 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
2691                         unsigned int len)
2692 {
2693         int i;
2694
2695         for (i = 0; i < len; i++) {
2696                 if (likely(pg_vec[i].buffer)) {
2697                         if (is_vmalloc_addr(pg_vec[i].buffer))
2698                                 vfree(pg_vec[i].buffer);
2699                         else
2700                                 free_pages((unsigned long)pg_vec[i].buffer,
2701                                            order);
2702                         pg_vec[i].buffer = NULL;
2703                 }
2704         }
2705         kfree(pg_vec);
2706 }
2707
2708 static inline char *alloc_one_pg_vec_page(unsigned long order)
2709 {
2710         char *buffer = NULL;
2711         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
2712                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
2713
2714         buffer = (char *) __get_free_pages(gfp_flags, order);
2715
2716         if (buffer)
2717                 return buffer;
2718
2719         /*
2720          * __get_free_pages failed, fall back to vmalloc
2721          */
2722         buffer = vzalloc((1 << order) * PAGE_SIZE);
2723
2724         if (buffer)
2725                 return buffer;
2726
2727         /*
2728          * vmalloc failed, lets dig into swap here
2729          */
2730         gfp_flags &= ~__GFP_NORETRY;
2731         buffer = (char *)__get_free_pages(gfp_flags, order);
2732         if (buffer)
2733                 return buffer;
2734
2735         /*
2736          * complete and utter failure
2737          */
2738         return NULL;
2739 }
2740
2741 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
2742 {
2743         unsigned int block_nr = req->tp_block_nr;
2744         struct pgv *pg_vec;
2745         int i;
2746
2747         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
2748         if (unlikely(!pg_vec))
2749                 goto out;
2750
2751         for (i = 0; i < block_nr; i++) {
2752                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
2753                 if (unlikely(!pg_vec[i].buffer))
2754                         goto out_free_pgvec;
2755         }
2756
2757 out:
2758         return pg_vec;
2759
2760 out_free_pgvec:
2761         free_pg_vec(pg_vec, order, block_nr);
2762         pg_vec = NULL;
2763         goto out;
2764 }
2765
2766 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2767                 int closing, int tx_ring)
2768 {
2769         struct pgv *pg_vec = NULL;
2770         struct packet_sock *po = pkt_sk(sk);
2771         int was_running, order = 0;
2772         struct packet_ring_buffer *rb;
2773         struct sk_buff_head *rb_queue;
2774         __be16 num;
2775         int err;
2776
2777         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2778         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2779
2780         err = -EBUSY;
2781         if (!closing) {
2782                 if (atomic_read(&po->mapped))
2783                         goto out;
2784                 if (atomic_read(&rb->pending))
2785                         goto out;
2786         }
2787
2788         if (req->tp_block_nr) {
2789                 /* Sanity tests and some calculations */
2790                 err = -EBUSY;
2791                 if (unlikely(rb->pg_vec))
2792                         goto out;
2793
2794                 switch (po->tp_version) {
2795                 case TPACKET_V1:
2796                         po->tp_hdrlen = TPACKET_HDRLEN;
2797                         break;
2798                 case TPACKET_V2:
2799                         po->tp_hdrlen = TPACKET2_HDRLEN;
2800                         break;
2801                 }
2802
2803                 err = -EINVAL;
2804                 if (unlikely((int)req->tp_block_size <= 0))
2805                         goto out;
2806                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2807                         goto out;
2808                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2809                                         po->tp_reserve))
2810                         goto out;
2811                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2812                         goto out;
2813
2814                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2815                 if (unlikely(rb->frames_per_block <= 0))
2816                         goto out;
2817                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2818                                         req->tp_frame_nr))
2819                         goto out;
2820
2821                 err = -ENOMEM;
2822                 order = get_order(req->tp_block_size);
2823                 pg_vec = alloc_pg_vec(req, order);
2824                 if (unlikely(!pg_vec))
2825                         goto out;
2826         }
2827         /* Done */
2828         else {
2829                 err = -EINVAL;
2830                 if (unlikely(req->tp_frame_nr))
2831                         goto out;
2832         }
2833
2834         lock_sock(sk);
2835
2836         /* Detach socket from network */
2837         spin_lock(&po->bind_lock);
2838         was_running = po->running;
2839         num = po->num;
2840         if (was_running) {
2841                 po->num = 0;
2842                 __unregister_prot_hook(sk, false);
2843         }
2844         spin_unlock(&po->bind_lock);
2845
2846         synchronize_net();
2847
2848         err = -EBUSY;
2849         mutex_lock(&po->pg_vec_lock);
2850         if (closing || atomic_read(&po->mapped) == 0) {
2851                 err = 0;
2852                 spin_lock_bh(&rb_queue->lock);
2853                 swap(rb->pg_vec, pg_vec);
2854                 rb->frame_max = (req->tp_frame_nr - 1);
2855                 rb->head = 0;
2856                 rb->frame_size = req->tp_frame_size;
2857                 spin_unlock_bh(&rb_queue->lock);
2858
2859                 swap(rb->pg_vec_order, order);
2860                 swap(rb->pg_vec_len, req->tp_block_nr);
2861
2862                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2863                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2864                                                 tpacket_rcv : packet_rcv;
2865                 skb_queue_purge(rb_queue);
2866                 if (atomic_read(&po->mapped))
2867                         pr_err("packet_mmap: vma is busy: %d\n",
2868                                atomic_read(&po->mapped));
2869         }
2870         mutex_unlock(&po->pg_vec_lock);
2871
2872         spin_lock(&po->bind_lock);
2873         if (was_running) {
2874                 po->num = num;
2875                 register_prot_hook(sk);
2876         }
2877         spin_unlock(&po->bind_lock);
2878
2879         release_sock(sk);
2880
2881         if (pg_vec)
2882                 free_pg_vec(pg_vec, order, req->tp_block_nr);
2883 out:
2884         return err;
2885 }
2886
2887 static int packet_mmap(struct file *file, struct socket *sock,
2888                 struct vm_area_struct *vma)
2889 {
2890         struct sock *sk = sock->sk;
2891         struct packet_sock *po = pkt_sk(sk);
2892         unsigned long size, expected_size;
2893         struct packet_ring_buffer *rb;
2894         unsigned long start;
2895         int err = -EINVAL;
2896         int i;
2897
2898         if (vma->vm_pgoff)
2899                 return -EINVAL;
2900
2901         mutex_lock(&po->pg_vec_lock);
2902
2903         expected_size = 0;
2904         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2905                 if (rb->pg_vec) {
2906                         expected_size += rb->pg_vec_len
2907                                                 * rb->pg_vec_pages
2908                                                 * PAGE_SIZE;
2909                 }
2910         }
2911
2912         if (expected_size == 0)
2913                 goto out;
2914
2915         size = vma->vm_end - vma->vm_start;
2916         if (size != expected_size)
2917                 goto out;
2918
2919         start = vma->vm_start;
2920         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2921                 if (rb->pg_vec == NULL)
2922                         continue;
2923
2924                 for (i = 0; i < rb->pg_vec_len; i++) {
2925                         struct page *page;
2926                         void *kaddr = rb->pg_vec[i].buffer;
2927                         int pg_num;
2928
2929                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
2930                                 page = pgv_to_page(kaddr);
2931                                 err = vm_insert_page(vma, start, page);
2932                                 if (unlikely(err))
2933                                         goto out;
2934                                 start += PAGE_SIZE;
2935                                 kaddr += PAGE_SIZE;
2936                         }
2937                 }
2938         }
2939
2940         atomic_inc(&po->mapped);
2941         vma->vm_ops = &packet_mmap_ops;
2942         err = 0;
2943
2944 out:
2945         mutex_unlock(&po->pg_vec_lock);
2946         return err;
2947 }
2948
2949 static const struct proto_ops packet_ops_spkt = {
2950         .family =       PF_PACKET,
2951         .owner =        THIS_MODULE,
2952         .release =      packet_release,
2953         .bind =         packet_bind_spkt,
2954         .connect =      sock_no_connect,
2955         .socketpair =   sock_no_socketpair,
2956         .accept =       sock_no_accept,
2957         .getname =      packet_getname_spkt,
2958         .poll =         datagram_poll,
2959         .ioctl =        packet_ioctl,
2960         .listen =       sock_no_listen,
2961         .shutdown =     sock_no_shutdown,
2962         .setsockopt =   sock_no_setsockopt,
2963         .getsockopt =   sock_no_getsockopt,
2964         .sendmsg =      packet_sendmsg_spkt,
2965         .recvmsg =      packet_recvmsg,
2966         .mmap =         sock_no_mmap,
2967         .sendpage =     sock_no_sendpage,
2968 };
2969
2970 static const struct proto_ops packet_ops = {
2971         .family =       PF_PACKET,
2972         .owner =        THIS_MODULE,
2973         .release =      packet_release,
2974         .bind =         packet_bind,
2975         .connect =      sock_no_connect,
2976         .socketpair =   sock_no_socketpair,
2977         .accept =       sock_no_accept,
2978         .getname =      packet_getname,
2979         .poll =         packet_poll,
2980         .ioctl =        packet_ioctl,
2981         .listen =       sock_no_listen,
2982         .shutdown =     sock_no_shutdown,
2983         .setsockopt =   packet_setsockopt,
2984         .getsockopt =   packet_getsockopt,
2985         .sendmsg =      packet_sendmsg,
2986         .recvmsg =      packet_recvmsg,
2987         .mmap =         packet_mmap,
2988         .sendpage =     sock_no_sendpage,
2989 };
2990
2991 static const struct net_proto_family packet_family_ops = {
2992         .family =       PF_PACKET,
2993         .create =       packet_create,
2994         .owner  =       THIS_MODULE,
2995 };
2996
2997 static struct notifier_block packet_netdev_notifier = {
2998         .notifier_call =        packet_notifier,
2999 };
3000
3001 #ifdef CONFIG_PROC_FS
3002
3003 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3004         __acquires(RCU)
3005 {
3006         struct net *net = seq_file_net(seq);
3007
3008         rcu_read_lock();
3009         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3010 }
3011
3012 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3013 {
3014         struct net *net = seq_file_net(seq);
3015         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3016 }
3017
3018 static void packet_seq_stop(struct seq_file *seq, void *v)
3019         __releases(RCU)
3020 {
3021         rcu_read_unlock();
3022 }
3023
3024 static int packet_seq_show(struct seq_file *seq, void *v)
3025 {
3026         if (v == SEQ_START_TOKEN)
3027                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3028         else {
3029                 struct sock *s = sk_entry(v);
3030                 const struct packet_sock *po = pkt_sk(s);
3031
3032                 seq_printf(seq,
3033                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3034                            s,
3035                            atomic_read(&s->sk_refcnt),
3036                            s->sk_type,
3037                            ntohs(po->num),
3038                            po->ifindex,
3039                            po->running,
3040                            atomic_read(&s->sk_rmem_alloc),
3041                            sock_i_uid(s),
3042                            sock_i_ino(s));
3043         }
3044
3045         return 0;
3046 }
3047
3048 static const struct seq_operations packet_seq_ops = {
3049         .start  = packet_seq_start,
3050         .next   = packet_seq_next,
3051         .stop   = packet_seq_stop,
3052         .show   = packet_seq_show,
3053 };
3054
3055 static int packet_seq_open(struct inode *inode, struct file *file)
3056 {
3057         return seq_open_net(inode, file, &packet_seq_ops,
3058                             sizeof(struct seq_net_private));
3059 }
3060
3061 static const struct file_operations packet_seq_fops = {
3062         .owner          = THIS_MODULE,
3063         .open           = packet_seq_open,
3064         .read           = seq_read,
3065         .llseek         = seq_lseek,
3066         .release        = seq_release_net,
3067 };
3068
3069 #endif
3070
3071 static int __net_init packet_net_init(struct net *net)
3072 {
3073         spin_lock_init(&net->packet.sklist_lock);
3074         INIT_HLIST_HEAD(&net->packet.sklist);
3075
3076         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3077                 return -ENOMEM;
3078
3079         return 0;
3080 }
3081
3082 static void __net_exit packet_net_exit(struct net *net)
3083 {
3084         proc_net_remove(net, "packet");
3085 }
3086
3087 static struct pernet_operations packet_net_ops = {
3088         .init = packet_net_init,
3089         .exit = packet_net_exit,
3090 };
3091
3092
3093 static void __exit packet_exit(void)
3094 {
3095         unregister_netdevice_notifier(&packet_netdev_notifier);
3096         unregister_pernet_subsys(&packet_net_ops);
3097         sock_unregister(PF_PACKET);
3098         proto_unregister(&packet_proto);
3099 }
3100
3101 static int __init packet_init(void)
3102 {
3103         int rc = proto_register(&packet_proto, 0);
3104
3105         if (rc != 0)
3106                 goto out;
3107
3108         sock_register(&packet_family_ops);
3109         register_pernet_subsys(&packet_net_ops);
3110         register_netdevice_notifier(&packet_netdev_notifier);
3111 out:
3112         return rc;
3113 }
3114
3115 module_init(packet_init);
3116 module_exit(packet_exit);
3117 MODULE_LICENSE("GPL");
3118 MODULE_ALIAS_NETPROTO(PF_PACKET);