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  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
  44  *                                      layer.
  45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
  46  *
  47  *
  48  *              This program is free software; you can redistribute it and/or
  49  *              modify it under the terms of the GNU General Public License
  50  *              as published by the Free Software Foundation; either version
  51  *              2 of the License, or (at your option) any later version.
  52  *
  53  */
  54
  55 #include <linux/types.h>
  56 #include <linux/mm.h>
  57 #include <linux/capability.h>
  58 #include <linux/fcntl.h>
  59 #include <linux/socket.h>
  60 #include <linux/in.h>
  61 #include <linux/inet.h>
  62 #include <linux/netdevice.h>
  63 #include <linux/if_packet.h>
  64 #include <linux/wireless.h>
  65 #include <linux/kernel.h>
  66 #include <linux/kmod.h>
  67 #include <linux/slab.h>
  68 #include <linux/vmalloc.h>
  69 #include <net/net_namespace.h>
  70 #include <net/ip.h>
  71 #include <net/protocol.h>
  72 #include <linux/skbuff.h>
  73 #include <net/sock.h>
  74 #include <linux/errno.h>
  75 #include <linux/timer.h>
  76 #include <asm/uaccess.h>
  77 #include <asm/ioctls.h>
  78 #include <asm/page.h>
  79 #include <asm/cacheflush.h>
  80 #include <asm/io.h>
  81 #include <linux/proc_fs.h>
  82 #include <linux/seq_file.h>
  83 #include <linux/poll.h>
  84 #include <linux/module.h>
  85 #include <linux/init.h>
  86 #include <linux/mutex.h>
  87 #include <linux/if_vlan.h>
  88 #include <linux/virtio_net.h>
  89 #include <linux/errqueue.h>
  90 #include <linux/net_tstamp.h>
  91 #include <linux/reciprocal_div.h>
  92 #ifdef CONFIG_INET
  93 #include <net/inet_common.h>
  94 #endif
  95
  96 #include "internal.h"
  97
  98 /*
  99    Assumptions:
 100    - if device has no dev->hard_header routine, it adds and removes ll header
 101      inside itself. In this case ll header is invisible outside of device,
 102      but higher levels still should reserve dev->hard_header_len.
 103      Some devices are enough clever to reallocate skb, when header
 104      will not fit to reserved space (tunnel), another ones are silly
 105      (PPP).
 106    - packet socket receives packets with pulled ll header,
 107      so that SOCK_RAW should push it back.
 108
 109 On receive:
 110 -----------
 111
 112 Incoming, dev->hard_header!=NULL
 113    mac_header -> ll header
 114    data       -> data
 115
 116 Outgoing, dev->hard_header!=NULL
 117    mac_header -> ll header
 118    data       -> ll header
 119
 120 Incoming, dev->hard_header==NULL
 121    mac_header -> UNKNOWN position. It is very likely, that it points to ll
 122                  header.  PPP makes it, that is wrong, because introduce
 123                  assymetry between rx and tx paths.
 124    data       -> data
 125
 126 Outgoing, dev->hard_header==NULL
 127    mac_header -> data. ll header is still not built!
 128    data       -> data
 129
 130 Resume
 131   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
 132
 133
 134 On transmit:
 135 ------------
 136
 137 dev->hard_header != NULL
 138    mac_header -> ll header
 139    data       -> ll header
 140
 141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
 142    mac_header -> data
 143    data       -> data
 144
 145    We should set nh.raw on output to correct posistion,
 146    packet classifier depends on it.
 147  */
 148
 149 /* Private packet socket structures. */
 150
 151 /* identical to struct packet_mreq except it has
 152  * a longer address field.
 153  */
 154 struct packet_mreq_max {
 155         int             mr_ifindex;
 156         unsigned short  mr_type;
 157         unsigned short  mr_alen;
 158         unsigned char   mr_address[MAX_ADDR_LEN];
 159 };
 160
 161 union tpacket_uhdr {
 162         struct tpacket_hdr  *h1;
 163         struct tpacket2_hdr *h2;
 164         struct tpacket3_hdr *h3;
 165         void *raw;
 166 };
 167
 168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
 169                 int closing, int tx_ring);
 170
 171 #define V3_ALIGNMENT    (8)
 172
 173 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
 174
 175 #define BLK_PLUS_PRIV(sz_of_priv) \
 176         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
 177
 178 #define PGV_FROM_VMALLOC 1
 179
 180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
 181 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
 182 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
 183 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
 184 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
 185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
 186 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
 187
 188 struct packet_sock;
 189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
 190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 191                        struct packet_type *pt, struct net_device *orig_dev);
 192
 193 static void *packet_previous_frame(struct packet_sock *po,
 194                 struct packet_ring_buffer *rb,
 195                 int status);
 196 static void packet_increment_head(struct packet_ring_buffer *buff);
 197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
 198                         struct tpacket_block_desc *);
 199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
 200                         struct packet_sock *);
 201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
 202                 struct packet_sock *, unsigned int status);
 203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 204 static void prb_open_block(struct tpacket_kbdq_core *,
 205                 struct tpacket_block_desc *);
 206 static void prb_retire_rx_blk_timer_expired(unsigned long);
 207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
 208 static void prb_init_blk_timer(struct packet_sock *,
 209                 struct tpacket_kbdq_core *,
 210                 void (*func) (unsigned long));
 211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 213                 struct tpacket3_hdr *);
 214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 215                 struct tpacket3_hdr *);
 216 static void packet_flush_mclist(struct sock *sk);
 217
 218 struct packet_skb_cb {
 219         unsigned int origlen;
 220         union {
 221                 struct sockaddr_pkt pkt;
 222                 struct sockaddr_ll ll;
 223         } sa;
 224 };
 225
 226 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
 227
 228 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 229 #define GET_PBLOCK_DESC(x, bid) \
 230         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
 231 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
 232         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
 233 #define GET_NEXT_PRB_BLK_NUM(x) \
 234         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
 235         ((x)->kactive_blk_num+1) : 0)
 236
 237 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
 238 static void __fanout_link(struct sock *sk, struct packet_sock *po);
 239
 240 static int packet_direct_xmit(struct sk_buff *skb)
 241 {
 242         struct net_device *dev = skb->dev;
 243         const struct net_device_ops *ops = dev->netdev_ops;
 244         netdev_features_t features;
 245         struct netdev_queue *txq;
 246         u16 queue_map;
 247         int ret;
 248
 249         if (unlikely(!netif_running(dev) ||
 250                      !netif_carrier_ok(dev))) {
 251                 kfree_skb(skb);
 252                 return NET_XMIT_DROP;
 253         }
 254
 255         features = netif_skb_features(skb);
 256         if (skb_needs_linearize(skb, features) &&
 257             __skb_linearize(skb)) {
 258                 kfree_skb(skb);
 259                 return NET_XMIT_DROP;
 260         }
 261
 262         queue_map = skb_get_queue_mapping(skb);
 263         txq = netdev_get_tx_queue(dev, queue_map);
 264
 265         __netif_tx_lock_bh(txq);
 266         if (unlikely(netif_xmit_frozen_or_stopped(txq))) {
 267                 ret = NETDEV_TX_BUSY;
 268                 kfree_skb(skb);
 269                 goto out;
 270         }
 271
 272         ret = ops->ndo_start_xmit(skb, dev);
 273         if (likely(dev_xmit_complete(ret)))
 274                 txq_trans_update(txq);
 275         else
 276                 kfree_skb(skb);
 277 out:
 278         __netif_tx_unlock_bh(txq);
 279         return ret;
 280 }
 281
 282 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 283 {
 284         struct net_device *dev;
 285
 286         rcu_read_lock();
 287         dev = rcu_dereference(po->cached_dev);
 288         if (likely(dev))
 289                 dev_hold(dev);
 290         rcu_read_unlock();
 291
 292         return dev;
 293 }
 294
 295 static void packet_cached_dev_assign(struct packet_sock *po,
 296                                      struct net_device *dev)
 297 {
 298         rcu_assign_pointer(po->cached_dev, dev);
 299 }
 300
 301 static void packet_cached_dev_reset(struct packet_sock *po)
 302 {
 303         RCU_INIT_POINTER(po->cached_dev, NULL);
 304 }
 305
 306 static bool packet_use_direct_xmit(const struct packet_sock *po)
 307 {
 308         return po->xmit == packet_direct_xmit;
 309 }
 310
 311 static u16 packet_pick_tx_queue(struct net_device *dev)
 312 {
 313         return (u16) smp_processor_id() % dev->real_num_tx_queues;
 314 }
 315
 316 /* register_prot_hook must be invoked with the po->bind_lock held,
 317  * or from a context in which asynchronous accesses to the packet
 318  * socket is not possible (packet_create()).
 319  */
 320 static void register_prot_hook(struct sock *sk)
 321 {
 322         struct packet_sock *po = pkt_sk(sk);
 323
 324         if (!po->running) {
 325                 if (po->fanout)
 326                         __fanout_link(sk, po);
 327                 else
 328                         dev_add_pack(&po->prot_hook);
 329
 330                 sock_hold(sk);
 331                 po->running = 1;
 332         }
 333 }
 334
 335 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 336  * held.   If the sync parameter is true, we will temporarily drop
 337  * the po->bind_lock and do a synchronize_net to make sure no
 338  * asynchronous packet processing paths still refer to the elements
 339  * of po->prot_hook.  If the sync parameter is false, it is the
 340  * callers responsibility to take care of this.
 341  */
 342 static void __unregister_prot_hook(struct sock *sk, bool sync)
 343 {
 344         struct packet_sock *po = pkt_sk(sk);
 345
 346         po->running = 0;
 347
 348         if (po->fanout)
 349                 __fanout_unlink(sk, po);
 350         else
 351                 __dev_remove_pack(&po->prot_hook);
 352
 353         __sock_put(sk);
 354
 355         if (sync) {
 356                 spin_unlock(&po->bind_lock);
 357                 synchronize_net();
 358                 spin_lock(&po->bind_lock);
 359         }
 360 }
 361
 362 static void unregister_prot_hook(struct sock *sk, bool sync)
 363 {
 364         struct packet_sock *po = pkt_sk(sk);
 365
 366         if (po->running)
 367                 __unregister_prot_hook(sk, sync);
 368 }
 369
 370 static inline __pure struct page *pgv_to_page(void *addr)
 371 {
 372         if (is_vmalloc_addr(addr))
 373                 return vmalloc_to_page(addr);
 374         return virt_to_page(addr);
 375 }
 376
 377 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 378 {
 379         union tpacket_uhdr h;
 380
 381         h.raw = frame;
 382         switch (po->tp_version) {
 383         case TPACKET_V1:
 384                 h.h1->tp_status = status;
 385                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 386                 break;
 387         case TPACKET_V2:
 388                 h.h2->tp_status = status;
 389                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 390                 break;
 391         case TPACKET_V3:
 392         default:
 393                 WARN(1, "TPACKET version not supported.\n");
 394                 BUG();
 395         }
 396
 397         smp_wmb();
 398 }
 399
 400 static int __packet_get_status(struct packet_sock *po, void *frame)
 401 {
 402         union tpacket_uhdr h;
 403
 404         smp_rmb();
 405
 406         h.raw = frame;
 407         switch (po->tp_version) {
 408         case TPACKET_V1:
 409                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 410                 return h.h1->tp_status;
 411         case TPACKET_V2:
 412                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 413                 return h.h2->tp_status;
 414         case TPACKET_V3:
 415         default:
 416                 WARN(1, "TPACKET version not supported.\n");
 417                 BUG();
 418                 return 0;
 419         }
 420 }
 421
 422 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 423                                    unsigned int flags)
 424 {
 425         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 426
 427         if (shhwtstamps) {
 428                 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
 429                     ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
 430                         return TP_STATUS_TS_SYS_HARDWARE;
 431                 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 432                     ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 433                         return TP_STATUS_TS_RAW_HARDWARE;
 434         }
 435
 436         if (ktime_to_timespec_cond(skb->tstamp, ts))
 437                 return TP_STATUS_TS_SOFTWARE;
 438
 439         return 0;
 440 }
 441
 442 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 443                                     struct sk_buff *skb)
 444 {
 445         union tpacket_uhdr h;
 446         struct timespec ts;
 447         __u32 ts_status;
 448
 449         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 450                 return 0;
 451
 452         h.raw = frame;
 453         switch (po->tp_version) {
 454         case TPACKET_V1:
 455                 h.h1->tp_sec = ts.tv_sec;
 456                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 457                 break;
 458         case TPACKET_V2:
 459                 h.h2->tp_sec = ts.tv_sec;
 460                 h.h2->tp_nsec = ts.tv_nsec;
 461                 break;
 462         case TPACKET_V3:
 463         default:
 464                 WARN(1, "TPACKET version not supported.\n");
 465                 BUG();
 466         }
 467
 468         /* one flush is safe, as both fields always lie on the same cacheline */
 469         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 470         smp_wmb();
 471
 472         return ts_status;
 473 }
 474
 475 static void *packet_lookup_frame(struct packet_sock *po,
 476                 struct packet_ring_buffer *rb,
 477                 unsigned int position,
 478                 int status)
 479 {
 480         unsigned int pg_vec_pos, frame_offset;
 481         union tpacket_uhdr h;
 482
 483         pg_vec_pos = position / rb->frames_per_block;
 484         frame_offset = position % rb->frames_per_block;
 485
 486         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 487                 (frame_offset * rb->frame_size);
 488
 489         if (status != __packet_get_status(po, h.raw))
 490                 return NULL;
 491
 492         return h.raw;
 493 }
 494
 495 static void *packet_current_frame(struct packet_sock *po,
 496                 struct packet_ring_buffer *rb,
 497                 int status)
 498 {
 499         return packet_lookup_frame(po, rb, rb->head, status);
 500 }
 501
 502 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 503 {
 504         del_timer_sync(&pkc->retire_blk_timer);
 505 }
 506
 507 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 508                 int tx_ring,
 509                 struct sk_buff_head *rb_queue)
 510 {
 511         struct tpacket_kbdq_core *pkc;
 512
 513         pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
 514                         GET_PBDQC_FROM_RB(&po->rx_ring);
 515
 516         spin_lock_bh(&rb_queue->lock);
 517         pkc->delete_blk_timer = 1;
 518         spin_unlock_bh(&rb_queue->lock);
 519
 520         prb_del_retire_blk_timer(pkc);
 521 }
 522
 523 static void prb_init_blk_timer(struct packet_sock *po,
 524                 struct tpacket_kbdq_core *pkc,
 525                 void (*func) (unsigned long))
 526 {
 527         init_timer(&pkc->retire_blk_timer);
 528         pkc->retire_blk_timer.data = (long)po;
 529         pkc->retire_blk_timer.function = func;
 530         pkc->retire_blk_timer.expires = jiffies;
 531 }
 532
 533 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 534 {
 535         struct tpacket_kbdq_core *pkc;
 536
 537         if (tx_ring)
 538                 BUG();
 539
 540         pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
 541                         GET_PBDQC_FROM_RB(&po->rx_ring);
 542         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 543 }
 544
 545 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 546                                 int blk_size_in_bytes)
 547 {
 548         struct net_device *dev;
 549         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 550         struct ethtool_cmd ecmd;
 551         int err;
 552         u32 speed;
 553
 554         rtnl_lock();
 555         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 556         if (unlikely(!dev)) {
 557                 rtnl_unlock();
 558                 return DEFAULT_PRB_RETIRE_TOV;
 559         }
 560         err = __ethtool_get_settings(dev, &ecmd);
 561         speed = ethtool_cmd_speed(&ecmd);
 562         rtnl_unlock();
 563         if (!err) {
 564                 /*
 565                  * If the link speed is so slow you don't really
 566                  * need to worry about perf anyways
 567                  */
 568                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 569                         return DEFAULT_PRB_RETIRE_TOV;
 570                 } else {
 571                         msec = 1;
 572                         div = speed / 1000;
 573                 }
 574         }
 575
 576         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 577
 578         if (div)
 579                 mbits /= div;
 580
 581         tmo = mbits * msec;
 582
 583         if (div)
 584                 return tmo+1;
 585         return tmo;
 586 }
 587
 588 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 589                         union tpacket_req_u *req_u)
 590 {
 591         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 592 }
 593
 594 static void init_prb_bdqc(struct packet_sock *po,
 595                         struct packet_ring_buffer *rb,
 596                         struct pgv *pg_vec,
 597                         union tpacket_req_u *req_u, int tx_ring)
 598 {
 599         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 600         struct tpacket_block_desc *pbd;
 601
 602         memset(p1, 0x0, sizeof(*p1));
 603
 604         p1->knxt_seq_num = 1;
 605         p1->pkbdq = pg_vec;
 606         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 607         p1->pkblk_start = pg_vec[0].buffer;
 608         p1->kblk_size = req_u->req3.tp_block_size;
 609         p1->knum_blocks = req_u->req3.tp_block_nr;
 610         p1->hdrlen = po->tp_hdrlen;
 611         p1->version = po->tp_version;
 612         p1->last_kactive_blk_num = 0;
 613         po->stats.stats3.tp_freeze_q_cnt = 0;
 614         if (req_u->req3.tp_retire_blk_tov)
 615                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 616         else
 617                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 618                                                 req_u->req3.tp_block_size);
 619         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 620         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 621
 622         prb_init_ft_ops(p1, req_u);
 623         prb_setup_retire_blk_timer(po, tx_ring);
 624         prb_open_block(p1, pbd);
 625 }
 626
 627 /*  Do NOT update the last_blk_num first.
 628  *  Assumes sk_buff_head lock is held.
 629  */
 630 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 631 {
 632         mod_timer(&pkc->retire_blk_timer,
 633                         jiffies + pkc->tov_in_jiffies);
 634         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 635 }
 636
 637 /*
 638  * Timer logic:
 639  * 1) We refresh the timer only when we open a block.
 640  *    By doing this we don't waste cycles refreshing the timer
 641  *        on packet-by-packet basis.
 642  *
 643  * With a 1MB block-size, on a 1Gbps line, it will take
 644  * i) ~8 ms to fill a block + ii) memcpy etc.
 645  * In this cut we are not accounting for the memcpy time.
 646  *
 647  * So, if the user sets the 'tmo' to 10ms then the timer
 648  * will never fire while the block is still getting filled
 649  * (which is what we want). However, the user could choose
 650  * to close a block early and that's fine.
 651  *
 652  * But when the timer does fire, we check whether or not to refresh it.
 653  * Since the tmo granularity is in msecs, it is not too expensive
 654  * to refresh the timer, lets say every '8' msecs.
 655  * Either the user can set the 'tmo' or we can derive it based on
 656  * a) line-speed and b) block-size.
 657  * prb_calc_retire_blk_tmo() calculates the tmo.
 658  *
 659  */
 660 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 661 {
 662         struct packet_sock *po = (struct packet_sock *)data;
 663         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 664         unsigned int frozen;
 665         struct tpacket_block_desc *pbd;
 666
 667         spin_lock(&po->sk.sk_receive_queue.lock);
 668
 669         frozen = prb_queue_frozen(pkc);
 670         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 671
 672         if (unlikely(pkc->delete_blk_timer))
 673                 goto out;
 674
 675         /* We only need to plug the race when the block is partially filled.
 676          * tpacket_rcv:
 677          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 678          *              copy_bits() is in progress ...
 679          *              timer fires on other cpu:
 680          *              we can't retire the current block because copy_bits
 681          *              is in progress.
 682          *
 683          */
 684         if (BLOCK_NUM_PKTS(pbd)) {
 685                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 686                         /* Waiting for skb_copy_bits to finish... */
 687                         cpu_relax();
 688                 }
 689         }
 690
 691         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 692                 if (!frozen) {
 693                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 694                         if (!prb_dispatch_next_block(pkc, po))
 695                                 goto refresh_timer;
 696                         else
 697                                 goto out;
 698                 } else {
 699                         /* Case 1. Queue was frozen because user-space was
 700                          *         lagging behind.
 701                          */
 702                         if (prb_curr_blk_in_use(pkc, pbd)) {
 703                                 /*
 704                                  * Ok, user-space is still behind.
 705                                  * So just refresh the timer.
 706                                  */
 707                                 goto refresh_timer;
 708                         } else {
 709                                /* Case 2. queue was frozen,user-space caught up,
 710                                 * now the link went idle && the timer fired.
 711                                 * We don't have a block to close.So we open this
 712                                 * block and restart the timer.
 713                                 * opening a block thaws the queue,restarts timer
 714                                 * Thawing/timer-refresh is a side effect.
 715                                 */
 716                                 prb_open_block(pkc, pbd);
 717                                 goto out;
 718                         }
 719                 }
 720         }
 721
 722 refresh_timer:
 723         _prb_refresh_rx_retire_blk_timer(pkc);
 724
 725 out:
 726         spin_unlock(&po->sk.sk_receive_queue.lock);
 727 }
 728
 729 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 730                 struct tpacket_block_desc *pbd1, __u32 status)
 731 {
 732         /* Flush everything minus the block header */
 733
 734 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 735         u8 *start, *end;
 736
 737         start = (u8 *)pbd1;
 738
 739         /* Skip the block header(we know header WILL fit in 4K) */
 740         start += PAGE_SIZE;
 741
 742         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 743         for (; start < end; start += PAGE_SIZE)
 744                 flush_dcache_page(pgv_to_page(start));
 745
 746         smp_wmb();
 747 #endif
 748
 749         /* Now update the block status. */
 750
 751         BLOCK_STATUS(pbd1) = status;
 752
 753         /* Flush the block header */
 754
 755 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 756         start = (u8 *)pbd1;
 757         flush_dcache_page(pgv_to_page(start));
 758
 759         smp_wmb();
 760 #endif
 761 }
 762
 763 /*
 764  * Side effect:
 765  *
 766  * 1) flush the block
 767  * 2) Increment active_blk_num
 768  *
 769  * Note:We DONT refresh the timer on purpose.
 770  *      Because almost always the next block will be opened.
 771  */
 772 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 773                 struct tpacket_block_desc *pbd1,
 774                 struct packet_sock *po, unsigned int stat)
 775 {
 776         __u32 status = TP_STATUS_USER | stat;
 777
 778         struct tpacket3_hdr *last_pkt;
 779         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 780
 781         if (po->stats.stats3.tp_drops)
 782                 status |= TP_STATUS_LOSING;
 783
 784         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 785         last_pkt->tp_next_offset = 0;
 786
 787         /* Get the ts of the last pkt */
 788         if (BLOCK_NUM_PKTS(pbd1)) {
 789                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 790                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 791         } else {
 792                 /* Ok, we tmo'd - so get the current time */
 793                 struct timespec ts;
 794                 getnstimeofday(&ts);
 795                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 796                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 797         }
 798
 799         smp_wmb();
 800
 801         /* Flush the block */
 802         prb_flush_block(pkc1, pbd1, status);
 803
 804         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 805 }
 806
 807 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 808 {
 809         pkc->reset_pending_on_curr_blk = 0;
 810 }
 811
 812 /*
 813  * Side effect of opening a block:
 814  *
 815  * 1) prb_queue is thawed.
 816  * 2) retire_blk_timer is refreshed.
 817  *
 818  */
 819 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 820         struct tpacket_block_desc *pbd1)
 821 {
 822         struct timespec ts;
 823         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 824
 825         smp_rmb();
 826
 827         /* We could have just memset this but we will lose the
 828          * flexibility of making the priv area sticky
 829          */
 830
 831         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 832         BLOCK_NUM_PKTS(pbd1) = 0;
 833         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 834
 835         getnstimeofday(&ts);
 836
 837         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 838         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 839
 840         pkc1->pkblk_start = (char *)pbd1;
 841         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 842
 843         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 844         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 845
 846         pbd1->version = pkc1->version;
 847         pkc1->prev = pkc1->nxt_offset;
 848         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 849
 850         prb_thaw_queue(pkc1);
 851         _prb_refresh_rx_retire_blk_timer(pkc1);
 852
 853         smp_wmb();
 854 }
 855
 856 /*
 857  * Queue freeze logic:
 858  * 1) Assume tp_block_nr = 8 blocks.
 859  * 2) At time 't0', user opens Rx ring.
 860  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 861  * 4) user-space is either sleeping or processing block '0'.
 862  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 863  *    it will close block-7,loop around and try to fill block '0'.
 864  *    call-flow:
 865  *    __packet_lookup_frame_in_block
 866  *      prb_retire_current_block()
 867  *      prb_dispatch_next_block()
 868  *        |->(BLOCK_STATUS == USER) evaluates to true
 869  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 870  * 6) Now there are two cases:
 871  *    6.1) Link goes idle right after the queue is frozen.
 872  *         But remember, the last open_block() refreshed the timer.
 873  *         When this timer expires,it will refresh itself so that we can
 874  *         re-open block-0 in near future.
 875  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 876  *         case and __packet_lookup_frame_in_block will check if block-0
 877  *         is free and can now be re-used.
 878  */
 879 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 880                                   struct packet_sock *po)
 881 {
 882         pkc->reset_pending_on_curr_blk = 1;
 883         po->stats.stats3.tp_freeze_q_cnt++;
 884 }
 885
 886 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 887
 888 /*
 889  * If the next block is free then we will dispatch it
 890  * and return a good offset.
 891  * Else, we will freeze the queue.
 892  * So, caller must check the return value.
 893  */
 894 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 895                 struct packet_sock *po)
 896 {
 897         struct tpacket_block_desc *pbd;
 898
 899         smp_rmb();
 900
 901         /* 1. Get current block num */
 902         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 903
 904         /* 2. If this block is currently in_use then freeze the queue */
 905         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 906                 prb_freeze_queue(pkc, po);
 907                 return NULL;
 908         }
 909
 910         /*
 911          * 3.
 912          * open this block and return the offset where the first packet
 913          * needs to get stored.
 914          */
 915         prb_open_block(pkc, pbd);
 916         return (void *)pkc->nxt_offset;
 917 }
 918
 919 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 920                 struct packet_sock *po, unsigned int status)
 921 {
 922         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 923
 924         /* retire/close the current block */
 925         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 926                 /*
 927                  * Plug the case where copy_bits() is in progress on
 928                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 929                  * have space to copy the pkt in the current block and
 930                  * called prb_retire_current_block()
 931                  *
 932                  * We don't need to worry about the TMO case because
 933                  * the timer-handler already handled this case.
 934                  */
 935                 if (!(status & TP_STATUS_BLK_TMO)) {
 936                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 937                                 /* Waiting for skb_copy_bits to finish... */
 938                                 cpu_relax();
 939                         }
 940                 }
 941                 prb_close_block(pkc, pbd, po, status);
 942                 return;
 943         }
 944 }
 945
 946 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 947                                       struct tpacket_block_desc *pbd)
 948 {
 949         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 950 }
 951
 952 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 953 {
 954         return pkc->reset_pending_on_curr_blk;
 955 }
 956
 957 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 958 {
 959         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 960         atomic_dec(&pkc->blk_fill_in_prog);
 961 }
 962
 963 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 964                         struct tpacket3_hdr *ppd)
 965 {
 966         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
 967 }
 968
 969 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 970                         struct tpacket3_hdr *ppd)
 971 {
 972         ppd->hv1.tp_rxhash = 0;
 973 }
 974
 975 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 976                         struct tpacket3_hdr *ppd)
 977 {
 978         if (vlan_tx_tag_present(pkc->skb)) {
 979                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 980                 ppd->tp_status = TP_STATUS_VLAN_VALID;
 981         } else {
 982                 ppd->hv1.tp_vlan_tci = 0;
 983                 ppd->tp_status = TP_STATUS_AVAILABLE;
 984         }
 985 }
 986
 987 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 988                         struct tpacket3_hdr *ppd)
 989 {
 990         prb_fill_vlan_info(pkc, ppd);
 991
 992         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 993                 prb_fill_rxhash(pkc, ppd);
 994         else
 995                 prb_clear_rxhash(pkc, ppd);
 996 }
 997
 998 static void prb_fill_curr_block(char *curr,
 999                                 struct tpacket_kbdq_core *pkc,
1000                                 struct tpacket_block_desc *pbd,
1001                                 unsigned int len)
1002 {
1003         struct tpacket3_hdr *ppd;
1004
1005         ppd  = (struct tpacket3_hdr *)curr;
1006         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1007         pkc->prev = curr;
1008         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1009         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1010         BLOCK_NUM_PKTS(pbd) += 1;
1011         atomic_inc(&pkc->blk_fill_in_prog);
1012         prb_run_all_ft_ops(pkc, ppd);
1013 }
1014
1015 /* Assumes caller has the sk->rx_queue.lock */
1016 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1017                                             struct sk_buff *skb,
1018                                                 int status,
1019                                             unsigned int len
1020                                             )
1021 {
1022         struct tpacket_kbdq_core *pkc;
1023         struct tpacket_block_desc *pbd;
1024         char *curr, *end;
1025
1026         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1027         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1028
1029         /* Queue is frozen when user space is lagging behind */
1030         if (prb_queue_frozen(pkc)) {
1031                 /*
1032                  * Check if that last block which caused the queue to freeze,
1033                  * is still in_use by user-space.
1034                  */
1035                 if (prb_curr_blk_in_use(pkc, pbd)) {
1036                         /* Can't record this packet */
1037                         return NULL;
1038                 } else {
1039                         /*
1040                          * Ok, the block was released by user-space.
1041                          * Now let's open that block.
1042                          * opening a block also thaws the queue.
1043                          * Thawing is a side effect.
1044                          */
1045                         prb_open_block(pkc, pbd);
1046                 }
1047         }
1048
1049         smp_mb();
1050         curr = pkc->nxt_offset;
1051         pkc->skb = skb;
1052         end = (char *)pbd + pkc->kblk_size;
1053
1054         /* first try the current block */
1055         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1056                 prb_fill_curr_block(curr, pkc, pbd, len);
1057                 return (void *)curr;
1058         }
1059
1060         /* Ok, close the current block */
1061         prb_retire_current_block(pkc, po, 0);
1062
1063         /* Now, try to dispatch the next block */
1064         curr = (char *)prb_dispatch_next_block(pkc, po);
1065         if (curr) {
1066                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1067                 prb_fill_curr_block(curr, pkc, pbd, len);
1068                 return (void *)curr;
1069         }
1070
1071         /*
1072          * No free blocks are available.user_space hasn't caught up yet.
1073          * Queue was just frozen and now this packet will get dropped.
1074          */
1075         return NULL;
1076 }
1077
1078 static void *packet_current_rx_frame(struct packet_sock *po,
1079                                             struct sk_buff *skb,
1080                                             int status, unsigned int len)
1081 {
1082         char *curr = NULL;
1083         switch (po->tp_version) {
1084         case TPACKET_V1:
1085         case TPACKET_V2:
1086                 curr = packet_lookup_frame(po, &po->rx_ring,
1087                                         po->rx_ring.head, status);
1088                 return curr;
1089         case TPACKET_V3:
1090                 return __packet_lookup_frame_in_block(po, skb, status, len);
1091         default:
1092                 WARN(1, "TPACKET version not supported\n");
1093                 BUG();
1094                 return NULL;
1095         }
1096 }
1097
1098 static void *prb_lookup_block(struct packet_sock *po,
1099                                      struct packet_ring_buffer *rb,
1100                                      unsigned int idx,
1101                                      int status)
1102 {
1103         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1104         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1105
1106         if (status != BLOCK_STATUS(pbd))
1107                 return NULL;
1108         return pbd;
1109 }
1110
1111 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1112 {
1113         unsigned int prev;
1114         if (rb->prb_bdqc.kactive_blk_num)
1115                 prev = rb->prb_bdqc.kactive_blk_num-1;
1116         else
1117                 prev = rb->prb_bdqc.knum_blocks-1;
1118         return prev;
1119 }
1120
1121 /* Assumes caller has held the rx_queue.lock */
1122 static void *__prb_previous_block(struct packet_sock *po,
1123                                          struct packet_ring_buffer *rb,
1124                                          int status)
1125 {
1126         unsigned int previous = prb_previous_blk_num(rb);
1127         return prb_lookup_block(po, rb, previous, status);
1128 }
1129
1130 static void *packet_previous_rx_frame(struct packet_sock *po,
1131                                              struct packet_ring_buffer *rb,
1132                                              int status)
1133 {
1134         if (po->tp_version <= TPACKET_V2)
1135                 return packet_previous_frame(po, rb, status);
1136
1137         return __prb_previous_block(po, rb, status);
1138 }
1139
1140 static void packet_increment_rx_head(struct packet_sock *po,
1141                                             struct packet_ring_buffer *rb)
1142 {
1143         switch (po->tp_version) {
1144         case TPACKET_V1:
1145         case TPACKET_V2:
1146                 return packet_increment_head(rb);
1147         case TPACKET_V3:
1148         default:
1149                 WARN(1, "TPACKET version not supported.\n");
1150                 BUG();
1151                 return;
1152         }
1153 }
1154
1155 static void *packet_previous_frame(struct packet_sock *po,
1156                 struct packet_ring_buffer *rb,
1157                 int status)
1158 {
1159         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1160         return packet_lookup_frame(po, rb, previous, status);
1161 }
1162
1163 static void packet_increment_head(struct packet_ring_buffer *buff)
1164 {
1165         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1166 }
1167
1168 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1169 {
1170         struct sock *sk = &po->sk;
1171         bool has_room;
1172
1173         if (po->prot_hook.func != tpacket_rcv)
1174                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1175                         <= sk->sk_rcvbuf;
1176
1177         spin_lock(&sk->sk_receive_queue.lock);
1178         if (po->tp_version == TPACKET_V3)
1179                 has_room = prb_lookup_block(po, &po->rx_ring,
1180                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1181                                             TP_STATUS_KERNEL);
1182         else
1183                 has_room = packet_lookup_frame(po, &po->rx_ring,
1184                                                po->rx_ring.head,
1185                                                TP_STATUS_KERNEL);
1186         spin_unlock(&sk->sk_receive_queue.lock);
1187
1188         return has_room;
1189 }
1190
1191 static void packet_sock_destruct(struct sock *sk)
1192 {
1193         skb_queue_purge(&sk->sk_error_queue);
1194
1195         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1196         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1197
1198         if (!sock_flag(sk, SOCK_DEAD)) {
1199                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1200                 return;
1201         }
1202
1203         sk_refcnt_debug_dec(sk);
1204 }
1205
1206 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1207 {
1208         int x = atomic_read(&f->rr_cur) + 1;
1209
1210         if (x >= num)
1211                 x = 0;
1212
1213         return x;
1214 }
1215
1216 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1217                                       struct sk_buff *skb,
1218                                       unsigned int num)
1219 {
1220         return reciprocal_divide(skb->rxhash, num);
1221 }
1222
1223 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1224                                     struct sk_buff *skb,
1225                                     unsigned int num)
1226 {
1227         int cur, old;
1228
1229         cur = atomic_read(&f->rr_cur);
1230         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1231                                      fanout_rr_next(f, num))) != cur)
1232                 cur = old;
1233         return cur;
1234 }
1235
1236 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1237                                      struct sk_buff *skb,
1238                                      unsigned int num)
1239 {
1240         return smp_processor_id() % num;
1241 }
1242
1243 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1244                                      struct sk_buff *skb,
1245                                      unsigned int num)
1246 {
1247         return reciprocal_divide(prandom_u32(), num);
1248 }
1249
1250 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1251                                           struct sk_buff *skb,
1252                                           unsigned int idx, unsigned int skip,
1253                                           unsigned int num)
1254 {
1255         unsigned int i, j;
1256
1257         i = j = min_t(int, f->next[idx], num - 1);
1258         do {
1259                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1260                         if (i != j)
1261                                 f->next[idx] = i;
1262                         return i;
1263                 }
1264                 if (++i == num)
1265                         i = 0;
1266         } while (i != j);
1267
1268         return idx;
1269 }
1270
1271 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1272 {
1273         return f->flags & (flag >> 8);
1274 }
1275
1276 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1277                              struct packet_type *pt, struct net_device *orig_dev)
1278 {
1279         struct packet_fanout *f = pt->af_packet_priv;
1280         unsigned int num = f->num_members;
1281         struct packet_sock *po;
1282         unsigned int idx;
1283
1284         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1285             !num) {
1286                 kfree_skb(skb);
1287                 return 0;
1288         }
1289
1290         switch (f->type) {
1291         case PACKET_FANOUT_HASH:
1292         default:
1293                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1294                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1295                         if (!skb)
1296                                 return 0;
1297                 }
1298                 skb_get_rxhash(skb);
1299                 idx = fanout_demux_hash(f, skb, num);
1300                 break;
1301         case PACKET_FANOUT_LB:
1302                 idx = fanout_demux_lb(f, skb, num);
1303                 break;
1304         case PACKET_FANOUT_CPU:
1305                 idx = fanout_demux_cpu(f, skb, num);
1306                 break;
1307         case PACKET_FANOUT_RND:
1308                 idx = fanout_demux_rnd(f, skb, num);
1309                 break;
1310         case PACKET_FANOUT_ROLLOVER:
1311                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1312                 break;
1313         }
1314
1315         po = pkt_sk(f->arr[idx]);
1316         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1317             unlikely(!packet_rcv_has_room(po, skb))) {
1318                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1319                 po = pkt_sk(f->arr[idx]);
1320         }
1321
1322         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1323 }
1324
1325 DEFINE_MUTEX(fanout_mutex);
1326 EXPORT_SYMBOL_GPL(fanout_mutex);
1327 static LIST_HEAD(fanout_list);
1328
1329 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1330 {
1331         struct packet_fanout *f = po->fanout;
1332
1333         spin_lock(&f->lock);
1334         f->arr[f->num_members] = sk;
1335         smp_wmb();
1336         f->num_members++;
1337         spin_unlock(&f->lock);
1338 }
1339
1340 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1341 {
1342         struct packet_fanout *f = po->fanout;
1343         int i;
1344
1345         spin_lock(&f->lock);
1346         for (i = 0; i < f->num_members; i++) {
1347                 if (f->arr[i] == sk)
1348                         break;
1349         }
1350         BUG_ON(i >= f->num_members);
1351         f->arr[i] = f->arr[f->num_members - 1];
1352         f->num_members--;
1353         spin_unlock(&f->lock);
1354 }
1355
1356 static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1357 {
1358         if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1359                 return true;
1360
1361         return false;
1362 }
1363
1364 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1365 {
1366         struct packet_sock *po = pkt_sk(sk);
1367         struct packet_fanout *f, *match;
1368         u8 type = type_flags & 0xff;
1369         u8 flags = type_flags >> 8;
1370         int err;
1371
1372         switch (type) {
1373         case PACKET_FANOUT_ROLLOVER:
1374                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1375                         return -EINVAL;
1376         case PACKET_FANOUT_HASH:
1377         case PACKET_FANOUT_LB:
1378         case PACKET_FANOUT_CPU:
1379         case PACKET_FANOUT_RND:
1380                 break;
1381         default:
1382                 return -EINVAL;
1383         }
1384
1385         if (!po->running)
1386                 return -EINVAL;
1387
1388         if (po->fanout)
1389                 return -EALREADY;
1390
1391         mutex_lock(&fanout_mutex);
1392         match = NULL;
1393         list_for_each_entry(f, &fanout_list, list) {
1394                 if (f->id == id &&
1395                     read_pnet(&f->net) == sock_net(sk)) {
1396                         match = f;
1397                         break;
1398                 }
1399         }
1400         err = -EINVAL;
1401         if (match && match->flags != flags)
1402                 goto out;
1403         if (!match) {
1404                 err = -ENOMEM;
1405                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1406                 if (!match)
1407                         goto out;
1408                 write_pnet(&match->net, sock_net(sk));
1409                 match->id = id;
1410                 match->type = type;
1411                 match->flags = flags;
1412                 atomic_set(&match->rr_cur, 0);
1413                 INIT_LIST_HEAD(&match->list);
1414                 spin_lock_init(&match->lock);
1415                 atomic_set(&match->sk_ref, 0);
1416                 match->prot_hook.type = po->prot_hook.type;
1417                 match->prot_hook.dev = po->prot_hook.dev;
1418                 match->prot_hook.func = packet_rcv_fanout;
1419                 match->prot_hook.af_packet_priv = match;
1420                 match->prot_hook.id_match = match_fanout_group;
1421                 dev_add_pack(&match->prot_hook);
1422                 list_add(&match->list, &fanout_list);
1423         }
1424         err = -EINVAL;
1425         if (match->type == type &&
1426             match->prot_hook.type == po->prot_hook.type &&
1427             match->prot_hook.dev == po->prot_hook.dev) {
1428                 err = -ENOSPC;
1429                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1430                         __dev_remove_pack(&po->prot_hook);
1431                         po->fanout = match;
1432                         atomic_inc(&match->sk_ref);
1433                         __fanout_link(sk, po);
1434                         err = 0;
1435                 }
1436         }
1437 out:
1438         mutex_unlock(&fanout_mutex);
1439         return err;
1440 }
1441
1442 static void fanout_release(struct sock *sk)
1443 {
1444         struct packet_sock *po = pkt_sk(sk);
1445         struct packet_fanout *f;
1446
1447         f = po->fanout;
1448         if (!f)
1449                 return;
1450
1451         mutex_lock(&fanout_mutex);
1452         po->fanout = NULL;
1453
1454         if (atomic_dec_and_test(&f->sk_ref)) {
1455                 list_del(&f->list);
1456                 dev_remove_pack(&f->prot_hook);
1457                 kfree(f);
1458         }
1459         mutex_unlock(&fanout_mutex);
1460 }
1461
1462 static const struct proto_ops packet_ops;
1463
1464 static const struct proto_ops packet_ops_spkt;
1465
1466 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1467                            struct packet_type *pt, struct net_device *orig_dev)
1468 {
1469         struct sock *sk;
1470         struct sockaddr_pkt *spkt;
1471
1472         /*
1473          *      When we registered the protocol we saved the socket in the data
1474          *      field for just this event.
1475          */
1476
1477         sk = pt->af_packet_priv;
1478
1479         /*
1480          *      Yank back the headers [hope the device set this
1481          *      right or kerboom...]
1482          *
1483          *      Incoming packets have ll header pulled,
1484          *      push it back.
1485          *
1486          *      For outgoing ones skb->data == skb_mac_header(skb)
1487          *      so that this procedure is noop.
1488          */
1489
1490         if (skb->pkt_type == PACKET_LOOPBACK)
1491                 goto out;
1492
1493         if (!net_eq(dev_net(dev), sock_net(sk)))
1494                 goto out;
1495
1496         skb = skb_share_check(skb, GFP_ATOMIC);
1497         if (skb == NULL)
1498                 goto oom;
1499
1500         /* drop any routing info */
1501         skb_dst_drop(skb);
1502
1503         /* drop conntrack reference */
1504         nf_reset(skb);
1505
1506         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1507
1508         skb_push(skb, skb->data - skb_mac_header(skb));
1509
1510         /*
1511          *      The SOCK_PACKET socket receives _all_ frames.
1512          */
1513
1514         spkt->spkt_family = dev->type;
1515         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1516         spkt->spkt_protocol = skb->protocol;
1517
1518         /*
1519          *      Charge the memory to the socket. This is done specifically
1520          *      to prevent sockets using all the memory up.
1521          */
1522
1523         if (sock_queue_rcv_skb(sk, skb) == 0)
1524                 return 0;
1525
1526 out:
1527         kfree_skb(skb);
1528 oom:
1529         return 0;
1530 }
1531
1532
1533 /*
1534  *      Output a raw packet to a device layer. This bypasses all the other
1535  *      protocol layers and you must therefore supply it with a complete frame
1536  */
1537
1538 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1539                                struct msghdr *msg, size_t len)
1540 {
1541         struct sock *sk = sock->sk;
1542         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1543         struct sk_buff *skb = NULL;
1544         struct net_device *dev;
1545         __be16 proto = 0;
1546         int err;
1547         int extra_len = 0;
1548
1549         /*
1550          *      Get and verify the address.
1551          */
1552
1553         if (saddr) {
1554                 if (msg->msg_namelen < sizeof(struct sockaddr))
1555                         return -EINVAL;
1556                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1557                         proto = saddr->spkt_protocol;
1558         } else
1559                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1560
1561         /*
1562          *      Find the device first to size check it
1563          */
1564
1565         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1566 retry:
1567         rcu_read_lock();
1568         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1569         err = -ENODEV;
1570         if (dev == NULL)
1571                 goto out_unlock;
1572
1573         err = -ENETDOWN;
1574         if (!(dev->flags & IFF_UP))
1575                 goto out_unlock;
1576
1577         /*
1578          * You may not queue a frame bigger than the mtu. This is the lowest level
1579          * raw protocol and you must do your own fragmentation at this level.
1580          */
1581
1582         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1583                 if (!netif_supports_nofcs(dev)) {
1584                         err = -EPROTONOSUPPORT;
1585                         goto out_unlock;
1586                 }
1587                 extra_len = 4; /* We're doing our own CRC */
1588         }
1589
1590         err = -EMSGSIZE;
1591         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1592                 goto out_unlock;
1593
1594         if (!skb) {
1595                 size_t reserved = LL_RESERVED_SPACE(dev);
1596                 int tlen = dev->needed_tailroom;
1597                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1598
1599                 rcu_read_unlock();
1600                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1601                 if (skb == NULL)
1602                         return -ENOBUFS;
1603                 /* FIXME: Save some space for broken drivers that write a hard
1604                  * header at transmission time by themselves. PPP is the notable
1605                  * one here. This should really be fixed at the driver level.
1606                  */
1607                 skb_reserve(skb, reserved);
1608                 skb_reset_network_header(skb);
1609
1610                 /* Try to align data part correctly */
1611                 if (hhlen) {
1612                         skb->data -= hhlen;
1613                         skb->tail -= hhlen;
1614                         if (len < hhlen)
1615                                 skb_reset_network_header(skb);
1616                 }
1617                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1618                 if (err)
1619                         goto out_free;
1620                 goto retry;
1621         }
1622
1623         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1624                 /* Earlier code assumed this would be a VLAN pkt,
1625                  * double-check this now that we have the actual
1626                  * packet in hand.
1627                  */
1628                 struct ethhdr *ehdr;
1629                 skb_reset_mac_header(skb);
1630                 ehdr = eth_hdr(skb);
1631                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1632                         err = -EMSGSIZE;
1633                         goto out_unlock;
1634                 }
1635         }
1636
1637         skb->protocol = proto;
1638         skb->dev = dev;
1639         skb->priority = sk->sk_priority;
1640         skb->mark = sk->sk_mark;
1641
1642         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1643
1644         if (unlikely(extra_len == 4))
1645                 skb->no_fcs = 1;
1646
1647         skb_probe_transport_header(skb, 0);
1648
1649         dev_queue_xmit(skb);
1650         rcu_read_unlock();
1651         return len;
1652
1653 out_unlock:
1654         rcu_read_unlock();
1655 out_free:
1656         kfree_skb(skb);
1657         return err;
1658 }
1659
1660 static unsigned int run_filter(const struct sk_buff *skb,
1661                                       const struct sock *sk,
1662                                       unsigned int res)
1663 {
1664         struct sk_filter *filter;
1665
1666         rcu_read_lock();
1667         filter = rcu_dereference(sk->sk_filter);
1668         if (filter != NULL)
1669                 res = SK_RUN_FILTER(filter, skb);
1670         rcu_read_unlock();
1671
1672         return res;
1673 }
1674
1675 /*
1676  * This function makes lazy skb cloning in hope that most of packets
1677  * are discarded by BPF.
1678  *
1679  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1680  * and skb->cb are mangled. It works because (and until) packets
1681  * falling here are owned by current CPU. Output packets are cloned
1682  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1683  * sequencially, so that if we return skb to original state on exit,
1684  * we will not harm anyone.
1685  */
1686
1687 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1688                       struct packet_type *pt, struct net_device *orig_dev)
1689 {
1690         struct sock *sk;
1691         struct sockaddr_ll *sll;
1692         struct packet_sock *po;
1693         u8 *skb_head = skb->data;
1694         int skb_len = skb->len;
1695         unsigned int snaplen, res;
1696
1697         if (skb->pkt_type == PACKET_LOOPBACK)
1698                 goto drop;
1699
1700         sk = pt->af_packet_priv;
1701         po = pkt_sk(sk);
1702
1703         if (!net_eq(dev_net(dev), sock_net(sk)))
1704                 goto drop;
1705
1706         skb->dev = dev;
1707
1708         if (dev->header_ops) {
1709                 /* The device has an explicit notion of ll header,
1710                  * exported to higher levels.
1711                  *
1712                  * Otherwise, the device hides details of its frame
1713                  * structure, so that corresponding packet head is
1714                  * never delivered to user.
1715                  */
1716                 if (sk->sk_type != SOCK_DGRAM)
1717                         skb_push(skb, skb->data - skb_mac_header(skb));
1718                 else if (skb->pkt_type == PACKET_OUTGOING) {
1719                         /* Special case: outgoing packets have ll header at head */
1720                         skb_pull(skb, skb_network_offset(skb));
1721                 }
1722         }
1723
1724         snaplen = skb->len;
1725
1726         res = run_filter(skb, sk, snaplen);
1727         if (!res)
1728                 goto drop_n_restore;
1729         if (snaplen > res)
1730                 snaplen = res;
1731
1732         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1733                 goto drop_n_acct;
1734
1735         if (skb_shared(skb)) {
1736                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1737                 if (nskb == NULL)
1738                         goto drop_n_acct;
1739
1740                 if (skb_head != skb->data) {
1741                         skb->data = skb_head;
1742                         skb->len = skb_len;
1743                 }
1744                 consume_skb(skb);
1745                 skb = nskb;
1746         }
1747
1748         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1749                      sizeof(skb->cb));
1750
1751         sll = &PACKET_SKB_CB(skb)->sa.ll;
1752         sll->sll_family = AF_PACKET;
1753         sll->sll_hatype = dev->type;
1754         sll->sll_protocol = skb->protocol;
1755         sll->sll_pkttype = skb->pkt_type;
1756         if (unlikely(po->origdev))
1757                 sll->sll_ifindex = orig_dev->ifindex;
1758         else
1759                 sll->sll_ifindex = dev->ifindex;
1760
1761         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1762
1763         PACKET_SKB_CB(skb)->origlen = skb->len;
1764
1765         if (pskb_trim(skb, snaplen))
1766                 goto drop_n_acct;
1767
1768         skb_set_owner_r(skb, sk);
1769         skb->dev = NULL;
1770         skb_dst_drop(skb);
1771
1772         /* drop conntrack reference */
1773         nf_reset(skb);
1774
1775         spin_lock(&sk->sk_receive_queue.lock);
1776         po->stats.stats1.tp_packets++;
1777         skb->dropcount = atomic_read(&sk->sk_drops);
1778         __skb_queue_tail(&sk->sk_receive_queue, skb);
1779         spin_unlock(&sk->sk_receive_queue.lock);
1780         sk->sk_data_ready(sk, skb->len);
1781         return 0;
1782
1783 drop_n_acct:
1784         spin_lock(&sk->sk_receive_queue.lock);
1785         po->stats.stats1.tp_drops++;
1786         atomic_inc(&sk->sk_drops);
1787         spin_unlock(&sk->sk_receive_queue.lock);
1788
1789 drop_n_restore:
1790         if (skb_head != skb->data && skb_shared(skb)) {
1791                 skb->data = skb_head;
1792                 skb->len = skb_len;
1793         }
1794 drop:
1795         consume_skb(skb);
1796         return 0;
1797 }
1798
1799 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1800                        struct packet_type *pt, struct net_device *orig_dev)
1801 {
1802         struct sock *sk;
1803         struct packet_sock *po;
1804         struct sockaddr_ll *sll;
1805         union tpacket_uhdr h;
1806         u8 *skb_head = skb->data;
1807         int skb_len = skb->len;
1808         unsigned int snaplen, res;
1809         unsigned long status = TP_STATUS_USER;
1810         unsigned short macoff, netoff, hdrlen;
1811         struct sk_buff *copy_skb = NULL;
1812         struct timespec ts;
1813         __u32 ts_status;
1814
1815         if (skb->pkt_type == PACKET_LOOPBACK)
1816                 goto drop;
1817
1818         sk = pt->af_packet_priv;
1819         po = pkt_sk(sk);
1820
1821         if (!net_eq(dev_net(dev), sock_net(sk)))
1822                 goto drop;
1823
1824         if (dev->header_ops) {
1825                 if (sk->sk_type != SOCK_DGRAM)
1826                         skb_push(skb, skb->data - skb_mac_header(skb));
1827                 else if (skb->pkt_type == PACKET_OUTGOING) {
1828                         /* Special case: outgoing packets have ll header at head */
1829                         skb_pull(skb, skb_network_offset(skb));
1830                 }
1831         }
1832
1833         if (skb->ip_summed == CHECKSUM_PARTIAL)
1834                 status |= TP_STATUS_CSUMNOTREADY;
1835
1836         snaplen = skb->len;
1837
1838         res = run_filter(skb, sk, snaplen);
1839         if (!res)
1840                 goto drop_n_restore;
1841         if (snaplen > res)
1842                 snaplen = res;
1843
1844         if (sk->sk_type == SOCK_DGRAM) {
1845                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1846                                   po->tp_reserve;
1847         } else {
1848                 unsigned int maclen = skb_network_offset(skb);
1849                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1850                                        (maclen < 16 ? 16 : maclen)) +
1851                         po->tp_reserve;
1852                 macoff = netoff - maclen;
1853         }
1854         if (po->tp_version <= TPACKET_V2) {
1855                 if (macoff + snaplen > po->rx_ring.frame_size) {
1856                         if (po->copy_thresh &&
1857                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1858                                 if (skb_shared(skb)) {
1859                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1860                                 } else {
1861                                         copy_skb = skb_get(skb);
1862                                         skb_head = skb->data;
1863                                 }
1864                                 if (copy_skb)
1865                                         skb_set_owner_r(copy_skb, sk);
1866                         }
1867                         snaplen = po->rx_ring.frame_size - macoff;
1868                         if ((int)snaplen < 0)
1869                                 snaplen = 0;
1870                 }
1871         }
1872         spin_lock(&sk->sk_receive_queue.lock);
1873         h.raw = packet_current_rx_frame(po, skb,
1874                                         TP_STATUS_KERNEL, (macoff+snaplen));
1875         if (!h.raw)
1876                 goto ring_is_full;
1877         if (po->tp_version <= TPACKET_V2) {
1878                 packet_increment_rx_head(po, &po->rx_ring);
1879         /*
1880          * LOSING will be reported till you read the stats,
1881          * because it's COR - Clear On Read.
1882          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1883          * at packet level.
1884          */
1885                 if (po->stats.stats1.tp_drops)
1886                         status |= TP_STATUS_LOSING;
1887         }
1888         po->stats.stats1.tp_packets++;
1889         if (copy_skb) {
1890                 status |= TP_STATUS_COPY;
1891                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1892         }
1893         spin_unlock(&sk->sk_receive_queue.lock);
1894
1895         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1896
1897         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1898                 getnstimeofday(&ts);
1899
1900         status |= ts_status;
1901
1902         switch (po->tp_version) {
1903         case TPACKET_V1:
1904                 h.h1->tp_len = skb->len;
1905                 h.h1->tp_snaplen = snaplen;
1906                 h.h1->tp_mac = macoff;
1907                 h.h1->tp_net = netoff;
1908                 h.h1->tp_sec = ts.tv_sec;
1909                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1910                 hdrlen = sizeof(*h.h1);
1911                 break;
1912         case TPACKET_V2:
1913                 h.h2->tp_len = skb->len;
1914                 h.h2->tp_snaplen = snaplen;
1915                 h.h2->tp_mac = macoff;
1916                 h.h2->tp_net = netoff;
1917                 h.h2->tp_sec = ts.tv_sec;
1918                 h.h2->tp_nsec = ts.tv_nsec;
1919                 if (vlan_tx_tag_present(skb)) {
1920                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1921                         status |= TP_STATUS_VLAN_VALID;
1922                 } else {
1923                         h.h2->tp_vlan_tci = 0;
1924                 }
1925                 h.h2->tp_padding = 0;
1926                 hdrlen = sizeof(*h.h2);
1927                 break;
1928         case TPACKET_V3:
1929                 /* tp_nxt_offset,vlan are already populated above.
1930                  * So DONT clear those fields here
1931                  */
1932                 h.h3->tp_status |= status;
1933                 h.h3->tp_len = skb->len;
1934                 h.h3->tp_snaplen = snaplen;
1935                 h.h3->tp_mac = macoff;
1936                 h.h3->tp_net = netoff;
1937                 h.h3->tp_sec  = ts.tv_sec;
1938                 h.h3->tp_nsec = ts.tv_nsec;
1939                 hdrlen = sizeof(*h.h3);
1940                 break;
1941         default:
1942                 BUG();
1943         }
1944
1945         sll = h.raw + TPACKET_ALIGN(hdrlen);
1946         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1947         sll->sll_family = AF_PACKET;
1948         sll->sll_hatype = dev->type;
1949         sll->sll_protocol = skb->protocol;
1950         sll->sll_pkttype = skb->pkt_type;
1951         if (unlikely(po->origdev))
1952                 sll->sll_ifindex = orig_dev->ifindex;
1953         else
1954                 sll->sll_ifindex = dev->ifindex;
1955
1956         smp_mb();
1957 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1958         {
1959                 u8 *start, *end;
1960
1961                 if (po->tp_version <= TPACKET_V2) {
1962                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1963                                 + macoff + snaplen);
1964                         for (start = h.raw; start < end; start += PAGE_SIZE)
1965                                 flush_dcache_page(pgv_to_page(start));
1966                 }
1967                 smp_wmb();
1968         }
1969 #endif
1970         if (po->tp_version <= TPACKET_V2)
1971                 __packet_set_status(po, h.raw, status);
1972         else
1973                 prb_clear_blk_fill_status(&po->rx_ring);
1974
1975         sk->sk_data_ready(sk, 0);
1976
1977 drop_n_restore:
1978         if (skb_head != skb->data && skb_shared(skb)) {
1979                 skb->data = skb_head;
1980                 skb->len = skb_len;
1981         }
1982 drop:
1983         kfree_skb(skb);
1984         return 0;
1985
1986 ring_is_full:
1987         po->stats.stats1.tp_drops++;
1988         spin_unlock(&sk->sk_receive_queue.lock);
1989
1990         sk->sk_data_ready(sk, 0);
1991         kfree_skb(copy_skb);
1992         goto drop_n_restore;
1993 }
1994
1995 static void tpacket_destruct_skb(struct sk_buff *skb)
1996 {
1997         struct packet_sock *po = pkt_sk(skb->sk);
1998         void *ph;
1999
2000         if (likely(po->tx_ring.pg_vec)) {
2001                 __u32 ts;
2002
2003                 ph = skb_shinfo(skb)->destructor_arg;
2004                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
2005                 atomic_dec(&po->tx_ring.pending);
2006
2007                 ts = __packet_set_timestamp(po, ph, skb);
2008                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2009         }
2010
2011         sock_wfree(skb);
2012 }
2013
2014 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2015                 void *frame, struct net_device *dev, int size_max,
2016                 __be16 proto, unsigned char *addr, int hlen)
2017 {
2018         union tpacket_uhdr ph;
2019         int to_write, offset, len, tp_len, nr_frags, len_max;
2020         struct socket *sock = po->sk.sk_socket;
2021         struct page *page;
2022         void *data;
2023         int err;
2024
2025         ph.raw = frame;
2026
2027         skb->protocol = proto;
2028         skb->dev = dev;
2029         skb->priority = po->sk.sk_priority;
2030         skb->mark = po->sk.sk_mark;
2031         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2032         skb_shinfo(skb)->destructor_arg = ph.raw;
2033
2034         switch (po->tp_version) {
2035         case TPACKET_V2:
2036                 tp_len = ph.h2->tp_len;
2037                 break;
2038         default:
2039                 tp_len = ph.h1->tp_len;
2040                 break;
2041         }
2042         if (unlikely(tp_len > size_max)) {
2043                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2044                 return -EMSGSIZE;
2045         }
2046
2047         skb_reserve(skb, hlen);
2048         skb_reset_network_header(skb);
2049
2050         if (!packet_use_direct_xmit(po))
2051                 skb_probe_transport_header(skb, 0);
2052         if (unlikely(po->tp_tx_has_off)) {
2053                 int off_min, off_max, off;
2054                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2055                 off_max = po->tx_ring.frame_size - tp_len;
2056                 if (sock->type == SOCK_DGRAM) {
2057                         switch (po->tp_version) {
2058                         case TPACKET_V2:
2059                                 off = ph.h2->tp_net;
2060                                 break;
2061                         default:
2062                                 off = ph.h1->tp_net;
2063                                 break;
2064                         }
2065                 } else {
2066                         switch (po->tp_version) {
2067                         case TPACKET_V2:
2068                                 off = ph.h2->tp_mac;
2069                                 break;
2070                         default:
2071                                 off = ph.h1->tp_mac;
2072                                 break;
2073                         }
2074                 }
2075                 if (unlikely((off < off_min) || (off_max < off)))
2076                         return -EINVAL;
2077                 data = ph.raw + off;
2078         } else {
2079                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2080         }
2081         to_write = tp_len;
2082
2083         if (sock->type == SOCK_DGRAM) {
2084                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2085                                 NULL, tp_len);
2086                 if (unlikely(err < 0))
2087                         return -EINVAL;
2088         } else if (dev->hard_header_len) {
2089                 /* net device doesn't like empty head */
2090                 if (unlikely(tp_len <= dev->hard_header_len)) {
2091                         pr_err("packet size is too short (%d < %d)\n",
2092                                tp_len, dev->hard_header_len);
2093                         return -EINVAL;
2094                 }
2095
2096                 skb_push(skb, dev->hard_header_len);
2097                 err = skb_store_bits(skb, 0, data,
2098                                 dev->hard_header_len);
2099                 if (unlikely(err))
2100                         return err;
2101
2102                 data += dev->hard_header_len;
2103                 to_write -= dev->hard_header_len;
2104         }
2105
2106         offset = offset_in_page(data);
2107         len_max = PAGE_SIZE - offset;
2108         len = ((to_write > len_max) ? len_max : to_write);
2109
2110         skb->data_len = to_write;
2111         skb->len += to_write;
2112         skb->truesize += to_write;
2113         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2114
2115         while (likely(to_write)) {
2116                 nr_frags = skb_shinfo(skb)->nr_frags;
2117
2118                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2119                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2120                                MAX_SKB_FRAGS);
2121                         return -EFAULT;
2122                 }
2123
2124                 page = pgv_to_page(data);
2125                 data += len;
2126                 flush_dcache_page(page);
2127                 get_page(page);
2128                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2129                 to_write -= len;
2130                 offset = 0;
2131                 len_max = PAGE_SIZE;
2132                 len = ((to_write > len_max) ? len_max : to_write);
2133         }
2134
2135         return tp_len;
2136 }
2137
2138 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2139 {
2140         struct sk_buff *skb;
2141         struct net_device *dev;
2142         __be16 proto;
2143         int err, reserve = 0;
2144         void *ph;
2145         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2146         int tp_len, size_max;
2147         unsigned char *addr;
2148         int len_sum = 0;
2149         int status = TP_STATUS_AVAILABLE;
2150         int hlen, tlen;
2151
2152         mutex_lock(&po->pg_vec_lock);
2153
2154         if (likely(saddr == NULL)) {
2155                 dev     = packet_cached_dev_get(po);
2156                 proto   = po->num;
2157                 addr    = NULL;
2158         } else {
2159                 err = -EINVAL;
2160                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2161                         goto out;
2162                 if (msg->msg_namelen < (saddr->sll_halen
2163                                         + offsetof(struct sockaddr_ll,
2164                                                 sll_addr)))
2165                         goto out;
2166                 proto   = saddr->sll_protocol;
2167                 addr    = saddr->sll_addr;
2168                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2169         }
2170
2171         err = -ENXIO;
2172         if (unlikely(dev == NULL))
2173                 goto out;
2174         err = -ENETDOWN;
2175         if (unlikely(!(dev->flags & IFF_UP)))
2176                 goto out_put;
2177
2178         reserve = dev->hard_header_len;
2179
2180         size_max = po->tx_ring.frame_size
2181                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2182
2183         if (size_max > dev->mtu + reserve)
2184                 size_max = dev->mtu + reserve;
2185
2186         do {
2187                 ph = packet_current_frame(po, &po->tx_ring,
2188                                 TP_STATUS_SEND_REQUEST);
2189
2190                 if (unlikely(ph == NULL)) {
2191                         schedule();
2192                         continue;
2193                 }
2194
2195                 status = TP_STATUS_SEND_REQUEST;
2196                 hlen = LL_RESERVED_SPACE(dev);
2197                 tlen = dev->needed_tailroom;
2198                 skb = sock_alloc_send_skb(&po->sk,
2199                                 hlen + tlen + sizeof(struct sockaddr_ll),
2200                                 0, &err);
2201
2202                 if (unlikely(skb == NULL))
2203                         goto out_status;
2204
2205                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2206                                 addr, hlen);
2207
2208                 if (unlikely(tp_len < 0)) {
2209                         if (po->tp_loss) {
2210                                 __packet_set_status(po, ph,
2211                                                 TP_STATUS_AVAILABLE);
2212                                 packet_increment_head(&po->tx_ring);
2213                                 kfree_skb(skb);
2214                                 continue;
2215                         } else {
2216                                 status = TP_STATUS_WRONG_FORMAT;
2217                                 err = tp_len;
2218                                 goto out_status;
2219                         }
2220                 }
2221
2222                 skb_set_queue_mapping(skb, packet_pick_tx_queue(dev));
2223                 skb->destructor = tpacket_destruct_skb;
2224                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2225                 atomic_inc(&po->tx_ring.pending);
2226
2227                 status = TP_STATUS_SEND_REQUEST;
2228                 err = po->xmit(skb);
2229                 if (unlikely(err > 0)) {
2230                         err = net_xmit_errno(err);
2231                         if (err && __packet_get_status(po, ph) ==
2232                                    TP_STATUS_AVAILABLE) {
2233                                 /* skb was destructed already */
2234                                 skb = NULL;
2235                                 goto out_status;
2236                         }
2237                         /*
2238                          * skb was dropped but not destructed yet;
2239                          * let's treat it like congestion or err < 0
2240                          */
2241                         err = 0;
2242                 }
2243                 packet_increment_head(&po->tx_ring);
2244                 len_sum += tp_len;
2245         } while (likely((ph != NULL) ||
2246                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2247                          (atomic_read(&po->tx_ring.pending))))
2248                 );
2249
2250         err = len_sum;
2251         goto out_put;
2252
2253 out_status:
2254         __packet_set_status(po, ph, status);
2255         kfree_skb(skb);
2256 out_put:
2257         dev_put(dev);
2258 out:
2259         mutex_unlock(&po->pg_vec_lock);
2260         return err;
2261 }
2262
2263 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2264                                         size_t reserve, size_t len,
2265                                         size_t linear, int noblock,
2266                                         int *err)
2267 {
2268         struct sk_buff *skb;
2269
2270         /* Under a page?  Don't bother with paged skb. */
2271         if (prepad + len < PAGE_SIZE || !linear)
2272                 linear = len;
2273
2274         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2275                                    err, 0);
2276         if (!skb)
2277                 return NULL;
2278
2279         skb_reserve(skb, reserve);
2280         skb_put(skb, linear);
2281         skb->data_len = len - linear;
2282         skb->len += len - linear;
2283
2284         return skb;
2285 }
2286
2287 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2288 {
2289         struct sock *sk = sock->sk;
2290         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2291         struct sk_buff *skb;
2292         struct net_device *dev;
2293         __be16 proto;
2294         unsigned char *addr;
2295         int err, reserve = 0;
2296         struct virtio_net_hdr vnet_hdr = { 0 };
2297         int offset = 0;
2298         int vnet_hdr_len;
2299         struct packet_sock *po = pkt_sk(sk);
2300         unsigned short gso_type = 0;
2301         int hlen, tlen;
2302         int extra_len = 0;
2303
2304         /*
2305          *      Get and verify the address.
2306          */
2307
2308         if (likely(saddr == NULL)) {
2309                 dev     = packet_cached_dev_get(po);
2310                 proto   = po->num;
2311                 addr    = NULL;
2312         } else {
2313                 err = -EINVAL;
2314                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2315                         goto out;
2316                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2317                         goto out;
2318                 proto   = saddr->sll_protocol;
2319                 addr    = saddr->sll_addr;
2320                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2321         }
2322
2323         err = -ENXIO;
2324         if (unlikely(dev == NULL))
2325                 goto out_unlock;
2326         err = -ENETDOWN;
2327         if (unlikely(!(dev->flags & IFF_UP)))
2328                 goto out_unlock;
2329
2330         if (sock->type == SOCK_RAW)
2331                 reserve = dev->hard_header_len;
2332         if (po->has_vnet_hdr) {
2333                 vnet_hdr_len = sizeof(vnet_hdr);
2334
2335                 err = -EINVAL;
2336                 if (len < vnet_hdr_len)
2337                         goto out_unlock;
2338
2339                 len -= vnet_hdr_len;
2340
2341                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2342                                        vnet_hdr_len);
2343                 if (err < 0)
2344                         goto out_unlock;
2345
2346                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2347                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2348                       vnet_hdr.hdr_len))
2349                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2350                                                  vnet_hdr.csum_offset + 2;
2351
2352                 err = -EINVAL;
2353                 if (vnet_hdr.hdr_len > len)
2354                         goto out_unlock;
2355
2356                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2357                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2358                         case VIRTIO_NET_HDR_GSO_TCPV4:
2359                                 gso_type = SKB_GSO_TCPV4;
2360                                 break;
2361                         case VIRTIO_NET_HDR_GSO_TCPV6:
2362                                 gso_type = SKB_GSO_TCPV6;
2363                                 break;
2364                         case VIRTIO_NET_HDR_GSO_UDP:
2365                                 gso_type = SKB_GSO_UDP;
2366                                 break;
2367                         default:
2368                                 goto out_unlock;
2369                         }
2370
2371                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2372                                 gso_type |= SKB_GSO_TCP_ECN;
2373
2374                         if (vnet_hdr.gso_size == 0)
2375                                 goto out_unlock;
2376
2377                 }
2378         }
2379
2380         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2381                 if (!netif_supports_nofcs(dev)) {
2382                         err = -EPROTONOSUPPORT;
2383                         goto out_unlock;
2384                 }
2385                 extra_len = 4; /* We're doing our own CRC */
2386         }
2387
2388         err = -EMSGSIZE;
2389         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2390                 goto out_unlock;
2391
2392         err = -ENOBUFS;
2393         hlen = LL_RESERVED_SPACE(dev);
2394         tlen = dev->needed_tailroom;
2395         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2396                                msg->msg_flags & MSG_DONTWAIT, &err);
2397         if (skb == NULL)
2398                 goto out_unlock;
2399
2400         skb_set_network_header(skb, reserve);
2401
2402         err = -EINVAL;
2403         if (sock->type == SOCK_DGRAM &&
2404             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2405                 goto out_free;
2406
2407         /* Returns -EFAULT on error */
2408         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2409         if (err)
2410                 goto out_free;
2411
2412         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2413
2414         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2415                 /* Earlier code assumed this would be a VLAN pkt,
2416                  * double-check this now that we have the actual
2417                  * packet in hand.
2418                  */
2419                 struct ethhdr *ehdr;
2420                 skb_reset_mac_header(skb);
2421                 ehdr = eth_hdr(skb);
2422                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2423                         err = -EMSGSIZE;
2424                         goto out_free;
2425                 }
2426         }
2427
2428         skb->protocol = proto;
2429         skb->dev = dev;
2430         skb->priority = sk->sk_priority;
2431         skb->mark = sk->sk_mark;
2432         skb_set_queue_mapping(skb, packet_pick_tx_queue(dev));
2433
2434         if (po->has_vnet_hdr) {
2435                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2436                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2437                                                   vnet_hdr.csum_offset)) {
2438                                 err = -EINVAL;
2439                                 goto out_free;
2440                         }
2441                 }
2442
2443                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2444                 skb_shinfo(skb)->gso_type = gso_type;
2445
2446                 /* Header must be checked, and gso_segs computed. */
2447                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2448                 skb_shinfo(skb)->gso_segs = 0;
2449
2450                 len += vnet_hdr_len;
2451         }
2452
2453         if (!packet_use_direct_xmit(po))
2454                 skb_probe_transport_header(skb, reserve);
2455         if (unlikely(extra_len == 4))
2456                 skb->no_fcs = 1;
2457
2458         err = po->xmit(skb);
2459         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2460                 goto out_unlock;
2461
2462         dev_put(dev);
2463
2464         return len;
2465
2466 out_free:
2467         kfree_skb(skb);
2468 out_unlock:
2469         if (dev)
2470                 dev_put(dev);
2471 out:
2472         return err;
2473 }
2474
2475 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2476                 struct msghdr *msg, size_t len)
2477 {
2478         struct sock *sk = sock->sk;
2479         struct packet_sock *po = pkt_sk(sk);
2480
2481         if (po->tx_ring.pg_vec)
2482                 return tpacket_snd(po, msg);
2483         else
2484                 return packet_snd(sock, msg, len);
2485 }
2486
2487 /*
2488  *      Close a PACKET socket. This is fairly simple. We immediately go
2489  *      to 'closed' state and remove our protocol entry in the device list.
2490  */
2491
2492 static int packet_release(struct socket *sock)
2493 {
2494         struct sock *sk = sock->sk;
2495         struct packet_sock *po;
2496         struct net *net;
2497         union tpacket_req_u req_u;
2498
2499         if (!sk)
2500                 return 0;
2501
2502         net = sock_net(sk);
2503         po = pkt_sk(sk);
2504
2505         mutex_lock(&net->packet.sklist_lock);
2506         sk_del_node_init_rcu(sk);
2507         mutex_unlock(&net->packet.sklist_lock);
2508
2509         preempt_disable();
2510         sock_prot_inuse_add(net, sk->sk_prot, -1);
2511         preempt_enable();
2512
2513         spin_lock(&po->bind_lock);
2514         unregister_prot_hook(sk, false);
2515         packet_cached_dev_reset(po);
2516
2517         if (po->prot_hook.dev) {
2518                 dev_put(po->prot_hook.dev);
2519                 po->prot_hook.dev = NULL;
2520         }
2521         spin_unlock(&po->bind_lock);
2522
2523         packet_flush_mclist(sk);
2524
2525         if (po->rx_ring.pg_vec) {
2526                 memset(&req_u, 0, sizeof(req_u));
2527                 packet_set_ring(sk, &req_u, 1, 0);
2528         }
2529
2530         if (po->tx_ring.pg_vec) {
2531                 memset(&req_u, 0, sizeof(req_u));
2532                 packet_set_ring(sk, &req_u, 1, 1);
2533         }
2534
2535         fanout_release(sk);
2536
2537         synchronize_net();
2538         /*
2539          *      Now the socket is dead. No more input will appear.
2540          */
2541         sock_orphan(sk);
2542         sock->sk = NULL;
2543
2544         /* Purge queues */
2545
2546         skb_queue_purge(&sk->sk_receive_queue);
2547         sk_refcnt_debug_release(sk);
2548
2549         sock_put(sk);
2550         return 0;
2551 }
2552
2553 /*
2554  *      Attach a packet hook.
2555  */
2556
2557 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2558 {
2559         struct packet_sock *po = pkt_sk(sk);
2560
2561         if (po->fanout) {
2562                 if (dev)
2563                         dev_put(dev);
2564
2565                 return -EINVAL;
2566         }
2567
2568         lock_sock(sk);
2569
2570         spin_lock(&po->bind_lock);
2571         unregister_prot_hook(sk, true);
2572
2573         po->num = protocol;
2574         po->prot_hook.type = protocol;
2575         if (po->prot_hook.dev)
2576                 dev_put(po->prot_hook.dev);
2577
2578         po->prot_hook.dev = dev;
2579         po->ifindex = dev ? dev->ifindex : 0;
2580
2581         packet_cached_dev_assign(po, dev);
2582
2583         if (protocol == 0)
2584                 goto out_unlock;
2585
2586         if (!dev || (dev->flags & IFF_UP)) {
2587                 register_prot_hook(sk);
2588         } else {
2589                 sk->sk_err = ENETDOWN;
2590                 if (!sock_flag(sk, SOCK_DEAD))
2591                         sk->sk_error_report(sk);
2592         }
2593
2594 out_unlock:
2595         spin_unlock(&po->bind_lock);
2596         release_sock(sk);
2597         return 0;
2598 }
2599
2600 /*
2601  *      Bind a packet socket to a device
2602  */
2603
2604 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2605                             int addr_len)
2606 {
2607         struct sock *sk = sock->sk;
2608         char name[15];
2609         struct net_device *dev;
2610         int err = -ENODEV;
2611
2612         /*
2613          *      Check legality
2614          */
2615
2616         if (addr_len != sizeof(struct sockaddr))
2617                 return -EINVAL;
2618         strlcpy(name, uaddr->sa_data, sizeof(name));
2619
2620         dev = dev_get_by_name(sock_net(sk), name);
2621         if (dev)
2622                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2623         return err;
2624 }
2625
2626 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2627 {
2628         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2629         struct sock *sk = sock->sk;
2630         struct net_device *dev = NULL;
2631         int err;
2632
2633
2634         /*
2635          *      Check legality
2636          */
2637
2638         if (addr_len < sizeof(struct sockaddr_ll))
2639                 return -EINVAL;
2640         if (sll->sll_family != AF_PACKET)
2641                 return -EINVAL;
2642
2643         if (sll->sll_ifindex) {
2644                 err = -ENODEV;
2645                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2646                 if (dev == NULL)
2647                         goto out;
2648         }
2649         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2650
2651 out:
2652         return err;
2653 }
2654
2655 static struct proto packet_proto = {
2656         .name     = "PACKET",
2657         .owner    = THIS_MODULE,
2658         .obj_size = sizeof(struct packet_sock),
2659 };
2660
2661 /*
2662  *      Create a packet of type SOCK_PACKET.
2663  */
2664
2665 static int packet_create(struct net *net, struct socket *sock, int protocol,
2666                          int kern)
2667 {
2668         struct sock *sk;
2669         struct packet_sock *po;
2670         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2671         int err;
2672
2673         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2674                 return -EPERM;
2675         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2676             sock->type != SOCK_PACKET)
2677                 return -ESOCKTNOSUPPORT;
2678
2679         sock->state = SS_UNCONNECTED;
2680
2681         err = -ENOBUFS;
2682         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2683         if (sk == NULL)
2684                 goto out;
2685
2686         sock->ops = &packet_ops;
2687         if (sock->type == SOCK_PACKET)
2688                 sock->ops = &packet_ops_spkt;
2689
2690         sock_init_data(sock, sk);
2691
2692         po = pkt_sk(sk);
2693         sk->sk_family = PF_PACKET;
2694         po->num = proto;
2695         po->xmit = dev_queue_xmit;
2696
2697         packet_cached_dev_reset(po);
2698
2699         sk->sk_destruct = packet_sock_destruct;
2700         sk_refcnt_debug_inc(sk);
2701
2702         /*
2703          *      Attach a protocol block
2704          */
2705
2706         spin_lock_init(&po->bind_lock);
2707         mutex_init(&po->pg_vec_lock);
2708         po->prot_hook.func = packet_rcv;
2709
2710         if (sock->type == SOCK_PACKET)
2711                 po->prot_hook.func = packet_rcv_spkt;
2712
2713         po->prot_hook.af_packet_priv = sk;
2714
2715         if (proto) {
2716                 po->prot_hook.type = proto;
2717                 register_prot_hook(sk);
2718         }
2719
2720         mutex_lock(&net->packet.sklist_lock);
2721         sk_add_node_rcu(sk, &net->packet.sklist);
2722         mutex_unlock(&net->packet.sklist_lock);
2723
2724         preempt_disable();
2725         sock_prot_inuse_add(net, &packet_proto, 1);
2726         preempt_enable();
2727
2728         return 0;
2729 out:
2730         return err;
2731 }
2732
2733 /*
2734  *      Pull a packet from our receive queue and hand it to the user.
2735  *      If necessary we block.
2736  */
2737
2738 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2739                           struct msghdr *msg, size_t len, int flags)
2740 {
2741         struct sock *sk = sock->sk;
2742         struct sk_buff *skb;
2743         int copied, err;
2744         int vnet_hdr_len = 0;
2745
2746         err = -EINVAL;
2747         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2748                 goto out;
2749
2750 #if 0
2751         /* What error should we return now? EUNATTACH? */
2752         if (pkt_sk(sk)->ifindex < 0)
2753                 return -ENODEV;
2754 #endif
2755
2756         if (flags & MSG_ERRQUEUE) {
2757                 err = sock_recv_errqueue(sk, msg, len,
2758                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
2759                 goto out;
2760         }
2761
2762         /*
2763          *      Call the generic datagram receiver. This handles all sorts
2764          *      of horrible races and re-entrancy so we can forget about it
2765          *      in the protocol layers.
2766          *
2767          *      Now it will return ENETDOWN, if device have just gone down,
2768          *      but then it will block.
2769          */
2770
2771         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2772
2773         /*
2774          *      An error occurred so return it. Because skb_recv_datagram()
2775          *      handles the blocking we don't see and worry about blocking
2776          *      retries.
2777          */
2778
2779         if (skb == NULL)
2780                 goto out;
2781
2782         if (pkt_sk(sk)->has_vnet_hdr) {
2783                 struct virtio_net_hdr vnet_hdr = { 0 };
2784
2785                 err = -EINVAL;
2786                 vnet_hdr_len = sizeof(vnet_hdr);
2787                 if (len < vnet_hdr_len)
2788                         goto out_free;
2789
2790                 len -= vnet_hdr_len;
2791
2792                 if (skb_is_gso(skb)) {
2793                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2794
2795                         /* This is a hint as to how much should be linear. */
2796                         vnet_hdr.hdr_len = skb_headlen(skb);
2797                         vnet_hdr.gso_size = sinfo->gso_size;
2798                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2799                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2800                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2801                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2802                         else if (sinfo->gso_type & SKB_GSO_UDP)
2803                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2804                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2805                                 goto out_free;
2806                         else
2807                                 BUG();
2808                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2809                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2810                 } else
2811                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2812
2813                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2814                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2815                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2816                         vnet_hdr.csum_offset = skb->csum_offset;
2817                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2818                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2819                 } /* else everything is zero */
2820
2821                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2822                                      vnet_hdr_len);
2823                 if (err < 0)
2824                         goto out_free;
2825         }
2826
2827         /* You lose any data beyond the buffer you gave. If it worries
2828          * a user program they can ask the device for its MTU
2829          * anyway.
2830          */
2831         copied = skb->len;
2832         if (copied > len) {
2833                 copied = len;
2834                 msg->msg_flags |= MSG_TRUNC;
2835         }
2836
2837         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2838         if (err)
2839                 goto out_free;
2840
2841         sock_recv_ts_and_drops(msg, sk, skb);
2842
2843         if (msg->msg_name) {
2844                 /* If the address length field is there to be filled
2845                  * in, we fill it in now.
2846                  */
2847                 if (sock->type == SOCK_PACKET) {
2848                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2849                 } else {
2850                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2851                         msg->msg_namelen = sll->sll_halen +
2852                                 offsetof(struct sockaddr_ll, sll_addr);
2853                 }
2854                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2855                        msg->msg_namelen);
2856         }
2857
2858         if (pkt_sk(sk)->auxdata) {
2859                 struct tpacket_auxdata aux;
2860
2861                 aux.tp_status = TP_STATUS_USER;
2862                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2863                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2864                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2865                 aux.tp_snaplen = skb->len;
2866                 aux.tp_mac = 0;
2867                 aux.tp_net = skb_network_offset(skb);
2868                 if (vlan_tx_tag_present(skb)) {
2869                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2870                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2871                 } else {
2872                         aux.tp_vlan_tci = 0;
2873                 }
2874                 aux.tp_padding = 0;
2875                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2876         }
2877
2878         /*
2879          *      Free or return the buffer as appropriate. Again this
2880          *      hides all the races and re-entrancy issues from us.
2881          */
2882         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2883
2884 out_free:
2885         skb_free_datagram(sk, skb);
2886 out:
2887         return err;
2888 }
2889
2890 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2891                                int *uaddr_len, int peer)
2892 {
2893         struct net_device *dev;
2894         struct sock *sk = sock->sk;
2895
2896         if (peer)
2897                 return -EOPNOTSUPP;
2898
2899         uaddr->sa_family = AF_PACKET;
2900         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2901         rcu_read_lock();
2902         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2903         if (dev)
2904                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2905         rcu_read_unlock();
2906         *uaddr_len = sizeof(*uaddr);
2907
2908         return 0;
2909 }
2910
2911 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2912                           int *uaddr_len, int peer)
2913 {
2914         struct net_device *dev;
2915         struct sock *sk = sock->sk;
2916         struct packet_sock *po = pkt_sk(sk);
2917         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2918
2919         if (peer)
2920                 return -EOPNOTSUPP;
2921
2922         sll->sll_family = AF_PACKET;
2923         sll->sll_ifindex = po->ifindex;
2924         sll->sll_protocol = po->num;
2925         sll->sll_pkttype = 0;
2926         rcu_read_lock();
2927         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2928         if (dev) {
2929                 sll->sll_hatype = dev->type;
2930                 sll->sll_halen = dev->addr_len;
2931                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2932         } else {
2933                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2934                 sll->sll_halen = 0;
2935         }
2936         rcu_read_unlock();
2937         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2938
2939         return 0;
2940 }
2941
2942 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2943                          int what)
2944 {
2945         switch (i->type) {
2946         case PACKET_MR_MULTICAST:
2947                 if (i->alen != dev->addr_len)
2948                         return -EINVAL;
2949                 if (what > 0)
2950                         return dev_mc_add(dev, i->addr);
2951                 else
2952                         return dev_mc_del(dev, i->addr);
2953                 break;
2954         case PACKET_MR_PROMISC:
2955                 return dev_set_promiscuity(dev, what);
2956                 break;
2957         case PACKET_MR_ALLMULTI:
2958                 return dev_set_allmulti(dev, what);
2959                 break;
2960         case PACKET_MR_UNICAST:
2961                 if (i->alen != dev->addr_len)
2962                         return -EINVAL;
2963                 if (what > 0)
2964                         return dev_uc_add(dev, i->addr);
2965                 else
2966                         return dev_uc_del(dev, i->addr);
2967                 break;
2968         default:
2969                 break;
2970         }
2971         return 0;
2972 }
2973
2974 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2975 {
2976         for ( ; i; i = i->next) {
2977                 if (i->ifindex == dev->ifindex)
2978                         packet_dev_mc(dev, i, what);
2979         }
2980 }
2981
2982 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2983 {
2984         struct packet_sock *po = pkt_sk(sk);
2985         struct packet_mclist *ml, *i;
2986         struct net_device *dev;
2987         int err;
2988
2989         rtnl_lock();
2990
2991         err = -ENODEV;
2992         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2993         if (!dev)
2994                 goto done;
2995
2996         err = -EINVAL;
2997         if (mreq->mr_alen > dev->addr_len)
2998                 goto done;
2999
3000         err = -ENOBUFS;
3001         i = kmalloc(sizeof(*i), GFP_KERNEL);
3002         if (i == NULL)
3003                 goto done;
3004
3005         err = 0;
3006         for (ml = po->mclist; ml; ml = ml->next) {
3007                 if (ml->ifindex == mreq->mr_ifindex &&
3008                     ml->type == mreq->mr_type &&
3009                     ml->alen == mreq->mr_alen &&
3010                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3011                         ml->count++;
3012                         /* Free the new element ... */
3013                         kfree(i);
3014                         goto done;
3015                 }
3016         }
3017
3018         i->type = mreq->mr_type;
3019         i->ifindex = mreq->mr_ifindex;
3020         i->alen = mreq->mr_alen;
3021         memcpy(i->addr, mreq->mr_address, i->alen);
3022         i->count = 1;
3023         i->next = po->mclist;
3024         po->mclist = i;
3025         err = packet_dev_mc(dev, i, 1);
3026         if (err) {
3027                 po->mclist = i->next;
3028                 kfree(i);
3029         }
3030
3031 done:
3032         rtnl_unlock();
3033         return err;
3034 }
3035
3036 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3037 {
3038         struct packet_mclist *ml, **mlp;
3039
3040         rtnl_lock();
3041
3042         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3043                 if (ml->ifindex == mreq->mr_ifindex &&
3044                     ml->type == mreq->mr_type &&
3045                     ml->alen == mreq->mr_alen &&
3046                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3047                         if (--ml->count == 0) {
3048                                 struct net_device *dev;
3049                                 *mlp = ml->next;
3050                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3051                                 if (dev)
3052                                         packet_dev_mc(dev, ml, -1);
3053                                 kfree(ml);
3054                         }
3055                         rtnl_unlock();
3056                         return 0;
3057                 }
3058         }
3059         rtnl_unlock();
3060         return -EADDRNOTAVAIL;
3061 }
3062
3063 static void packet_flush_mclist(struct sock *sk)
3064 {
3065         struct packet_sock *po = pkt_sk(sk);
3066         struct packet_mclist *ml;
3067
3068         if (!po->mclist)
3069                 return;
3070
3071         rtnl_lock();
3072         while ((ml = po->mclist) != NULL) {
3073                 struct net_device *dev;
3074
3075                 po->mclist = ml->next;
3076                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3077                 if (dev != NULL)
3078                         packet_dev_mc(dev, ml, -1);
3079                 kfree(ml);
3080         }
3081         rtnl_unlock();
3082 }
3083
3084 static int
3085 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3086 {
3087         struct sock *sk = sock->sk;
3088         struct packet_sock *po = pkt_sk(sk);
3089         int ret;
3090
3091         if (level != SOL_PACKET)
3092                 return -ENOPROTOOPT;
3093
3094         switch (optname) {
3095         case PACKET_ADD_MEMBERSHIP:
3096         case PACKET_DROP_MEMBERSHIP:
3097         {
3098                 struct packet_mreq_max mreq;
3099                 int len = optlen;
3100                 memset(&mreq, 0, sizeof(mreq));
3101                 if (len < sizeof(struct packet_mreq))
3102                         return -EINVAL;
3103                 if (len > sizeof(mreq))
3104                         len = sizeof(mreq);
3105                 if (copy_from_user(&mreq, optval, len))
3106                         return -EFAULT;
3107                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3108                         return -EINVAL;
3109                 if (optname == PACKET_ADD_MEMBERSHIP)
3110                         ret = packet_mc_add(sk, &mreq);
3111                 else
3112                         ret = packet_mc_drop(sk, &mreq);
3113                 return ret;
3114         }
3115
3116         case PACKET_RX_RING:
3117         case PACKET_TX_RING:
3118         {
3119                 union tpacket_req_u req_u;
3120                 int len;
3121
3122                 switch (po->tp_version) {
3123                 case TPACKET_V1:
3124                 case TPACKET_V2:
3125                         len = sizeof(req_u.req);
3126                         break;
3127                 case TPACKET_V3:
3128                 default:
3129                         len = sizeof(req_u.req3);
3130                         break;
3131                 }
3132                 if (optlen < len)
3133                         return -EINVAL;
3134                 if (pkt_sk(sk)->has_vnet_hdr)
3135                         return -EINVAL;
3136                 if (copy_from_user(&req_u.req, optval, len))
3137                         return -EFAULT;
3138                 return packet_set_ring(sk, &req_u, 0,
3139                         optname == PACKET_TX_RING);
3140         }
3141         case PACKET_COPY_THRESH:
3142         {
3143                 int val;
3144
3145                 if (optlen != sizeof(val))
3146                         return -EINVAL;
3147                 if (copy_from_user(&val, optval, sizeof(val)))
3148                         return -EFAULT;
3149
3150                 pkt_sk(sk)->copy_thresh = val;
3151                 return 0;
3152         }
3153         case PACKET_VERSION:
3154         {
3155                 int val;
3156
3157                 if (optlen != sizeof(val))
3158                         return -EINVAL;
3159                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3160                         return -EBUSY;
3161                 if (copy_from_user(&val, optval, sizeof(val)))
3162                         return -EFAULT;
3163                 switch (val) {
3164                 case TPACKET_V1:
3165                 case TPACKET_V2:
3166                 case TPACKET_V3:
3167                         po->tp_version = val;
3168                         return 0;
3169                 default:
3170                         return -EINVAL;
3171                 }
3172         }
3173         case PACKET_RESERVE:
3174         {
3175                 unsigned int val;
3176
3177                 if (optlen != sizeof(val))
3178                         return -EINVAL;
3179                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3180                         return -EBUSY;
3181                 if (copy_from_user(&val, optval, sizeof(val)))
3182                         return -EFAULT;
3183                 po->tp_reserve = val;
3184                 return 0;
3185         }
3186         case PACKET_LOSS:
3187         {
3188                 unsigned int val;
3189
3190                 if (optlen != sizeof(val))
3191                         return -EINVAL;
3192                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3193                         return -EBUSY;
3194                 if (copy_from_user(&val, optval, sizeof(val)))
3195                         return -EFAULT;
3196                 po->tp_loss = !!val;
3197                 return 0;
3198         }
3199         case PACKET_AUXDATA:
3200         {
3201                 int val;
3202
3203                 if (optlen < sizeof(val))
3204                         return -EINVAL;
3205                 if (copy_from_user(&val, optval, sizeof(val)))
3206                         return -EFAULT;
3207
3208                 po->auxdata = !!val;
3209                 return 0;
3210         }
3211         case PACKET_ORIGDEV:
3212         {
3213                 int val;
3214
3215                 if (optlen < sizeof(val))
3216                         return -EINVAL;
3217                 if (copy_from_user(&val, optval, sizeof(val)))
3218                         return -EFAULT;
3219
3220                 po->origdev = !!val;
3221                 return 0;
3222         }
3223         case PACKET_VNET_HDR:
3224         {
3225                 int val;
3226
3227                 if (sock->type != SOCK_RAW)
3228                         return -EINVAL;
3229                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3230                         return -EBUSY;
3231                 if (optlen < sizeof(val))
3232                         return -EINVAL;
3233                 if (copy_from_user(&val, optval, sizeof(val)))
3234                         return -EFAULT;
3235
3236                 po->has_vnet_hdr = !!val;
3237                 return 0;
3238         }
3239         case PACKET_TIMESTAMP:
3240         {
3241                 int val;
3242
3243                 if (optlen != sizeof(val))
3244                         return -EINVAL;
3245                 if (copy_from_user(&val, optval, sizeof(val)))
3246                         return -EFAULT;
3247
3248                 po->tp_tstamp = val;
3249                 return 0;
3250         }
3251         case PACKET_FANOUT:
3252         {
3253                 int val;
3254
3255                 if (optlen != sizeof(val))
3256                         return -EINVAL;
3257                 if (copy_from_user(&val, optval, sizeof(val)))
3258                         return -EFAULT;
3259
3260                 return fanout_add(sk, val & 0xffff, val >> 16);
3261         }
3262         case PACKET_TX_HAS_OFF:
3263         {
3264                 unsigned int val;
3265
3266                 if (optlen != sizeof(val))
3267                         return -EINVAL;
3268                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3269                         return -EBUSY;
3270                 if (copy_from_user(&val, optval, sizeof(val)))
3271                         return -EFAULT;
3272                 po->tp_tx_has_off = !!val;
3273                 return 0;
3274         }
3275         case PACKET_QDISC_BYPASS:
3276         {
3277                 int val;
3278
3279                 if (optlen != sizeof(val))
3280                         return -EINVAL;
3281                 if (copy_from_user(&val, optval, sizeof(val)))
3282                         return -EFAULT;
3283
3284                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3285                 return 0;
3286         }
3287         default:
3288                 return -ENOPROTOOPT;
3289         }
3290 }
3291
3292 static int packet_getsockopt(struct socket *sock, int level, int optname,
3293                              char __user *optval, int __user *optlen)
3294 {
3295         int len;
3296         int val, lv = sizeof(val);
3297         struct sock *sk = sock->sk;
3298         struct packet_sock *po = pkt_sk(sk);
3299         void *data = &val;
3300         union tpacket_stats_u st;
3301
3302         if (level != SOL_PACKET)
3303                 return -ENOPROTOOPT;
3304
3305         if (get_user(len, optlen))
3306                 return -EFAULT;
3307
3308         if (len < 0)
3309                 return -EINVAL;
3310
3311         switch (optname) {
3312         case PACKET_STATISTICS:
3313                 spin_lock_bh(&sk->sk_receive_queue.lock);
3314                 memcpy(&st, &po->stats, sizeof(st));
3315                 memset(&po->stats, 0, sizeof(po->stats));
3316                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3317
3318                 if (po->tp_version == TPACKET_V3) {
3319                         lv = sizeof(struct tpacket_stats_v3);
3320                         st.stats3.tp_packets += st.stats3.tp_drops;
3321                         data = &st.stats3;
3322                 } else {
3323                         lv = sizeof(struct tpacket_stats);
3324                         st.stats1.tp_packets += st.stats1.tp_drops;
3325                         data = &st.stats1;
3326                 }
3327
3328                 break;
3329         case PACKET_AUXDATA:
3330                 val = po->auxdata;
3331                 break;
3332         case PACKET_ORIGDEV:
3333                 val = po->origdev;
3334                 break;
3335         case PACKET_VNET_HDR:
3336                 val = po->has_vnet_hdr;
3337                 break;
3338         case PACKET_VERSION:
3339                 val = po->tp_version;
3340                 break;
3341         case PACKET_HDRLEN:
3342                 if (len > sizeof(int))
3343                         len = sizeof(int);
3344                 if (copy_from_user(&val, optval, len))
3345                         return -EFAULT;
3346                 switch (val) {
3347                 case TPACKET_V1:
3348                         val = sizeof(struct tpacket_hdr);
3349                         break;
3350                 case TPACKET_V2:
3351                         val = sizeof(struct tpacket2_hdr);
3352                         break;
3353                 case TPACKET_V3:
3354                         val = sizeof(struct tpacket3_hdr);
3355                         break;
3356                 default:
3357                         return -EINVAL;
3358                 }
3359                 break;
3360         case PACKET_RESERVE:
3361                 val = po->tp_reserve;
3362                 break;
3363         case PACKET_LOSS:
3364                 val = po->tp_loss;
3365                 break;
3366         case PACKET_TIMESTAMP:
3367                 val = po->tp_tstamp;
3368                 break;
3369         case PACKET_FANOUT:
3370                 val = (po->fanout ?
3371                        ((u32)po->fanout->id |
3372                         ((u32)po->fanout->type << 16) |
3373                         ((u32)po->fanout->flags << 24)) :
3374                        0);
3375                 break;
3376         case PACKET_TX_HAS_OFF:
3377                 val = po->tp_tx_has_off;
3378                 break;
3379         case PACKET_QDISC_BYPASS:
3380                 val = packet_use_direct_xmit(po);
3381                 break;
3382         default:
3383                 return -ENOPROTOOPT;
3384         }
3385
3386         if (len > lv)
3387                 len = lv;
3388         if (put_user(len, optlen))
3389                 return -EFAULT;
3390         if (copy_to_user(optval, data, len))
3391                 return -EFAULT;
3392         return 0;
3393 }
3394
3395
3396 static int packet_notifier(struct notifier_block *this,
3397                            unsigned long msg, void *ptr)
3398 {
3399         struct sock *sk;
3400         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3401         struct net *net = dev_net(dev);
3402
3403         rcu_read_lock();
3404         sk_for_each_rcu(sk, &net->packet.sklist) {
3405                 struct packet_sock *po = pkt_sk(sk);
3406
3407                 switch (msg) {
3408                 case NETDEV_UNREGISTER:
3409                         if (po->mclist)
3410                                 packet_dev_mclist(dev, po->mclist, -1);
3411                         /* fallthrough */
3412
3413                 case NETDEV_DOWN:
3414                         if (dev->ifindex == po->ifindex) {
3415                                 spin_lock(&po->bind_lock);
3416                                 if (po->running) {
3417                                         __unregister_prot_hook(sk, false);
3418                                         sk->sk_err = ENETDOWN;
3419                                         if (!sock_flag(sk, SOCK_DEAD))
3420                                                 sk->sk_error_report(sk);
3421                                 }
3422                                 if (msg == NETDEV_UNREGISTER) {
3423                                         packet_cached_dev_reset(po);
3424                                         po->ifindex = -1;
3425                                         if (po->prot_hook.dev)
3426                                                 dev_put(po->prot_hook.dev);
3427                                         po->prot_hook.dev = NULL;
3428                                 }
3429                                 spin_unlock(&po->bind_lock);
3430                         }
3431                         break;
3432                 case NETDEV_UP:
3433                         if (dev->ifindex == po->ifindex) {
3434                                 spin_lock(&po->bind_lock);
3435                                 if (po->num)
3436                                         register_prot_hook(sk);
3437                                 spin_unlock(&po->bind_lock);
3438                         }
3439                         break;
3440                 }
3441         }
3442         rcu_read_unlock();
3443         return NOTIFY_DONE;
3444 }
3445
3446
3447 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3448                         unsigned long arg)
3449 {
3450         struct sock *sk = sock->sk;
3451
3452         switch (cmd) {
3453         case SIOCOUTQ:
3454         {
3455                 int amount = sk_wmem_alloc_get(sk);
3456
3457                 return put_user(amount, (int __user *)arg);
3458         }
3459         case SIOCINQ:
3460         {
3461                 struct sk_buff *skb;
3462                 int amount = 0;
3463
3464                 spin_lock_bh(&sk->sk_receive_queue.lock);
3465                 skb = skb_peek(&sk->sk_receive_queue);
3466                 if (skb)
3467                         amount = skb->len;
3468                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3469                 return put_user(amount, (int __user *)arg);
3470         }
3471         case SIOCGSTAMP:
3472                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3473         case SIOCGSTAMPNS:
3474                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3475
3476 #ifdef CONFIG_INET
3477         case SIOCADDRT:
3478         case SIOCDELRT:
3479         case SIOCDARP:
3480         case SIOCGARP:
3481         case SIOCSARP:
3482         case SIOCGIFADDR:
3483         case SIOCSIFADDR:
3484         case SIOCGIFBRDADDR:
3485         case SIOCSIFBRDADDR:
3486         case SIOCGIFNETMASK:
3487         case SIOCSIFNETMASK:
3488         case SIOCGIFDSTADDR:
3489         case SIOCSIFDSTADDR:
3490         case SIOCSIFFLAGS:
3491                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3492 #endif
3493
3494         default:
3495                 return -ENOIOCTLCMD;
3496         }
3497         return 0;
3498 }
3499
3500 static unsigned int packet_poll(struct file *file, struct socket *sock,
3501                                 poll_table *wait)
3502 {
3503         struct sock *sk = sock->sk;
3504         struct packet_sock *po = pkt_sk(sk);
3505         unsigned int mask = datagram_poll(file, sock, wait);
3506
3507         spin_lock_bh(&sk->sk_receive_queue.lock);
3508         if (po->rx_ring.pg_vec) {
3509                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3510                         TP_STATUS_KERNEL))
3511                         mask |= POLLIN | POLLRDNORM;
3512         }
3513         spin_unlock_bh(&sk->sk_receive_queue.lock);
3514         spin_lock_bh(&sk->sk_write_queue.lock);
3515         if (po->tx_ring.pg_vec) {
3516                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3517                         mask |= POLLOUT | POLLWRNORM;
3518         }
3519         spin_unlock_bh(&sk->sk_write_queue.lock);
3520         return mask;
3521 }
3522
3523
3524 /* Dirty? Well, I still did not learn better way to account
3525  * for user mmaps.
3526  */
3527
3528 static void packet_mm_open(struct vm_area_struct *vma)
3529 {
3530         struct file *file = vma->vm_file;
3531         struct socket *sock = file->private_data;
3532         struct sock *sk = sock->sk;
3533
3534         if (sk)
3535                 atomic_inc(&pkt_sk(sk)->mapped);
3536 }
3537
3538 static void packet_mm_close(struct vm_area_struct *vma)
3539 {
3540         struct file *file = vma->vm_file;
3541         struct socket *sock = file->private_data;
3542         struct sock *sk = sock->sk;
3543
3544         if (sk)
3545                 atomic_dec(&pkt_sk(sk)->mapped);
3546 }
3547
3548 static const struct vm_operations_struct packet_mmap_ops = {
3549         .open   =       packet_mm_open,
3550         .close  =       packet_mm_close,
3551 };
3552
3553 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3554                         unsigned int len)
3555 {
3556         int i;
3557
3558         for (i = 0; i < len; i++) {
3559                 if (likely(pg_vec[i].buffer)) {
3560                         if (is_vmalloc_addr(pg_vec[i].buffer))
3561                                 vfree(pg_vec[i].buffer);
3562                         else
3563                                 free_pages((unsigned long)pg_vec[i].buffer,
3564                                            order);
3565                         pg_vec[i].buffer = NULL;
3566                 }
3567         }
3568         kfree(pg_vec);
3569 }
3570
3571 static char *alloc_one_pg_vec_page(unsigned long order)
3572 {
3573         char *buffer = NULL;
3574         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3575                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3576
3577         buffer = (char *) __get_free_pages(gfp_flags, order);
3578
3579         if (buffer)
3580                 return buffer;
3581
3582         /*
3583          * __get_free_pages failed, fall back to vmalloc
3584          */
3585         buffer = vzalloc((1 << order) * PAGE_SIZE);
3586
3587         if (buffer)
3588                 return buffer;
3589
3590         /*
3591          * vmalloc failed, lets dig into swap here
3592          */
3593         gfp_flags &= ~__GFP_NORETRY;
3594         buffer = (char *)__get_free_pages(gfp_flags, order);
3595         if (buffer)
3596                 return buffer;
3597
3598         /*
3599          * complete and utter failure
3600          */
3601         return NULL;
3602 }
3603
3604 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3605 {
3606         unsigned int block_nr = req->tp_block_nr;
3607         struct pgv *pg_vec;
3608         int i;
3609
3610         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3611         if (unlikely(!pg_vec))
3612                 goto out;
3613
3614         for (i = 0; i < block_nr; i++) {
3615                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3616                 if (unlikely(!pg_vec[i].buffer))
3617                         goto out_free_pgvec;
3618         }
3619
3620 out:
3621         return pg_vec;
3622
3623 out_free_pgvec:
3624         free_pg_vec(pg_vec, order, block_nr);
3625         pg_vec = NULL;
3626         goto out;
3627 }
3628
3629 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3630                 int closing, int tx_ring)
3631 {
3632         struct pgv *pg_vec = NULL;
3633         struct packet_sock *po = pkt_sk(sk);
3634         int was_running, order = 0;
3635         struct packet_ring_buffer *rb;
3636         struct sk_buff_head *rb_queue;
3637         __be16 num;
3638         int err = -EINVAL;
3639         /* Added to avoid minimal code churn */
3640         struct tpacket_req *req = &req_u->req;
3641
3642         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3643         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3644                 WARN(1, "Tx-ring is not supported.\n");
3645                 goto out;
3646         }
3647
3648         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3649         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3650
3651         err = -EBUSY;
3652         if (!closing) {
3653                 if (atomic_read(&po->mapped))
3654                         goto out;
3655                 if (atomic_read(&rb->pending))
3656                         goto out;
3657         }
3658
3659         if (req->tp_block_nr) {
3660                 /* Sanity tests and some calculations */
3661                 err = -EBUSY;
3662                 if (unlikely(rb->pg_vec))
3663                         goto out;
3664
3665                 switch (po->tp_version) {
3666                 case TPACKET_V1:
3667                         po->tp_hdrlen = TPACKET_HDRLEN;
3668                         break;
3669                 case TPACKET_V2:
3670                         po->tp_hdrlen = TPACKET2_HDRLEN;
3671                         break;
3672                 case TPACKET_V3:
3673                         po->tp_hdrlen = TPACKET3_HDRLEN;
3674                         break;
3675                 }
3676
3677                 err = -EINVAL;
3678                 if (unlikely((int)req->tp_block_size <= 0))
3679                         goto out;
3680                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3681                         goto out;
3682                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3683                                         po->tp_reserve))
3684                         goto out;
3685                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3686                         goto out;
3687
3688                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3689                 if (unlikely(rb->frames_per_block <= 0))
3690                         goto out;
3691                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3692                                         req->tp_frame_nr))
3693                         goto out;
3694
3695                 err = -ENOMEM;
3696                 order = get_order(req->tp_block_size);
3697                 pg_vec = alloc_pg_vec(req, order);
3698                 if (unlikely(!pg_vec))
3699                         goto out;
3700                 switch (po->tp_version) {
3701                 case TPACKET_V3:
3702                 /* Transmit path is not supported. We checked
3703                  * it above but just being paranoid
3704                  */
3705                         if (!tx_ring)
3706                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3707                                 break;
3708                 default:
3709                         break;
3710                 }
3711         }
3712         /* Done */
3713         else {
3714                 err = -EINVAL;
3715                 if (unlikely(req->tp_frame_nr))
3716                         goto out;
3717         }
3718
3719         lock_sock(sk);
3720
3721         /* Detach socket from network */
3722         spin_lock(&po->bind_lock);
3723         was_running = po->running;
3724         num = po->num;
3725         if (was_running) {
3726                 po->num = 0;
3727                 __unregister_prot_hook(sk, false);
3728         }
3729         spin_unlock(&po->bind_lock);
3730
3731         synchronize_net();
3732
3733         err = -EBUSY;
3734         mutex_lock(&po->pg_vec_lock);
3735         if (closing || atomic_read(&po->mapped) == 0) {
3736                 err = 0;
3737                 spin_lock_bh(&rb_queue->lock);
3738                 swap(rb->pg_vec, pg_vec);
3739                 rb->frame_max = (req->tp_frame_nr - 1);
3740                 rb->head = 0;
3741                 rb->frame_size = req->tp_frame_size;
3742                 spin_unlock_bh(&rb_queue->lock);
3743
3744                 swap(rb->pg_vec_order, order);
3745                 swap(rb->pg_vec_len, req->tp_block_nr);
3746
3747                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3748                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3749                                                 tpacket_rcv : packet_rcv;
3750                 skb_queue_purge(rb_queue);
3751                 if (atomic_read(&po->mapped))
3752                         pr_err("packet_mmap: vma is busy: %d\n",
3753                                atomic_read(&po->mapped));
3754         }
3755         mutex_unlock(&po->pg_vec_lock);
3756
3757         spin_lock(&po->bind_lock);
3758         if (was_running) {
3759                 po->num = num;
3760                 register_prot_hook(sk);
3761         }
3762         spin_unlock(&po->bind_lock);
3763         if (closing && (po->tp_version > TPACKET_V2)) {
3764                 /* Because we don't support block-based V3 on tx-ring */
3765                 if (!tx_ring)
3766                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3767         }
3768         release_sock(sk);
3769
3770         if (pg_vec)
3771                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3772 out:
3773         return err;
3774 }
3775
3776 static int packet_mmap(struct file *file, struct socket *sock,
3777                 struct vm_area_struct *vma)
3778 {
3779         struct sock *sk = sock->sk;
3780         struct packet_sock *po = pkt_sk(sk);
3781         unsigned long size, expected_size;
3782         struct packet_ring_buffer *rb;
3783         unsigned long start;
3784         int err = -EINVAL;
3785         int i;
3786
3787         if (vma->vm_pgoff)
3788                 return -EINVAL;
3789
3790         mutex_lock(&po->pg_vec_lock);
3791
3792         expected_size = 0;
3793         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3794                 if (rb->pg_vec) {
3795                         expected_size += rb->pg_vec_len
3796                                                 * rb->pg_vec_pages
3797                                                 * PAGE_SIZE;
3798                 }
3799         }
3800
3801         if (expected_size == 0)
3802                 goto out;
3803
3804         size = vma->vm_end - vma->vm_start;
3805         if (size != expected_size)
3806                 goto out;
3807
3808         start = vma->vm_start;
3809         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3810                 if (rb->pg_vec == NULL)
3811                         continue;
3812
3813                 for (i = 0; i < rb->pg_vec_len; i++) {
3814                         struct page *page;
3815                         void *kaddr = rb->pg_vec[i].buffer;
3816                         int pg_num;
3817
3818                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3819                                 page = pgv_to_page(kaddr);
3820                                 err = vm_insert_page(vma, start, page);
3821                                 if (unlikely(err))
3822                                         goto out;
3823                                 start += PAGE_SIZE;
3824                                 kaddr += PAGE_SIZE;
3825                         }
3826                 }
3827         }
3828
3829         atomic_inc(&po->mapped);
3830         vma->vm_ops = &packet_mmap_ops;
3831         err = 0;
3832
3833 out:
3834         mutex_unlock(&po->pg_vec_lock);
3835         return err;
3836 }
3837
3838 static const struct proto_ops packet_ops_spkt = {
3839         .family =       PF_PACKET,
3840         .owner =        THIS_MODULE,
3841         .release =      packet_release,
3842         .bind =         packet_bind_spkt,
3843         .connect =      sock_no_connect,
3844         .socketpair =   sock_no_socketpair,
3845         .accept =       sock_no_accept,
3846         .getname =      packet_getname_spkt,
3847         .poll =         datagram_poll,
3848         .ioctl =        packet_ioctl,
3849         .listen =       sock_no_listen,
3850         .shutdown =     sock_no_shutdown,
3851         .setsockopt =   sock_no_setsockopt,
3852         .getsockopt =   sock_no_getsockopt,
3853         .sendmsg =      packet_sendmsg_spkt,
3854         .recvmsg =      packet_recvmsg,
3855         .mmap =         sock_no_mmap,
3856         .sendpage =     sock_no_sendpage,
3857 };
3858
3859 static const struct proto_ops packet_ops = {
3860         .family =       PF_PACKET,
3861         .owner =        THIS_MODULE,
3862         .release =      packet_release,
3863         .bind =         packet_bind,
3864         .connect =      sock_no_connect,
3865         .socketpair =   sock_no_socketpair,
3866         .accept =       sock_no_accept,
3867         .getname =      packet_getname,
3868         .poll =         packet_poll,
3869         .ioctl =        packet_ioctl,
3870         .listen =       sock_no_listen,
3871         .shutdown =     sock_no_shutdown,
3872         .setsockopt =   packet_setsockopt,
3873         .getsockopt =   packet_getsockopt,
3874         .sendmsg =      packet_sendmsg,
3875         .recvmsg =      packet_recvmsg,
3876         .mmap =         packet_mmap,
3877         .sendpage =     sock_no_sendpage,
3878 };
3879
3880 static const struct net_proto_family packet_family_ops = {
3881         .family =       PF_PACKET,
3882         .create =       packet_create,
3883         .owner  =       THIS_MODULE,
3884 };
3885
3886 static struct notifier_block packet_netdev_notifier = {
3887         .notifier_call =        packet_notifier,
3888 };
3889
3890 #ifdef CONFIG_PROC_FS
3891
3892 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3893         __acquires(RCU)
3894 {
3895         struct net *net = seq_file_net(seq);
3896
3897         rcu_read_lock();
3898         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3899 }
3900
3901 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3902 {
3903         struct net *net = seq_file_net(seq);
3904         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3905 }
3906
3907 static void packet_seq_stop(struct seq_file *seq, void *v)
3908         __releases(RCU)
3909 {
3910         rcu_read_unlock();
3911 }
3912
3913 static int packet_seq_show(struct seq_file *seq, void *v)
3914 {
3915         if (v == SEQ_START_TOKEN)
3916                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3917         else {
3918                 struct sock *s = sk_entry(v);
3919                 const struct packet_sock *po = pkt_sk(s);
3920
3921                 seq_printf(seq,
3922                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3923                            s,
3924                            atomic_read(&s->sk_refcnt),
3925                            s->sk_type,
3926                            ntohs(po->num),
3927                            po->ifindex,
3928                            po->running,
3929                            atomic_read(&s->sk_rmem_alloc),
3930                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3931                            sock_i_ino(s));
3932         }
3933
3934         return 0;
3935 }
3936
3937 static const struct seq_operations packet_seq_ops = {
3938         .start  = packet_seq_start,
3939         .next   = packet_seq_next,
3940         .stop   = packet_seq_stop,
3941         .show   = packet_seq_show,
3942 };
3943
3944 static int packet_seq_open(struct inode *inode, struct file *file)
3945 {
3946         return seq_open_net(inode, file, &packet_seq_ops,
3947                             sizeof(struct seq_net_private));
3948 }
3949
3950 static const struct file_operations packet_seq_fops = {
3951         .owner          = THIS_MODULE,
3952         .open           = packet_seq_open,
3953         .read           = seq_read,
3954         .llseek         = seq_lseek,
3955         .release        = seq_release_net,
3956 };
3957
3958 #endif
3959
3960 static int __net_init packet_net_init(struct net *net)
3961 {
3962         mutex_init(&net->packet.sklist_lock);
3963         INIT_HLIST_HEAD(&net->packet.sklist);
3964
3965         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3966                 return -ENOMEM;
3967
3968         return 0;
3969 }
3970
3971 static void __net_exit packet_net_exit(struct net *net)
3972 {
3973         remove_proc_entry("packet", net->proc_net);
3974 }
3975
3976 static struct pernet_operations packet_net_ops = {
3977         .init = packet_net_init,
3978         .exit = packet_net_exit,
3979 };
3980
3981
3982 static void __exit packet_exit(void)
3983 {
3984         unregister_netdevice_notifier(&packet_netdev_notifier);
3985         unregister_pernet_subsys(&packet_net_ops);
3986         sock_unregister(PF_PACKET);
3987         proto_unregister(&packet_proto);
3988 }
3989
3990 static int __init packet_init(void)
3991 {
3992         int rc = proto_register(&packet_proto, 0);
3993
3994         if (rc != 0)
3995                 goto out;
3996
3997         sock_register(&packet_family_ops);
3998         register_pernet_subsys(&packet_net_ops);
3999         register_netdevice_notifier(&packet_netdev_notifier);
4000 out:
4001         return rc;
4002 }
4003
4004 module_init(packet_init);
4005 module_exit(packet_exit);
4006 MODULE_LICENSE("GPL");
4007 MODULE_ALIAS_NETPROTO(PF_PACKET);