Pileus Git - ~andy/linux/blob - net/packet/af_packet.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              PACKET - implements raw packet sockets.
   7  *
   8  * Authors:     Ross Biro
   9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  11  *
  12  * Fixes:
  13  *              Alan Cox        :       verify_area() now used correctly
  14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
  15  *              Alan Cox        :       tidied skbuff lists.
  16  *              Alan Cox        :       Now uses generic datagram routines I
  17  *                                      added. Also fixed the peek/read crash
  18  *                                      from all old Linux datagram code.
  19  *              Alan Cox        :       Uses the improved datagram code.
  20  *              Alan Cox        :       Added NULL's for socket options.
  21  *              Alan Cox        :       Re-commented the code.
  22  *              Alan Cox        :       Use new kernel side addressing
  23  *              Rob Janssen     :       Correct MTU usage.
  24  *              Dave Platt      :       Counter leaks caused by incorrect
  25  *                                      interrupt locking and some slightly
  26  *                                      dubious gcc output. Can you read
  27  *                                      compiler: it said _VOLATILE_
  28  *      Richard Kooijman        :       Timestamp fixes.
  29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
  30  *              Alan Cox        :       sendmsg/recvmsg support.
  31  *              Alan Cox        :       Protocol setting support
  32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
  33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
  34  *      Michal Ostrowski        :       Module initialization cleanup.
  35  *         Ulises Alonso        :       Frame number limit removal and
  36  *                                      packet_set_ring memory leak.
  37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
  38  *                                      The convention is that longer addresses
  39  *                                      will simply extend the hardware address
  40  *                                      byte arrays at the end of sockaddr_ll
  41  *                                      and packet_mreq.
  42  *              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) raw_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_hash(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->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
 981                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
 982         } else {
 983                 ppd->hv1.tp_vlan_tci = 0;
 984                 ppd->hv1.tp_vlan_tpid = 0;
 985                 ppd->tp_status = TP_STATUS_AVAILABLE;
 986         }
 987 }
 988
 989 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 990                         struct tpacket3_hdr *ppd)
 991 {
 992         ppd->hv1.tp_padding = 0;
 993         prb_fill_vlan_info(pkc, ppd);
 994
 995         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
 996                 prb_fill_rxhash(pkc, ppd);
 997         else
 998                 prb_clear_rxhash(pkc, ppd);
 999 }
1000
1001 static void prb_fill_curr_block(char *curr,
1002                                 struct tpacket_kbdq_core *pkc,
1003                                 struct tpacket_block_desc *pbd,
1004                                 unsigned int len)
1005 {
1006         struct tpacket3_hdr *ppd;
1007
1008         ppd  = (struct tpacket3_hdr *)curr;
1009         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1010         pkc->prev = curr;
1011         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1012         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1013         BLOCK_NUM_PKTS(pbd) += 1;
1014         atomic_inc(&pkc->blk_fill_in_prog);
1015         prb_run_all_ft_ops(pkc, ppd);
1016 }
1017
1018 /* Assumes caller has the sk->rx_queue.lock */
1019 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1020                                             struct sk_buff *skb,
1021                                                 int status,
1022                                             unsigned int len
1023                                             )
1024 {
1025         struct tpacket_kbdq_core *pkc;
1026         struct tpacket_block_desc *pbd;
1027         char *curr, *end;
1028
1029         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1030         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1031
1032         /* Queue is frozen when user space is lagging behind */
1033         if (prb_queue_frozen(pkc)) {
1034                 /*
1035                  * Check if that last block which caused the queue to freeze,
1036                  * is still in_use by user-space.
1037                  */
1038                 if (prb_curr_blk_in_use(pkc, pbd)) {
1039                         /* Can't record this packet */
1040                         return NULL;
1041                 } else {
1042                         /*
1043                          * Ok, the block was released by user-space.
1044                          * Now let's open that block.
1045                          * opening a block also thaws the queue.
1046                          * Thawing is a side effect.
1047                          */
1048                         prb_open_block(pkc, pbd);
1049                 }
1050         }
1051
1052         smp_mb();
1053         curr = pkc->nxt_offset;
1054         pkc->skb = skb;
1055         end = (char *)pbd + pkc->kblk_size;
1056
1057         /* first try the current block */
1058         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1059                 prb_fill_curr_block(curr, pkc, pbd, len);
1060                 return (void *)curr;
1061         }
1062
1063         /* Ok, close the current block */
1064         prb_retire_current_block(pkc, po, 0);
1065
1066         /* Now, try to dispatch the next block */
1067         curr = (char *)prb_dispatch_next_block(pkc, po);
1068         if (curr) {
1069                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1070                 prb_fill_curr_block(curr, pkc, pbd, len);
1071                 return (void *)curr;
1072         }
1073
1074         /*
1075          * No free blocks are available.user_space hasn't caught up yet.
1076          * Queue was just frozen and now this packet will get dropped.
1077          */
1078         return NULL;
1079 }
1080
1081 static void *packet_current_rx_frame(struct packet_sock *po,
1082                                             struct sk_buff *skb,
1083                                             int status, unsigned int len)
1084 {
1085         char *curr = NULL;
1086         switch (po->tp_version) {
1087         case TPACKET_V1:
1088         case TPACKET_V2:
1089                 curr = packet_lookup_frame(po, &po->rx_ring,
1090                                         po->rx_ring.head, status);
1091                 return curr;
1092         case TPACKET_V3:
1093                 return __packet_lookup_frame_in_block(po, skb, status, len);
1094         default:
1095                 WARN(1, "TPACKET version not supported\n");
1096                 BUG();
1097                 return NULL;
1098         }
1099 }
1100
1101 static void *prb_lookup_block(struct packet_sock *po,
1102                                      struct packet_ring_buffer *rb,
1103                                      unsigned int idx,
1104                                      int status)
1105 {
1106         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1107         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1108
1109         if (status != BLOCK_STATUS(pbd))
1110                 return NULL;
1111         return pbd;
1112 }
1113
1114 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1115 {
1116         unsigned int prev;
1117         if (rb->prb_bdqc.kactive_blk_num)
1118                 prev = rb->prb_bdqc.kactive_blk_num-1;
1119         else
1120                 prev = rb->prb_bdqc.knum_blocks-1;
1121         return prev;
1122 }
1123
1124 /* Assumes caller has held the rx_queue.lock */
1125 static void *__prb_previous_block(struct packet_sock *po,
1126                                          struct packet_ring_buffer *rb,
1127                                          int status)
1128 {
1129         unsigned int previous = prb_previous_blk_num(rb);
1130         return prb_lookup_block(po, rb, previous, status);
1131 }
1132
1133 static void *packet_previous_rx_frame(struct packet_sock *po,
1134                                              struct packet_ring_buffer *rb,
1135                                              int status)
1136 {
1137         if (po->tp_version <= TPACKET_V2)
1138                 return packet_previous_frame(po, rb, status);
1139
1140         return __prb_previous_block(po, rb, status);
1141 }
1142
1143 static void packet_increment_rx_head(struct packet_sock *po,
1144                                             struct packet_ring_buffer *rb)
1145 {
1146         switch (po->tp_version) {
1147         case TPACKET_V1:
1148         case TPACKET_V2:
1149                 return packet_increment_head(rb);
1150         case TPACKET_V3:
1151         default:
1152                 WARN(1, "TPACKET version not supported.\n");
1153                 BUG();
1154                 return;
1155         }
1156 }
1157
1158 static void *packet_previous_frame(struct packet_sock *po,
1159                 struct packet_ring_buffer *rb,
1160                 int status)
1161 {
1162         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1163         return packet_lookup_frame(po, rb, previous, status);
1164 }
1165
1166 static void packet_increment_head(struct packet_ring_buffer *buff)
1167 {
1168         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1169 }
1170
1171 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1172 {
1173         struct sock *sk = &po->sk;
1174         bool has_room;
1175
1176         if (po->prot_hook.func != tpacket_rcv)
1177                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1178                         <= sk->sk_rcvbuf;
1179
1180         spin_lock(&sk->sk_receive_queue.lock);
1181         if (po->tp_version == TPACKET_V3)
1182                 has_room = prb_lookup_block(po, &po->rx_ring,
1183                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1184                                             TP_STATUS_KERNEL);
1185         else
1186                 has_room = packet_lookup_frame(po, &po->rx_ring,
1187                                                po->rx_ring.head,
1188                                                TP_STATUS_KERNEL);
1189         spin_unlock(&sk->sk_receive_queue.lock);
1190
1191         return has_room;
1192 }
1193
1194 static void packet_sock_destruct(struct sock *sk)
1195 {
1196         skb_queue_purge(&sk->sk_error_queue);
1197
1198         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1199         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1200
1201         if (!sock_flag(sk, SOCK_DEAD)) {
1202                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1203                 return;
1204         }
1205
1206         sk_refcnt_debug_dec(sk);
1207 }
1208
1209 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1210 {
1211         int x = atomic_read(&f->rr_cur) + 1;
1212
1213         if (x >= num)
1214                 x = 0;
1215
1216         return x;
1217 }
1218
1219 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1220                                       struct sk_buff *skb,
1221                                       unsigned int num)
1222 {
1223         return reciprocal_divide(skb->rxhash, num);
1224 }
1225
1226 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1227                                     struct sk_buff *skb,
1228                                     unsigned int num)
1229 {
1230         int cur, old;
1231
1232         cur = atomic_read(&f->rr_cur);
1233         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1234                                      fanout_rr_next(f, num))) != cur)
1235                 cur = old;
1236         return cur;
1237 }
1238
1239 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1240                                      struct sk_buff *skb,
1241                                      unsigned int num)
1242 {
1243         return smp_processor_id() % num;
1244 }
1245
1246 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1247                                      struct sk_buff *skb,
1248                                      unsigned int num)
1249 {
1250         return reciprocal_divide(prandom_u32(), num);
1251 }
1252
1253 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1254                                           struct sk_buff *skb,
1255                                           unsigned int idx, unsigned int skip,
1256                                           unsigned int num)
1257 {
1258         unsigned int i, j;
1259
1260         i = j = min_t(int, f->next[idx], num - 1);
1261         do {
1262                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1263                         if (i != j)
1264                                 f->next[idx] = i;
1265                         return i;
1266                 }
1267                 if (++i == num)
1268                         i = 0;
1269         } while (i != j);
1270
1271         return idx;
1272 }
1273
1274 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1275 {
1276         return f->flags & (flag >> 8);
1277 }
1278
1279 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1280                              struct packet_type *pt, struct net_device *orig_dev)
1281 {
1282         struct packet_fanout *f = pt->af_packet_priv;
1283         unsigned int num = f->num_members;
1284         struct packet_sock *po;
1285         unsigned int idx;
1286
1287         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1288             !num) {
1289                 kfree_skb(skb);
1290                 return 0;
1291         }
1292
1293         switch (f->type) {
1294         case PACKET_FANOUT_HASH:
1295         default:
1296                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1297                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1298                         if (!skb)
1299                                 return 0;
1300                 }
1301                 skb_get_hash(skb);
1302                 idx = fanout_demux_hash(f, skb, num);
1303                 break;
1304         case PACKET_FANOUT_LB:
1305                 idx = fanout_demux_lb(f, skb, num);
1306                 break;
1307         case PACKET_FANOUT_CPU:
1308                 idx = fanout_demux_cpu(f, skb, num);
1309                 break;
1310         case PACKET_FANOUT_RND:
1311                 idx = fanout_demux_rnd(f, skb, num);
1312                 break;
1313         case PACKET_FANOUT_ROLLOVER:
1314                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1315                 break;
1316         }
1317
1318         po = pkt_sk(f->arr[idx]);
1319         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1320             unlikely(!packet_rcv_has_room(po, skb))) {
1321                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1322                 po = pkt_sk(f->arr[idx]);
1323         }
1324
1325         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1326 }
1327
1328 DEFINE_MUTEX(fanout_mutex);
1329 EXPORT_SYMBOL_GPL(fanout_mutex);
1330 static LIST_HEAD(fanout_list);
1331
1332 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1333 {
1334         struct packet_fanout *f = po->fanout;
1335
1336         spin_lock(&f->lock);
1337         f->arr[f->num_members] = sk;
1338         smp_wmb();
1339         f->num_members++;
1340         spin_unlock(&f->lock);
1341 }
1342
1343 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1344 {
1345         struct packet_fanout *f = po->fanout;
1346         int i;
1347
1348         spin_lock(&f->lock);
1349         for (i = 0; i < f->num_members; i++) {
1350                 if (f->arr[i] == sk)
1351                         break;
1352         }
1353         BUG_ON(i >= f->num_members);
1354         f->arr[i] = f->arr[f->num_members - 1];
1355         f->num_members--;
1356         spin_unlock(&f->lock);
1357 }
1358
1359 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1360 {
1361         if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1362                 return true;
1363
1364         return false;
1365 }
1366
1367 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1368 {
1369         struct packet_sock *po = pkt_sk(sk);
1370         struct packet_fanout *f, *match;
1371         u8 type = type_flags & 0xff;
1372         u8 flags = type_flags >> 8;
1373         int err;
1374
1375         switch (type) {
1376         case PACKET_FANOUT_ROLLOVER:
1377                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1378                         return -EINVAL;
1379         case PACKET_FANOUT_HASH:
1380         case PACKET_FANOUT_LB:
1381         case PACKET_FANOUT_CPU:
1382         case PACKET_FANOUT_RND:
1383                 break;
1384         default:
1385                 return -EINVAL;
1386         }
1387
1388         if (!po->running)
1389                 return -EINVAL;
1390
1391         if (po->fanout)
1392                 return -EALREADY;
1393
1394         mutex_lock(&fanout_mutex);
1395         match = NULL;
1396         list_for_each_entry(f, &fanout_list, list) {
1397                 if (f->id == id &&
1398                     read_pnet(&f->net) == sock_net(sk)) {
1399                         match = f;
1400                         break;
1401                 }
1402         }
1403         err = -EINVAL;
1404         if (match && match->flags != flags)
1405                 goto out;
1406         if (!match) {
1407                 err = -ENOMEM;
1408                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1409                 if (!match)
1410                         goto out;
1411                 write_pnet(&match->net, sock_net(sk));
1412                 match->id = id;
1413                 match->type = type;
1414                 match->flags = flags;
1415                 atomic_set(&match->rr_cur, 0);
1416                 INIT_LIST_HEAD(&match->list);
1417                 spin_lock_init(&match->lock);
1418                 atomic_set(&match->sk_ref, 0);
1419                 match->prot_hook.type = po->prot_hook.type;
1420                 match->prot_hook.dev = po->prot_hook.dev;
1421                 match->prot_hook.func = packet_rcv_fanout;
1422                 match->prot_hook.af_packet_priv = match;
1423                 match->prot_hook.id_match = match_fanout_group;
1424                 dev_add_pack(&match->prot_hook);
1425                 list_add(&match->list, &fanout_list);
1426         }
1427         err = -EINVAL;
1428         if (match->type == type &&
1429             match->prot_hook.type == po->prot_hook.type &&
1430             match->prot_hook.dev == po->prot_hook.dev) {
1431                 err = -ENOSPC;
1432                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1433                         __dev_remove_pack(&po->prot_hook);
1434                         po->fanout = match;
1435                         atomic_inc(&match->sk_ref);
1436                         __fanout_link(sk, po);
1437                         err = 0;
1438                 }
1439         }
1440 out:
1441         mutex_unlock(&fanout_mutex);
1442         return err;
1443 }
1444
1445 static void fanout_release(struct sock *sk)
1446 {
1447         struct packet_sock *po = pkt_sk(sk);
1448         struct packet_fanout *f;
1449
1450         f = po->fanout;
1451         if (!f)
1452                 return;
1453
1454         mutex_lock(&fanout_mutex);
1455         po->fanout = NULL;
1456
1457         if (atomic_dec_and_test(&f->sk_ref)) {
1458                 list_del(&f->list);
1459                 dev_remove_pack(&f->prot_hook);
1460                 kfree(f);
1461         }
1462         mutex_unlock(&fanout_mutex);
1463 }
1464
1465 static const struct proto_ops packet_ops;
1466
1467 static const struct proto_ops packet_ops_spkt;
1468
1469 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1470                            struct packet_type *pt, struct net_device *orig_dev)
1471 {
1472         struct sock *sk;
1473         struct sockaddr_pkt *spkt;
1474
1475         /*
1476          *      When we registered the protocol we saved the socket in the data
1477          *      field for just this event.
1478          */
1479
1480         sk = pt->af_packet_priv;
1481
1482         /*
1483          *      Yank back the headers [hope the device set this
1484          *      right or kerboom...]
1485          *
1486          *      Incoming packets have ll header pulled,
1487          *      push it back.
1488          *
1489          *      For outgoing ones skb->data == skb_mac_header(skb)
1490          *      so that this procedure is noop.
1491          */
1492
1493         if (skb->pkt_type == PACKET_LOOPBACK)
1494                 goto out;
1495
1496         if (!net_eq(dev_net(dev), sock_net(sk)))
1497                 goto out;
1498
1499         skb = skb_share_check(skb, GFP_ATOMIC);
1500         if (skb == NULL)
1501                 goto oom;
1502
1503         /* drop any routing info */
1504         skb_dst_drop(skb);
1505
1506         /* drop conntrack reference */
1507         nf_reset(skb);
1508
1509         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1510
1511         skb_push(skb, skb->data - skb_mac_header(skb));
1512
1513         /*
1514          *      The SOCK_PACKET socket receives _all_ frames.
1515          */
1516
1517         spkt->spkt_family = dev->type;
1518         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1519         spkt->spkt_protocol = skb->protocol;
1520
1521         /*
1522          *      Charge the memory to the socket. This is done specifically
1523          *      to prevent sockets using all the memory up.
1524          */
1525
1526         if (sock_queue_rcv_skb(sk, skb) == 0)
1527                 return 0;
1528
1529 out:
1530         kfree_skb(skb);
1531 oom:
1532         return 0;
1533 }
1534
1535
1536 /*
1537  *      Output a raw packet to a device layer. This bypasses all the other
1538  *      protocol layers and you must therefore supply it with a complete frame
1539  */
1540
1541 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1542                                struct msghdr *msg, size_t len)
1543 {
1544         struct sock *sk = sock->sk;
1545         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1546         struct sk_buff *skb = NULL;
1547         struct net_device *dev;
1548         __be16 proto = 0;
1549         int err;
1550         int extra_len = 0;
1551
1552         /*
1553          *      Get and verify the address.
1554          */
1555
1556         if (saddr) {
1557                 if (msg->msg_namelen < sizeof(struct sockaddr))
1558                         return -EINVAL;
1559                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1560                         proto = saddr->spkt_protocol;
1561         } else
1562                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1563
1564         /*
1565          *      Find the device first to size check it
1566          */
1567
1568         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1569 retry:
1570         rcu_read_lock();
1571         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1572         err = -ENODEV;
1573         if (dev == NULL)
1574                 goto out_unlock;
1575
1576         err = -ENETDOWN;
1577         if (!(dev->flags & IFF_UP))
1578                 goto out_unlock;
1579
1580         /*
1581          * You may not queue a frame bigger than the mtu. This is the lowest level
1582          * raw protocol and you must do your own fragmentation at this level.
1583          */
1584
1585         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1586                 if (!netif_supports_nofcs(dev)) {
1587                         err = -EPROTONOSUPPORT;
1588                         goto out_unlock;
1589                 }
1590                 extra_len = 4; /* We're doing our own CRC */
1591         }
1592
1593         err = -EMSGSIZE;
1594         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1595                 goto out_unlock;
1596
1597         if (!skb) {
1598                 size_t reserved = LL_RESERVED_SPACE(dev);
1599                 int tlen = dev->needed_tailroom;
1600                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1601
1602                 rcu_read_unlock();
1603                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1604                 if (skb == NULL)
1605                         return -ENOBUFS;
1606                 /* FIXME: Save some space for broken drivers that write a hard
1607                  * header at transmission time by themselves. PPP is the notable
1608                  * one here. This should really be fixed at the driver level.
1609                  */
1610                 skb_reserve(skb, reserved);
1611                 skb_reset_network_header(skb);
1612
1613                 /* Try to align data part correctly */
1614                 if (hhlen) {
1615                         skb->data -= hhlen;
1616                         skb->tail -= hhlen;
1617                         if (len < hhlen)
1618                                 skb_reset_network_header(skb);
1619                 }
1620                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1621                 if (err)
1622                         goto out_free;
1623                 goto retry;
1624         }
1625
1626         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1627                 /* Earlier code assumed this would be a VLAN pkt,
1628                  * double-check this now that we have the actual
1629                  * packet in hand.
1630                  */
1631                 struct ethhdr *ehdr;
1632                 skb_reset_mac_header(skb);
1633                 ehdr = eth_hdr(skb);
1634                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1635                         err = -EMSGSIZE;
1636                         goto out_unlock;
1637                 }
1638         }
1639
1640         skb->protocol = proto;
1641         skb->dev = dev;
1642         skb->priority = sk->sk_priority;
1643         skb->mark = sk->sk_mark;
1644
1645         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1646
1647         if (unlikely(extra_len == 4))
1648                 skb->no_fcs = 1;
1649
1650         skb_probe_transport_header(skb, 0);
1651
1652         dev_queue_xmit(skb);
1653         rcu_read_unlock();
1654         return len;
1655
1656 out_unlock:
1657         rcu_read_unlock();
1658 out_free:
1659         kfree_skb(skb);
1660         return err;
1661 }
1662
1663 static unsigned int run_filter(const struct sk_buff *skb,
1664                                       const struct sock *sk,
1665                                       unsigned int res)
1666 {
1667         struct sk_filter *filter;
1668
1669         rcu_read_lock();
1670         filter = rcu_dereference(sk->sk_filter);
1671         if (filter != NULL)
1672                 res = SK_RUN_FILTER(filter, skb);
1673         rcu_read_unlock();
1674
1675         return res;
1676 }
1677
1678 /*
1679  * This function makes lazy skb cloning in hope that most of packets
1680  * are discarded by BPF.
1681  *
1682  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1683  * and skb->cb are mangled. It works because (and until) packets
1684  * falling here are owned by current CPU. Output packets are cloned
1685  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1686  * sequencially, so that if we return skb to original state on exit,
1687  * we will not harm anyone.
1688  */
1689
1690 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1691                       struct packet_type *pt, struct net_device *orig_dev)
1692 {
1693         struct sock *sk;
1694         struct sockaddr_ll *sll;
1695         struct packet_sock *po;
1696         u8 *skb_head = skb->data;
1697         int skb_len = skb->len;
1698         unsigned int snaplen, res;
1699
1700         if (skb->pkt_type == PACKET_LOOPBACK)
1701                 goto drop;
1702
1703         sk = pt->af_packet_priv;
1704         po = pkt_sk(sk);
1705
1706         if (!net_eq(dev_net(dev), sock_net(sk)))
1707                 goto drop;
1708
1709         skb->dev = dev;
1710
1711         if (dev->header_ops) {
1712                 /* The device has an explicit notion of ll header,
1713                  * exported to higher levels.
1714                  *
1715                  * Otherwise, the device hides details of its frame
1716                  * structure, so that corresponding packet head is
1717                  * never delivered to user.
1718                  */
1719                 if (sk->sk_type != SOCK_DGRAM)
1720                         skb_push(skb, skb->data - skb_mac_header(skb));
1721                 else if (skb->pkt_type == PACKET_OUTGOING) {
1722                         /* Special case: outgoing packets have ll header at head */
1723                         skb_pull(skb, skb_network_offset(skb));
1724                 }
1725         }
1726
1727         snaplen = skb->len;
1728
1729         res = run_filter(skb, sk, snaplen);
1730         if (!res)
1731                 goto drop_n_restore;
1732         if (snaplen > res)
1733                 snaplen = res;
1734
1735         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1736                 goto drop_n_acct;
1737
1738         if (skb_shared(skb)) {
1739                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1740                 if (nskb == NULL)
1741                         goto drop_n_acct;
1742
1743                 if (skb_head != skb->data) {
1744                         skb->data = skb_head;
1745                         skb->len = skb_len;
1746                 }
1747                 consume_skb(skb);
1748                 skb = nskb;
1749         }
1750
1751         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1752                      sizeof(skb->cb));
1753
1754         sll = &PACKET_SKB_CB(skb)->sa.ll;
1755         sll->sll_family = AF_PACKET;
1756         sll->sll_hatype = dev->type;
1757         sll->sll_protocol = skb->protocol;
1758         sll->sll_pkttype = skb->pkt_type;
1759         if (unlikely(po->origdev))
1760                 sll->sll_ifindex = orig_dev->ifindex;
1761         else
1762                 sll->sll_ifindex = dev->ifindex;
1763
1764         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1765
1766         PACKET_SKB_CB(skb)->origlen = skb->len;
1767
1768         if (pskb_trim(skb, snaplen))
1769                 goto drop_n_acct;
1770
1771         skb_set_owner_r(skb, sk);
1772         skb->dev = NULL;
1773         skb_dst_drop(skb);
1774
1775         /* drop conntrack reference */
1776         nf_reset(skb);
1777
1778         spin_lock(&sk->sk_receive_queue.lock);
1779         po->stats.stats1.tp_packets++;
1780         skb->dropcount = atomic_read(&sk->sk_drops);
1781         __skb_queue_tail(&sk->sk_receive_queue, skb);
1782         spin_unlock(&sk->sk_receive_queue.lock);
1783         sk->sk_data_ready(sk, skb->len);
1784         return 0;
1785
1786 drop_n_acct:
1787         spin_lock(&sk->sk_receive_queue.lock);
1788         po->stats.stats1.tp_drops++;
1789         atomic_inc(&sk->sk_drops);
1790         spin_unlock(&sk->sk_receive_queue.lock);
1791
1792 drop_n_restore:
1793         if (skb_head != skb->data && skb_shared(skb)) {
1794                 skb->data = skb_head;
1795                 skb->len = skb_len;
1796         }
1797 drop:
1798         consume_skb(skb);
1799         return 0;
1800 }
1801
1802 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1803                        struct packet_type *pt, struct net_device *orig_dev)
1804 {
1805         struct sock *sk;
1806         struct packet_sock *po;
1807         struct sockaddr_ll *sll;
1808         union tpacket_uhdr h;
1809         u8 *skb_head = skb->data;
1810         int skb_len = skb->len;
1811         unsigned int snaplen, res;
1812         unsigned long status = TP_STATUS_USER;
1813         unsigned short macoff, netoff, hdrlen;
1814         struct sk_buff *copy_skb = NULL;
1815         struct timespec ts;
1816         __u32 ts_status;
1817
1818         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1819          * We may add members to them until current aligned size without forcing
1820          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1821          */
1822         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1823         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1824
1825         if (skb->pkt_type == PACKET_LOOPBACK)
1826                 goto drop;
1827
1828         sk = pt->af_packet_priv;
1829         po = pkt_sk(sk);
1830
1831         if (!net_eq(dev_net(dev), sock_net(sk)))
1832                 goto drop;
1833
1834         if (dev->header_ops) {
1835                 if (sk->sk_type != SOCK_DGRAM)
1836                         skb_push(skb, skb->data - skb_mac_header(skb));
1837                 else if (skb->pkt_type == PACKET_OUTGOING) {
1838                         /* Special case: outgoing packets have ll header at head */
1839                         skb_pull(skb, skb_network_offset(skb));
1840                 }
1841         }
1842
1843         if (skb->ip_summed == CHECKSUM_PARTIAL)
1844                 status |= TP_STATUS_CSUMNOTREADY;
1845
1846         snaplen = skb->len;
1847
1848         res = run_filter(skb, sk, snaplen);
1849         if (!res)
1850                 goto drop_n_restore;
1851         if (snaplen > res)
1852                 snaplen = res;
1853
1854         if (sk->sk_type == SOCK_DGRAM) {
1855                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1856                                   po->tp_reserve;
1857         } else {
1858                 unsigned int maclen = skb_network_offset(skb);
1859                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1860                                        (maclen < 16 ? 16 : maclen)) +
1861                         po->tp_reserve;
1862                 macoff = netoff - maclen;
1863         }
1864         if (po->tp_version <= TPACKET_V2) {
1865                 if (macoff + snaplen > po->rx_ring.frame_size) {
1866                         if (po->copy_thresh &&
1867                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1868                                 if (skb_shared(skb)) {
1869                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1870                                 } else {
1871                                         copy_skb = skb_get(skb);
1872                                         skb_head = skb->data;
1873                                 }
1874                                 if (copy_skb)
1875                                         skb_set_owner_r(copy_skb, sk);
1876                         }
1877                         snaplen = po->rx_ring.frame_size - macoff;
1878                         if ((int)snaplen < 0)
1879                                 snaplen = 0;
1880                 }
1881         }
1882         spin_lock(&sk->sk_receive_queue.lock);
1883         h.raw = packet_current_rx_frame(po, skb,
1884                                         TP_STATUS_KERNEL, (macoff+snaplen));
1885         if (!h.raw)
1886                 goto ring_is_full;
1887         if (po->tp_version <= TPACKET_V2) {
1888                 packet_increment_rx_head(po, &po->rx_ring);
1889         /*
1890          * LOSING will be reported till you read the stats,
1891          * because it's COR - Clear On Read.
1892          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1893          * at packet level.
1894          */
1895                 if (po->stats.stats1.tp_drops)
1896                         status |= TP_STATUS_LOSING;
1897         }
1898         po->stats.stats1.tp_packets++;
1899         if (copy_skb) {
1900                 status |= TP_STATUS_COPY;
1901                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1902         }
1903         spin_unlock(&sk->sk_receive_queue.lock);
1904
1905         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1906
1907         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1908                 getnstimeofday(&ts);
1909
1910         status |= ts_status;
1911
1912         switch (po->tp_version) {
1913         case TPACKET_V1:
1914                 h.h1->tp_len = skb->len;
1915                 h.h1->tp_snaplen = snaplen;
1916                 h.h1->tp_mac = macoff;
1917                 h.h1->tp_net = netoff;
1918                 h.h1->tp_sec = ts.tv_sec;
1919                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1920                 hdrlen = sizeof(*h.h1);
1921                 break;
1922         case TPACKET_V2:
1923                 h.h2->tp_len = skb->len;
1924                 h.h2->tp_snaplen = snaplen;
1925                 h.h2->tp_mac = macoff;
1926                 h.h2->tp_net = netoff;
1927                 h.h2->tp_sec = ts.tv_sec;
1928                 h.h2->tp_nsec = ts.tv_nsec;
1929                 if (vlan_tx_tag_present(skb)) {
1930                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1931                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
1932                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1933                 } else {
1934                         h.h2->tp_vlan_tci = 0;
1935                         h.h2->tp_vlan_tpid = 0;
1936                 }
1937                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
1938                 hdrlen = sizeof(*h.h2);
1939                 break;
1940         case TPACKET_V3:
1941                 /* tp_nxt_offset,vlan are already populated above.
1942                  * So DONT clear those fields here
1943                  */
1944                 h.h3->tp_status |= status;
1945                 h.h3->tp_len = skb->len;
1946                 h.h3->tp_snaplen = snaplen;
1947                 h.h3->tp_mac = macoff;
1948                 h.h3->tp_net = netoff;
1949                 h.h3->tp_sec  = ts.tv_sec;
1950                 h.h3->tp_nsec = ts.tv_nsec;
1951                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
1952                 hdrlen = sizeof(*h.h3);
1953                 break;
1954         default:
1955                 BUG();
1956         }
1957
1958         sll = h.raw + TPACKET_ALIGN(hdrlen);
1959         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1960         sll->sll_family = AF_PACKET;
1961         sll->sll_hatype = dev->type;
1962         sll->sll_protocol = skb->protocol;
1963         sll->sll_pkttype = skb->pkt_type;
1964         if (unlikely(po->origdev))
1965                 sll->sll_ifindex = orig_dev->ifindex;
1966         else
1967                 sll->sll_ifindex = dev->ifindex;
1968
1969         smp_mb();
1970 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1971         {
1972                 u8 *start, *end;
1973
1974                 if (po->tp_version <= TPACKET_V2) {
1975                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1976                                 + macoff + snaplen);
1977                         for (start = h.raw; start < end; start += PAGE_SIZE)
1978                                 flush_dcache_page(pgv_to_page(start));
1979                 }
1980                 smp_wmb();
1981         }
1982 #endif
1983         if (po->tp_version <= TPACKET_V2)
1984                 __packet_set_status(po, h.raw, status);
1985         else
1986                 prb_clear_blk_fill_status(&po->rx_ring);
1987
1988         sk->sk_data_ready(sk, 0);
1989
1990 drop_n_restore:
1991         if (skb_head != skb->data && skb_shared(skb)) {
1992                 skb->data = skb_head;
1993                 skb->len = skb_len;
1994         }
1995 drop:
1996         kfree_skb(skb);
1997         return 0;
1998
1999 ring_is_full:
2000         po->stats.stats1.tp_drops++;
2001         spin_unlock(&sk->sk_receive_queue.lock);
2002
2003         sk->sk_data_ready(sk, 0);
2004         kfree_skb(copy_skb);
2005         goto drop_n_restore;
2006 }
2007
2008 static void tpacket_destruct_skb(struct sk_buff *skb)
2009 {
2010         struct packet_sock *po = pkt_sk(skb->sk);
2011         void *ph;
2012
2013         if (likely(po->tx_ring.pg_vec)) {
2014                 __u32 ts;
2015
2016                 ph = skb_shinfo(skb)->destructor_arg;
2017                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
2018                 atomic_dec(&po->tx_ring.pending);
2019
2020                 ts = __packet_set_timestamp(po, ph, skb);
2021                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2022         }
2023
2024         sock_wfree(skb);
2025 }
2026
2027 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2028                 void *frame, struct net_device *dev, int size_max,
2029                 __be16 proto, unsigned char *addr, int hlen)
2030 {
2031         union tpacket_uhdr ph;
2032         int to_write, offset, len, tp_len, nr_frags, len_max;
2033         struct socket *sock = po->sk.sk_socket;
2034         struct page *page;
2035         void *data;
2036         int err;
2037
2038         ph.raw = frame;
2039
2040         skb->protocol = proto;
2041         skb->dev = dev;
2042         skb->priority = po->sk.sk_priority;
2043         skb->mark = po->sk.sk_mark;
2044         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2045         skb_shinfo(skb)->destructor_arg = ph.raw;
2046
2047         switch (po->tp_version) {
2048         case TPACKET_V2:
2049                 tp_len = ph.h2->tp_len;
2050                 break;
2051         default:
2052                 tp_len = ph.h1->tp_len;
2053                 break;
2054         }
2055         if (unlikely(tp_len > size_max)) {
2056                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2057                 return -EMSGSIZE;
2058         }
2059
2060         skb_reserve(skb, hlen);
2061         skb_reset_network_header(skb);
2062
2063         if (!packet_use_direct_xmit(po))
2064                 skb_probe_transport_header(skb, 0);
2065         if (unlikely(po->tp_tx_has_off)) {
2066                 int off_min, off_max, off;
2067                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2068                 off_max = po->tx_ring.frame_size - tp_len;
2069                 if (sock->type == SOCK_DGRAM) {
2070                         switch (po->tp_version) {
2071                         case TPACKET_V2:
2072                                 off = ph.h2->tp_net;
2073                                 break;
2074                         default:
2075                                 off = ph.h1->tp_net;
2076                                 break;
2077                         }
2078                 } else {
2079                         switch (po->tp_version) {
2080                         case TPACKET_V2:
2081                                 off = ph.h2->tp_mac;
2082                                 break;
2083                         default:
2084                                 off = ph.h1->tp_mac;
2085                                 break;
2086                         }
2087                 }
2088                 if (unlikely((off < off_min) || (off_max < off)))
2089                         return -EINVAL;
2090                 data = ph.raw + off;
2091         } else {
2092                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2093         }
2094         to_write = tp_len;
2095
2096         if (sock->type == SOCK_DGRAM) {
2097                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2098                                 NULL, tp_len);
2099                 if (unlikely(err < 0))
2100                         return -EINVAL;
2101         } else if (dev->hard_header_len) {
2102                 /* net device doesn't like empty head */
2103                 if (unlikely(tp_len <= dev->hard_header_len)) {
2104                         pr_err("packet size is too short (%d < %d)\n",
2105                                tp_len, dev->hard_header_len);
2106                         return -EINVAL;
2107                 }
2108
2109                 skb_push(skb, dev->hard_header_len);
2110                 err = skb_store_bits(skb, 0, data,
2111                                 dev->hard_header_len);
2112                 if (unlikely(err))
2113                         return err;
2114
2115                 data += dev->hard_header_len;
2116                 to_write -= dev->hard_header_len;
2117         }
2118
2119         offset = offset_in_page(data);
2120         len_max = PAGE_SIZE - offset;
2121         len = ((to_write > len_max) ? len_max : to_write);
2122
2123         skb->data_len = to_write;
2124         skb->len += to_write;
2125         skb->truesize += to_write;
2126         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2127
2128         while (likely(to_write)) {
2129                 nr_frags = skb_shinfo(skb)->nr_frags;
2130
2131                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2132                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2133                                MAX_SKB_FRAGS);
2134                         return -EFAULT;
2135                 }
2136
2137                 page = pgv_to_page(data);
2138                 data += len;
2139                 flush_dcache_page(page);
2140                 get_page(page);
2141                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2142                 to_write -= len;
2143                 offset = 0;
2144                 len_max = PAGE_SIZE;
2145                 len = ((to_write > len_max) ? len_max : to_write);
2146         }
2147
2148         return tp_len;
2149 }
2150
2151 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2152 {
2153         struct sk_buff *skb;
2154         struct net_device *dev;
2155         __be16 proto;
2156         int err, reserve = 0;
2157         void *ph;
2158         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2159         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2160         int tp_len, size_max;
2161         unsigned char *addr;
2162         int len_sum = 0;
2163         int status = TP_STATUS_AVAILABLE;
2164         int hlen, tlen;
2165
2166         mutex_lock(&po->pg_vec_lock);
2167
2168         if (likely(saddr == NULL)) {
2169                 dev     = packet_cached_dev_get(po);
2170                 proto   = po->num;
2171                 addr    = NULL;
2172         } else {
2173                 err = -EINVAL;
2174                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2175                         goto out;
2176                 if (msg->msg_namelen < (saddr->sll_halen
2177                                         + offsetof(struct sockaddr_ll,
2178                                                 sll_addr)))
2179                         goto out;
2180                 proto   = saddr->sll_protocol;
2181                 addr    = saddr->sll_addr;
2182                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2183         }
2184
2185         err = -ENXIO;
2186         if (unlikely(dev == NULL))
2187                 goto out;
2188         err = -ENETDOWN;
2189         if (unlikely(!(dev->flags & IFF_UP)))
2190                 goto out_put;
2191
2192         reserve = dev->hard_header_len;
2193
2194         size_max = po->tx_ring.frame_size
2195                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2196
2197         if (size_max > dev->mtu + reserve)
2198                 size_max = dev->mtu + reserve;
2199
2200         do {
2201                 ph = packet_current_frame(po, &po->tx_ring,
2202                                           TP_STATUS_SEND_REQUEST);
2203                 if (unlikely(ph == NULL)) {
2204                         if (need_wait && need_resched())
2205                                 schedule();
2206                         continue;
2207                 }
2208
2209                 status = TP_STATUS_SEND_REQUEST;
2210                 hlen = LL_RESERVED_SPACE(dev);
2211                 tlen = dev->needed_tailroom;
2212                 skb = sock_alloc_send_skb(&po->sk,
2213                                 hlen + tlen + sizeof(struct sockaddr_ll),
2214                                 0, &err);
2215
2216                 if (unlikely(skb == NULL))
2217                         goto out_status;
2218
2219                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2220                                 addr, hlen);
2221
2222                 if (unlikely(tp_len < 0)) {
2223                         if (po->tp_loss) {
2224                                 __packet_set_status(po, ph,
2225                                                 TP_STATUS_AVAILABLE);
2226                                 packet_increment_head(&po->tx_ring);
2227                                 kfree_skb(skb);
2228                                 continue;
2229                         } else {
2230                                 status = TP_STATUS_WRONG_FORMAT;
2231                                 err = tp_len;
2232                                 goto out_status;
2233                         }
2234                 }
2235
2236                 skb_set_queue_mapping(skb, packet_pick_tx_queue(dev));
2237                 skb->destructor = tpacket_destruct_skb;
2238                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2239                 atomic_inc(&po->tx_ring.pending);
2240
2241                 status = TP_STATUS_SEND_REQUEST;
2242                 err = po->xmit(skb);
2243                 if (unlikely(err > 0)) {
2244                         err = net_xmit_errno(err);
2245                         if (err && __packet_get_status(po, ph) ==
2246                                    TP_STATUS_AVAILABLE) {
2247                                 /* skb was destructed already */
2248                                 skb = NULL;
2249                                 goto out_status;
2250                         }
2251                         /*
2252                          * skb was dropped but not destructed yet;
2253                          * let's treat it like congestion or err < 0
2254                          */
2255                         err = 0;
2256                 }
2257                 packet_increment_head(&po->tx_ring);
2258                 len_sum += tp_len;
2259         } while (likely((ph != NULL) || (need_wait &&
2260                                          atomic_read(&po->tx_ring.pending))));
2261
2262         err = len_sum;
2263         goto out_put;
2264
2265 out_status:
2266         __packet_set_status(po, ph, status);
2267         kfree_skb(skb);
2268 out_put:
2269         dev_put(dev);
2270 out:
2271         mutex_unlock(&po->pg_vec_lock);
2272         return err;
2273 }
2274
2275 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2276                                         size_t reserve, size_t len,
2277                                         size_t linear, int noblock,
2278                                         int *err)
2279 {
2280         struct sk_buff *skb;
2281
2282         /* Under a page?  Don't bother with paged skb. */
2283         if (prepad + len < PAGE_SIZE || !linear)
2284                 linear = len;
2285
2286         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2287                                    err, 0);
2288         if (!skb)
2289                 return NULL;
2290
2291         skb_reserve(skb, reserve);
2292         skb_put(skb, linear);
2293         skb->data_len = len - linear;
2294         skb->len += len - linear;
2295
2296         return skb;
2297 }
2298
2299 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2300 {
2301         struct sock *sk = sock->sk;
2302         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2303         struct sk_buff *skb;
2304         struct net_device *dev;
2305         __be16 proto;
2306         unsigned char *addr;
2307         int err, reserve = 0;
2308         struct virtio_net_hdr vnet_hdr = { 0 };
2309         int offset = 0;
2310         int vnet_hdr_len;
2311         struct packet_sock *po = pkt_sk(sk);
2312         unsigned short gso_type = 0;
2313         int hlen, tlen;
2314         int extra_len = 0;
2315
2316         /*
2317          *      Get and verify the address.
2318          */
2319
2320         if (likely(saddr == NULL)) {
2321                 dev     = packet_cached_dev_get(po);
2322                 proto   = po->num;
2323                 addr    = NULL;
2324         } else {
2325                 err = -EINVAL;
2326                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2327                         goto out;
2328                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2329                         goto out;
2330                 proto   = saddr->sll_protocol;
2331                 addr    = saddr->sll_addr;
2332                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2333         }
2334
2335         err = -ENXIO;
2336         if (unlikely(dev == NULL))
2337                 goto out_unlock;
2338         err = -ENETDOWN;
2339         if (unlikely(!(dev->flags & IFF_UP)))
2340                 goto out_unlock;
2341
2342         if (sock->type == SOCK_RAW)
2343                 reserve = dev->hard_header_len;
2344         if (po->has_vnet_hdr) {
2345                 vnet_hdr_len = sizeof(vnet_hdr);
2346
2347                 err = -EINVAL;
2348                 if (len < vnet_hdr_len)
2349                         goto out_unlock;
2350
2351                 len -= vnet_hdr_len;
2352
2353                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2354                                        vnet_hdr_len);
2355                 if (err < 0)
2356                         goto out_unlock;
2357
2358                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2359                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2360                       vnet_hdr.hdr_len))
2361                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2362                                                  vnet_hdr.csum_offset + 2;
2363
2364                 err = -EINVAL;
2365                 if (vnet_hdr.hdr_len > len)
2366                         goto out_unlock;
2367
2368                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2369                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2370                         case VIRTIO_NET_HDR_GSO_TCPV4:
2371                                 gso_type = SKB_GSO_TCPV4;
2372                                 break;
2373                         case VIRTIO_NET_HDR_GSO_TCPV6:
2374                                 gso_type = SKB_GSO_TCPV6;
2375                                 break;
2376                         case VIRTIO_NET_HDR_GSO_UDP:
2377                                 gso_type = SKB_GSO_UDP;
2378                                 break;
2379                         default:
2380                                 goto out_unlock;
2381                         }
2382
2383                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2384                                 gso_type |= SKB_GSO_TCP_ECN;
2385
2386                         if (vnet_hdr.gso_size == 0)
2387                                 goto out_unlock;
2388
2389                 }
2390         }
2391
2392         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2393                 if (!netif_supports_nofcs(dev)) {
2394                         err = -EPROTONOSUPPORT;
2395                         goto out_unlock;
2396                 }
2397                 extra_len = 4; /* We're doing our own CRC */
2398         }
2399
2400         err = -EMSGSIZE;
2401         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2402                 goto out_unlock;
2403
2404         err = -ENOBUFS;
2405         hlen = LL_RESERVED_SPACE(dev);
2406         tlen = dev->needed_tailroom;
2407         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2408                                msg->msg_flags & MSG_DONTWAIT, &err);
2409         if (skb == NULL)
2410                 goto out_unlock;
2411
2412         skb_set_network_header(skb, reserve);
2413
2414         err = -EINVAL;
2415         if (sock->type == SOCK_DGRAM &&
2416             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2417                 goto out_free;
2418
2419         /* Returns -EFAULT on error */
2420         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2421         if (err)
2422                 goto out_free;
2423
2424         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2425
2426         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2427                 /* Earlier code assumed this would be a VLAN pkt,
2428                  * double-check this now that we have the actual
2429                  * packet in hand.
2430                  */
2431                 struct ethhdr *ehdr;
2432                 skb_reset_mac_header(skb);
2433                 ehdr = eth_hdr(skb);
2434                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2435                         err = -EMSGSIZE;
2436                         goto out_free;
2437                 }
2438         }
2439
2440         skb->protocol = proto;
2441         skb->dev = dev;
2442         skb->priority = sk->sk_priority;
2443         skb->mark = sk->sk_mark;
2444         skb_set_queue_mapping(skb, packet_pick_tx_queue(dev));
2445
2446         if (po->has_vnet_hdr) {
2447                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2448                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2449                                                   vnet_hdr.csum_offset)) {
2450                                 err = -EINVAL;
2451                                 goto out_free;
2452                         }
2453                 }
2454
2455                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2456                 skb_shinfo(skb)->gso_type = gso_type;
2457
2458                 /* Header must be checked, and gso_segs computed. */
2459                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2460                 skb_shinfo(skb)->gso_segs = 0;
2461
2462                 len += vnet_hdr_len;
2463         }
2464
2465         if (!packet_use_direct_xmit(po))
2466                 skb_probe_transport_header(skb, reserve);
2467         if (unlikely(extra_len == 4))
2468                 skb->no_fcs = 1;
2469
2470         err = po->xmit(skb);
2471         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2472                 goto out_unlock;
2473
2474         dev_put(dev);
2475
2476         return len;
2477
2478 out_free:
2479         kfree_skb(skb);
2480 out_unlock:
2481         if (dev)
2482                 dev_put(dev);
2483 out:
2484         return err;
2485 }
2486
2487 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2488                 struct msghdr *msg, size_t len)
2489 {
2490         struct sock *sk = sock->sk;
2491         struct packet_sock *po = pkt_sk(sk);
2492
2493         if (po->tx_ring.pg_vec)
2494                 return tpacket_snd(po, msg);
2495         else
2496                 return packet_snd(sock, msg, len);
2497 }
2498
2499 /*
2500  *      Close a PACKET socket. This is fairly simple. We immediately go
2501  *      to 'closed' state and remove our protocol entry in the device list.
2502  */
2503
2504 static int packet_release(struct socket *sock)
2505 {
2506         struct sock *sk = sock->sk;
2507         struct packet_sock *po;
2508         struct net *net;
2509         union tpacket_req_u req_u;
2510
2511         if (!sk)
2512                 return 0;
2513
2514         net = sock_net(sk);
2515         po = pkt_sk(sk);
2516
2517         mutex_lock(&net->packet.sklist_lock);
2518         sk_del_node_init_rcu(sk);
2519         mutex_unlock(&net->packet.sklist_lock);
2520
2521         preempt_disable();
2522         sock_prot_inuse_add(net, sk->sk_prot, -1);
2523         preempt_enable();
2524
2525         spin_lock(&po->bind_lock);
2526         unregister_prot_hook(sk, false);
2527         packet_cached_dev_reset(po);
2528
2529         if (po->prot_hook.dev) {
2530                 dev_put(po->prot_hook.dev);
2531                 po->prot_hook.dev = NULL;
2532         }
2533         spin_unlock(&po->bind_lock);
2534
2535         packet_flush_mclist(sk);
2536
2537         if (po->rx_ring.pg_vec) {
2538                 memset(&req_u, 0, sizeof(req_u));
2539                 packet_set_ring(sk, &req_u, 1, 0);
2540         }
2541
2542         if (po->tx_ring.pg_vec) {
2543                 memset(&req_u, 0, sizeof(req_u));
2544                 packet_set_ring(sk, &req_u, 1, 1);
2545         }
2546
2547         fanout_release(sk);
2548
2549         synchronize_net();
2550         /*
2551          *      Now the socket is dead. No more input will appear.
2552          */
2553         sock_orphan(sk);
2554         sock->sk = NULL;
2555
2556         /* Purge queues */
2557
2558         skb_queue_purge(&sk->sk_receive_queue);
2559         sk_refcnt_debug_release(sk);
2560
2561         sock_put(sk);
2562         return 0;
2563 }
2564
2565 /*
2566  *      Attach a packet hook.
2567  */
2568
2569 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2570 {
2571         struct packet_sock *po = pkt_sk(sk);
2572         const struct net_device *dev_curr;
2573         __be16 proto_curr;
2574         bool need_rehook;
2575
2576         if (po->fanout) {
2577                 if (dev)
2578                         dev_put(dev);
2579
2580                 return -EINVAL;
2581         }
2582
2583         lock_sock(sk);
2584         spin_lock(&po->bind_lock);
2585
2586         proto_curr = po->prot_hook.type;
2587         dev_curr = po->prot_hook.dev;
2588
2589         need_rehook = proto_curr != proto || dev_curr != dev;
2590
2591         if (need_rehook) {
2592                 unregister_prot_hook(sk, true);
2593
2594                 po->num = proto;
2595                 po->prot_hook.type = proto;
2596
2597                 if (po->prot_hook.dev)
2598                         dev_put(po->prot_hook.dev);
2599
2600                 po->prot_hook.dev = dev;
2601
2602                 po->ifindex = dev ? dev->ifindex : 0;
2603                 packet_cached_dev_assign(po, dev);
2604         }
2605
2606         if (proto == 0 || !need_rehook)
2607                 goto out_unlock;
2608
2609         if (!dev || (dev->flags & IFF_UP)) {
2610                 register_prot_hook(sk);
2611         } else {
2612                 sk->sk_err = ENETDOWN;
2613                 if (!sock_flag(sk, SOCK_DEAD))
2614                         sk->sk_error_report(sk);
2615         }
2616
2617 out_unlock:
2618         spin_unlock(&po->bind_lock);
2619         release_sock(sk);
2620         return 0;
2621 }
2622
2623 /*
2624  *      Bind a packet socket to a device
2625  */
2626
2627 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2628                             int addr_len)
2629 {
2630         struct sock *sk = sock->sk;
2631         char name[15];
2632         struct net_device *dev;
2633         int err = -ENODEV;
2634
2635         /*
2636          *      Check legality
2637          */
2638
2639         if (addr_len != sizeof(struct sockaddr))
2640                 return -EINVAL;
2641         strlcpy(name, uaddr->sa_data, sizeof(name));
2642
2643         dev = dev_get_by_name(sock_net(sk), name);
2644         if (dev)
2645                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2646         return err;
2647 }
2648
2649 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2650 {
2651         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2652         struct sock *sk = sock->sk;
2653         struct net_device *dev = NULL;
2654         int err;
2655
2656
2657         /*
2658          *      Check legality
2659          */
2660
2661         if (addr_len < sizeof(struct sockaddr_ll))
2662                 return -EINVAL;
2663         if (sll->sll_family != AF_PACKET)
2664                 return -EINVAL;
2665
2666         if (sll->sll_ifindex) {
2667                 err = -ENODEV;
2668                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2669                 if (dev == NULL)
2670                         goto out;
2671         }
2672         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2673
2674 out:
2675         return err;
2676 }
2677
2678 static struct proto packet_proto = {
2679         .name     = "PACKET",
2680         .owner    = THIS_MODULE,
2681         .obj_size = sizeof(struct packet_sock),
2682 };
2683
2684 /*
2685  *      Create a packet of type SOCK_PACKET.
2686  */
2687
2688 static int packet_create(struct net *net, struct socket *sock, int protocol,
2689                          int kern)
2690 {
2691         struct sock *sk;
2692         struct packet_sock *po;
2693         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2694         int err;
2695
2696         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2697                 return -EPERM;
2698         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2699             sock->type != SOCK_PACKET)
2700                 return -ESOCKTNOSUPPORT;
2701
2702         sock->state = SS_UNCONNECTED;
2703
2704         err = -ENOBUFS;
2705         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2706         if (sk == NULL)
2707                 goto out;
2708
2709         sock->ops = &packet_ops;
2710         if (sock->type == SOCK_PACKET)
2711                 sock->ops = &packet_ops_spkt;
2712
2713         sock_init_data(sock, sk);
2714
2715         po = pkt_sk(sk);
2716         sk->sk_family = PF_PACKET;
2717         po->num = proto;
2718         po->xmit = dev_queue_xmit;
2719
2720         packet_cached_dev_reset(po);
2721
2722         sk->sk_destruct = packet_sock_destruct;
2723         sk_refcnt_debug_inc(sk);
2724
2725         /*
2726          *      Attach a protocol block
2727          */
2728
2729         spin_lock_init(&po->bind_lock);
2730         mutex_init(&po->pg_vec_lock);
2731         po->prot_hook.func = packet_rcv;
2732
2733         if (sock->type == SOCK_PACKET)
2734                 po->prot_hook.func = packet_rcv_spkt;
2735
2736         po->prot_hook.af_packet_priv = sk;
2737
2738         if (proto) {
2739                 po->prot_hook.type = proto;
2740                 register_prot_hook(sk);
2741         }
2742
2743         mutex_lock(&net->packet.sklist_lock);
2744         sk_add_node_rcu(sk, &net->packet.sklist);
2745         mutex_unlock(&net->packet.sklist_lock);
2746
2747         preempt_disable();
2748         sock_prot_inuse_add(net, &packet_proto, 1);
2749         preempt_enable();
2750
2751         return 0;
2752 out:
2753         return err;
2754 }
2755
2756 /*
2757  *      Pull a packet from our receive queue and hand it to the user.
2758  *      If necessary we block.
2759  */
2760
2761 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2762                           struct msghdr *msg, size_t len, int flags)
2763 {
2764         struct sock *sk = sock->sk;
2765         struct sk_buff *skb;
2766         int copied, err;
2767         int vnet_hdr_len = 0;
2768
2769         err = -EINVAL;
2770         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2771                 goto out;
2772
2773 #if 0
2774         /* What error should we return now? EUNATTACH? */
2775         if (pkt_sk(sk)->ifindex < 0)
2776                 return -ENODEV;
2777 #endif
2778
2779         if (flags & MSG_ERRQUEUE) {
2780                 err = sock_recv_errqueue(sk, msg, len,
2781                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
2782                 goto out;
2783         }
2784
2785         /*
2786          *      Call the generic datagram receiver. This handles all sorts
2787          *      of horrible races and re-entrancy so we can forget about it
2788          *      in the protocol layers.
2789          *
2790          *      Now it will return ENETDOWN, if device have just gone down,
2791          *      but then it will block.
2792          */
2793
2794         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2795
2796         /*
2797          *      An error occurred so return it. Because skb_recv_datagram()
2798          *      handles the blocking we don't see and worry about blocking
2799          *      retries.
2800          */
2801
2802         if (skb == NULL)
2803                 goto out;
2804
2805         if (pkt_sk(sk)->has_vnet_hdr) {
2806                 struct virtio_net_hdr vnet_hdr = { 0 };
2807
2808                 err = -EINVAL;
2809                 vnet_hdr_len = sizeof(vnet_hdr);
2810                 if (len < vnet_hdr_len)
2811                         goto out_free;
2812
2813                 len -= vnet_hdr_len;
2814
2815                 if (skb_is_gso(skb)) {
2816                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2817
2818                         /* This is a hint as to how much should be linear. */
2819                         vnet_hdr.hdr_len = skb_headlen(skb);
2820                         vnet_hdr.gso_size = sinfo->gso_size;
2821                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2822                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2823                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2824                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2825                         else if (sinfo->gso_type & SKB_GSO_UDP)
2826                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2827                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2828                                 goto out_free;
2829                         else
2830                                 BUG();
2831                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2832                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2833                 } else
2834                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2835
2836                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2837                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2838                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2839                         vnet_hdr.csum_offset = skb->csum_offset;
2840                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2841                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2842                 } /* else everything is zero */
2843
2844                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2845                                      vnet_hdr_len);
2846                 if (err < 0)
2847                         goto out_free;
2848         }
2849
2850         /* You lose any data beyond the buffer you gave. If it worries
2851          * a user program they can ask the device for its MTU
2852          * anyway.
2853          */
2854         copied = skb->len;
2855         if (copied > len) {
2856                 copied = len;
2857                 msg->msg_flags |= MSG_TRUNC;
2858         }
2859
2860         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2861         if (err)
2862                 goto out_free;
2863
2864         sock_recv_ts_and_drops(msg, sk, skb);
2865
2866         if (msg->msg_name) {
2867                 /* If the address length field is there to be filled
2868                  * in, we fill it in now.
2869                  */
2870                 if (sock->type == SOCK_PACKET) {
2871                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2872                 } else {
2873                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2874                         msg->msg_namelen = sll->sll_halen +
2875                                 offsetof(struct sockaddr_ll, sll_addr);
2876                 }
2877                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2878                        msg->msg_namelen);
2879         }
2880
2881         if (pkt_sk(sk)->auxdata) {
2882                 struct tpacket_auxdata aux;
2883
2884                 aux.tp_status = TP_STATUS_USER;
2885                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2886                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2887                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2888                 aux.tp_snaplen = skb->len;
2889                 aux.tp_mac = 0;
2890                 aux.tp_net = skb_network_offset(skb);
2891                 if (vlan_tx_tag_present(skb)) {
2892                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2893                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
2894                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2895                 } else {
2896                         aux.tp_vlan_tci = 0;
2897                         aux.tp_vlan_tpid = 0;
2898                 }
2899                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2900         }
2901
2902         /*
2903          *      Free or return the buffer as appropriate. Again this
2904          *      hides all the races and re-entrancy issues from us.
2905          */
2906         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2907
2908 out_free:
2909         skb_free_datagram(sk, skb);
2910 out:
2911         return err;
2912 }
2913
2914 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2915                                int *uaddr_len, int peer)
2916 {
2917         struct net_device *dev;
2918         struct sock *sk = sock->sk;
2919
2920         if (peer)
2921                 return -EOPNOTSUPP;
2922
2923         uaddr->sa_family = AF_PACKET;
2924         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2925         rcu_read_lock();
2926         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2927         if (dev)
2928                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2929         rcu_read_unlock();
2930         *uaddr_len = sizeof(*uaddr);
2931
2932         return 0;
2933 }
2934
2935 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2936                           int *uaddr_len, int peer)
2937 {
2938         struct net_device *dev;
2939         struct sock *sk = sock->sk;
2940         struct packet_sock *po = pkt_sk(sk);
2941         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2942
2943         if (peer)
2944                 return -EOPNOTSUPP;
2945
2946         sll->sll_family = AF_PACKET;
2947         sll->sll_ifindex = po->ifindex;
2948         sll->sll_protocol = po->num;
2949         sll->sll_pkttype = 0;
2950         rcu_read_lock();
2951         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2952         if (dev) {
2953                 sll->sll_hatype = dev->type;
2954                 sll->sll_halen = dev->addr_len;
2955                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2956         } else {
2957                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2958                 sll->sll_halen = 0;
2959         }
2960         rcu_read_unlock();
2961         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2962
2963         return 0;
2964 }
2965
2966 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2967                          int what)
2968 {
2969         switch (i->type) {
2970         case PACKET_MR_MULTICAST:
2971                 if (i->alen != dev->addr_len)
2972                         return -EINVAL;
2973                 if (what > 0)
2974                         return dev_mc_add(dev, i->addr);
2975                 else
2976                         return dev_mc_del(dev, i->addr);
2977                 break;
2978         case PACKET_MR_PROMISC:
2979                 return dev_set_promiscuity(dev, what);
2980                 break;
2981         case PACKET_MR_ALLMULTI:
2982                 return dev_set_allmulti(dev, what);
2983                 break;
2984         case PACKET_MR_UNICAST:
2985                 if (i->alen != dev->addr_len)
2986                         return -EINVAL;
2987                 if (what > 0)
2988                         return dev_uc_add(dev, i->addr);
2989                 else
2990                         return dev_uc_del(dev, i->addr);
2991                 break;
2992         default:
2993                 break;
2994         }
2995         return 0;
2996 }
2997
2998 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2999 {
3000         for ( ; i; i = i->next) {
3001                 if (i->ifindex == dev->ifindex)
3002                         packet_dev_mc(dev, i, what);
3003         }
3004 }
3005
3006 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3007 {
3008         struct packet_sock *po = pkt_sk(sk);
3009         struct packet_mclist *ml, *i;
3010         struct net_device *dev;
3011         int err;
3012
3013         rtnl_lock();
3014
3015         err = -ENODEV;
3016         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3017         if (!dev)
3018                 goto done;
3019
3020         err = -EINVAL;
3021         if (mreq->mr_alen > dev->addr_len)
3022                 goto done;
3023
3024         err = -ENOBUFS;
3025         i = kmalloc(sizeof(*i), GFP_KERNEL);
3026         if (i == NULL)
3027                 goto done;
3028
3029         err = 0;
3030         for (ml = po->mclist; ml; ml = ml->next) {
3031                 if (ml->ifindex == mreq->mr_ifindex &&
3032                     ml->type == mreq->mr_type &&
3033                     ml->alen == mreq->mr_alen &&
3034                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3035                         ml->count++;
3036                         /* Free the new element ... */
3037                         kfree(i);
3038                         goto done;
3039                 }
3040         }
3041
3042         i->type = mreq->mr_type;
3043         i->ifindex = mreq->mr_ifindex;
3044         i->alen = mreq->mr_alen;
3045         memcpy(i->addr, mreq->mr_address, i->alen);
3046         i->count = 1;
3047         i->next = po->mclist;
3048         po->mclist = i;
3049         err = packet_dev_mc(dev, i, 1);
3050         if (err) {
3051                 po->mclist = i->next;
3052                 kfree(i);
3053         }
3054
3055 done:
3056         rtnl_unlock();
3057         return err;
3058 }
3059
3060 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3061 {
3062         struct packet_mclist *ml, **mlp;
3063
3064         rtnl_lock();
3065
3066         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3067                 if (ml->ifindex == mreq->mr_ifindex &&
3068                     ml->type == mreq->mr_type &&
3069                     ml->alen == mreq->mr_alen &&
3070                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3071                         if (--ml->count == 0) {
3072                                 struct net_device *dev;
3073                                 *mlp = ml->next;
3074                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3075                                 if (dev)
3076                                         packet_dev_mc(dev, ml, -1);
3077                                 kfree(ml);
3078                         }
3079                         rtnl_unlock();
3080                         return 0;
3081                 }
3082         }
3083         rtnl_unlock();
3084         return -EADDRNOTAVAIL;
3085 }
3086
3087 static void packet_flush_mclist(struct sock *sk)
3088 {
3089         struct packet_sock *po = pkt_sk(sk);
3090         struct packet_mclist *ml;
3091
3092         if (!po->mclist)
3093                 return;
3094
3095         rtnl_lock();
3096         while ((ml = po->mclist) != NULL) {
3097                 struct net_device *dev;
3098
3099                 po->mclist = ml->next;
3100                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3101                 if (dev != NULL)
3102                         packet_dev_mc(dev, ml, -1);
3103                 kfree(ml);
3104         }
3105         rtnl_unlock();
3106 }
3107
3108 static int
3109 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3110 {
3111         struct sock *sk = sock->sk;
3112         struct packet_sock *po = pkt_sk(sk);
3113         int ret;
3114
3115         if (level != SOL_PACKET)
3116                 return -ENOPROTOOPT;
3117
3118         switch (optname) {
3119         case PACKET_ADD_MEMBERSHIP:
3120         case PACKET_DROP_MEMBERSHIP:
3121         {
3122                 struct packet_mreq_max mreq;
3123                 int len = optlen;
3124                 memset(&mreq, 0, sizeof(mreq));
3125                 if (len < sizeof(struct packet_mreq))
3126                         return -EINVAL;
3127                 if (len > sizeof(mreq))
3128                         len = sizeof(mreq);
3129                 if (copy_from_user(&mreq, optval, len))
3130                         return -EFAULT;
3131                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3132                         return -EINVAL;
3133                 if (optname == PACKET_ADD_MEMBERSHIP)
3134                         ret = packet_mc_add(sk, &mreq);
3135                 else
3136                         ret = packet_mc_drop(sk, &mreq);
3137                 return ret;
3138         }
3139
3140         case PACKET_RX_RING:
3141         case PACKET_TX_RING:
3142         {
3143                 union tpacket_req_u req_u;
3144                 int len;
3145
3146                 switch (po->tp_version) {
3147                 case TPACKET_V1:
3148                 case TPACKET_V2:
3149                         len = sizeof(req_u.req);
3150                         break;
3151                 case TPACKET_V3:
3152                 default:
3153                         len = sizeof(req_u.req3);
3154                         break;
3155                 }
3156                 if (optlen < len)
3157                         return -EINVAL;
3158                 if (pkt_sk(sk)->has_vnet_hdr)
3159                         return -EINVAL;
3160                 if (copy_from_user(&req_u.req, optval, len))
3161                         return -EFAULT;
3162                 return packet_set_ring(sk, &req_u, 0,
3163                         optname == PACKET_TX_RING);
3164         }
3165         case PACKET_COPY_THRESH:
3166         {
3167                 int val;
3168
3169                 if (optlen != sizeof(val))
3170                         return -EINVAL;
3171                 if (copy_from_user(&val, optval, sizeof(val)))
3172                         return -EFAULT;
3173
3174                 pkt_sk(sk)->copy_thresh = val;
3175                 return 0;
3176         }
3177         case PACKET_VERSION:
3178         {
3179                 int val;
3180
3181                 if (optlen != sizeof(val))
3182                         return -EINVAL;
3183                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3184                         return -EBUSY;
3185                 if (copy_from_user(&val, optval, sizeof(val)))
3186                         return -EFAULT;
3187                 switch (val) {
3188                 case TPACKET_V1:
3189                 case TPACKET_V2:
3190                 case TPACKET_V3:
3191                         po->tp_version = val;
3192                         return 0;
3193                 default:
3194                         return -EINVAL;
3195                 }
3196         }
3197         case PACKET_RESERVE:
3198         {
3199                 unsigned int val;
3200
3201                 if (optlen != sizeof(val))
3202                         return -EINVAL;
3203                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3204                         return -EBUSY;
3205                 if (copy_from_user(&val, optval, sizeof(val)))
3206                         return -EFAULT;
3207                 po->tp_reserve = val;
3208                 return 0;
3209         }
3210         case PACKET_LOSS:
3211         {
3212                 unsigned int val;
3213
3214                 if (optlen != sizeof(val))
3215                         return -EINVAL;
3216                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3217                         return -EBUSY;
3218                 if (copy_from_user(&val, optval, sizeof(val)))
3219                         return -EFAULT;
3220                 po->tp_loss = !!val;
3221                 return 0;
3222         }
3223         case PACKET_AUXDATA:
3224         {
3225                 int val;
3226
3227                 if (optlen < sizeof(val))
3228                         return -EINVAL;
3229                 if (copy_from_user(&val, optval, sizeof(val)))
3230                         return -EFAULT;
3231
3232                 po->auxdata = !!val;
3233                 return 0;
3234         }
3235         case PACKET_ORIGDEV:
3236         {
3237                 int val;
3238
3239                 if (optlen < sizeof(val))
3240                         return -EINVAL;
3241                 if (copy_from_user(&val, optval, sizeof(val)))
3242                         return -EFAULT;
3243
3244                 po->origdev = !!val;
3245                 return 0;
3246         }
3247         case PACKET_VNET_HDR:
3248         {
3249                 int val;
3250
3251                 if (sock->type != SOCK_RAW)
3252                         return -EINVAL;
3253                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3254                         return -EBUSY;
3255                 if (optlen < sizeof(val))
3256                         return -EINVAL;
3257                 if (copy_from_user(&val, optval, sizeof(val)))
3258                         return -EFAULT;
3259
3260                 po->has_vnet_hdr = !!val;
3261                 return 0;
3262         }
3263         case PACKET_TIMESTAMP:
3264         {
3265                 int val;
3266
3267                 if (optlen != sizeof(val))
3268                         return -EINVAL;
3269                 if (copy_from_user(&val, optval, sizeof(val)))
3270                         return -EFAULT;
3271
3272                 po->tp_tstamp = val;
3273                 return 0;
3274         }
3275         case PACKET_FANOUT:
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                 return fanout_add(sk, val & 0xffff, val >> 16);
3285         }
3286         case PACKET_TX_HAS_OFF:
3287         {
3288                 unsigned int val;
3289
3290                 if (optlen != sizeof(val))
3291                         return -EINVAL;
3292                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3293                         return -EBUSY;
3294                 if (copy_from_user(&val, optval, sizeof(val)))
3295                         return -EFAULT;
3296                 po->tp_tx_has_off = !!val;
3297                 return 0;
3298         }
3299         case PACKET_QDISC_BYPASS:
3300         {
3301                 int val;
3302
3303                 if (optlen != sizeof(val))
3304                         return -EINVAL;
3305                 if (copy_from_user(&val, optval, sizeof(val)))
3306                         return -EFAULT;
3307
3308                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3309                 return 0;
3310         }
3311         default:
3312                 return -ENOPROTOOPT;
3313         }
3314 }
3315
3316 static int packet_getsockopt(struct socket *sock, int level, int optname,
3317                              char __user *optval, int __user *optlen)
3318 {
3319         int len;
3320         int val, lv = sizeof(val);
3321         struct sock *sk = sock->sk;
3322         struct packet_sock *po = pkt_sk(sk);
3323         void *data = &val;
3324         union tpacket_stats_u st;
3325
3326         if (level != SOL_PACKET)
3327                 return -ENOPROTOOPT;
3328
3329         if (get_user(len, optlen))
3330                 return -EFAULT;
3331
3332         if (len < 0)
3333                 return -EINVAL;
3334
3335         switch (optname) {
3336         case PACKET_STATISTICS:
3337                 spin_lock_bh(&sk->sk_receive_queue.lock);
3338                 memcpy(&st, &po->stats, sizeof(st));
3339                 memset(&po->stats, 0, sizeof(po->stats));
3340                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3341
3342                 if (po->tp_version == TPACKET_V3) {
3343                         lv = sizeof(struct tpacket_stats_v3);
3344                         st.stats3.tp_packets += st.stats3.tp_drops;
3345                         data = &st.stats3;
3346                 } else {
3347                         lv = sizeof(struct tpacket_stats);
3348                         st.stats1.tp_packets += st.stats1.tp_drops;
3349                         data = &st.stats1;
3350                 }
3351
3352                 break;
3353         case PACKET_AUXDATA:
3354                 val = po->auxdata;
3355                 break;
3356         case PACKET_ORIGDEV:
3357                 val = po->origdev;
3358                 break;
3359         case PACKET_VNET_HDR:
3360                 val = po->has_vnet_hdr;
3361                 break;
3362         case PACKET_VERSION:
3363                 val = po->tp_version;
3364                 break;
3365         case PACKET_HDRLEN:
3366                 if (len > sizeof(int))
3367                         len = sizeof(int);
3368                 if (copy_from_user(&val, optval, len))
3369                         return -EFAULT;
3370                 switch (val) {
3371                 case TPACKET_V1:
3372                         val = sizeof(struct tpacket_hdr);
3373                         break;
3374                 case TPACKET_V2:
3375                         val = sizeof(struct tpacket2_hdr);
3376                         break;
3377                 case TPACKET_V3:
3378                         val = sizeof(struct tpacket3_hdr);
3379                         break;
3380                 default:
3381                         return -EINVAL;
3382                 }
3383                 break;
3384         case PACKET_RESERVE:
3385                 val = po->tp_reserve;
3386                 break;
3387         case PACKET_LOSS:
3388                 val = po->tp_loss;
3389                 break;
3390         case PACKET_TIMESTAMP:
3391                 val = po->tp_tstamp;
3392                 break;
3393         case PACKET_FANOUT:
3394                 val = (po->fanout ?
3395                        ((u32)po->fanout->id |
3396                         ((u32)po->fanout->type << 16) |
3397                         ((u32)po->fanout->flags << 24)) :
3398                        0);
3399                 break;
3400         case PACKET_TX_HAS_OFF:
3401                 val = po->tp_tx_has_off;
3402                 break;
3403         case PACKET_QDISC_BYPASS:
3404                 val = packet_use_direct_xmit(po);
3405                 break;
3406         default:
3407                 return -ENOPROTOOPT;
3408         }
3409
3410         if (len > lv)
3411                 len = lv;
3412         if (put_user(len, optlen))
3413                 return -EFAULT;
3414         if (copy_to_user(optval, data, len))
3415                 return -EFAULT;
3416         return 0;
3417 }
3418
3419
3420 static int packet_notifier(struct notifier_block *this,
3421                            unsigned long msg, void *ptr)
3422 {
3423         struct sock *sk;
3424         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3425         struct net *net = dev_net(dev);
3426
3427         rcu_read_lock();
3428         sk_for_each_rcu(sk, &net->packet.sklist) {
3429                 struct packet_sock *po = pkt_sk(sk);
3430
3431                 switch (msg) {
3432                 case NETDEV_UNREGISTER:
3433                         if (po->mclist)
3434                                 packet_dev_mclist(dev, po->mclist, -1);
3435                         /* fallthrough */
3436
3437                 case NETDEV_DOWN:
3438                         if (dev->ifindex == po->ifindex) {
3439                                 spin_lock(&po->bind_lock);
3440                                 if (po->running) {
3441                                         __unregister_prot_hook(sk, false);
3442                                         sk->sk_err = ENETDOWN;
3443                                         if (!sock_flag(sk, SOCK_DEAD))
3444                                                 sk->sk_error_report(sk);
3445                                 }
3446                                 if (msg == NETDEV_UNREGISTER) {
3447                                         packet_cached_dev_reset(po);
3448                                         po->ifindex = -1;
3449                                         if (po->prot_hook.dev)
3450                                                 dev_put(po->prot_hook.dev);
3451                                         po->prot_hook.dev = NULL;
3452                                 }
3453                                 spin_unlock(&po->bind_lock);
3454                         }
3455                         break;
3456                 case NETDEV_UP:
3457                         if (dev->ifindex == po->ifindex) {
3458                                 spin_lock(&po->bind_lock);
3459                                 if (po->num)
3460                                         register_prot_hook(sk);
3461                                 spin_unlock(&po->bind_lock);
3462                         }
3463                         break;
3464                 }
3465         }
3466         rcu_read_unlock();
3467         return NOTIFY_DONE;
3468 }
3469
3470
3471 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3472                         unsigned long arg)
3473 {
3474         struct sock *sk = sock->sk;
3475
3476         switch (cmd) {
3477         case SIOCOUTQ:
3478         {
3479                 int amount = sk_wmem_alloc_get(sk);
3480
3481                 return put_user(amount, (int __user *)arg);
3482         }
3483         case SIOCINQ:
3484         {
3485                 struct sk_buff *skb;
3486                 int amount = 0;
3487
3488                 spin_lock_bh(&sk->sk_receive_queue.lock);
3489                 skb = skb_peek(&sk->sk_receive_queue);
3490                 if (skb)
3491                         amount = skb->len;
3492                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3493                 return put_user(amount, (int __user *)arg);
3494         }
3495         case SIOCGSTAMP:
3496                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3497         case SIOCGSTAMPNS:
3498                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3499
3500 #ifdef CONFIG_INET
3501         case SIOCADDRT:
3502         case SIOCDELRT:
3503         case SIOCDARP:
3504         case SIOCGARP:
3505         case SIOCSARP:
3506         case SIOCGIFADDR:
3507         case SIOCSIFADDR:
3508         case SIOCGIFBRDADDR:
3509         case SIOCSIFBRDADDR:
3510         case SIOCGIFNETMASK:
3511         case SIOCSIFNETMASK:
3512         case SIOCGIFDSTADDR:
3513         case SIOCSIFDSTADDR:
3514         case SIOCSIFFLAGS:
3515                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3516 #endif
3517
3518         default:
3519                 return -ENOIOCTLCMD;
3520         }
3521         return 0;
3522 }
3523
3524 static unsigned int packet_poll(struct file *file, struct socket *sock,
3525                                 poll_table *wait)
3526 {
3527         struct sock *sk = sock->sk;
3528         struct packet_sock *po = pkt_sk(sk);
3529         unsigned int mask = datagram_poll(file, sock, wait);
3530
3531         spin_lock_bh(&sk->sk_receive_queue.lock);
3532         if (po->rx_ring.pg_vec) {
3533                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3534                         TP_STATUS_KERNEL))
3535                         mask |= POLLIN | POLLRDNORM;
3536         }
3537         spin_unlock_bh(&sk->sk_receive_queue.lock);
3538         spin_lock_bh(&sk->sk_write_queue.lock);
3539         if (po->tx_ring.pg_vec) {
3540                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3541                         mask |= POLLOUT | POLLWRNORM;
3542         }
3543         spin_unlock_bh(&sk->sk_write_queue.lock);
3544         return mask;
3545 }
3546
3547
3548 /* Dirty? Well, I still did not learn better way to account
3549  * for user mmaps.
3550  */
3551
3552 static void packet_mm_open(struct vm_area_struct *vma)
3553 {
3554         struct file *file = vma->vm_file;
3555         struct socket *sock = file->private_data;
3556         struct sock *sk = sock->sk;
3557
3558         if (sk)
3559                 atomic_inc(&pkt_sk(sk)->mapped);
3560 }
3561
3562 static void packet_mm_close(struct vm_area_struct *vma)
3563 {
3564         struct file *file = vma->vm_file;
3565         struct socket *sock = file->private_data;
3566         struct sock *sk = sock->sk;
3567
3568         if (sk)
3569                 atomic_dec(&pkt_sk(sk)->mapped);
3570 }
3571
3572 static const struct vm_operations_struct packet_mmap_ops = {
3573         .open   =       packet_mm_open,
3574         .close  =       packet_mm_close,
3575 };
3576
3577 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3578                         unsigned int len)
3579 {
3580         int i;
3581
3582         for (i = 0; i < len; i++) {
3583                 if (likely(pg_vec[i].buffer)) {
3584                         if (is_vmalloc_addr(pg_vec[i].buffer))
3585                                 vfree(pg_vec[i].buffer);
3586                         else
3587                                 free_pages((unsigned long)pg_vec[i].buffer,
3588                                            order);
3589                         pg_vec[i].buffer = NULL;
3590                 }
3591         }
3592         kfree(pg_vec);
3593 }
3594
3595 static char *alloc_one_pg_vec_page(unsigned long order)
3596 {
3597         char *buffer = NULL;
3598         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3599                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3600
3601         buffer = (char *) __get_free_pages(gfp_flags, order);
3602
3603         if (buffer)
3604                 return buffer;
3605
3606         /*
3607          * __get_free_pages failed, fall back to vmalloc
3608          */
3609         buffer = vzalloc((1 << order) * PAGE_SIZE);
3610
3611         if (buffer)
3612                 return buffer;
3613
3614         /*
3615          * vmalloc failed, lets dig into swap here
3616          */
3617         gfp_flags &= ~__GFP_NORETRY;
3618         buffer = (char *)__get_free_pages(gfp_flags, order);
3619         if (buffer)
3620                 return buffer;
3621
3622         /*
3623          * complete and utter failure
3624          */
3625         return NULL;
3626 }
3627
3628 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3629 {
3630         unsigned int block_nr = req->tp_block_nr;
3631         struct pgv *pg_vec;
3632         int i;
3633
3634         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3635         if (unlikely(!pg_vec))
3636                 goto out;
3637
3638         for (i = 0; i < block_nr; i++) {
3639                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3640                 if (unlikely(!pg_vec[i].buffer))
3641                         goto out_free_pgvec;
3642         }
3643
3644 out:
3645         return pg_vec;
3646
3647 out_free_pgvec:
3648         free_pg_vec(pg_vec, order, block_nr);
3649         pg_vec = NULL;
3650         goto out;
3651 }
3652
3653 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3654                 int closing, int tx_ring)
3655 {
3656         struct pgv *pg_vec = NULL;
3657         struct packet_sock *po = pkt_sk(sk);
3658         int was_running, order = 0;
3659         struct packet_ring_buffer *rb;
3660         struct sk_buff_head *rb_queue;
3661         __be16 num;
3662         int err = -EINVAL;
3663         /* Added to avoid minimal code churn */
3664         struct tpacket_req *req = &req_u->req;
3665
3666         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3667         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3668                 WARN(1, "Tx-ring is not supported.\n");
3669                 goto out;
3670         }
3671
3672         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3673         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3674
3675         err = -EBUSY;
3676         if (!closing) {
3677                 if (atomic_read(&po->mapped))
3678                         goto out;
3679                 if (atomic_read(&rb->pending))
3680                         goto out;
3681         }
3682
3683         if (req->tp_block_nr) {
3684                 /* Sanity tests and some calculations */
3685                 err = -EBUSY;
3686                 if (unlikely(rb->pg_vec))
3687                         goto out;
3688
3689                 switch (po->tp_version) {
3690                 case TPACKET_V1:
3691                         po->tp_hdrlen = TPACKET_HDRLEN;
3692                         break;
3693                 case TPACKET_V2:
3694                         po->tp_hdrlen = TPACKET2_HDRLEN;
3695                         break;
3696                 case TPACKET_V3:
3697                         po->tp_hdrlen = TPACKET3_HDRLEN;
3698                         break;
3699                 }
3700
3701                 err = -EINVAL;
3702                 if (unlikely((int)req->tp_block_size <= 0))
3703                         goto out;
3704                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3705                         goto out;
3706                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3707                                         po->tp_reserve))
3708                         goto out;
3709                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3710                         goto out;
3711
3712                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3713                 if (unlikely(rb->frames_per_block <= 0))
3714                         goto out;
3715                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3716                                         req->tp_frame_nr))
3717                         goto out;
3718
3719                 err = -ENOMEM;
3720                 order = get_order(req->tp_block_size);
3721                 pg_vec = alloc_pg_vec(req, order);
3722                 if (unlikely(!pg_vec))
3723                         goto out;
3724                 switch (po->tp_version) {
3725                 case TPACKET_V3:
3726                 /* Transmit path is not supported. We checked
3727                  * it above but just being paranoid
3728                  */
3729                         if (!tx_ring)
3730                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3731                                 break;
3732                 default:
3733                         break;
3734                 }
3735         }
3736         /* Done */
3737         else {
3738                 err = -EINVAL;
3739                 if (unlikely(req->tp_frame_nr))
3740                         goto out;
3741         }
3742
3743         lock_sock(sk);
3744
3745         /* Detach socket from network */
3746         spin_lock(&po->bind_lock);
3747         was_running = po->running;
3748         num = po->num;
3749         if (was_running) {
3750                 po->num = 0;
3751                 __unregister_prot_hook(sk, false);
3752         }
3753         spin_unlock(&po->bind_lock);
3754
3755         synchronize_net();
3756
3757         err = -EBUSY;
3758         mutex_lock(&po->pg_vec_lock);
3759         if (closing || atomic_read(&po->mapped) == 0) {
3760                 err = 0;
3761                 spin_lock_bh(&rb_queue->lock);
3762                 swap(rb->pg_vec, pg_vec);
3763                 rb->frame_max = (req->tp_frame_nr - 1);
3764                 rb->head = 0;
3765                 rb->frame_size = req->tp_frame_size;
3766                 spin_unlock_bh(&rb_queue->lock);
3767
3768                 swap(rb->pg_vec_order, order);
3769                 swap(rb->pg_vec_len, req->tp_block_nr);
3770
3771                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3772                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3773                                                 tpacket_rcv : packet_rcv;
3774                 skb_queue_purge(rb_queue);
3775                 if (atomic_read(&po->mapped))
3776                         pr_err("packet_mmap: vma is busy: %d\n",
3777                                atomic_read(&po->mapped));
3778         }
3779         mutex_unlock(&po->pg_vec_lock);
3780
3781         spin_lock(&po->bind_lock);
3782         if (was_running) {
3783                 po->num = num;
3784                 register_prot_hook(sk);
3785         }
3786         spin_unlock(&po->bind_lock);
3787         if (closing && (po->tp_version > TPACKET_V2)) {
3788                 /* Because we don't support block-based V3 on tx-ring */
3789                 if (!tx_ring)
3790                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3791         }
3792         release_sock(sk);
3793
3794         if (pg_vec)
3795                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3796 out:
3797         return err;
3798 }
3799
3800 static int packet_mmap(struct file *file, struct socket *sock,
3801                 struct vm_area_struct *vma)
3802 {
3803         struct sock *sk = sock->sk;
3804         struct packet_sock *po = pkt_sk(sk);
3805         unsigned long size, expected_size;
3806         struct packet_ring_buffer *rb;
3807         unsigned long start;
3808         int err = -EINVAL;
3809         int i;
3810
3811         if (vma->vm_pgoff)
3812                 return -EINVAL;
3813
3814         mutex_lock(&po->pg_vec_lock);
3815
3816         expected_size = 0;
3817         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3818                 if (rb->pg_vec) {
3819                         expected_size += rb->pg_vec_len
3820                                                 * rb->pg_vec_pages
3821                                                 * PAGE_SIZE;
3822                 }
3823         }
3824
3825         if (expected_size == 0)
3826                 goto out;
3827
3828         size = vma->vm_end - vma->vm_start;
3829         if (size != expected_size)
3830                 goto out;
3831
3832         start = vma->vm_start;
3833         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3834                 if (rb->pg_vec == NULL)
3835                         continue;
3836
3837                 for (i = 0; i < rb->pg_vec_len; i++) {
3838                         struct page *page;
3839                         void *kaddr = rb->pg_vec[i].buffer;
3840                         int pg_num;
3841
3842                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3843                                 page = pgv_to_page(kaddr);
3844                                 err = vm_insert_page(vma, start, page);
3845                                 if (unlikely(err))
3846                                         goto out;
3847                                 start += PAGE_SIZE;
3848                                 kaddr += PAGE_SIZE;
3849                         }
3850                 }
3851         }
3852
3853         atomic_inc(&po->mapped);
3854         vma->vm_ops = &packet_mmap_ops;
3855         err = 0;
3856
3857 out:
3858         mutex_unlock(&po->pg_vec_lock);
3859         return err;
3860 }
3861
3862 static const struct proto_ops packet_ops_spkt = {
3863         .family =       PF_PACKET,
3864         .owner =        THIS_MODULE,
3865         .release =      packet_release,
3866         .bind =         packet_bind_spkt,
3867         .connect =      sock_no_connect,
3868         .socketpair =   sock_no_socketpair,
3869         .accept =       sock_no_accept,
3870         .getname =      packet_getname_spkt,
3871         .poll =         datagram_poll,
3872         .ioctl =        packet_ioctl,
3873         .listen =       sock_no_listen,
3874         .shutdown =     sock_no_shutdown,
3875         .setsockopt =   sock_no_setsockopt,
3876         .getsockopt =   sock_no_getsockopt,
3877         .sendmsg =      packet_sendmsg_spkt,
3878         .recvmsg =      packet_recvmsg,
3879         .mmap =         sock_no_mmap,
3880         .sendpage =     sock_no_sendpage,
3881 };
3882
3883 static const struct proto_ops packet_ops = {
3884         .family =       PF_PACKET,
3885         .owner =        THIS_MODULE,
3886         .release =      packet_release,
3887         .bind =         packet_bind,
3888         .connect =      sock_no_connect,
3889         .socketpair =   sock_no_socketpair,
3890         .accept =       sock_no_accept,
3891         .getname =      packet_getname,
3892         .poll =         packet_poll,
3893         .ioctl =        packet_ioctl,
3894         .listen =       sock_no_listen,
3895         .shutdown =     sock_no_shutdown,
3896         .setsockopt =   packet_setsockopt,
3897         .getsockopt =   packet_getsockopt,
3898         .sendmsg =      packet_sendmsg,
3899         .recvmsg =      packet_recvmsg,
3900         .mmap =         packet_mmap,
3901         .sendpage =     sock_no_sendpage,
3902 };
3903
3904 static const struct net_proto_family packet_family_ops = {
3905         .family =       PF_PACKET,
3906         .create =       packet_create,
3907         .owner  =       THIS_MODULE,
3908 };
3909
3910 static struct notifier_block packet_netdev_notifier = {
3911         .notifier_call =        packet_notifier,
3912 };
3913
3914 #ifdef CONFIG_PROC_FS
3915
3916 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3917         __acquires(RCU)
3918 {
3919         struct net *net = seq_file_net(seq);
3920
3921         rcu_read_lock();
3922         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3923 }
3924
3925 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3926 {
3927         struct net *net = seq_file_net(seq);
3928         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3929 }
3930
3931 static void packet_seq_stop(struct seq_file *seq, void *v)
3932         __releases(RCU)
3933 {
3934         rcu_read_unlock();
3935 }
3936
3937 static int packet_seq_show(struct seq_file *seq, void *v)
3938 {
3939         if (v == SEQ_START_TOKEN)
3940                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3941         else {
3942                 struct sock *s = sk_entry(v);
3943                 const struct packet_sock *po = pkt_sk(s);
3944
3945                 seq_printf(seq,
3946                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3947                            s,
3948                            atomic_read(&s->sk_refcnt),
3949                            s->sk_type,
3950                            ntohs(po->num),
3951                            po->ifindex,
3952                            po->running,
3953                            atomic_read(&s->sk_rmem_alloc),
3954                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
3955                            sock_i_ino(s));
3956         }
3957
3958         return 0;
3959 }
3960
3961 static const struct seq_operations packet_seq_ops = {
3962         .start  = packet_seq_start,
3963         .next   = packet_seq_next,
3964         .stop   = packet_seq_stop,
3965         .show   = packet_seq_show,
3966 };
3967
3968 static int packet_seq_open(struct inode *inode, struct file *file)
3969 {
3970         return seq_open_net(inode, file, &packet_seq_ops,
3971                             sizeof(struct seq_net_private));
3972 }
3973
3974 static const struct file_operations packet_seq_fops = {
3975         .owner          = THIS_MODULE,
3976         .open           = packet_seq_open,
3977         .read           = seq_read,
3978         .llseek         = seq_lseek,
3979         .release        = seq_release_net,
3980 };
3981
3982 #endif
3983
3984 static int __net_init packet_net_init(struct net *net)
3985 {
3986         mutex_init(&net->packet.sklist_lock);
3987         INIT_HLIST_HEAD(&net->packet.sklist);
3988
3989         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
3990                 return -ENOMEM;
3991
3992         return 0;
3993 }
3994
3995 static void __net_exit packet_net_exit(struct net *net)
3996 {
3997         remove_proc_entry("packet", net->proc_net);
3998 }
3999
4000 static struct pernet_operations packet_net_ops = {
4001         .init = packet_net_init,
4002         .exit = packet_net_exit,
4003 };
4004
4005
4006 static void __exit packet_exit(void)
4007 {
4008         unregister_netdevice_notifier(&packet_netdev_notifier);
4009         unregister_pernet_subsys(&packet_net_ops);
4010         sock_unregister(PF_PACKET);
4011         proto_unregister(&packet_proto);
4012 }
4013
4014 static int __init packet_init(void)
4015 {
4016         int rc = proto_register(&packet_proto, 0);
4017
4018         if (rc != 0)
4019                 goto out;
4020
4021         sock_register(&packet_family_ops);
4022         register_pernet_subsys(&packet_net_ops);
4023         register_netdevice_notifier(&packet_netdev_notifier);
4024 out:
4025         return rc;
4026 }
4027
4028 module_init(packet_init);
4029 module_exit(packet_exit);
4030 MODULE_LICENSE("GPL");
4031 MODULE_ALIAS_NETPROTO(PF_PACKET);