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/percpu.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         netdev_features_t features;
 244         struct netdev_queue *txq;
 245         int ret = NETDEV_TX_BUSY;
 246
 247         if (unlikely(!netif_running(dev) ||
 248                      !netif_carrier_ok(dev)))
 249                 goto drop;
 250
 251         features = netif_skb_features(skb);
 252         if (skb_needs_linearize(skb, features) &&
 253             __skb_linearize(skb))
 254                 goto drop;
 255
 256         txq = skb_get_tx_queue(dev, skb);
 257
 258         local_bh_disable();
 259
 260         HARD_TX_LOCK(dev, txq, smp_processor_id());
 261         if (!netif_xmit_frozen_or_drv_stopped(txq))
 262                 ret = netdev_start_xmit(skb, dev, txq, false);
 263         HARD_TX_UNLOCK(dev, txq);
 264
 265         local_bh_enable();
 266
 267         if (!dev_xmit_complete(ret))
 268                 kfree_skb(skb);
 269
 270         return ret;
 271 drop:
 272         atomic_long_inc(&dev->tx_dropped);
 273         kfree_skb(skb);
 274         return NET_XMIT_DROP;
 275 }
 276
 277 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
 278 {
 279         struct net_device *dev;
 280
 281         rcu_read_lock();
 282         dev = rcu_dereference(po->cached_dev);
 283         if (likely(dev))
 284                 dev_hold(dev);
 285         rcu_read_unlock();
 286
 287         return dev;
 288 }
 289
 290 static void packet_cached_dev_assign(struct packet_sock *po,
 291                                      struct net_device *dev)
 292 {
 293         rcu_assign_pointer(po->cached_dev, dev);
 294 }
 295
 296 static void packet_cached_dev_reset(struct packet_sock *po)
 297 {
 298         RCU_INIT_POINTER(po->cached_dev, NULL);
 299 }
 300
 301 static bool packet_use_direct_xmit(const struct packet_sock *po)
 302 {
 303         return po->xmit == packet_direct_xmit;
 304 }
 305
 306 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 307 {
 308         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
 309 }
 310
 311 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
 312 {
 313         const struct net_device_ops *ops = dev->netdev_ops;
 314         u16 queue_index;
 315
 316         if (ops->ndo_select_queue) {
 317                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
 318                                                     __packet_pick_tx_queue);
 319                 queue_index = netdev_cap_txqueue(dev, queue_index);
 320         } else {
 321                 queue_index = __packet_pick_tx_queue(dev, skb);
 322         }
 323
 324         skb_set_queue_mapping(skb, queue_index);
 325 }
 326
 327 /* register_prot_hook must be invoked with the po->bind_lock held,
 328  * or from a context in which asynchronous accesses to the packet
 329  * socket is not possible (packet_create()).
 330  */
 331 static void register_prot_hook(struct sock *sk)
 332 {
 333         struct packet_sock *po = pkt_sk(sk);
 334
 335         if (!po->running) {
 336                 if (po->fanout)
 337                         __fanout_link(sk, po);
 338                 else
 339                         dev_add_pack(&po->prot_hook);
 340
 341                 sock_hold(sk);
 342                 po->running = 1;
 343         }
 344 }
 345
 346 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
 347  * held.   If the sync parameter is true, we will temporarily drop
 348  * the po->bind_lock and do a synchronize_net to make sure no
 349  * asynchronous packet processing paths still refer to the elements
 350  * of po->prot_hook.  If the sync parameter is false, it is the
 351  * callers responsibility to take care of this.
 352  */
 353 static void __unregister_prot_hook(struct sock *sk, bool sync)
 354 {
 355         struct packet_sock *po = pkt_sk(sk);
 356
 357         po->running = 0;
 358
 359         if (po->fanout)
 360                 __fanout_unlink(sk, po);
 361         else
 362                 __dev_remove_pack(&po->prot_hook);
 363
 364         __sock_put(sk);
 365
 366         if (sync) {
 367                 spin_unlock(&po->bind_lock);
 368                 synchronize_net();
 369                 spin_lock(&po->bind_lock);
 370         }
 371 }
 372
 373 static void unregister_prot_hook(struct sock *sk, bool sync)
 374 {
 375         struct packet_sock *po = pkt_sk(sk);
 376
 377         if (po->running)
 378                 __unregister_prot_hook(sk, sync);
 379 }
 380
 381 static inline struct page * __pure pgv_to_page(void *addr)
 382 {
 383         if (is_vmalloc_addr(addr))
 384                 return vmalloc_to_page(addr);
 385         return virt_to_page(addr);
 386 }
 387
 388 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
 389 {
 390         union tpacket_uhdr h;
 391
 392         h.raw = frame;
 393         switch (po->tp_version) {
 394         case TPACKET_V1:
 395                 h.h1->tp_status = status;
 396                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 397                 break;
 398         case TPACKET_V2:
 399                 h.h2->tp_status = status;
 400                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 401                 break;
 402         case TPACKET_V3:
 403         default:
 404                 WARN(1, "TPACKET version not supported.\n");
 405                 BUG();
 406         }
 407
 408         smp_wmb();
 409 }
 410
 411 static int __packet_get_status(struct packet_sock *po, void *frame)
 412 {
 413         union tpacket_uhdr h;
 414
 415         smp_rmb();
 416
 417         h.raw = frame;
 418         switch (po->tp_version) {
 419         case TPACKET_V1:
 420                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
 421                 return h.h1->tp_status;
 422         case TPACKET_V2:
 423                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
 424                 return h.h2->tp_status;
 425         case TPACKET_V3:
 426         default:
 427                 WARN(1, "TPACKET version not supported.\n");
 428                 BUG();
 429                 return 0;
 430         }
 431 }
 432
 433 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
 434                                    unsigned int flags)
 435 {
 436         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
 437
 438         if (shhwtstamps &&
 439             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
 440             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
 441                 return TP_STATUS_TS_RAW_HARDWARE;
 442
 443         if (ktime_to_timespec_cond(skb->tstamp, ts))
 444                 return TP_STATUS_TS_SOFTWARE;
 445
 446         return 0;
 447 }
 448
 449 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
 450                                     struct sk_buff *skb)
 451 {
 452         union tpacket_uhdr h;
 453         struct timespec ts;
 454         __u32 ts_status;
 455
 456         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
 457                 return 0;
 458
 459         h.raw = frame;
 460         switch (po->tp_version) {
 461         case TPACKET_V1:
 462                 h.h1->tp_sec = ts.tv_sec;
 463                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
 464                 break;
 465         case TPACKET_V2:
 466                 h.h2->tp_sec = ts.tv_sec;
 467                 h.h2->tp_nsec = ts.tv_nsec;
 468                 break;
 469         case TPACKET_V3:
 470         default:
 471                 WARN(1, "TPACKET version not supported.\n");
 472                 BUG();
 473         }
 474
 475         /* one flush is safe, as both fields always lie on the same cacheline */
 476         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
 477         smp_wmb();
 478
 479         return ts_status;
 480 }
 481
 482 static void *packet_lookup_frame(struct packet_sock *po,
 483                 struct packet_ring_buffer *rb,
 484                 unsigned int position,
 485                 int status)
 486 {
 487         unsigned int pg_vec_pos, frame_offset;
 488         union tpacket_uhdr h;
 489
 490         pg_vec_pos = position / rb->frames_per_block;
 491         frame_offset = position % rb->frames_per_block;
 492
 493         h.raw = rb->pg_vec[pg_vec_pos].buffer +
 494                 (frame_offset * rb->frame_size);
 495
 496         if (status != __packet_get_status(po, h.raw))
 497                 return NULL;
 498
 499         return h.raw;
 500 }
 501
 502 static void *packet_current_frame(struct packet_sock *po,
 503                 struct packet_ring_buffer *rb,
 504                 int status)
 505 {
 506         return packet_lookup_frame(po, rb, rb->head, status);
 507 }
 508
 509 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 510 {
 511         del_timer_sync(&pkc->retire_blk_timer);
 512 }
 513
 514 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
 515                 int tx_ring,
 516                 struct sk_buff_head *rb_queue)
 517 {
 518         struct tpacket_kbdq_core *pkc;
 519
 520         pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
 521                         GET_PBDQC_FROM_RB(&po->rx_ring);
 522
 523         spin_lock_bh(&rb_queue->lock);
 524         pkc->delete_blk_timer = 1;
 525         spin_unlock_bh(&rb_queue->lock);
 526
 527         prb_del_retire_blk_timer(pkc);
 528 }
 529
 530 static void prb_init_blk_timer(struct packet_sock *po,
 531                 struct tpacket_kbdq_core *pkc,
 532                 void (*func) (unsigned long))
 533 {
 534         init_timer(&pkc->retire_blk_timer);
 535         pkc->retire_blk_timer.data = (long)po;
 536         pkc->retire_blk_timer.function = func;
 537         pkc->retire_blk_timer.expires = jiffies;
 538 }
 539
 540 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
 541 {
 542         struct tpacket_kbdq_core *pkc;
 543
 544         if (tx_ring)
 545                 BUG();
 546
 547         pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
 548                         GET_PBDQC_FROM_RB(&po->rx_ring);
 549         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
 550 }
 551
 552 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
 553                                 int blk_size_in_bytes)
 554 {
 555         struct net_device *dev;
 556         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
 557         struct ethtool_cmd ecmd;
 558         int err;
 559         u32 speed;
 560
 561         rtnl_lock();
 562         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
 563         if (unlikely(!dev)) {
 564                 rtnl_unlock();
 565                 return DEFAULT_PRB_RETIRE_TOV;
 566         }
 567         err = __ethtool_get_settings(dev, &ecmd);
 568         speed = ethtool_cmd_speed(&ecmd);
 569         rtnl_unlock();
 570         if (!err) {
 571                 /*
 572                  * If the link speed is so slow you don't really
 573                  * need to worry about perf anyways
 574                  */
 575                 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
 576                         return DEFAULT_PRB_RETIRE_TOV;
 577                 } else {
 578                         msec = 1;
 579                         div = speed / 1000;
 580                 }
 581         }
 582
 583         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
 584
 585         if (div)
 586                 mbits /= div;
 587
 588         tmo = mbits * msec;
 589
 590         if (div)
 591                 return tmo+1;
 592         return tmo;
 593 }
 594
 595 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
 596                         union tpacket_req_u *req_u)
 597 {
 598         p1->feature_req_word = req_u->req3.tp_feature_req_word;
 599 }
 600
 601 static void init_prb_bdqc(struct packet_sock *po,
 602                         struct packet_ring_buffer *rb,
 603                         struct pgv *pg_vec,
 604                         union tpacket_req_u *req_u, int tx_ring)
 605 {
 606         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
 607         struct tpacket_block_desc *pbd;
 608
 609         memset(p1, 0x0, sizeof(*p1));
 610
 611         p1->knxt_seq_num = 1;
 612         p1->pkbdq = pg_vec;
 613         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
 614         p1->pkblk_start = pg_vec[0].buffer;
 615         p1->kblk_size = req_u->req3.tp_block_size;
 616         p1->knum_blocks = req_u->req3.tp_block_nr;
 617         p1->hdrlen = po->tp_hdrlen;
 618         p1->version = po->tp_version;
 619         p1->last_kactive_blk_num = 0;
 620         po->stats.stats3.tp_freeze_q_cnt = 0;
 621         if (req_u->req3.tp_retire_blk_tov)
 622                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
 623         else
 624                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
 625                                                 req_u->req3.tp_block_size);
 626         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
 627         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 628
 629         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
 630         prb_init_ft_ops(p1, req_u);
 631         prb_setup_retire_blk_timer(po, tx_ring);
 632         prb_open_block(p1, pbd);
 633 }
 634
 635 /*  Do NOT update the last_blk_num first.
 636  *  Assumes sk_buff_head lock is held.
 637  */
 638 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 639 {
 640         mod_timer(&pkc->retire_blk_timer,
 641                         jiffies + pkc->tov_in_jiffies);
 642         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 643 }
 644
 645 /*
 646  * Timer logic:
 647  * 1) We refresh the timer only when we open a block.
 648  *    By doing this we don't waste cycles refreshing the timer
 649  *        on packet-by-packet basis.
 650  *
 651  * With a 1MB block-size, on a 1Gbps line, it will take
 652  * i) ~8 ms to fill a block + ii) memcpy etc.
 653  * In this cut we are not accounting for the memcpy time.
 654  *
 655  * So, if the user sets the 'tmo' to 10ms then the timer
 656  * will never fire while the block is still getting filled
 657  * (which is what we want). However, the user could choose
 658  * to close a block early and that's fine.
 659  *
 660  * But when the timer does fire, we check whether or not to refresh it.
 661  * Since the tmo granularity is in msecs, it is not too expensive
 662  * to refresh the timer, lets say every '8' msecs.
 663  * Either the user can set the 'tmo' or we can derive it based on
 664  * a) line-speed and b) block-size.
 665  * prb_calc_retire_blk_tmo() calculates the tmo.
 666  *
 667  */
 668 static void prb_retire_rx_blk_timer_expired(unsigned long data)
 669 {
 670         struct packet_sock *po = (struct packet_sock *)data;
 671         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
 672         unsigned int frozen;
 673         struct tpacket_block_desc *pbd;
 674
 675         spin_lock(&po->sk.sk_receive_queue.lock);
 676
 677         frozen = prb_queue_frozen(pkc);
 678         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 679
 680         if (unlikely(pkc->delete_blk_timer))
 681                 goto out;
 682
 683         /* We only need to plug the race when the block is partially filled.
 684          * tpacket_rcv:
 685          *              lock(); increment BLOCK_NUM_PKTS; unlock()
 686          *              copy_bits() is in progress ...
 687          *              timer fires on other cpu:
 688          *              we can't retire the current block because copy_bits
 689          *              is in progress.
 690          *
 691          */
 692         if (BLOCK_NUM_PKTS(pbd)) {
 693                 while (atomic_read(&pkc->blk_fill_in_prog)) {
 694                         /* Waiting for skb_copy_bits to finish... */
 695                         cpu_relax();
 696                 }
 697         }
 698
 699         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
 700                 if (!frozen) {
 701                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
 702                         if (!prb_dispatch_next_block(pkc, po))
 703                                 goto refresh_timer;
 704                         else
 705                                 goto out;
 706                 } else {
 707                         /* Case 1. Queue was frozen because user-space was
 708                          *         lagging behind.
 709                          */
 710                         if (prb_curr_blk_in_use(pkc, pbd)) {
 711                                 /*
 712                                  * Ok, user-space is still behind.
 713                                  * So just refresh the timer.
 714                                  */
 715                                 goto refresh_timer;
 716                         } else {
 717                                /* Case 2. queue was frozen,user-space caught up,
 718                                 * now the link went idle && the timer fired.
 719                                 * We don't have a block to close.So we open this
 720                                 * block and restart the timer.
 721                                 * opening a block thaws the queue,restarts timer
 722                                 * Thawing/timer-refresh is a side effect.
 723                                 */
 724                                 prb_open_block(pkc, pbd);
 725                                 goto out;
 726                         }
 727                 }
 728         }
 729
 730 refresh_timer:
 731         _prb_refresh_rx_retire_blk_timer(pkc);
 732
 733 out:
 734         spin_unlock(&po->sk.sk_receive_queue.lock);
 735 }
 736
 737 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
 738                 struct tpacket_block_desc *pbd1, __u32 status)
 739 {
 740         /* Flush everything minus the block header */
 741
 742 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 743         u8 *start, *end;
 744
 745         start = (u8 *)pbd1;
 746
 747         /* Skip the block header(we know header WILL fit in 4K) */
 748         start += PAGE_SIZE;
 749
 750         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
 751         for (; start < end; start += PAGE_SIZE)
 752                 flush_dcache_page(pgv_to_page(start));
 753
 754         smp_wmb();
 755 #endif
 756
 757         /* Now update the block status. */
 758
 759         BLOCK_STATUS(pbd1) = status;
 760
 761         /* Flush the block header */
 762
 763 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
 764         start = (u8 *)pbd1;
 765         flush_dcache_page(pgv_to_page(start));
 766
 767         smp_wmb();
 768 #endif
 769 }
 770
 771 /*
 772  * Side effect:
 773  *
 774  * 1) flush the block
 775  * 2) Increment active_blk_num
 776  *
 777  * Note:We DONT refresh the timer on purpose.
 778  *      Because almost always the next block will be opened.
 779  */
 780 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 781                 struct tpacket_block_desc *pbd1,
 782                 struct packet_sock *po, unsigned int stat)
 783 {
 784         __u32 status = TP_STATUS_USER | stat;
 785
 786         struct tpacket3_hdr *last_pkt;
 787         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 788
 789         if (po->stats.stats3.tp_drops)
 790                 status |= TP_STATUS_LOSING;
 791
 792         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
 793         last_pkt->tp_next_offset = 0;
 794
 795         /* Get the ts of the last pkt */
 796         if (BLOCK_NUM_PKTS(pbd1)) {
 797                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
 798                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
 799         } else {
 800                 /* Ok, we tmo'd - so get the current time */
 801                 struct timespec ts;
 802                 getnstimeofday(&ts);
 803                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
 804                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
 805         }
 806
 807         smp_wmb();
 808
 809         /* Flush the block */
 810         prb_flush_block(pkc1, pbd1, status);
 811
 812         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
 813 }
 814
 815 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
 816 {
 817         pkc->reset_pending_on_curr_blk = 0;
 818 }
 819
 820 /*
 821  * Side effect of opening a block:
 822  *
 823  * 1) prb_queue is thawed.
 824  * 2) retire_blk_timer is refreshed.
 825  *
 826  */
 827 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 828         struct tpacket_block_desc *pbd1)
 829 {
 830         struct timespec ts;
 831         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 832
 833         smp_rmb();
 834
 835         /* We could have just memset this but we will lose the
 836          * flexibility of making the priv area sticky
 837          */
 838
 839         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
 840         BLOCK_NUM_PKTS(pbd1) = 0;
 841         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 842
 843         getnstimeofday(&ts);
 844
 845         h1->ts_first_pkt.ts_sec = ts.tv_sec;
 846         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
 847
 848         pkc1->pkblk_start = (char *)pbd1;
 849         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 850
 851         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
 852         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
 853
 854         pbd1->version = pkc1->version;
 855         pkc1->prev = pkc1->nxt_offset;
 856         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 857
 858         prb_thaw_queue(pkc1);
 859         _prb_refresh_rx_retire_blk_timer(pkc1);
 860
 861         smp_wmb();
 862 }
 863
 864 /*
 865  * Queue freeze logic:
 866  * 1) Assume tp_block_nr = 8 blocks.
 867  * 2) At time 't0', user opens Rx ring.
 868  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
 869  * 4) user-space is either sleeping or processing block '0'.
 870  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
 871  *    it will close block-7,loop around and try to fill block '0'.
 872  *    call-flow:
 873  *    __packet_lookup_frame_in_block
 874  *      prb_retire_current_block()
 875  *      prb_dispatch_next_block()
 876  *        |->(BLOCK_STATUS == USER) evaluates to true
 877  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
 878  * 6) Now there are two cases:
 879  *    6.1) Link goes idle right after the queue is frozen.
 880  *         But remember, the last open_block() refreshed the timer.
 881  *         When this timer expires,it will refresh itself so that we can
 882  *         re-open block-0 in near future.
 883  *    6.2) Link is busy and keeps on receiving packets. This is a simple
 884  *         case and __packet_lookup_frame_in_block will check if block-0
 885  *         is free and can now be re-used.
 886  */
 887 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
 888                                   struct packet_sock *po)
 889 {
 890         pkc->reset_pending_on_curr_blk = 1;
 891         po->stats.stats3.tp_freeze_q_cnt++;
 892 }
 893
 894 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
 895
 896 /*
 897  * If the next block is free then we will dispatch it
 898  * and return a good offset.
 899  * Else, we will freeze the queue.
 900  * So, caller must check the return value.
 901  */
 902 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
 903                 struct packet_sock *po)
 904 {
 905         struct tpacket_block_desc *pbd;
 906
 907         smp_rmb();
 908
 909         /* 1. Get current block num */
 910         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 911
 912         /* 2. If this block is currently in_use then freeze the queue */
 913         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
 914                 prb_freeze_queue(pkc, po);
 915                 return NULL;
 916         }
 917
 918         /*
 919          * 3.
 920          * open this block and return the offset where the first packet
 921          * needs to get stored.
 922          */
 923         prb_open_block(pkc, pbd);
 924         return (void *)pkc->nxt_offset;
 925 }
 926
 927 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
 928                 struct packet_sock *po, unsigned int status)
 929 {
 930         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 931
 932         /* retire/close the current block */
 933         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
 934                 /*
 935                  * Plug the case where copy_bits() is in progress on
 936                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
 937                  * have space to copy the pkt in the current block and
 938                  * called prb_retire_current_block()
 939                  *
 940                  * We don't need to worry about the TMO case because
 941                  * the timer-handler already handled this case.
 942                  */
 943                 if (!(status & TP_STATUS_BLK_TMO)) {
 944                         while (atomic_read(&pkc->blk_fill_in_prog)) {
 945                                 /* Waiting for skb_copy_bits to finish... */
 946                                 cpu_relax();
 947                         }
 948                 }
 949                 prb_close_block(pkc, pbd, po, status);
 950                 return;
 951         }
 952 }
 953
 954 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
 955                                       struct tpacket_block_desc *pbd)
 956 {
 957         return TP_STATUS_USER & BLOCK_STATUS(pbd);
 958 }
 959
 960 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
 961 {
 962         return pkc->reset_pending_on_curr_blk;
 963 }
 964
 965 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
 966 {
 967         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
 968         atomic_dec(&pkc->blk_fill_in_prog);
 969 }
 970
 971 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
 972                         struct tpacket3_hdr *ppd)
 973 {
 974         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
 975 }
 976
 977 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
 978                         struct tpacket3_hdr *ppd)
 979 {
 980         ppd->hv1.tp_rxhash = 0;
 981 }
 982
 983 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
 984                         struct tpacket3_hdr *ppd)
 985 {
 986         if (vlan_tx_tag_present(pkc->skb)) {
 987                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
 988                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
 989                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
 990         } else {
 991                 ppd->hv1.tp_vlan_tci = 0;
 992                 ppd->hv1.tp_vlan_tpid = 0;
 993                 ppd->tp_status = TP_STATUS_AVAILABLE;
 994         }
 995 }
 996
 997 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
 998                         struct tpacket3_hdr *ppd)
 999 {
1000         ppd->hv1.tp_padding = 0;
1001         prb_fill_vlan_info(pkc, ppd);
1002
1003         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1004                 prb_fill_rxhash(pkc, ppd);
1005         else
1006                 prb_clear_rxhash(pkc, ppd);
1007 }
1008
1009 static void prb_fill_curr_block(char *curr,
1010                                 struct tpacket_kbdq_core *pkc,
1011                                 struct tpacket_block_desc *pbd,
1012                                 unsigned int len)
1013 {
1014         struct tpacket3_hdr *ppd;
1015
1016         ppd  = (struct tpacket3_hdr *)curr;
1017         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1018         pkc->prev = curr;
1019         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1020         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1021         BLOCK_NUM_PKTS(pbd) += 1;
1022         atomic_inc(&pkc->blk_fill_in_prog);
1023         prb_run_all_ft_ops(pkc, ppd);
1024 }
1025
1026 /* Assumes caller has the sk->rx_queue.lock */
1027 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1028                                             struct sk_buff *skb,
1029                                                 int status,
1030                                             unsigned int len
1031                                             )
1032 {
1033         struct tpacket_kbdq_core *pkc;
1034         struct tpacket_block_desc *pbd;
1035         char *curr, *end;
1036
1037         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1038         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1039
1040         /* Queue is frozen when user space is lagging behind */
1041         if (prb_queue_frozen(pkc)) {
1042                 /*
1043                  * Check if that last block which caused the queue to freeze,
1044                  * is still in_use by user-space.
1045                  */
1046                 if (prb_curr_blk_in_use(pkc, pbd)) {
1047                         /* Can't record this packet */
1048                         return NULL;
1049                 } else {
1050                         /*
1051                          * Ok, the block was released by user-space.
1052                          * Now let's open that block.
1053                          * opening a block also thaws the queue.
1054                          * Thawing is a side effect.
1055                          */
1056                         prb_open_block(pkc, pbd);
1057                 }
1058         }
1059
1060         smp_mb();
1061         curr = pkc->nxt_offset;
1062         pkc->skb = skb;
1063         end = (char *)pbd + pkc->kblk_size;
1064
1065         /* first try the current block */
1066         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1067                 prb_fill_curr_block(curr, pkc, pbd, len);
1068                 return (void *)curr;
1069         }
1070
1071         /* Ok, close the current block */
1072         prb_retire_current_block(pkc, po, 0);
1073
1074         /* Now, try to dispatch the next block */
1075         curr = (char *)prb_dispatch_next_block(pkc, po);
1076         if (curr) {
1077                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1078                 prb_fill_curr_block(curr, pkc, pbd, len);
1079                 return (void *)curr;
1080         }
1081
1082         /*
1083          * No free blocks are available.user_space hasn't caught up yet.
1084          * Queue was just frozen and now this packet will get dropped.
1085          */
1086         return NULL;
1087 }
1088
1089 static void *packet_current_rx_frame(struct packet_sock *po,
1090                                             struct sk_buff *skb,
1091                                             int status, unsigned int len)
1092 {
1093         char *curr = NULL;
1094         switch (po->tp_version) {
1095         case TPACKET_V1:
1096         case TPACKET_V2:
1097                 curr = packet_lookup_frame(po, &po->rx_ring,
1098                                         po->rx_ring.head, status);
1099                 return curr;
1100         case TPACKET_V3:
1101                 return __packet_lookup_frame_in_block(po, skb, status, len);
1102         default:
1103                 WARN(1, "TPACKET version not supported\n");
1104                 BUG();
1105                 return NULL;
1106         }
1107 }
1108
1109 static void *prb_lookup_block(struct packet_sock *po,
1110                                      struct packet_ring_buffer *rb,
1111                                      unsigned int idx,
1112                                      int status)
1113 {
1114         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1115         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1116
1117         if (status != BLOCK_STATUS(pbd))
1118                 return NULL;
1119         return pbd;
1120 }
1121
1122 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1123 {
1124         unsigned int prev;
1125         if (rb->prb_bdqc.kactive_blk_num)
1126                 prev = rb->prb_bdqc.kactive_blk_num-1;
1127         else
1128                 prev = rb->prb_bdqc.knum_blocks-1;
1129         return prev;
1130 }
1131
1132 /* Assumes caller has held the rx_queue.lock */
1133 static void *__prb_previous_block(struct packet_sock *po,
1134                                          struct packet_ring_buffer *rb,
1135                                          int status)
1136 {
1137         unsigned int previous = prb_previous_blk_num(rb);
1138         return prb_lookup_block(po, rb, previous, status);
1139 }
1140
1141 static void *packet_previous_rx_frame(struct packet_sock *po,
1142                                              struct packet_ring_buffer *rb,
1143                                              int status)
1144 {
1145         if (po->tp_version <= TPACKET_V2)
1146                 return packet_previous_frame(po, rb, status);
1147
1148         return __prb_previous_block(po, rb, status);
1149 }
1150
1151 static void packet_increment_rx_head(struct packet_sock *po,
1152                                             struct packet_ring_buffer *rb)
1153 {
1154         switch (po->tp_version) {
1155         case TPACKET_V1:
1156         case TPACKET_V2:
1157                 return packet_increment_head(rb);
1158         case TPACKET_V3:
1159         default:
1160                 WARN(1, "TPACKET version not supported.\n");
1161                 BUG();
1162                 return;
1163         }
1164 }
1165
1166 static void *packet_previous_frame(struct packet_sock *po,
1167                 struct packet_ring_buffer *rb,
1168                 int status)
1169 {
1170         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1171         return packet_lookup_frame(po, rb, previous, status);
1172 }
1173
1174 static void packet_increment_head(struct packet_ring_buffer *buff)
1175 {
1176         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1177 }
1178
1179 static void packet_inc_pending(struct packet_ring_buffer *rb)
1180 {
1181         this_cpu_inc(*rb->pending_refcnt);
1182 }
1183
1184 static void packet_dec_pending(struct packet_ring_buffer *rb)
1185 {
1186         this_cpu_dec(*rb->pending_refcnt);
1187 }
1188
1189 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1190 {
1191         unsigned int refcnt = 0;
1192         int cpu;
1193
1194         /* We don't use pending refcount in rx_ring. */
1195         if (rb->pending_refcnt == NULL)
1196                 return 0;
1197
1198         for_each_possible_cpu(cpu)
1199                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1200
1201         return refcnt;
1202 }
1203
1204 static int packet_alloc_pending(struct packet_sock *po)
1205 {
1206         po->rx_ring.pending_refcnt = NULL;
1207
1208         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1209         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1210                 return -ENOBUFS;
1211
1212         return 0;
1213 }
1214
1215 static void packet_free_pending(struct packet_sock *po)
1216 {
1217         free_percpu(po->tx_ring.pending_refcnt);
1218 }
1219
1220 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1221 {
1222         struct sock *sk = &po->sk;
1223         bool has_room;
1224
1225         if (po->prot_hook.func != tpacket_rcv)
1226                 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1227                         <= sk->sk_rcvbuf;
1228
1229         spin_lock(&sk->sk_receive_queue.lock);
1230         if (po->tp_version == TPACKET_V3)
1231                 has_room = prb_lookup_block(po, &po->rx_ring,
1232                                             po->rx_ring.prb_bdqc.kactive_blk_num,
1233                                             TP_STATUS_KERNEL);
1234         else
1235                 has_room = packet_lookup_frame(po, &po->rx_ring,
1236                                                po->rx_ring.head,
1237                                                TP_STATUS_KERNEL);
1238         spin_unlock(&sk->sk_receive_queue.lock);
1239
1240         return has_room;
1241 }
1242
1243 static void packet_sock_destruct(struct sock *sk)
1244 {
1245         skb_queue_purge(&sk->sk_error_queue);
1246
1247         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1248         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1249
1250         if (!sock_flag(sk, SOCK_DEAD)) {
1251                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1252                 return;
1253         }
1254
1255         sk_refcnt_debug_dec(sk);
1256 }
1257
1258 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1259 {
1260         int x = atomic_read(&f->rr_cur) + 1;
1261
1262         if (x >= num)
1263                 x = 0;
1264
1265         return x;
1266 }
1267
1268 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1269                                       struct sk_buff *skb,
1270                                       unsigned int num)
1271 {
1272         return reciprocal_scale(skb_get_hash(skb), num);
1273 }
1274
1275 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1276                                     struct sk_buff *skb,
1277                                     unsigned int num)
1278 {
1279         int cur, old;
1280
1281         cur = atomic_read(&f->rr_cur);
1282         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1283                                      fanout_rr_next(f, num))) != cur)
1284                 cur = old;
1285         return cur;
1286 }
1287
1288 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1289                                      struct sk_buff *skb,
1290                                      unsigned int num)
1291 {
1292         return smp_processor_id() % num;
1293 }
1294
1295 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1296                                      struct sk_buff *skb,
1297                                      unsigned int num)
1298 {
1299         return prandom_u32_max(num);
1300 }
1301
1302 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1303                                           struct sk_buff *skb,
1304                                           unsigned int idx, unsigned int skip,
1305                                           unsigned int num)
1306 {
1307         unsigned int i, j;
1308
1309         i = j = min_t(int, f->next[idx], num - 1);
1310         do {
1311                 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1312                         if (i != j)
1313                                 f->next[idx] = i;
1314                         return i;
1315                 }
1316                 if (++i == num)
1317                         i = 0;
1318         } while (i != j);
1319
1320         return idx;
1321 }
1322
1323 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1324                                     struct sk_buff *skb,
1325                                     unsigned int num)
1326 {
1327         return skb_get_queue_mapping(skb) % num;
1328 }
1329
1330 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1331 {
1332         return f->flags & (flag >> 8);
1333 }
1334
1335 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1336                              struct packet_type *pt, struct net_device *orig_dev)
1337 {
1338         struct packet_fanout *f = pt->af_packet_priv;
1339         unsigned int num = f->num_members;
1340         struct packet_sock *po;
1341         unsigned int idx;
1342
1343         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1344             !num) {
1345                 kfree_skb(skb);
1346                 return 0;
1347         }
1348
1349         switch (f->type) {
1350         case PACKET_FANOUT_HASH:
1351         default:
1352                 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1353                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1354                         if (!skb)
1355                                 return 0;
1356                 }
1357                 idx = fanout_demux_hash(f, skb, num);
1358                 break;
1359         case PACKET_FANOUT_LB:
1360                 idx = fanout_demux_lb(f, skb, num);
1361                 break;
1362         case PACKET_FANOUT_CPU:
1363                 idx = fanout_demux_cpu(f, skb, num);
1364                 break;
1365         case PACKET_FANOUT_RND:
1366                 idx = fanout_demux_rnd(f, skb, num);
1367                 break;
1368         case PACKET_FANOUT_QM:
1369                 idx = fanout_demux_qm(f, skb, num);
1370                 break;
1371         case PACKET_FANOUT_ROLLOVER:
1372                 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1373                 break;
1374         }
1375
1376         po = pkt_sk(f->arr[idx]);
1377         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1378             unlikely(!packet_rcv_has_room(po, skb))) {
1379                 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1380                 po = pkt_sk(f->arr[idx]);
1381         }
1382
1383         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1384 }
1385
1386 DEFINE_MUTEX(fanout_mutex);
1387 EXPORT_SYMBOL_GPL(fanout_mutex);
1388 static LIST_HEAD(fanout_list);
1389
1390 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1391 {
1392         struct packet_fanout *f = po->fanout;
1393
1394         spin_lock(&f->lock);
1395         f->arr[f->num_members] = sk;
1396         smp_wmb();
1397         f->num_members++;
1398         spin_unlock(&f->lock);
1399 }
1400
1401 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1402 {
1403         struct packet_fanout *f = po->fanout;
1404         int i;
1405
1406         spin_lock(&f->lock);
1407         for (i = 0; i < f->num_members; i++) {
1408                 if (f->arr[i] == sk)
1409                         break;
1410         }
1411         BUG_ON(i >= f->num_members);
1412         f->arr[i] = f->arr[f->num_members - 1];
1413         f->num_members--;
1414         spin_unlock(&f->lock);
1415 }
1416
1417 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1418 {
1419         if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1420                 return true;
1421
1422         return false;
1423 }
1424
1425 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1426 {
1427         struct packet_sock *po = pkt_sk(sk);
1428         struct packet_fanout *f, *match;
1429         u8 type = type_flags & 0xff;
1430         u8 flags = type_flags >> 8;
1431         int err;
1432
1433         switch (type) {
1434         case PACKET_FANOUT_ROLLOVER:
1435                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1436                         return -EINVAL;
1437         case PACKET_FANOUT_HASH:
1438         case PACKET_FANOUT_LB:
1439         case PACKET_FANOUT_CPU:
1440         case PACKET_FANOUT_RND:
1441         case PACKET_FANOUT_QM:
1442                 break;
1443         default:
1444                 return -EINVAL;
1445         }
1446
1447         if (!po->running)
1448                 return -EINVAL;
1449
1450         if (po->fanout)
1451                 return -EALREADY;
1452
1453         mutex_lock(&fanout_mutex);
1454         match = NULL;
1455         list_for_each_entry(f, &fanout_list, list) {
1456                 if (f->id == id &&
1457                     read_pnet(&f->net) == sock_net(sk)) {
1458                         match = f;
1459                         break;
1460                 }
1461         }
1462         err = -EINVAL;
1463         if (match && match->flags != flags)
1464                 goto out;
1465         if (!match) {
1466                 err = -ENOMEM;
1467                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1468                 if (!match)
1469                         goto out;
1470                 write_pnet(&match->net, sock_net(sk));
1471                 match->id = id;
1472                 match->type = type;
1473                 match->flags = flags;
1474                 atomic_set(&match->rr_cur, 0);
1475                 INIT_LIST_HEAD(&match->list);
1476                 spin_lock_init(&match->lock);
1477                 atomic_set(&match->sk_ref, 0);
1478                 match->prot_hook.type = po->prot_hook.type;
1479                 match->prot_hook.dev = po->prot_hook.dev;
1480                 match->prot_hook.func = packet_rcv_fanout;
1481                 match->prot_hook.af_packet_priv = match;
1482                 match->prot_hook.id_match = match_fanout_group;
1483                 dev_add_pack(&match->prot_hook);
1484                 list_add(&match->list, &fanout_list);
1485         }
1486         err = -EINVAL;
1487         if (match->type == type &&
1488             match->prot_hook.type == po->prot_hook.type &&
1489             match->prot_hook.dev == po->prot_hook.dev) {
1490                 err = -ENOSPC;
1491                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1492                         __dev_remove_pack(&po->prot_hook);
1493                         po->fanout = match;
1494                         atomic_inc(&match->sk_ref);
1495                         __fanout_link(sk, po);
1496                         err = 0;
1497                 }
1498         }
1499 out:
1500         mutex_unlock(&fanout_mutex);
1501         return err;
1502 }
1503
1504 static void fanout_release(struct sock *sk)
1505 {
1506         struct packet_sock *po = pkt_sk(sk);
1507         struct packet_fanout *f;
1508
1509         f = po->fanout;
1510         if (!f)
1511                 return;
1512
1513         mutex_lock(&fanout_mutex);
1514         po->fanout = NULL;
1515
1516         if (atomic_dec_and_test(&f->sk_ref)) {
1517                 list_del(&f->list);
1518                 dev_remove_pack(&f->prot_hook);
1519                 kfree(f);
1520         }
1521         mutex_unlock(&fanout_mutex);
1522 }
1523
1524 static const struct proto_ops packet_ops;
1525
1526 static const struct proto_ops packet_ops_spkt;
1527
1528 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1529                            struct packet_type *pt, struct net_device *orig_dev)
1530 {
1531         struct sock *sk;
1532         struct sockaddr_pkt *spkt;
1533
1534         /*
1535          *      When we registered the protocol we saved the socket in the data
1536          *      field for just this event.
1537          */
1538
1539         sk = pt->af_packet_priv;
1540
1541         /*
1542          *      Yank back the headers [hope the device set this
1543          *      right or kerboom...]
1544          *
1545          *      Incoming packets have ll header pulled,
1546          *      push it back.
1547          *
1548          *      For outgoing ones skb->data == skb_mac_header(skb)
1549          *      so that this procedure is noop.
1550          */
1551
1552         if (skb->pkt_type == PACKET_LOOPBACK)
1553                 goto out;
1554
1555         if (!net_eq(dev_net(dev), sock_net(sk)))
1556                 goto out;
1557
1558         skb = skb_share_check(skb, GFP_ATOMIC);
1559         if (skb == NULL)
1560                 goto oom;
1561
1562         /* drop any routing info */
1563         skb_dst_drop(skb);
1564
1565         /* drop conntrack reference */
1566         nf_reset(skb);
1567
1568         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1569
1570         skb_push(skb, skb->data - skb_mac_header(skb));
1571
1572         /*
1573          *      The SOCK_PACKET socket receives _all_ frames.
1574          */
1575
1576         spkt->spkt_family = dev->type;
1577         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1578         spkt->spkt_protocol = skb->protocol;
1579
1580         /*
1581          *      Charge the memory to the socket. This is done specifically
1582          *      to prevent sockets using all the memory up.
1583          */
1584
1585         if (sock_queue_rcv_skb(sk, skb) == 0)
1586                 return 0;
1587
1588 out:
1589         kfree_skb(skb);
1590 oom:
1591         return 0;
1592 }
1593
1594
1595 /*
1596  *      Output a raw packet to a device layer. This bypasses all the other
1597  *      protocol layers and you must therefore supply it with a complete frame
1598  */
1599
1600 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1601                                struct msghdr *msg, size_t len)
1602 {
1603         struct sock *sk = sock->sk;
1604         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1605         struct sk_buff *skb = NULL;
1606         struct net_device *dev;
1607         __be16 proto = 0;
1608         int err;
1609         int extra_len = 0;
1610
1611         /*
1612          *      Get and verify the address.
1613          */
1614
1615         if (saddr) {
1616                 if (msg->msg_namelen < sizeof(struct sockaddr))
1617                         return -EINVAL;
1618                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1619                         proto = saddr->spkt_protocol;
1620         } else
1621                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1622
1623         /*
1624          *      Find the device first to size check it
1625          */
1626
1627         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1628 retry:
1629         rcu_read_lock();
1630         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1631         err = -ENODEV;
1632         if (dev == NULL)
1633                 goto out_unlock;
1634
1635         err = -ENETDOWN;
1636         if (!(dev->flags & IFF_UP))
1637                 goto out_unlock;
1638
1639         /*
1640          * You may not queue a frame bigger than the mtu. This is the lowest level
1641          * raw protocol and you must do your own fragmentation at this level.
1642          */
1643
1644         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1645                 if (!netif_supports_nofcs(dev)) {
1646                         err = -EPROTONOSUPPORT;
1647                         goto out_unlock;
1648                 }
1649                 extra_len = 4; /* We're doing our own CRC */
1650         }
1651
1652         err = -EMSGSIZE;
1653         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1654                 goto out_unlock;
1655
1656         if (!skb) {
1657                 size_t reserved = LL_RESERVED_SPACE(dev);
1658                 int tlen = dev->needed_tailroom;
1659                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1660
1661                 rcu_read_unlock();
1662                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1663                 if (skb == NULL)
1664                         return -ENOBUFS;
1665                 /* FIXME: Save some space for broken drivers that write a hard
1666                  * header at transmission time by themselves. PPP is the notable
1667                  * one here. This should really be fixed at the driver level.
1668                  */
1669                 skb_reserve(skb, reserved);
1670                 skb_reset_network_header(skb);
1671
1672                 /* Try to align data part correctly */
1673                 if (hhlen) {
1674                         skb->data -= hhlen;
1675                         skb->tail -= hhlen;
1676                         if (len < hhlen)
1677                                 skb_reset_network_header(skb);
1678                 }
1679                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1680                 if (err)
1681                         goto out_free;
1682                 goto retry;
1683         }
1684
1685         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1686                 /* Earlier code assumed this would be a VLAN pkt,
1687                  * double-check this now that we have the actual
1688                  * packet in hand.
1689                  */
1690                 struct ethhdr *ehdr;
1691                 skb_reset_mac_header(skb);
1692                 ehdr = eth_hdr(skb);
1693                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1694                         err = -EMSGSIZE;
1695                         goto out_unlock;
1696                 }
1697         }
1698
1699         skb->protocol = proto;
1700         skb->dev = dev;
1701         skb->priority = sk->sk_priority;
1702         skb->mark = sk->sk_mark;
1703
1704         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1705
1706         if (unlikely(extra_len == 4))
1707                 skb->no_fcs = 1;
1708
1709         skb_probe_transport_header(skb, 0);
1710
1711         dev_queue_xmit(skb);
1712         rcu_read_unlock();
1713         return len;
1714
1715 out_unlock:
1716         rcu_read_unlock();
1717 out_free:
1718         kfree_skb(skb);
1719         return err;
1720 }
1721
1722 static unsigned int run_filter(const struct sk_buff *skb,
1723                                       const struct sock *sk,
1724                                       unsigned int res)
1725 {
1726         struct sk_filter *filter;
1727
1728         rcu_read_lock();
1729         filter = rcu_dereference(sk->sk_filter);
1730         if (filter != NULL)
1731                 res = SK_RUN_FILTER(filter, skb);
1732         rcu_read_unlock();
1733
1734         return res;
1735 }
1736
1737 /*
1738  * This function makes lazy skb cloning in hope that most of packets
1739  * are discarded by BPF.
1740  *
1741  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1742  * and skb->cb are mangled. It works because (and until) packets
1743  * falling here are owned by current CPU. Output packets are cloned
1744  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1745  * sequencially, so that if we return skb to original state on exit,
1746  * we will not harm anyone.
1747  */
1748
1749 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1750                       struct packet_type *pt, struct net_device *orig_dev)
1751 {
1752         struct sock *sk;
1753         struct sockaddr_ll *sll;
1754         struct packet_sock *po;
1755         u8 *skb_head = skb->data;
1756         int skb_len = skb->len;
1757         unsigned int snaplen, res;
1758
1759         if (skb->pkt_type == PACKET_LOOPBACK)
1760                 goto drop;
1761
1762         sk = pt->af_packet_priv;
1763         po = pkt_sk(sk);
1764
1765         if (!net_eq(dev_net(dev), sock_net(sk)))
1766                 goto drop;
1767
1768         skb->dev = dev;
1769
1770         if (dev->header_ops) {
1771                 /* The device has an explicit notion of ll header,
1772                  * exported to higher levels.
1773                  *
1774                  * Otherwise, the device hides details of its frame
1775                  * structure, so that corresponding packet head is
1776                  * never delivered to user.
1777                  */
1778                 if (sk->sk_type != SOCK_DGRAM)
1779                         skb_push(skb, skb->data - skb_mac_header(skb));
1780                 else if (skb->pkt_type == PACKET_OUTGOING) {
1781                         /* Special case: outgoing packets have ll header at head */
1782                         skb_pull(skb, skb_network_offset(skb));
1783                 }
1784         }
1785
1786         snaplen = skb->len;
1787
1788         res = run_filter(skb, sk, snaplen);
1789         if (!res)
1790                 goto drop_n_restore;
1791         if (snaplen > res)
1792                 snaplen = res;
1793
1794         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1795                 goto drop_n_acct;
1796
1797         if (skb_shared(skb)) {
1798                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1799                 if (nskb == NULL)
1800                         goto drop_n_acct;
1801
1802                 if (skb_head != skb->data) {
1803                         skb->data = skb_head;
1804                         skb->len = skb_len;
1805                 }
1806                 consume_skb(skb);
1807                 skb = nskb;
1808         }
1809
1810         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1811                      sizeof(skb->cb));
1812
1813         sll = &PACKET_SKB_CB(skb)->sa.ll;
1814         sll->sll_family = AF_PACKET;
1815         sll->sll_hatype = dev->type;
1816         sll->sll_protocol = skb->protocol;
1817         sll->sll_pkttype = skb->pkt_type;
1818         if (unlikely(po->origdev))
1819                 sll->sll_ifindex = orig_dev->ifindex;
1820         else
1821                 sll->sll_ifindex = dev->ifindex;
1822
1823         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1824
1825         PACKET_SKB_CB(skb)->origlen = skb->len;
1826
1827         if (pskb_trim(skb, snaplen))
1828                 goto drop_n_acct;
1829
1830         skb_set_owner_r(skb, sk);
1831         skb->dev = NULL;
1832         skb_dst_drop(skb);
1833
1834         /* drop conntrack reference */
1835         nf_reset(skb);
1836
1837         spin_lock(&sk->sk_receive_queue.lock);
1838         po->stats.stats1.tp_packets++;
1839         skb->dropcount = atomic_read(&sk->sk_drops);
1840         __skb_queue_tail(&sk->sk_receive_queue, skb);
1841         spin_unlock(&sk->sk_receive_queue.lock);
1842         sk->sk_data_ready(sk);
1843         return 0;
1844
1845 drop_n_acct:
1846         spin_lock(&sk->sk_receive_queue.lock);
1847         po->stats.stats1.tp_drops++;
1848         atomic_inc(&sk->sk_drops);
1849         spin_unlock(&sk->sk_receive_queue.lock);
1850
1851 drop_n_restore:
1852         if (skb_head != skb->data && skb_shared(skb)) {
1853                 skb->data = skb_head;
1854                 skb->len = skb_len;
1855         }
1856 drop:
1857         consume_skb(skb);
1858         return 0;
1859 }
1860
1861 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1862                        struct packet_type *pt, struct net_device *orig_dev)
1863 {
1864         struct sock *sk;
1865         struct packet_sock *po;
1866         struct sockaddr_ll *sll;
1867         union tpacket_uhdr h;
1868         u8 *skb_head = skb->data;
1869         int skb_len = skb->len;
1870         unsigned int snaplen, res;
1871         unsigned long status = TP_STATUS_USER;
1872         unsigned short macoff, netoff, hdrlen;
1873         struct sk_buff *copy_skb = NULL;
1874         struct timespec ts;
1875         __u32 ts_status;
1876
1877         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1878          * We may add members to them until current aligned size without forcing
1879          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1880          */
1881         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1882         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1883
1884         if (skb->pkt_type == PACKET_LOOPBACK)
1885                 goto drop;
1886
1887         sk = pt->af_packet_priv;
1888         po = pkt_sk(sk);
1889
1890         if (!net_eq(dev_net(dev), sock_net(sk)))
1891                 goto drop;
1892
1893         if (dev->header_ops) {
1894                 if (sk->sk_type != SOCK_DGRAM)
1895                         skb_push(skb, skb->data - skb_mac_header(skb));
1896                 else if (skb->pkt_type == PACKET_OUTGOING) {
1897                         /* Special case: outgoing packets have ll header at head */
1898                         skb_pull(skb, skb_network_offset(skb));
1899                 }
1900         }
1901
1902         if (skb->ip_summed == CHECKSUM_PARTIAL)
1903                 status |= TP_STATUS_CSUMNOTREADY;
1904
1905         snaplen = skb->len;
1906
1907         res = run_filter(skb, sk, snaplen);
1908         if (!res)
1909                 goto drop_n_restore;
1910         if (snaplen > res)
1911                 snaplen = res;
1912
1913         if (sk->sk_type == SOCK_DGRAM) {
1914                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1915                                   po->tp_reserve;
1916         } else {
1917                 unsigned int maclen = skb_network_offset(skb);
1918                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1919                                        (maclen < 16 ? 16 : maclen)) +
1920                         po->tp_reserve;
1921                 macoff = netoff - maclen;
1922         }
1923         if (po->tp_version <= TPACKET_V2) {
1924                 if (macoff + snaplen > po->rx_ring.frame_size) {
1925                         if (po->copy_thresh &&
1926                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1927                                 if (skb_shared(skb)) {
1928                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1929                                 } else {
1930                                         copy_skb = skb_get(skb);
1931                                         skb_head = skb->data;
1932                                 }
1933                                 if (copy_skb)
1934                                         skb_set_owner_r(copy_skb, sk);
1935                         }
1936                         snaplen = po->rx_ring.frame_size - macoff;
1937                         if ((int)snaplen < 0)
1938                                 snaplen = 0;
1939                 }
1940         } else if (unlikely(macoff + snaplen >
1941                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1942                 u32 nval;
1943
1944                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1945                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1946                             snaplen, nval, macoff);
1947                 snaplen = nval;
1948                 if (unlikely((int)snaplen < 0)) {
1949                         snaplen = 0;
1950                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1951                 }
1952         }
1953         spin_lock(&sk->sk_receive_queue.lock);
1954         h.raw = packet_current_rx_frame(po, skb,
1955                                         TP_STATUS_KERNEL, (macoff+snaplen));
1956         if (!h.raw)
1957                 goto ring_is_full;
1958         if (po->tp_version <= TPACKET_V2) {
1959                 packet_increment_rx_head(po, &po->rx_ring);
1960         /*
1961          * LOSING will be reported till you read the stats,
1962          * because it's COR - Clear On Read.
1963          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1964          * at packet level.
1965          */
1966                 if (po->stats.stats1.tp_drops)
1967                         status |= TP_STATUS_LOSING;
1968         }
1969         po->stats.stats1.tp_packets++;
1970         if (copy_skb) {
1971                 status |= TP_STATUS_COPY;
1972                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1973         }
1974         spin_unlock(&sk->sk_receive_queue.lock);
1975
1976         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1977
1978         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1979                 getnstimeofday(&ts);
1980
1981         status |= ts_status;
1982
1983         switch (po->tp_version) {
1984         case TPACKET_V1:
1985                 h.h1->tp_len = skb->len;
1986                 h.h1->tp_snaplen = snaplen;
1987                 h.h1->tp_mac = macoff;
1988                 h.h1->tp_net = netoff;
1989                 h.h1->tp_sec = ts.tv_sec;
1990                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1991                 hdrlen = sizeof(*h.h1);
1992                 break;
1993         case TPACKET_V2:
1994                 h.h2->tp_len = skb->len;
1995                 h.h2->tp_snaplen = snaplen;
1996                 h.h2->tp_mac = macoff;
1997                 h.h2->tp_net = netoff;
1998                 h.h2->tp_sec = ts.tv_sec;
1999                 h.h2->tp_nsec = ts.tv_nsec;
2000                 if (vlan_tx_tag_present(skb)) {
2001                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
2002                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2003                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2004                 } else {
2005                         h.h2->tp_vlan_tci = 0;
2006                         h.h2->tp_vlan_tpid = 0;
2007                 }
2008                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2009                 hdrlen = sizeof(*h.h2);
2010                 break;
2011         case TPACKET_V3:
2012                 /* tp_nxt_offset,vlan are already populated above.
2013                  * So DONT clear those fields here
2014                  */
2015                 h.h3->tp_status |= status;
2016                 h.h3->tp_len = skb->len;
2017                 h.h3->tp_snaplen = snaplen;
2018                 h.h3->tp_mac = macoff;
2019                 h.h3->tp_net = netoff;
2020                 h.h3->tp_sec  = ts.tv_sec;
2021                 h.h3->tp_nsec = ts.tv_nsec;
2022                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2023                 hdrlen = sizeof(*h.h3);
2024                 break;
2025         default:
2026                 BUG();
2027         }
2028
2029         sll = h.raw + TPACKET_ALIGN(hdrlen);
2030         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2031         sll->sll_family = AF_PACKET;
2032         sll->sll_hatype = dev->type;
2033         sll->sll_protocol = skb->protocol;
2034         sll->sll_pkttype = skb->pkt_type;
2035         if (unlikely(po->origdev))
2036                 sll->sll_ifindex = orig_dev->ifindex;
2037         else
2038                 sll->sll_ifindex = dev->ifindex;
2039
2040         smp_mb();
2041
2042 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2043         if (po->tp_version <= TPACKET_V2) {
2044                 u8 *start, *end;
2045
2046                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2047                                         macoff + snaplen);
2048
2049                 for (start = h.raw; start < end; start += PAGE_SIZE)
2050                         flush_dcache_page(pgv_to_page(start));
2051         }
2052         smp_wmb();
2053 #endif
2054
2055         if (po->tp_version <= TPACKET_V2)
2056                 __packet_set_status(po, h.raw, status);
2057         else
2058                 prb_clear_blk_fill_status(&po->rx_ring);
2059
2060         sk->sk_data_ready(sk);
2061
2062 drop_n_restore:
2063         if (skb_head != skb->data && skb_shared(skb)) {
2064                 skb->data = skb_head;
2065                 skb->len = skb_len;
2066         }
2067 drop:
2068         kfree_skb(skb);
2069         return 0;
2070
2071 ring_is_full:
2072         po->stats.stats1.tp_drops++;
2073         spin_unlock(&sk->sk_receive_queue.lock);
2074
2075         sk->sk_data_ready(sk);
2076         kfree_skb(copy_skb);
2077         goto drop_n_restore;
2078 }
2079
2080 static void tpacket_destruct_skb(struct sk_buff *skb)
2081 {
2082         struct packet_sock *po = pkt_sk(skb->sk);
2083
2084         if (likely(po->tx_ring.pg_vec)) {
2085                 void *ph;
2086                 __u32 ts;
2087
2088                 ph = skb_shinfo(skb)->destructor_arg;
2089                 packet_dec_pending(&po->tx_ring);
2090
2091                 ts = __packet_set_timestamp(po, ph, skb);
2092                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2093         }
2094
2095         sock_wfree(skb);
2096 }
2097
2098 static bool ll_header_truncated(const struct net_device *dev, int len)
2099 {
2100         /* net device doesn't like empty head */
2101         if (unlikely(len <= dev->hard_header_len)) {
2102                 net_warn_ratelimited("%s: packet size is too short (%d < %d)\n",
2103                                      current->comm, len, dev->hard_header_len);
2104                 return true;
2105         }
2106
2107         return false;
2108 }
2109
2110 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2111                 void *frame, struct net_device *dev, int size_max,
2112                 __be16 proto, unsigned char *addr, int hlen)
2113 {
2114         union tpacket_uhdr ph;
2115         int to_write, offset, len, tp_len, nr_frags, len_max;
2116         struct socket *sock = po->sk.sk_socket;
2117         struct page *page;
2118         void *data;
2119         int err;
2120
2121         ph.raw = frame;
2122
2123         skb->protocol = proto;
2124         skb->dev = dev;
2125         skb->priority = po->sk.sk_priority;
2126         skb->mark = po->sk.sk_mark;
2127         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2128         skb_shinfo(skb)->destructor_arg = ph.raw;
2129
2130         switch (po->tp_version) {
2131         case TPACKET_V2:
2132                 tp_len = ph.h2->tp_len;
2133                 break;
2134         default:
2135                 tp_len = ph.h1->tp_len;
2136                 break;
2137         }
2138         if (unlikely(tp_len > size_max)) {
2139                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2140                 return -EMSGSIZE;
2141         }
2142
2143         skb_reserve(skb, hlen);
2144         skb_reset_network_header(skb);
2145
2146         if (!packet_use_direct_xmit(po))
2147                 skb_probe_transport_header(skb, 0);
2148         if (unlikely(po->tp_tx_has_off)) {
2149                 int off_min, off_max, off;
2150                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2151                 off_max = po->tx_ring.frame_size - tp_len;
2152                 if (sock->type == SOCK_DGRAM) {
2153                         switch (po->tp_version) {
2154                         case TPACKET_V2:
2155                                 off = ph.h2->tp_net;
2156                                 break;
2157                         default:
2158                                 off = ph.h1->tp_net;
2159                                 break;
2160                         }
2161                 } else {
2162                         switch (po->tp_version) {
2163                         case TPACKET_V2:
2164                                 off = ph.h2->tp_mac;
2165                                 break;
2166                         default:
2167                                 off = ph.h1->tp_mac;
2168                                 break;
2169                         }
2170                 }
2171                 if (unlikely((off < off_min) || (off_max < off)))
2172                         return -EINVAL;
2173                 data = ph.raw + off;
2174         } else {
2175                 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2176         }
2177         to_write = tp_len;
2178
2179         if (sock->type == SOCK_DGRAM) {
2180                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2181                                 NULL, tp_len);
2182                 if (unlikely(err < 0))
2183                         return -EINVAL;
2184         } else if (dev->hard_header_len) {
2185                 if (ll_header_truncated(dev, tp_len))
2186                         return -EINVAL;
2187
2188                 skb_push(skb, dev->hard_header_len);
2189                 err = skb_store_bits(skb, 0, data,
2190                                 dev->hard_header_len);
2191                 if (unlikely(err))
2192                         return err;
2193
2194                 data += dev->hard_header_len;
2195                 to_write -= dev->hard_header_len;
2196         }
2197
2198         offset = offset_in_page(data);
2199         len_max = PAGE_SIZE - offset;
2200         len = ((to_write > len_max) ? len_max : to_write);
2201
2202         skb->data_len = to_write;
2203         skb->len += to_write;
2204         skb->truesize += to_write;
2205         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2206
2207         while (likely(to_write)) {
2208                 nr_frags = skb_shinfo(skb)->nr_frags;
2209
2210                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2211                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2212                                MAX_SKB_FRAGS);
2213                         return -EFAULT;
2214                 }
2215
2216                 page = pgv_to_page(data);
2217                 data += len;
2218                 flush_dcache_page(page);
2219                 get_page(page);
2220                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2221                 to_write -= len;
2222                 offset = 0;
2223                 len_max = PAGE_SIZE;
2224                 len = ((to_write > len_max) ? len_max : to_write);
2225         }
2226
2227         return tp_len;
2228 }
2229
2230 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2231 {
2232         struct sk_buff *skb;
2233         struct net_device *dev;
2234         __be16 proto;
2235         int err, reserve = 0;
2236         void *ph;
2237         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2238         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2239         int tp_len, size_max;
2240         unsigned char *addr;
2241         int len_sum = 0;
2242         int status = TP_STATUS_AVAILABLE;
2243         int hlen, tlen;
2244
2245         mutex_lock(&po->pg_vec_lock);
2246
2247         if (likely(saddr == NULL)) {
2248                 dev     = packet_cached_dev_get(po);
2249                 proto   = po->num;
2250                 addr    = NULL;
2251         } else {
2252                 err = -EINVAL;
2253                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2254                         goto out;
2255                 if (msg->msg_namelen < (saddr->sll_halen
2256                                         + offsetof(struct sockaddr_ll,
2257                                                 sll_addr)))
2258                         goto out;
2259                 proto   = saddr->sll_protocol;
2260                 addr    = saddr->sll_addr;
2261                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2262         }
2263
2264         err = -ENXIO;
2265         if (unlikely(dev == NULL))
2266                 goto out;
2267         err = -ENETDOWN;
2268         if (unlikely(!(dev->flags & IFF_UP)))
2269                 goto out_put;
2270
2271         reserve = dev->hard_header_len + VLAN_HLEN;
2272         size_max = po->tx_ring.frame_size
2273                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2274
2275         if (size_max > dev->mtu + reserve)
2276                 size_max = dev->mtu + reserve;
2277
2278         do {
2279                 ph = packet_current_frame(po, &po->tx_ring,
2280                                           TP_STATUS_SEND_REQUEST);
2281                 if (unlikely(ph == NULL)) {
2282                         if (need_wait && need_resched())
2283                                 schedule();
2284                         continue;
2285                 }
2286
2287                 status = TP_STATUS_SEND_REQUEST;
2288                 hlen = LL_RESERVED_SPACE(dev);
2289                 tlen = dev->needed_tailroom;
2290                 skb = sock_alloc_send_skb(&po->sk,
2291                                 hlen + tlen + sizeof(struct sockaddr_ll),
2292                                 0, &err);
2293
2294                 if (unlikely(skb == NULL))
2295                         goto out_status;
2296
2297                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2298                                           addr, hlen);
2299                 if (tp_len > dev->mtu + dev->hard_header_len) {
2300                         struct ethhdr *ehdr;
2301                         /* Earlier code assumed this would be a VLAN pkt,
2302                          * double-check this now that we have the actual
2303                          * packet in hand.
2304                          */
2305
2306                         skb_reset_mac_header(skb);
2307                         ehdr = eth_hdr(skb);
2308                         if (ehdr->h_proto != htons(ETH_P_8021Q))
2309                                 tp_len = -EMSGSIZE;
2310                 }
2311                 if (unlikely(tp_len < 0)) {
2312                         if (po->tp_loss) {
2313                                 __packet_set_status(po, ph,
2314                                                 TP_STATUS_AVAILABLE);
2315                                 packet_increment_head(&po->tx_ring);
2316                                 kfree_skb(skb);
2317                                 continue;
2318                         } else {
2319                                 status = TP_STATUS_WRONG_FORMAT;
2320                                 err = tp_len;
2321                                 goto out_status;
2322                         }
2323                 }
2324
2325                 packet_pick_tx_queue(dev, skb);
2326
2327                 skb->destructor = tpacket_destruct_skb;
2328                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2329                 packet_inc_pending(&po->tx_ring);
2330
2331                 status = TP_STATUS_SEND_REQUEST;
2332                 err = po->xmit(skb);
2333                 if (unlikely(err > 0)) {
2334                         err = net_xmit_errno(err);
2335                         if (err && __packet_get_status(po, ph) ==
2336                                    TP_STATUS_AVAILABLE) {
2337                                 /* skb was destructed already */
2338                                 skb = NULL;
2339                                 goto out_status;
2340                         }
2341                         /*
2342                          * skb was dropped but not destructed yet;
2343                          * let's treat it like congestion or err < 0
2344                          */
2345                         err = 0;
2346                 }
2347                 packet_increment_head(&po->tx_ring);
2348                 len_sum += tp_len;
2349         } while (likely((ph != NULL) ||
2350                 /* Note: packet_read_pending() might be slow if we have
2351                  * to call it as it's per_cpu variable, but in fast-path
2352                  * we already short-circuit the loop with the first
2353                  * condition, and luckily don't have to go that path
2354                  * anyway.
2355                  */
2356                  (need_wait && packet_read_pending(&po->tx_ring))));
2357
2358         err = len_sum;
2359         goto out_put;
2360
2361 out_status:
2362         __packet_set_status(po, ph, status);
2363         kfree_skb(skb);
2364 out_put:
2365         dev_put(dev);
2366 out:
2367         mutex_unlock(&po->pg_vec_lock);
2368         return err;
2369 }
2370
2371 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2372                                         size_t reserve, size_t len,
2373                                         size_t linear, int noblock,
2374                                         int *err)
2375 {
2376         struct sk_buff *skb;
2377
2378         /* Under a page?  Don't bother with paged skb. */
2379         if (prepad + len < PAGE_SIZE || !linear)
2380                 linear = len;
2381
2382         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2383                                    err, 0);
2384         if (!skb)
2385                 return NULL;
2386
2387         skb_reserve(skb, reserve);
2388         skb_put(skb, linear);
2389         skb->data_len = len - linear;
2390         skb->len += len - linear;
2391
2392         return skb;
2393 }
2394
2395 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2396 {
2397         struct sock *sk = sock->sk;
2398         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2399         struct sk_buff *skb;
2400         struct net_device *dev;
2401         __be16 proto;
2402         unsigned char *addr;
2403         int err, reserve = 0;
2404         struct virtio_net_hdr vnet_hdr = { 0 };
2405         int offset = 0;
2406         int vnet_hdr_len;
2407         struct packet_sock *po = pkt_sk(sk);
2408         unsigned short gso_type = 0;
2409         int hlen, tlen;
2410         int extra_len = 0;
2411         ssize_t n;
2412
2413         /*
2414          *      Get and verify the address.
2415          */
2416
2417         if (likely(saddr == NULL)) {
2418                 dev     = packet_cached_dev_get(po);
2419                 proto   = po->num;
2420                 addr    = NULL;
2421         } else {
2422                 err = -EINVAL;
2423                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2424                         goto out;
2425                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2426                         goto out;
2427                 proto   = saddr->sll_protocol;
2428                 addr    = saddr->sll_addr;
2429                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2430         }
2431
2432         err = -ENXIO;
2433         if (unlikely(dev == NULL))
2434                 goto out_unlock;
2435         err = -ENETDOWN;
2436         if (unlikely(!(dev->flags & IFF_UP)))
2437                 goto out_unlock;
2438
2439         if (sock->type == SOCK_RAW)
2440                 reserve = dev->hard_header_len;
2441         if (po->has_vnet_hdr) {
2442                 vnet_hdr_len = sizeof(vnet_hdr);
2443
2444                 err = -EINVAL;
2445                 if (len < vnet_hdr_len)
2446                         goto out_unlock;
2447
2448                 len -= vnet_hdr_len;
2449
2450                 err = -EFAULT;
2451                 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2452                 if (n != vnet_hdr_len)
2453                         goto out_unlock;
2454
2455                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2456                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2457                       vnet_hdr.hdr_len))
2458                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2459                                                  vnet_hdr.csum_offset + 2;
2460
2461                 err = -EINVAL;
2462                 if (vnet_hdr.hdr_len > len)
2463                         goto out_unlock;
2464
2465                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2466                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2467                         case VIRTIO_NET_HDR_GSO_TCPV4:
2468                                 gso_type = SKB_GSO_TCPV4;
2469                                 break;
2470                         case VIRTIO_NET_HDR_GSO_TCPV6:
2471                                 gso_type = SKB_GSO_TCPV6;
2472                                 break;
2473                         case VIRTIO_NET_HDR_GSO_UDP:
2474                                 gso_type = SKB_GSO_UDP;
2475                                 break;
2476                         default:
2477                                 goto out_unlock;
2478                         }
2479
2480                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2481                                 gso_type |= SKB_GSO_TCP_ECN;
2482
2483                         if (vnet_hdr.gso_size == 0)
2484                                 goto out_unlock;
2485
2486                 }
2487         }
2488
2489         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2490                 if (!netif_supports_nofcs(dev)) {
2491                         err = -EPROTONOSUPPORT;
2492                         goto out_unlock;
2493                 }
2494                 extra_len = 4; /* We're doing our own CRC */
2495         }
2496
2497         err = -EMSGSIZE;
2498         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2499                 goto out_unlock;
2500
2501         err = -ENOBUFS;
2502         hlen = LL_RESERVED_SPACE(dev);
2503         tlen = dev->needed_tailroom;
2504         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2505                                msg->msg_flags & MSG_DONTWAIT, &err);
2506         if (skb == NULL)
2507                 goto out_unlock;
2508
2509         skb_set_network_header(skb, reserve);
2510
2511         err = -EINVAL;
2512         if (sock->type == SOCK_DGRAM) {
2513                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2514                 if (unlikely(offset) < 0)
2515                         goto out_free;
2516         } else {
2517                 if (ll_header_truncated(dev, len))
2518                         goto out_free;
2519         }
2520
2521         /* Returns -EFAULT on error */
2522         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2523         if (err)
2524                 goto out_free;
2525
2526         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2527
2528         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2529                 /* Earlier code assumed this would be a VLAN pkt,
2530                  * double-check this now that we have the actual
2531                  * packet in hand.
2532                  */
2533                 struct ethhdr *ehdr;
2534                 skb_reset_mac_header(skb);
2535                 ehdr = eth_hdr(skb);
2536                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2537                         err = -EMSGSIZE;
2538                         goto out_free;
2539                 }
2540         }
2541
2542         skb->protocol = proto;
2543         skb->dev = dev;
2544         skb->priority = sk->sk_priority;
2545         skb->mark = sk->sk_mark;
2546
2547         packet_pick_tx_queue(dev, skb);
2548
2549         if (po->has_vnet_hdr) {
2550                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2551                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2552                                                   vnet_hdr.csum_offset)) {
2553                                 err = -EINVAL;
2554                                 goto out_free;
2555                         }
2556                 }
2557
2558                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2559                 skb_shinfo(skb)->gso_type = gso_type;
2560
2561                 /* Header must be checked, and gso_segs computed. */
2562                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2563                 skb_shinfo(skb)->gso_segs = 0;
2564
2565                 len += vnet_hdr_len;
2566         }
2567
2568         if (!packet_use_direct_xmit(po))
2569                 skb_probe_transport_header(skb, reserve);
2570         if (unlikely(extra_len == 4))
2571                 skb->no_fcs = 1;
2572
2573         err = po->xmit(skb);
2574         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2575                 goto out_unlock;
2576
2577         dev_put(dev);
2578
2579         return len;
2580
2581 out_free:
2582         kfree_skb(skb);
2583 out_unlock:
2584         if (dev)
2585                 dev_put(dev);
2586 out:
2587         return err;
2588 }
2589
2590 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2591                 struct msghdr *msg, size_t len)
2592 {
2593         struct sock *sk = sock->sk;
2594         struct packet_sock *po = pkt_sk(sk);
2595
2596         if (po->tx_ring.pg_vec)
2597                 return tpacket_snd(po, msg);
2598         else
2599                 return packet_snd(sock, msg, len);
2600 }
2601
2602 /*
2603  *      Close a PACKET socket. This is fairly simple. We immediately go
2604  *      to 'closed' state and remove our protocol entry in the device list.
2605  */
2606
2607 static int packet_release(struct socket *sock)
2608 {
2609         struct sock *sk = sock->sk;
2610         struct packet_sock *po;
2611         struct net *net;
2612         union tpacket_req_u req_u;
2613
2614         if (!sk)
2615                 return 0;
2616
2617         net = sock_net(sk);
2618         po = pkt_sk(sk);
2619
2620         mutex_lock(&net->packet.sklist_lock);
2621         sk_del_node_init_rcu(sk);
2622         mutex_unlock(&net->packet.sklist_lock);
2623
2624         preempt_disable();
2625         sock_prot_inuse_add(net, sk->sk_prot, -1);
2626         preempt_enable();
2627
2628         spin_lock(&po->bind_lock);
2629         unregister_prot_hook(sk, false);
2630         packet_cached_dev_reset(po);
2631
2632         if (po->prot_hook.dev) {
2633                 dev_put(po->prot_hook.dev);
2634                 po->prot_hook.dev = NULL;
2635         }
2636         spin_unlock(&po->bind_lock);
2637
2638         packet_flush_mclist(sk);
2639
2640         if (po->rx_ring.pg_vec) {
2641                 memset(&req_u, 0, sizeof(req_u));
2642                 packet_set_ring(sk, &req_u, 1, 0);
2643         }
2644
2645         if (po->tx_ring.pg_vec) {
2646                 memset(&req_u, 0, sizeof(req_u));
2647                 packet_set_ring(sk, &req_u, 1, 1);
2648         }
2649
2650         fanout_release(sk);
2651
2652         synchronize_net();
2653         /*
2654          *      Now the socket is dead. No more input will appear.
2655          */
2656         sock_orphan(sk);
2657         sock->sk = NULL;
2658
2659         /* Purge queues */
2660
2661         skb_queue_purge(&sk->sk_receive_queue);
2662         packet_free_pending(po);
2663         sk_refcnt_debug_release(sk);
2664
2665         sock_put(sk);
2666         return 0;
2667 }
2668
2669 /*
2670  *      Attach a packet hook.
2671  */
2672
2673 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2674 {
2675         struct packet_sock *po = pkt_sk(sk);
2676         const struct net_device *dev_curr;
2677         __be16 proto_curr;
2678         bool need_rehook;
2679
2680         if (po->fanout) {
2681                 if (dev)
2682                         dev_put(dev);
2683
2684                 return -EINVAL;
2685         }
2686
2687         lock_sock(sk);
2688         spin_lock(&po->bind_lock);
2689
2690         proto_curr = po->prot_hook.type;
2691         dev_curr = po->prot_hook.dev;
2692
2693         need_rehook = proto_curr != proto || dev_curr != dev;
2694
2695         if (need_rehook) {
2696                 unregister_prot_hook(sk, true);
2697
2698                 po->num = proto;
2699                 po->prot_hook.type = proto;
2700
2701                 if (po->prot_hook.dev)
2702                         dev_put(po->prot_hook.dev);
2703
2704                 po->prot_hook.dev = dev;
2705
2706                 po->ifindex = dev ? dev->ifindex : 0;
2707                 packet_cached_dev_assign(po, dev);
2708         }
2709
2710         if (proto == 0 || !need_rehook)
2711                 goto out_unlock;
2712
2713         if (!dev || (dev->flags & IFF_UP)) {
2714                 register_prot_hook(sk);
2715         } else {
2716                 sk->sk_err = ENETDOWN;
2717                 if (!sock_flag(sk, SOCK_DEAD))
2718                         sk->sk_error_report(sk);
2719         }
2720
2721 out_unlock:
2722         spin_unlock(&po->bind_lock);
2723         release_sock(sk);
2724         return 0;
2725 }
2726
2727 /*
2728  *      Bind a packet socket to a device
2729  */
2730
2731 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2732                             int addr_len)
2733 {
2734         struct sock *sk = sock->sk;
2735         char name[15];
2736         struct net_device *dev;
2737         int err = -ENODEV;
2738
2739         /*
2740          *      Check legality
2741          */
2742
2743         if (addr_len != sizeof(struct sockaddr))
2744                 return -EINVAL;
2745         strlcpy(name, uaddr->sa_data, sizeof(name));
2746
2747         dev = dev_get_by_name(sock_net(sk), name);
2748         if (dev)
2749                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2750         return err;
2751 }
2752
2753 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2754 {
2755         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2756         struct sock *sk = sock->sk;
2757         struct net_device *dev = NULL;
2758         int err;
2759
2760
2761         /*
2762          *      Check legality
2763          */
2764
2765         if (addr_len < sizeof(struct sockaddr_ll))
2766                 return -EINVAL;
2767         if (sll->sll_family != AF_PACKET)
2768                 return -EINVAL;
2769
2770         if (sll->sll_ifindex) {
2771                 err = -ENODEV;
2772                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2773                 if (dev == NULL)
2774                         goto out;
2775         }
2776         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2777
2778 out:
2779         return err;
2780 }
2781
2782 static struct proto packet_proto = {
2783         .name     = "PACKET",
2784         .owner    = THIS_MODULE,
2785         .obj_size = sizeof(struct packet_sock),
2786 };
2787
2788 /*
2789  *      Create a packet of type SOCK_PACKET.
2790  */
2791
2792 static int packet_create(struct net *net, struct socket *sock, int protocol,
2793                          int kern)
2794 {
2795         struct sock *sk;
2796         struct packet_sock *po;
2797         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2798         int err;
2799
2800         if (!ns_capable(net->user_ns, CAP_NET_RAW))
2801                 return -EPERM;
2802         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2803             sock->type != SOCK_PACKET)
2804                 return -ESOCKTNOSUPPORT;
2805
2806         sock->state = SS_UNCONNECTED;
2807
2808         err = -ENOBUFS;
2809         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2810         if (sk == NULL)
2811                 goto out;
2812
2813         sock->ops = &packet_ops;
2814         if (sock->type == SOCK_PACKET)
2815                 sock->ops = &packet_ops_spkt;
2816
2817         sock_init_data(sock, sk);
2818
2819         po = pkt_sk(sk);
2820         sk->sk_family = PF_PACKET;
2821         po->num = proto;
2822         po->xmit = dev_queue_xmit;
2823
2824         err = packet_alloc_pending(po);
2825         if (err)
2826                 goto out2;
2827
2828         packet_cached_dev_reset(po);
2829
2830         sk->sk_destruct = packet_sock_destruct;
2831         sk_refcnt_debug_inc(sk);
2832
2833         /*
2834          *      Attach a protocol block
2835          */
2836
2837         spin_lock_init(&po->bind_lock);
2838         mutex_init(&po->pg_vec_lock);
2839         po->prot_hook.func = packet_rcv;
2840
2841         if (sock->type == SOCK_PACKET)
2842                 po->prot_hook.func = packet_rcv_spkt;
2843
2844         po->prot_hook.af_packet_priv = sk;
2845
2846         if (proto) {
2847                 po->prot_hook.type = proto;
2848                 register_prot_hook(sk);
2849         }
2850
2851         mutex_lock(&net->packet.sklist_lock);
2852         sk_add_node_rcu(sk, &net->packet.sklist);
2853         mutex_unlock(&net->packet.sklist_lock);
2854
2855         preempt_disable();
2856         sock_prot_inuse_add(net, &packet_proto, 1);
2857         preempt_enable();
2858
2859         return 0;
2860 out2:
2861         sk_free(sk);
2862 out:
2863         return err;
2864 }
2865
2866 /*
2867  *      Pull a packet from our receive queue and hand it to the user.
2868  *      If necessary we block.
2869  */
2870
2871 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2872                           struct msghdr *msg, size_t len, int flags)
2873 {
2874         struct sock *sk = sock->sk;
2875         struct sk_buff *skb;
2876         int copied, err;
2877         int vnet_hdr_len = 0;
2878
2879         err = -EINVAL;
2880         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2881                 goto out;
2882
2883 #if 0
2884         /* What error should we return now? EUNATTACH? */
2885         if (pkt_sk(sk)->ifindex < 0)
2886                 return -ENODEV;
2887 #endif
2888
2889         if (flags & MSG_ERRQUEUE) {
2890                 err = sock_recv_errqueue(sk, msg, len,
2891                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
2892                 goto out;
2893         }
2894
2895         /*
2896          *      Call the generic datagram receiver. This handles all sorts
2897          *      of horrible races and re-entrancy so we can forget about it
2898          *      in the protocol layers.
2899          *
2900          *      Now it will return ENETDOWN, if device have just gone down,
2901          *      but then it will block.
2902          */
2903
2904         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2905
2906         /*
2907          *      An error occurred so return it. Because skb_recv_datagram()
2908          *      handles the blocking we don't see and worry about blocking
2909          *      retries.
2910          */
2911
2912         if (skb == NULL)
2913                 goto out;
2914
2915         if (pkt_sk(sk)->has_vnet_hdr) {
2916                 struct virtio_net_hdr vnet_hdr = { 0 };
2917
2918                 err = -EINVAL;
2919                 vnet_hdr_len = sizeof(vnet_hdr);
2920                 if (len < vnet_hdr_len)
2921                         goto out_free;
2922
2923                 len -= vnet_hdr_len;
2924
2925                 if (skb_is_gso(skb)) {
2926                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2927
2928                         /* This is a hint as to how much should be linear. */
2929                         vnet_hdr.hdr_len = skb_headlen(skb);
2930                         vnet_hdr.gso_size = sinfo->gso_size;
2931                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2932                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2933                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2934                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2935                         else if (sinfo->gso_type & SKB_GSO_UDP)
2936                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2937                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2938                                 goto out_free;
2939                         else
2940                                 BUG();
2941                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2942                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2943                 } else
2944                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2945
2946                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2947                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2948                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2949                         vnet_hdr.csum_offset = skb->csum_offset;
2950                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2951                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2952                 } /* else everything is zero */
2953
2954                 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
2955                 if (err < 0)
2956                         goto out_free;
2957         }
2958
2959         /* You lose any data beyond the buffer you gave. If it worries
2960          * a user program they can ask the device for its MTU
2961          * anyway.
2962          */
2963         copied = skb->len;
2964         if (copied > len) {
2965                 copied = len;
2966                 msg->msg_flags |= MSG_TRUNC;
2967         }
2968
2969         err = skb_copy_datagram_msg(skb, 0, msg, copied);
2970         if (err)
2971                 goto out_free;
2972
2973         sock_recv_ts_and_drops(msg, sk, skb);
2974
2975         if (msg->msg_name) {
2976                 /* If the address length field is there to be filled
2977                  * in, we fill it in now.
2978                  */
2979                 if (sock->type == SOCK_PACKET) {
2980                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
2981                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2982                 } else {
2983                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2984                         msg->msg_namelen = sll->sll_halen +
2985                                 offsetof(struct sockaddr_ll, sll_addr);
2986                 }
2987                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2988                        msg->msg_namelen);
2989         }
2990
2991         if (pkt_sk(sk)->auxdata) {
2992                 struct tpacket_auxdata aux;
2993
2994                 aux.tp_status = TP_STATUS_USER;
2995                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2996                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2997                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2998                 aux.tp_snaplen = skb->len;
2999                 aux.tp_mac = 0;
3000                 aux.tp_net = skb_network_offset(skb);
3001                 if (vlan_tx_tag_present(skb)) {
3002                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
3003                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3004                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3005                 } else {
3006                         aux.tp_vlan_tci = 0;
3007                         aux.tp_vlan_tpid = 0;
3008                 }
3009                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3010         }
3011
3012         /*
3013          *      Free or return the buffer as appropriate. Again this
3014          *      hides all the races and re-entrancy issues from us.
3015          */
3016         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3017
3018 out_free:
3019         skb_free_datagram(sk, skb);
3020 out:
3021         return err;
3022 }
3023
3024 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3025                                int *uaddr_len, int peer)
3026 {
3027         struct net_device *dev;
3028         struct sock *sk = sock->sk;
3029
3030         if (peer)
3031                 return -EOPNOTSUPP;
3032
3033         uaddr->sa_family = AF_PACKET;
3034         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3035         rcu_read_lock();
3036         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3037         if (dev)
3038                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3039         rcu_read_unlock();
3040         *uaddr_len = sizeof(*uaddr);
3041
3042         return 0;
3043 }
3044
3045 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3046                           int *uaddr_len, int peer)
3047 {
3048         struct net_device *dev;
3049         struct sock *sk = sock->sk;
3050         struct packet_sock *po = pkt_sk(sk);
3051         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3052
3053         if (peer)
3054                 return -EOPNOTSUPP;
3055
3056         sll->sll_family = AF_PACKET;
3057         sll->sll_ifindex = po->ifindex;
3058         sll->sll_protocol = po->num;
3059         sll->sll_pkttype = 0;
3060         rcu_read_lock();
3061         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3062         if (dev) {
3063                 sll->sll_hatype = dev->type;
3064                 sll->sll_halen = dev->addr_len;
3065                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3066         } else {
3067                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3068                 sll->sll_halen = 0;
3069         }
3070         rcu_read_unlock();
3071         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3072
3073         return 0;
3074 }
3075
3076 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3077                          int what)
3078 {
3079         switch (i->type) {
3080         case PACKET_MR_MULTICAST:
3081                 if (i->alen != dev->addr_len)
3082                         return -EINVAL;
3083                 if (what > 0)
3084                         return dev_mc_add(dev, i->addr);
3085                 else
3086                         return dev_mc_del(dev, i->addr);
3087                 break;
3088         case PACKET_MR_PROMISC:
3089                 return dev_set_promiscuity(dev, what);
3090         case PACKET_MR_ALLMULTI:
3091                 return dev_set_allmulti(dev, what);
3092         case PACKET_MR_UNICAST:
3093                 if (i->alen != dev->addr_len)
3094                         return -EINVAL;
3095                 if (what > 0)
3096                         return dev_uc_add(dev, i->addr);
3097                 else
3098                         return dev_uc_del(dev, i->addr);
3099                 break;
3100         default:
3101                 break;
3102         }
3103         return 0;
3104 }
3105
3106 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
3107 {
3108         for ( ; i; i = i->next) {
3109                 if (i->ifindex == dev->ifindex)
3110                         packet_dev_mc(dev, i, what);
3111         }
3112 }
3113
3114 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3115 {
3116         struct packet_sock *po = pkt_sk(sk);
3117         struct packet_mclist *ml, *i;
3118         struct net_device *dev;
3119         int err;
3120
3121         rtnl_lock();
3122
3123         err = -ENODEV;
3124         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3125         if (!dev)
3126                 goto done;
3127
3128         err = -EINVAL;
3129         if (mreq->mr_alen > dev->addr_len)
3130                 goto done;
3131
3132         err = -ENOBUFS;
3133         i = kmalloc(sizeof(*i), GFP_KERNEL);
3134         if (i == NULL)
3135                 goto done;
3136
3137         err = 0;
3138         for (ml = po->mclist; ml; ml = ml->next) {
3139                 if (ml->ifindex == mreq->mr_ifindex &&
3140                     ml->type == mreq->mr_type &&
3141                     ml->alen == mreq->mr_alen &&
3142                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3143                         ml->count++;
3144                         /* Free the new element ... */
3145                         kfree(i);
3146                         goto done;
3147                 }
3148         }
3149
3150         i->type = mreq->mr_type;
3151         i->ifindex = mreq->mr_ifindex;
3152         i->alen = mreq->mr_alen;
3153         memcpy(i->addr, mreq->mr_address, i->alen);
3154         i->count = 1;
3155         i->next = po->mclist;
3156         po->mclist = i;
3157         err = packet_dev_mc(dev, i, 1);
3158         if (err) {
3159                 po->mclist = i->next;
3160                 kfree(i);
3161         }
3162
3163 done:
3164         rtnl_unlock();
3165         return err;
3166 }
3167
3168 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3169 {
3170         struct packet_mclist *ml, **mlp;
3171
3172         rtnl_lock();
3173
3174         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3175                 if (ml->ifindex == mreq->mr_ifindex &&
3176                     ml->type == mreq->mr_type &&
3177                     ml->alen == mreq->mr_alen &&
3178                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3179                         if (--ml->count == 0) {
3180                                 struct net_device *dev;
3181                                 *mlp = ml->next;
3182                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3183                                 if (dev)
3184                                         packet_dev_mc(dev, ml, -1);
3185                                 kfree(ml);
3186                         }
3187                         rtnl_unlock();
3188                         return 0;
3189                 }
3190         }
3191         rtnl_unlock();
3192         return -EADDRNOTAVAIL;
3193 }
3194
3195 static void packet_flush_mclist(struct sock *sk)
3196 {
3197         struct packet_sock *po = pkt_sk(sk);
3198         struct packet_mclist *ml;
3199
3200         if (!po->mclist)
3201                 return;
3202
3203         rtnl_lock();
3204         while ((ml = po->mclist) != NULL) {
3205                 struct net_device *dev;
3206
3207                 po->mclist = ml->next;
3208                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3209                 if (dev != NULL)
3210                         packet_dev_mc(dev, ml, -1);
3211                 kfree(ml);
3212         }
3213         rtnl_unlock();
3214 }
3215
3216 static int
3217 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3218 {
3219         struct sock *sk = sock->sk;
3220         struct packet_sock *po = pkt_sk(sk);
3221         int ret;
3222
3223         if (level != SOL_PACKET)
3224                 return -ENOPROTOOPT;
3225
3226         switch (optname) {
3227         case PACKET_ADD_MEMBERSHIP:
3228         case PACKET_DROP_MEMBERSHIP:
3229         {
3230                 struct packet_mreq_max mreq;
3231                 int len = optlen;
3232                 memset(&mreq, 0, sizeof(mreq));
3233                 if (len < sizeof(struct packet_mreq))
3234                         return -EINVAL;
3235                 if (len > sizeof(mreq))
3236                         len = sizeof(mreq);
3237                 if (copy_from_user(&mreq, optval, len))
3238                         return -EFAULT;
3239                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3240                         return -EINVAL;
3241                 if (optname == PACKET_ADD_MEMBERSHIP)
3242                         ret = packet_mc_add(sk, &mreq);
3243                 else
3244                         ret = packet_mc_drop(sk, &mreq);
3245                 return ret;
3246         }
3247
3248         case PACKET_RX_RING:
3249         case PACKET_TX_RING:
3250         {
3251                 union tpacket_req_u req_u;
3252                 int len;
3253
3254                 switch (po->tp_version) {
3255                 case TPACKET_V1:
3256                 case TPACKET_V2:
3257                         len = sizeof(req_u.req);
3258                         break;
3259                 case TPACKET_V3:
3260                 default:
3261                         len = sizeof(req_u.req3);
3262                         break;
3263                 }
3264                 if (optlen < len)
3265                         return -EINVAL;
3266                 if (pkt_sk(sk)->has_vnet_hdr)
3267                         return -EINVAL;
3268                 if (copy_from_user(&req_u.req, optval, len))
3269                         return -EFAULT;
3270                 return packet_set_ring(sk, &req_u, 0,
3271                         optname == PACKET_TX_RING);
3272         }
3273         case PACKET_COPY_THRESH:
3274         {
3275                 int val;
3276
3277                 if (optlen != sizeof(val))
3278                         return -EINVAL;
3279                 if (copy_from_user(&val, optval, sizeof(val)))
3280                         return -EFAULT;
3281
3282                 pkt_sk(sk)->copy_thresh = val;
3283                 return 0;
3284         }
3285         case PACKET_VERSION:
3286         {
3287                 int val;
3288
3289                 if (optlen != sizeof(val))
3290                         return -EINVAL;
3291                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3292                         return -EBUSY;
3293                 if (copy_from_user(&val, optval, sizeof(val)))
3294                         return -EFAULT;
3295                 switch (val) {
3296                 case TPACKET_V1:
3297                 case TPACKET_V2:
3298                 case TPACKET_V3:
3299                         po->tp_version = val;
3300                         return 0;
3301                 default:
3302                         return -EINVAL;
3303                 }
3304         }
3305         case PACKET_RESERVE:
3306         {
3307                 unsigned int val;
3308
3309                 if (optlen != sizeof(val))
3310                         return -EINVAL;
3311                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3312                         return -EBUSY;
3313                 if (copy_from_user(&val, optval, sizeof(val)))
3314                         return -EFAULT;
3315                 po->tp_reserve = val;
3316                 return 0;
3317         }
3318         case PACKET_LOSS:
3319         {
3320                 unsigned int val;
3321
3322                 if (optlen != sizeof(val))
3323                         return -EINVAL;
3324                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3325                         return -EBUSY;
3326                 if (copy_from_user(&val, optval, sizeof(val)))
3327                         return -EFAULT;
3328                 po->tp_loss = !!val;
3329                 return 0;
3330         }
3331         case PACKET_AUXDATA:
3332         {
3333                 int val;
3334
3335                 if (optlen < sizeof(val))
3336                         return -EINVAL;
3337                 if (copy_from_user(&val, optval, sizeof(val)))
3338                         return -EFAULT;
3339
3340                 po->auxdata = !!val;
3341                 return 0;
3342         }
3343         case PACKET_ORIGDEV:
3344         {
3345                 int val;
3346
3347                 if (optlen < sizeof(val))
3348                         return -EINVAL;
3349                 if (copy_from_user(&val, optval, sizeof(val)))
3350                         return -EFAULT;
3351
3352                 po->origdev = !!val;
3353                 return 0;
3354         }
3355         case PACKET_VNET_HDR:
3356         {
3357                 int val;
3358
3359                 if (sock->type != SOCK_RAW)
3360                         return -EINVAL;
3361                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3362                         return -EBUSY;
3363                 if (optlen < sizeof(val))
3364                         return -EINVAL;
3365                 if (copy_from_user(&val, optval, sizeof(val)))
3366                         return -EFAULT;
3367
3368                 po->has_vnet_hdr = !!val;
3369                 return 0;
3370         }
3371         case PACKET_TIMESTAMP:
3372         {
3373                 int val;
3374
3375                 if (optlen != sizeof(val))
3376                         return -EINVAL;
3377                 if (copy_from_user(&val, optval, sizeof(val)))
3378                         return -EFAULT;
3379
3380                 po->tp_tstamp = val;
3381                 return 0;
3382         }
3383         case PACKET_FANOUT:
3384         {
3385                 int val;
3386
3387                 if (optlen != sizeof(val))
3388                         return -EINVAL;
3389                 if (copy_from_user(&val, optval, sizeof(val)))
3390                         return -EFAULT;
3391
3392                 return fanout_add(sk, val & 0xffff, val >> 16);
3393         }
3394         case PACKET_TX_HAS_OFF:
3395         {
3396                 unsigned int val;
3397
3398                 if (optlen != sizeof(val))
3399                         return -EINVAL;
3400                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3401                         return -EBUSY;
3402                 if (copy_from_user(&val, optval, sizeof(val)))
3403                         return -EFAULT;
3404                 po->tp_tx_has_off = !!val;
3405                 return 0;
3406         }
3407         case PACKET_QDISC_BYPASS:
3408         {
3409                 int val;
3410
3411                 if (optlen != sizeof(val))
3412                         return -EINVAL;
3413                 if (copy_from_user(&val, optval, sizeof(val)))
3414                         return -EFAULT;
3415
3416                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3417                 return 0;
3418         }
3419         default:
3420                 return -ENOPROTOOPT;
3421         }
3422 }
3423
3424 static int packet_getsockopt(struct socket *sock, int level, int optname,
3425                              char __user *optval, int __user *optlen)
3426 {
3427         int len;
3428         int val, lv = sizeof(val);
3429         struct sock *sk = sock->sk;
3430         struct packet_sock *po = pkt_sk(sk);
3431         void *data = &val;
3432         union tpacket_stats_u st;
3433
3434         if (level != SOL_PACKET)
3435                 return -ENOPROTOOPT;
3436
3437         if (get_user(len, optlen))
3438                 return -EFAULT;
3439
3440         if (len < 0)
3441                 return -EINVAL;
3442
3443         switch (optname) {
3444         case PACKET_STATISTICS:
3445                 spin_lock_bh(&sk->sk_receive_queue.lock);
3446                 memcpy(&st, &po->stats, sizeof(st));
3447                 memset(&po->stats, 0, sizeof(po->stats));
3448                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3449
3450                 if (po->tp_version == TPACKET_V3) {
3451                         lv = sizeof(struct tpacket_stats_v3);
3452                         st.stats3.tp_packets += st.stats3.tp_drops;
3453                         data = &st.stats3;
3454                 } else {
3455                         lv = sizeof(struct tpacket_stats);
3456                         st.stats1.tp_packets += st.stats1.tp_drops;
3457                         data = &st.stats1;
3458                 }
3459
3460                 break;
3461         case PACKET_AUXDATA:
3462                 val = po->auxdata;
3463                 break;
3464         case PACKET_ORIGDEV:
3465                 val = po->origdev;
3466                 break;
3467         case PACKET_VNET_HDR:
3468                 val = po->has_vnet_hdr;
3469                 break;
3470         case PACKET_VERSION:
3471                 val = po->tp_version;
3472                 break;
3473         case PACKET_HDRLEN:
3474                 if (len > sizeof(int))
3475                         len = sizeof(int);
3476                 if (copy_from_user(&val, optval, len))
3477                         return -EFAULT;
3478                 switch (val) {
3479                 case TPACKET_V1:
3480                         val = sizeof(struct tpacket_hdr);
3481                         break;
3482                 case TPACKET_V2:
3483                         val = sizeof(struct tpacket2_hdr);
3484                         break;
3485                 case TPACKET_V3:
3486                         val = sizeof(struct tpacket3_hdr);
3487                         break;
3488                 default:
3489                         return -EINVAL;
3490                 }
3491                 break;
3492         case PACKET_RESERVE:
3493                 val = po->tp_reserve;
3494                 break;
3495         case PACKET_LOSS:
3496                 val = po->tp_loss;
3497                 break;
3498         case PACKET_TIMESTAMP:
3499                 val = po->tp_tstamp;
3500                 break;
3501         case PACKET_FANOUT:
3502                 val = (po->fanout ?
3503                        ((u32)po->fanout->id |
3504                         ((u32)po->fanout->type << 16) |
3505                         ((u32)po->fanout->flags << 24)) :
3506                        0);
3507                 break;
3508         case PACKET_TX_HAS_OFF:
3509                 val = po->tp_tx_has_off;
3510                 break;
3511         case PACKET_QDISC_BYPASS:
3512                 val = packet_use_direct_xmit(po);
3513                 break;
3514         default:
3515                 return -ENOPROTOOPT;
3516         }
3517
3518         if (len > lv)
3519                 len = lv;
3520         if (put_user(len, optlen))
3521                 return -EFAULT;
3522         if (copy_to_user(optval, data, len))
3523                 return -EFAULT;
3524         return 0;
3525 }
3526
3527
3528 static int packet_notifier(struct notifier_block *this,
3529                            unsigned long msg, void *ptr)
3530 {
3531         struct sock *sk;
3532         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3533         struct net *net = dev_net(dev);
3534
3535         rcu_read_lock();
3536         sk_for_each_rcu(sk, &net->packet.sklist) {
3537                 struct packet_sock *po = pkt_sk(sk);
3538
3539                 switch (msg) {
3540                 case NETDEV_UNREGISTER:
3541                         if (po->mclist)
3542                                 packet_dev_mclist(dev, po->mclist, -1);
3543                         /* fallthrough */
3544
3545                 case NETDEV_DOWN:
3546                         if (dev->ifindex == po->ifindex) {
3547                                 spin_lock(&po->bind_lock);
3548                                 if (po->running) {
3549                                         __unregister_prot_hook(sk, false);
3550                                         sk->sk_err = ENETDOWN;
3551                                         if (!sock_flag(sk, SOCK_DEAD))
3552                                                 sk->sk_error_report(sk);
3553                                 }
3554                                 if (msg == NETDEV_UNREGISTER) {
3555                                         packet_cached_dev_reset(po);
3556                                         po->ifindex = -1;
3557                                         if (po->prot_hook.dev)
3558                                                 dev_put(po->prot_hook.dev);
3559                                         po->prot_hook.dev = NULL;
3560                                 }
3561                                 spin_unlock(&po->bind_lock);
3562                         }
3563                         break;
3564                 case NETDEV_UP:
3565                         if (dev->ifindex == po->ifindex) {
3566                                 spin_lock(&po->bind_lock);
3567                                 if (po->num)
3568                                         register_prot_hook(sk);
3569                                 spin_unlock(&po->bind_lock);
3570                         }
3571                         break;
3572                 }
3573         }
3574         rcu_read_unlock();
3575         return NOTIFY_DONE;
3576 }
3577
3578
3579 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3580                         unsigned long arg)
3581 {
3582         struct sock *sk = sock->sk;
3583
3584         switch (cmd) {
3585         case SIOCOUTQ:
3586         {
3587                 int amount = sk_wmem_alloc_get(sk);
3588
3589                 return put_user(amount, (int __user *)arg);
3590         }
3591         case SIOCINQ:
3592         {
3593                 struct sk_buff *skb;
3594                 int amount = 0;
3595
3596                 spin_lock_bh(&sk->sk_receive_queue.lock);
3597                 skb = skb_peek(&sk->sk_receive_queue);
3598                 if (skb)
3599                         amount = skb->len;
3600                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3601                 return put_user(amount, (int __user *)arg);
3602         }
3603         case SIOCGSTAMP:
3604                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3605         case SIOCGSTAMPNS:
3606                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3607
3608 #ifdef CONFIG_INET
3609         case SIOCADDRT:
3610         case SIOCDELRT:
3611         case SIOCDARP:
3612         case SIOCGARP:
3613         case SIOCSARP:
3614         case SIOCGIFADDR:
3615         case SIOCSIFADDR:
3616         case SIOCGIFBRDADDR:
3617         case SIOCSIFBRDADDR:
3618         case SIOCGIFNETMASK:
3619         case SIOCSIFNETMASK:
3620         case SIOCGIFDSTADDR:
3621         case SIOCSIFDSTADDR:
3622         case SIOCSIFFLAGS:
3623                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3624 #endif
3625
3626         default:
3627                 return -ENOIOCTLCMD;
3628         }
3629         return 0;
3630 }
3631
3632 static unsigned int packet_poll(struct file *file, struct socket *sock,
3633                                 poll_table *wait)
3634 {
3635         struct sock *sk = sock->sk;
3636         struct packet_sock *po = pkt_sk(sk);
3637         unsigned int mask = datagram_poll(file, sock, wait);
3638
3639         spin_lock_bh(&sk->sk_receive_queue.lock);
3640         if (po->rx_ring.pg_vec) {
3641                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3642                         TP_STATUS_KERNEL))
3643                         mask |= POLLIN | POLLRDNORM;
3644         }
3645         spin_unlock_bh(&sk->sk_receive_queue.lock);
3646         spin_lock_bh(&sk->sk_write_queue.lock);
3647         if (po->tx_ring.pg_vec) {
3648                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3649                         mask |= POLLOUT | POLLWRNORM;
3650         }
3651         spin_unlock_bh(&sk->sk_write_queue.lock);
3652         return mask;
3653 }
3654
3655
3656 /* Dirty? Well, I still did not learn better way to account
3657  * for user mmaps.
3658  */
3659
3660 static void packet_mm_open(struct vm_area_struct *vma)
3661 {
3662         struct file *file = vma->vm_file;
3663         struct socket *sock = file->private_data;
3664         struct sock *sk = sock->sk;
3665
3666         if (sk)
3667                 atomic_inc(&pkt_sk(sk)->mapped);
3668 }
3669
3670 static void packet_mm_close(struct vm_area_struct *vma)
3671 {
3672         struct file *file = vma->vm_file;
3673         struct socket *sock = file->private_data;
3674         struct sock *sk = sock->sk;
3675
3676         if (sk)
3677                 atomic_dec(&pkt_sk(sk)->mapped);
3678 }
3679
3680 static const struct vm_operations_struct packet_mmap_ops = {
3681         .open   =       packet_mm_open,
3682         .close  =       packet_mm_close,
3683 };
3684
3685 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3686                         unsigned int len)
3687 {
3688         int i;
3689
3690         for (i = 0; i < len; i++) {
3691                 if (likely(pg_vec[i].buffer)) {
3692                         if (is_vmalloc_addr(pg_vec[i].buffer))
3693                                 vfree(pg_vec[i].buffer);
3694                         else
3695                                 free_pages((unsigned long)pg_vec[i].buffer,
3696                                            order);
3697                         pg_vec[i].buffer = NULL;
3698                 }
3699         }
3700         kfree(pg_vec);
3701 }
3702
3703 static char *alloc_one_pg_vec_page(unsigned long order)
3704 {
3705         char *buffer;
3706         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3707                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3708
3709         buffer = (char *) __get_free_pages(gfp_flags, order);
3710         if (buffer)
3711                 return buffer;
3712
3713         /* __get_free_pages failed, fall back to vmalloc */
3714         buffer = vzalloc((1 << order) * PAGE_SIZE);
3715         if (buffer)
3716                 return buffer;
3717
3718         /* vmalloc failed, lets dig into swap here */
3719         gfp_flags &= ~__GFP_NORETRY;
3720         buffer = (char *) __get_free_pages(gfp_flags, order);
3721         if (buffer)
3722                 return buffer;
3723
3724         /* complete and utter failure */
3725         return NULL;
3726 }
3727
3728 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3729 {
3730         unsigned int block_nr = req->tp_block_nr;
3731         struct pgv *pg_vec;
3732         int i;
3733
3734         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3735         if (unlikely(!pg_vec))
3736                 goto out;
3737
3738         for (i = 0; i < block_nr; i++) {
3739                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3740                 if (unlikely(!pg_vec[i].buffer))
3741                         goto out_free_pgvec;
3742         }
3743
3744 out:
3745         return pg_vec;
3746
3747 out_free_pgvec:
3748         free_pg_vec(pg_vec, order, block_nr);
3749         pg_vec = NULL;
3750         goto out;
3751 }
3752
3753 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3754                 int closing, int tx_ring)
3755 {
3756         struct pgv *pg_vec = NULL;
3757         struct packet_sock *po = pkt_sk(sk);
3758         int was_running, order = 0;
3759         struct packet_ring_buffer *rb;
3760         struct sk_buff_head *rb_queue;
3761         __be16 num;
3762         int err = -EINVAL;
3763         /* Added to avoid minimal code churn */
3764         struct tpacket_req *req = &req_u->req;
3765
3766         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3767         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3768                 WARN(1, "Tx-ring is not supported.\n");
3769                 goto out;
3770         }
3771
3772         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3773         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3774
3775         err = -EBUSY;
3776         if (!closing) {
3777                 if (atomic_read(&po->mapped))
3778                         goto out;
3779                 if (packet_read_pending(rb))
3780                         goto out;
3781         }
3782
3783         if (req->tp_block_nr) {
3784                 /* Sanity tests and some calculations */
3785                 err = -EBUSY;
3786                 if (unlikely(rb->pg_vec))
3787                         goto out;
3788
3789                 switch (po->tp_version) {
3790                 case TPACKET_V1:
3791                         po->tp_hdrlen = TPACKET_HDRLEN;
3792                         break;
3793                 case TPACKET_V2:
3794                         po->tp_hdrlen = TPACKET2_HDRLEN;
3795                         break;
3796                 case TPACKET_V3:
3797                         po->tp_hdrlen = TPACKET3_HDRLEN;
3798                         break;
3799                 }
3800
3801                 err = -EINVAL;
3802                 if (unlikely((int)req->tp_block_size <= 0))
3803                         goto out;
3804                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3805                         goto out;
3806                 if (po->tp_version >= TPACKET_V3 &&
3807                     (int)(req->tp_block_size -
3808                           BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3809                         goto out;
3810                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3811                                         po->tp_reserve))
3812                         goto out;
3813                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3814                         goto out;
3815
3816                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3817                 if (unlikely(rb->frames_per_block <= 0))
3818                         goto out;
3819                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3820                                         req->tp_frame_nr))
3821                         goto out;
3822
3823                 err = -ENOMEM;
3824                 order = get_order(req->tp_block_size);
3825                 pg_vec = alloc_pg_vec(req, order);
3826                 if (unlikely(!pg_vec))
3827                         goto out;
3828                 switch (po->tp_version) {
3829                 case TPACKET_V3:
3830                 /* Transmit path is not supported. We checked
3831                  * it above but just being paranoid
3832                  */
3833                         if (!tx_ring)
3834                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3835                         break;
3836                 default:
3837                         break;
3838                 }
3839         }
3840         /* Done */
3841         else {
3842                 err = -EINVAL;
3843                 if (unlikely(req->tp_frame_nr))
3844                         goto out;
3845         }
3846
3847         lock_sock(sk);
3848
3849         /* Detach socket from network */
3850         spin_lock(&po->bind_lock);
3851         was_running = po->running;
3852         num = po->num;
3853         if (was_running) {
3854                 po->num = 0;
3855                 __unregister_prot_hook(sk, false);
3856         }
3857         spin_unlock(&po->bind_lock);
3858
3859         synchronize_net();
3860
3861         err = -EBUSY;
3862         mutex_lock(&po->pg_vec_lock);
3863         if (closing || atomic_read(&po->mapped) == 0) {
3864                 err = 0;
3865                 spin_lock_bh(&rb_queue->lock);
3866                 swap(rb->pg_vec, pg_vec);
3867                 rb->frame_max = (req->tp_frame_nr - 1);
3868                 rb->head = 0;
3869                 rb->frame_size = req->tp_frame_size;
3870                 spin_unlock_bh(&rb_queue->lock);
3871
3872                 swap(rb->pg_vec_order, order);
3873                 swap(rb->pg_vec_len, req->tp_block_nr);
3874
3875                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3876                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3877                                                 tpacket_rcv : packet_rcv;
3878                 skb_queue_purge(rb_queue);
3879                 if (atomic_read(&po->mapped))
3880                         pr_err("packet_mmap: vma is busy: %d\n",
3881                                atomic_read(&po->mapped));
3882         }
3883         mutex_unlock(&po->pg_vec_lock);
3884
3885         spin_lock(&po->bind_lock);
3886         if (was_running) {
3887                 po->num = num;
3888                 register_prot_hook(sk);
3889         }
3890         spin_unlock(&po->bind_lock);
3891         if (closing && (po->tp_version > TPACKET_V2)) {
3892                 /* Because we don't support block-based V3 on tx-ring */
3893                 if (!tx_ring)
3894                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3895         }
3896         release_sock(sk);
3897
3898         if (pg_vec)
3899                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3900 out:
3901         return err;
3902 }
3903
3904 static int packet_mmap(struct file *file, struct socket *sock,
3905                 struct vm_area_struct *vma)
3906 {
3907         struct sock *sk = sock->sk;
3908         struct packet_sock *po = pkt_sk(sk);
3909         unsigned long size, expected_size;
3910         struct packet_ring_buffer *rb;
3911         unsigned long start;
3912         int err = -EINVAL;
3913         int i;
3914
3915         if (vma->vm_pgoff)
3916                 return -EINVAL;
3917
3918         mutex_lock(&po->pg_vec_lock);
3919
3920         expected_size = 0;
3921         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3922                 if (rb->pg_vec) {
3923                         expected_size += rb->pg_vec_len
3924                                                 * rb->pg_vec_pages
3925                                                 * PAGE_SIZE;
3926                 }
3927         }
3928
3929         if (expected_size == 0)
3930                 goto out;
3931
3932         size = vma->vm_end - vma->vm_start;
3933         if (size != expected_size)
3934                 goto out;
3935
3936         start = vma->vm_start;
3937         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3938                 if (rb->pg_vec == NULL)
3939                         continue;
3940
3941                 for (i = 0; i < rb->pg_vec_len; i++) {
3942                         struct page *page;
3943                         void *kaddr = rb->pg_vec[i].buffer;
3944                         int pg_num;
3945
3946                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3947                                 page = pgv_to_page(kaddr);
3948                                 err = vm_insert_page(vma, start, page);
3949                                 if (unlikely(err))
3950                                         goto out;
3951                                 start += PAGE_SIZE;
3952                                 kaddr += PAGE_SIZE;
3953                         }
3954                 }
3955         }
3956
3957         atomic_inc(&po->mapped);
3958         vma->vm_ops = &packet_mmap_ops;
3959         err = 0;
3960
3961 out:
3962         mutex_unlock(&po->pg_vec_lock);
3963         return err;
3964 }
3965
3966 static const struct proto_ops packet_ops_spkt = {
3967         .family =       PF_PACKET,
3968         .owner =        THIS_MODULE,
3969         .release =      packet_release,
3970         .bind =         packet_bind_spkt,
3971         .connect =      sock_no_connect,
3972         .socketpair =   sock_no_socketpair,
3973         .accept =       sock_no_accept,
3974         .getname =      packet_getname_spkt,
3975         .poll =         datagram_poll,
3976         .ioctl =        packet_ioctl,
3977         .listen =       sock_no_listen,
3978         .shutdown =     sock_no_shutdown,
3979         .setsockopt =   sock_no_setsockopt,
3980         .getsockopt =   sock_no_getsockopt,
3981         .sendmsg =      packet_sendmsg_spkt,
3982         .recvmsg =      packet_recvmsg,
3983         .mmap =         sock_no_mmap,
3984         .sendpage =     sock_no_sendpage,
3985 };
3986
3987 static const struct proto_ops packet_ops = {
3988         .family =       PF_PACKET,
3989         .owner =        THIS_MODULE,
3990         .release =      packet_release,
3991         .bind =         packet_bind,
3992         .connect =      sock_no_connect,
3993         .socketpair =   sock_no_socketpair,
3994         .accept =       sock_no_accept,
3995         .getname =      packet_getname,
3996         .poll =         packet_poll,
3997         .ioctl =        packet_ioctl,
3998         .listen =       sock_no_listen,
3999         .shutdown =     sock_no_shutdown,
4000         .setsockopt =   packet_setsockopt,
4001         .getsockopt =   packet_getsockopt,
4002         .sendmsg =      packet_sendmsg,
4003         .recvmsg =      packet_recvmsg,
4004         .mmap =         packet_mmap,
4005         .sendpage =     sock_no_sendpage,
4006 };
4007
4008 static const struct net_proto_family packet_family_ops = {
4009         .family =       PF_PACKET,
4010         .create =       packet_create,
4011         .owner  =       THIS_MODULE,
4012 };
4013
4014 static struct notifier_block packet_netdev_notifier = {
4015         .notifier_call =        packet_notifier,
4016 };
4017
4018 #ifdef CONFIG_PROC_FS
4019
4020 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4021         __acquires(RCU)
4022 {
4023         struct net *net = seq_file_net(seq);
4024
4025         rcu_read_lock();
4026         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4027 }
4028
4029 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4030 {
4031         struct net *net = seq_file_net(seq);
4032         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4033 }
4034
4035 static void packet_seq_stop(struct seq_file *seq, void *v)
4036         __releases(RCU)
4037 {
4038         rcu_read_unlock();
4039 }
4040
4041 static int packet_seq_show(struct seq_file *seq, void *v)
4042 {
4043         if (v == SEQ_START_TOKEN)
4044                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4045         else {
4046                 struct sock *s = sk_entry(v);
4047                 const struct packet_sock *po = pkt_sk(s);
4048
4049                 seq_printf(seq,
4050                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4051                            s,
4052                            atomic_read(&s->sk_refcnt),
4053                            s->sk_type,
4054                            ntohs(po->num),
4055                            po->ifindex,
4056                            po->running,
4057                            atomic_read(&s->sk_rmem_alloc),
4058                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4059                            sock_i_ino(s));
4060         }
4061
4062         return 0;
4063 }
4064
4065 static const struct seq_operations packet_seq_ops = {
4066         .start  = packet_seq_start,
4067         .next   = packet_seq_next,
4068         .stop   = packet_seq_stop,
4069         .show   = packet_seq_show,
4070 };
4071
4072 static int packet_seq_open(struct inode *inode, struct file *file)
4073 {
4074         return seq_open_net(inode, file, &packet_seq_ops,
4075                             sizeof(struct seq_net_private));
4076 }
4077
4078 static const struct file_operations packet_seq_fops = {
4079         .owner          = THIS_MODULE,
4080         .open           = packet_seq_open,
4081         .read           = seq_read,
4082         .llseek         = seq_lseek,
4083         .release        = seq_release_net,
4084 };
4085
4086 #endif
4087
4088 static int __net_init packet_net_init(struct net *net)
4089 {
4090         mutex_init(&net->packet.sklist_lock);
4091         INIT_HLIST_HEAD(&net->packet.sklist);
4092
4093         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4094                 return -ENOMEM;
4095
4096         return 0;
4097 }
4098
4099 static void __net_exit packet_net_exit(struct net *net)
4100 {
4101         remove_proc_entry("packet", net->proc_net);
4102 }
4103
4104 static struct pernet_operations packet_net_ops = {
4105         .init = packet_net_init,
4106         .exit = packet_net_exit,
4107 };
4108
4109
4110 static void __exit packet_exit(void)
4111 {
4112         unregister_netdevice_notifier(&packet_netdev_notifier);
4113         unregister_pernet_subsys(&packet_net_ops);
4114         sock_unregister(PF_PACKET);
4115         proto_unregister(&packet_proto);
4116 }
4117
4118 static int __init packet_init(void)
4119 {
4120         int rc = proto_register(&packet_proto, 0);
4121
4122         if (rc != 0)
4123                 goto out;
4124
4125         sock_register(&packet_family_ops);
4126         register_pernet_subsys(&packet_net_ops);
4127         register_netdevice_notifier(&packet_netdev_notifier);
4128 out:
4129         return rc;
4130 }
4131
4132 module_init(packet_init);
4133 module_exit(packet_exit);
4134 MODULE_LICENSE("GPL");
4135 MODULE_ALIAS_NETPROTO(PF_PACKET);