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